• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*******************************************************************************
2 
3   Intel 10 Gigabit PCI Express Linux driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5 
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.
9 
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.
14 
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.
18 
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21 
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 
26 *******************************************************************************/
27 
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/interrupt.h>
36 #include <linux/ip.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
45 #include <linux/if.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
49 
50 #include "ixgbe.h"
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
54 
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
58 #ifdef IXGBE_FCOE
59 char ixgbe_default_device_descr[] =
60 			      "Intel(R) 10 Gigabit Network Connection";
61 #else
62 static char ixgbe_default_device_descr[] =
63 			      "Intel(R) 10 Gigabit Network Connection";
64 #endif
65 #define MAJ 3
66 #define MIN 8
67 #define BUILD 21
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 	__stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 				"Copyright (c) 1999-2012 Intel Corporation.";
73 
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 	[board_82598] = &ixgbe_82598_info,
76 	[board_82599] = &ixgbe_82599_info,
77 	[board_X540] = &ixgbe_X540_info,
78 };
79 
80 /* ixgbe_pci_tbl - PCI Device ID Table
81  *
82  * Wildcard entries (PCI_ANY_ID) should come last
83  * Last entry must be all 0s
84  *
85  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86  *   Class, Class Mask, private data (not used) }
87  */
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
118 	/* required last entry */
119 	{0, }
120 };
121 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
122 
123 #ifdef CONFIG_IXGBE_DCA
124 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
125 			    void *p);
126 static struct notifier_block dca_notifier = {
127 	.notifier_call = ixgbe_notify_dca,
128 	.next          = NULL,
129 	.priority      = 0
130 };
131 #endif
132 
133 #ifdef CONFIG_PCI_IOV
134 static unsigned int max_vfs;
135 module_param(max_vfs, uint, 0);
136 MODULE_PARM_DESC(max_vfs,
137 		 "Maximum number of virtual functions to allocate per physical function");
138 #endif /* CONFIG_PCI_IOV */
139 
140 static unsigned int allow_unsupported_sfp;
141 module_param(allow_unsupported_sfp, uint, 0);
142 MODULE_PARM_DESC(allow_unsupported_sfp,
143 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
144 
145 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
146 static int debug = -1;
147 module_param(debug, int, 0);
148 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
149 
150 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
151 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
152 MODULE_LICENSE("GPL");
153 MODULE_VERSION(DRV_VERSION);
154 
ixgbe_service_event_schedule(struct ixgbe_adapter * adapter)155 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
156 {
157 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
158 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
159 		schedule_work(&adapter->service_task);
160 }
161 
ixgbe_service_event_complete(struct ixgbe_adapter * adapter)162 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
163 {
164 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
165 
166 	/* flush memory to make sure state is correct before next watchdog */
167 	smp_mb__before_clear_bit();
168 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
169 }
170 
171 struct ixgbe_reg_info {
172 	u32 ofs;
173 	char *name;
174 };
175 
176 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
177 
178 	/* General Registers */
179 	{IXGBE_CTRL, "CTRL"},
180 	{IXGBE_STATUS, "STATUS"},
181 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
182 
183 	/* Interrupt Registers */
184 	{IXGBE_EICR, "EICR"},
185 
186 	/* RX Registers */
187 	{IXGBE_SRRCTL(0), "SRRCTL"},
188 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
189 	{IXGBE_RDLEN(0), "RDLEN"},
190 	{IXGBE_RDH(0), "RDH"},
191 	{IXGBE_RDT(0), "RDT"},
192 	{IXGBE_RXDCTL(0), "RXDCTL"},
193 	{IXGBE_RDBAL(0), "RDBAL"},
194 	{IXGBE_RDBAH(0), "RDBAH"},
195 
196 	/* TX Registers */
197 	{IXGBE_TDBAL(0), "TDBAL"},
198 	{IXGBE_TDBAH(0), "TDBAH"},
199 	{IXGBE_TDLEN(0), "TDLEN"},
200 	{IXGBE_TDH(0), "TDH"},
201 	{IXGBE_TDT(0), "TDT"},
202 	{IXGBE_TXDCTL(0), "TXDCTL"},
203 
204 	/* List Terminator */
205 	{}
206 };
207 
208 
209 /*
210  * ixgbe_regdump - register printout routine
211  */
ixgbe_regdump(struct ixgbe_hw * hw,struct ixgbe_reg_info * reginfo)212 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
213 {
214 	int i = 0, j = 0;
215 	char rname[16];
216 	u32 regs[64];
217 
218 	switch (reginfo->ofs) {
219 	case IXGBE_SRRCTL(0):
220 		for (i = 0; i < 64; i++)
221 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
222 		break;
223 	case IXGBE_DCA_RXCTRL(0):
224 		for (i = 0; i < 64; i++)
225 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
226 		break;
227 	case IXGBE_RDLEN(0):
228 		for (i = 0; i < 64; i++)
229 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
230 		break;
231 	case IXGBE_RDH(0):
232 		for (i = 0; i < 64; i++)
233 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
234 		break;
235 	case IXGBE_RDT(0):
236 		for (i = 0; i < 64; i++)
237 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
238 		break;
239 	case IXGBE_RXDCTL(0):
240 		for (i = 0; i < 64; i++)
241 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
242 		break;
243 	case IXGBE_RDBAL(0):
244 		for (i = 0; i < 64; i++)
245 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
246 		break;
247 	case IXGBE_RDBAH(0):
248 		for (i = 0; i < 64; i++)
249 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
250 		break;
251 	case IXGBE_TDBAL(0):
252 		for (i = 0; i < 64; i++)
253 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
254 		break;
255 	case IXGBE_TDBAH(0):
256 		for (i = 0; i < 64; i++)
257 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
258 		break;
259 	case IXGBE_TDLEN(0):
260 		for (i = 0; i < 64; i++)
261 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
262 		break;
263 	case IXGBE_TDH(0):
264 		for (i = 0; i < 64; i++)
265 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
266 		break;
267 	case IXGBE_TDT(0):
268 		for (i = 0; i < 64; i++)
269 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
270 		break;
271 	case IXGBE_TXDCTL(0):
272 		for (i = 0; i < 64; i++)
273 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
274 		break;
275 	default:
276 		pr_info("%-15s %08x\n", reginfo->name,
277 			IXGBE_READ_REG(hw, reginfo->ofs));
278 		return;
279 	}
280 
281 	for (i = 0; i < 8; i++) {
282 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
283 		pr_err("%-15s", rname);
284 		for (j = 0; j < 8; j++)
285 			pr_cont(" %08x", regs[i*8+j]);
286 		pr_cont("\n");
287 	}
288 
289 }
290 
291 /*
292  * ixgbe_dump - Print registers, tx-rings and rx-rings
293  */
ixgbe_dump(struct ixgbe_adapter * adapter)294 static void ixgbe_dump(struct ixgbe_adapter *adapter)
295 {
296 	struct net_device *netdev = adapter->netdev;
297 	struct ixgbe_hw *hw = &adapter->hw;
298 	struct ixgbe_reg_info *reginfo;
299 	int n = 0;
300 	struct ixgbe_ring *tx_ring;
301 	struct ixgbe_tx_buffer *tx_buffer;
302 	union ixgbe_adv_tx_desc *tx_desc;
303 	struct my_u0 { u64 a; u64 b; } *u0;
304 	struct ixgbe_ring *rx_ring;
305 	union ixgbe_adv_rx_desc *rx_desc;
306 	struct ixgbe_rx_buffer *rx_buffer_info;
307 	u32 staterr;
308 	int i = 0;
309 
310 	if (!netif_msg_hw(adapter))
311 		return;
312 
313 	/* Print netdevice Info */
314 	if (netdev) {
315 		dev_info(&adapter->pdev->dev, "Net device Info\n");
316 		pr_info("Device Name     state            "
317 			"trans_start      last_rx\n");
318 		pr_info("%-15s %016lX %016lX %016lX\n",
319 			netdev->name,
320 			netdev->state,
321 			netdev->trans_start,
322 			netdev->last_rx);
323 	}
324 
325 	/* Print Registers */
326 	dev_info(&adapter->pdev->dev, "Register Dump\n");
327 	pr_info(" Register Name   Value\n");
328 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
329 	     reginfo->name; reginfo++) {
330 		ixgbe_regdump(hw, reginfo);
331 	}
332 
333 	/* Print TX Ring Summary */
334 	if (!netdev || !netif_running(netdev))
335 		goto exit;
336 
337 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
338 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
339 	for (n = 0; n < adapter->num_tx_queues; n++) {
340 		tx_ring = adapter->tx_ring[n];
341 		tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
342 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
343 			   n, tx_ring->next_to_use, tx_ring->next_to_clean,
344 			   (u64)dma_unmap_addr(tx_buffer, dma),
345 			   dma_unmap_len(tx_buffer, len),
346 			   tx_buffer->next_to_watch,
347 			   (u64)tx_buffer->time_stamp);
348 	}
349 
350 	/* Print TX Rings */
351 	if (!netif_msg_tx_done(adapter))
352 		goto rx_ring_summary;
353 
354 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
355 
356 	/* Transmit Descriptor Formats
357 	 *
358 	 * Advanced Transmit Descriptor
359 	 *   +--------------------------------------------------------------+
360 	 * 0 |         Buffer Address [63:0]                                |
361 	 *   +--------------------------------------------------------------+
362 	 * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
363 	 *   +--------------------------------------------------------------+
364 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
365 	 */
366 
367 	for (n = 0; n < adapter->num_tx_queues; n++) {
368 		tx_ring = adapter->tx_ring[n];
369 		pr_info("------------------------------------\n");
370 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
371 		pr_info("------------------------------------\n");
372 		pr_info("T [desc]     [address 63:0  ] "
373 			"[PlPOIdStDDt Ln] [bi->dma       ] "
374 			"leng  ntw timestamp        bi->skb\n");
375 
376 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
377 			tx_desc = IXGBE_TX_DESC(tx_ring, i);
378 			tx_buffer = &tx_ring->tx_buffer_info[i];
379 			u0 = (struct my_u0 *)tx_desc;
380 			pr_info("T [0x%03X]    %016llX %016llX %016llX"
381 				" %04X  %p %016llX %p", i,
382 				le64_to_cpu(u0->a),
383 				le64_to_cpu(u0->b),
384 				(u64)dma_unmap_addr(tx_buffer, dma),
385 				dma_unmap_len(tx_buffer, len),
386 				tx_buffer->next_to_watch,
387 				(u64)tx_buffer->time_stamp,
388 				tx_buffer->skb);
389 			if (i == tx_ring->next_to_use &&
390 				i == tx_ring->next_to_clean)
391 				pr_cont(" NTC/U\n");
392 			else if (i == tx_ring->next_to_use)
393 				pr_cont(" NTU\n");
394 			else if (i == tx_ring->next_to_clean)
395 				pr_cont(" NTC\n");
396 			else
397 				pr_cont("\n");
398 
399 			if (netif_msg_pktdata(adapter) &&
400 			    dma_unmap_len(tx_buffer, len) != 0)
401 				print_hex_dump(KERN_INFO, "",
402 					DUMP_PREFIX_ADDRESS, 16, 1,
403 					phys_to_virt(dma_unmap_addr(tx_buffer,
404 								    dma)),
405 					dma_unmap_len(tx_buffer, len),
406 					true);
407 		}
408 	}
409 
410 	/* Print RX Rings Summary */
411 rx_ring_summary:
412 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
413 	pr_info("Queue [NTU] [NTC]\n");
414 	for (n = 0; n < adapter->num_rx_queues; n++) {
415 		rx_ring = adapter->rx_ring[n];
416 		pr_info("%5d %5X %5X\n",
417 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
418 	}
419 
420 	/* Print RX Rings */
421 	if (!netif_msg_rx_status(adapter))
422 		goto exit;
423 
424 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
425 
426 	/* Advanced Receive Descriptor (Read) Format
427 	 *    63                                           1        0
428 	 *    +-----------------------------------------------------+
429 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
430 	 *    +----------------------------------------------+------+
431 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
432 	 *    +-----------------------------------------------------+
433 	 *
434 	 *
435 	 * Advanced Receive Descriptor (Write-Back) Format
436 	 *
437 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
438 	 *   +------------------------------------------------------+
439 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
440 	 *   | Checksum   Ident  |   |           |    | Type | Type |
441 	 *   +------------------------------------------------------+
442 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
443 	 *   +------------------------------------------------------+
444 	 *   63       48 47    32 31            20 19               0
445 	 */
446 	for (n = 0; n < adapter->num_rx_queues; n++) {
447 		rx_ring = adapter->rx_ring[n];
448 		pr_info("------------------------------------\n");
449 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
450 		pr_info("------------------------------------\n");
451 		pr_info("R  [desc]      [ PktBuf     A0] "
452 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
453 			"<-- Adv Rx Read format\n");
454 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
455 			"[vl er S cks ln] ---------------- [bi->skb] "
456 			"<-- Adv Rx Write-Back format\n");
457 
458 		for (i = 0; i < rx_ring->count; i++) {
459 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
460 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
461 			u0 = (struct my_u0 *)rx_desc;
462 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
463 			if (staterr & IXGBE_RXD_STAT_DD) {
464 				/* Descriptor Done */
465 				pr_info("RWB[0x%03X]     %016llX "
466 					"%016llX ---------------- %p", i,
467 					le64_to_cpu(u0->a),
468 					le64_to_cpu(u0->b),
469 					rx_buffer_info->skb);
470 			} else {
471 				pr_info("R  [0x%03X]     %016llX "
472 					"%016llX %016llX %p", i,
473 					le64_to_cpu(u0->a),
474 					le64_to_cpu(u0->b),
475 					(u64)rx_buffer_info->dma,
476 					rx_buffer_info->skb);
477 
478 				if (netif_msg_pktdata(adapter)) {
479 					print_hex_dump(KERN_INFO, "",
480 					   DUMP_PREFIX_ADDRESS, 16, 1,
481 					   phys_to_virt(rx_buffer_info->dma),
482 					   ixgbe_rx_bufsz(rx_ring), true);
483 				}
484 			}
485 
486 			if (i == rx_ring->next_to_use)
487 				pr_cont(" NTU\n");
488 			else if (i == rx_ring->next_to_clean)
489 				pr_cont(" NTC\n");
490 			else
491 				pr_cont("\n");
492 
493 		}
494 	}
495 
496 exit:
497 	return;
498 }
499 
ixgbe_release_hw_control(struct ixgbe_adapter * adapter)500 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
501 {
502 	u32 ctrl_ext;
503 
504 	/* Let firmware take over control of h/w */
505 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
506 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
507 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
508 }
509 
ixgbe_get_hw_control(struct ixgbe_adapter * adapter)510 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
511 {
512 	u32 ctrl_ext;
513 
514 	/* Let firmware know the driver has taken over */
515 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
516 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
517 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
518 }
519 
520 /*
521  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
522  * @adapter: pointer to adapter struct
523  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
524  * @queue: queue to map the corresponding interrupt to
525  * @msix_vector: the vector to map to the corresponding queue
526  *
527  */
ixgbe_set_ivar(struct ixgbe_adapter * adapter,s8 direction,u8 queue,u8 msix_vector)528 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
529 			   u8 queue, u8 msix_vector)
530 {
531 	u32 ivar, index;
532 	struct ixgbe_hw *hw = &adapter->hw;
533 	switch (hw->mac.type) {
534 	case ixgbe_mac_82598EB:
535 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
536 		if (direction == -1)
537 			direction = 0;
538 		index = (((direction * 64) + queue) >> 2) & 0x1F;
539 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
540 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
541 		ivar |= (msix_vector << (8 * (queue & 0x3)));
542 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
543 		break;
544 	case ixgbe_mac_82599EB:
545 	case ixgbe_mac_X540:
546 		if (direction == -1) {
547 			/* other causes */
548 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
549 			index = ((queue & 1) * 8);
550 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
551 			ivar &= ~(0xFF << index);
552 			ivar |= (msix_vector << index);
553 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
554 			break;
555 		} else {
556 			/* tx or rx causes */
557 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
558 			index = ((16 * (queue & 1)) + (8 * direction));
559 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
560 			ivar &= ~(0xFF << index);
561 			ivar |= (msix_vector << index);
562 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
563 			break;
564 		}
565 	default:
566 		break;
567 	}
568 }
569 
ixgbe_irq_rearm_queues(struct ixgbe_adapter * adapter,u64 qmask)570 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
571 					  u64 qmask)
572 {
573 	u32 mask;
574 
575 	switch (adapter->hw.mac.type) {
576 	case ixgbe_mac_82598EB:
577 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
578 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
579 		break;
580 	case ixgbe_mac_82599EB:
581 	case ixgbe_mac_X540:
582 		mask = (qmask & 0xFFFFFFFF);
583 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
584 		mask = (qmask >> 32);
585 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
586 		break;
587 	default:
588 		break;
589 	}
590 }
591 
ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring * ring,struct ixgbe_tx_buffer * tx_buffer)592 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
593 				      struct ixgbe_tx_buffer *tx_buffer)
594 {
595 	if (tx_buffer->skb) {
596 		dev_kfree_skb_any(tx_buffer->skb);
597 		if (dma_unmap_len(tx_buffer, len))
598 			dma_unmap_single(ring->dev,
599 					 dma_unmap_addr(tx_buffer, dma),
600 					 dma_unmap_len(tx_buffer, len),
601 					 DMA_TO_DEVICE);
602 	} else if (dma_unmap_len(tx_buffer, len)) {
603 		dma_unmap_page(ring->dev,
604 			       dma_unmap_addr(tx_buffer, dma),
605 			       dma_unmap_len(tx_buffer, len),
606 			       DMA_TO_DEVICE);
607 	}
608 	tx_buffer->next_to_watch = NULL;
609 	tx_buffer->skb = NULL;
610 	dma_unmap_len_set(tx_buffer, len, 0);
611 	/* tx_buffer must be completely set up in the transmit path */
612 }
613 
ixgbe_update_xoff_received(struct ixgbe_adapter * adapter)614 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
615 {
616 	struct ixgbe_hw *hw = &adapter->hw;
617 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
618 	u32 data = 0;
619 	u32 xoff[8] = {0};
620 	int i;
621 
622 	if ((hw->fc.current_mode == ixgbe_fc_full) ||
623 	    (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
624 		switch (hw->mac.type) {
625 		case ixgbe_mac_82598EB:
626 			data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
627 			break;
628 		default:
629 			data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
630 		}
631 		hwstats->lxoffrxc += data;
632 
633 		/* refill credits (no tx hang) if we received xoff */
634 		if (!data)
635 			return;
636 
637 		for (i = 0; i < adapter->num_tx_queues; i++)
638 			clear_bit(__IXGBE_HANG_CHECK_ARMED,
639 				  &adapter->tx_ring[i]->state);
640 		return;
641 	} else if (!(adapter->dcb_cfg.pfc_mode_enable))
642 		return;
643 
644 	/* update stats for each tc, only valid with PFC enabled */
645 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
646 		switch (hw->mac.type) {
647 		case ixgbe_mac_82598EB:
648 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
649 			break;
650 		default:
651 			xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
652 		}
653 		hwstats->pxoffrxc[i] += xoff[i];
654 	}
655 
656 	/* disarm tx queues that have received xoff frames */
657 	for (i = 0; i < adapter->num_tx_queues; i++) {
658 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
659 		u8 tc = tx_ring->dcb_tc;
660 
661 		if (xoff[tc])
662 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
663 	}
664 }
665 
ixgbe_get_tx_completed(struct ixgbe_ring * ring)666 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
667 {
668 	return ring->stats.packets;
669 }
670 
ixgbe_get_tx_pending(struct ixgbe_ring * ring)671 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
672 {
673 	struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
674 	struct ixgbe_hw *hw = &adapter->hw;
675 
676 	u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
677 	u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
678 
679 	if (head != tail)
680 		return (head < tail) ?
681 			tail - head : (tail + ring->count - head);
682 
683 	return 0;
684 }
685 
ixgbe_check_tx_hang(struct ixgbe_ring * tx_ring)686 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
687 {
688 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
689 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
690 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
691 	bool ret = false;
692 
693 	clear_check_for_tx_hang(tx_ring);
694 
695 	/*
696 	 * Check for a hung queue, but be thorough. This verifies
697 	 * that a transmit has been completed since the previous
698 	 * check AND there is at least one packet pending. The
699 	 * ARMED bit is set to indicate a potential hang. The
700 	 * bit is cleared if a pause frame is received to remove
701 	 * false hang detection due to PFC or 802.3x frames. By
702 	 * requiring this to fail twice we avoid races with
703 	 * pfc clearing the ARMED bit and conditions where we
704 	 * run the check_tx_hang logic with a transmit completion
705 	 * pending but without time to complete it yet.
706 	 */
707 	if ((tx_done_old == tx_done) && tx_pending) {
708 		/* make sure it is true for two checks in a row */
709 		ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
710 				       &tx_ring->state);
711 	} else {
712 		/* update completed stats and continue */
713 		tx_ring->tx_stats.tx_done_old = tx_done;
714 		/* reset the countdown */
715 		clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
716 	}
717 
718 	return ret;
719 }
720 
721 /**
722  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
723  * @adapter: driver private struct
724  **/
ixgbe_tx_timeout_reset(struct ixgbe_adapter * adapter)725 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
726 {
727 
728 	/* Do the reset outside of interrupt context */
729 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
730 		adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
731 		ixgbe_service_event_schedule(adapter);
732 	}
733 }
734 
735 /**
736  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
737  * @q_vector: structure containing interrupt and ring information
738  * @tx_ring: tx ring to clean
739  **/
ixgbe_clean_tx_irq(struct ixgbe_q_vector * q_vector,struct ixgbe_ring * tx_ring)740 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
741 			       struct ixgbe_ring *tx_ring)
742 {
743 	struct ixgbe_adapter *adapter = q_vector->adapter;
744 	struct ixgbe_tx_buffer *tx_buffer;
745 	union ixgbe_adv_tx_desc *tx_desc;
746 	unsigned int total_bytes = 0, total_packets = 0;
747 	unsigned int budget = q_vector->tx.work_limit;
748 	unsigned int i = tx_ring->next_to_clean;
749 
750 	if (test_bit(__IXGBE_DOWN, &adapter->state))
751 		return true;
752 
753 	tx_buffer = &tx_ring->tx_buffer_info[i];
754 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
755 	i -= tx_ring->count;
756 
757 	do {
758 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
759 
760 		/* if next_to_watch is not set then there is no work pending */
761 		if (!eop_desc)
762 			break;
763 
764 		/* prevent any other reads prior to eop_desc */
765 		rmb();
766 
767 		/* if DD is not set pending work has not been completed */
768 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
769 			break;
770 
771 		/* clear next_to_watch to prevent false hangs */
772 		tx_buffer->next_to_watch = NULL;
773 
774 		/* update the statistics for this packet */
775 		total_bytes += tx_buffer->bytecount;
776 		total_packets += tx_buffer->gso_segs;
777 
778 		/* free the skb */
779 		dev_kfree_skb_any(tx_buffer->skb);
780 
781 		/* unmap skb header data */
782 		dma_unmap_single(tx_ring->dev,
783 				 dma_unmap_addr(tx_buffer, dma),
784 				 dma_unmap_len(tx_buffer, len),
785 				 DMA_TO_DEVICE);
786 
787 		/* clear tx_buffer data */
788 		tx_buffer->skb = NULL;
789 		dma_unmap_len_set(tx_buffer, len, 0);
790 
791 		/* unmap remaining buffers */
792 		while (tx_desc != eop_desc) {
793 			tx_buffer++;
794 			tx_desc++;
795 			i++;
796 			if (unlikely(!i)) {
797 				i -= tx_ring->count;
798 				tx_buffer = tx_ring->tx_buffer_info;
799 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
800 			}
801 
802 			/* unmap any remaining paged data */
803 			if (dma_unmap_len(tx_buffer, len)) {
804 				dma_unmap_page(tx_ring->dev,
805 					       dma_unmap_addr(tx_buffer, dma),
806 					       dma_unmap_len(tx_buffer, len),
807 					       DMA_TO_DEVICE);
808 				dma_unmap_len_set(tx_buffer, len, 0);
809 			}
810 		}
811 
812 		/* move us one more past the eop_desc for start of next pkt */
813 		tx_buffer++;
814 		tx_desc++;
815 		i++;
816 		if (unlikely(!i)) {
817 			i -= tx_ring->count;
818 			tx_buffer = tx_ring->tx_buffer_info;
819 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
820 		}
821 
822 		/* issue prefetch for next Tx descriptor */
823 		prefetch(tx_desc);
824 
825 		/* update budget accounting */
826 		budget--;
827 	} while (likely(budget));
828 
829 	i += tx_ring->count;
830 	tx_ring->next_to_clean = i;
831 	u64_stats_update_begin(&tx_ring->syncp);
832 	tx_ring->stats.bytes += total_bytes;
833 	tx_ring->stats.packets += total_packets;
834 	u64_stats_update_end(&tx_ring->syncp);
835 	q_vector->tx.total_bytes += total_bytes;
836 	q_vector->tx.total_packets += total_packets;
837 
838 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
839 		/* schedule immediate reset if we believe we hung */
840 		struct ixgbe_hw *hw = &adapter->hw;
841 		e_err(drv, "Detected Tx Unit Hang\n"
842 			"  Tx Queue             <%d>\n"
843 			"  TDH, TDT             <%x>, <%x>\n"
844 			"  next_to_use          <%x>\n"
845 			"  next_to_clean        <%x>\n"
846 			"tx_buffer_info[next_to_clean]\n"
847 			"  time_stamp           <%lx>\n"
848 			"  jiffies              <%lx>\n",
849 			tx_ring->queue_index,
850 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
851 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
852 			tx_ring->next_to_use, i,
853 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
854 
855 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
856 
857 		e_info(probe,
858 		       "tx hang %d detected on queue %d, resetting adapter\n",
859 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
860 
861 		/* schedule immediate reset if we believe we hung */
862 		ixgbe_tx_timeout_reset(adapter);
863 
864 		/* the adapter is about to reset, no point in enabling stuff */
865 		return true;
866 	}
867 
868 	netdev_tx_completed_queue(txring_txq(tx_ring),
869 				  total_packets, total_bytes);
870 
871 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
872 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
873 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
874 		/* Make sure that anybody stopping the queue after this
875 		 * sees the new next_to_clean.
876 		 */
877 		smp_mb();
878 		if (__netif_subqueue_stopped(tx_ring->netdev,
879 					     tx_ring->queue_index)
880 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
881 			netif_wake_subqueue(tx_ring->netdev,
882 					    tx_ring->queue_index);
883 			++tx_ring->tx_stats.restart_queue;
884 		}
885 	}
886 
887 	return !!budget;
888 }
889 
890 #ifdef CONFIG_IXGBE_DCA
ixgbe_update_tx_dca(struct ixgbe_adapter * adapter,struct ixgbe_ring * tx_ring,int cpu)891 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
892 				struct ixgbe_ring *tx_ring,
893 				int cpu)
894 {
895 	struct ixgbe_hw *hw = &adapter->hw;
896 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
897 	u16 reg_offset;
898 
899 	switch (hw->mac.type) {
900 	case ixgbe_mac_82598EB:
901 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
902 		break;
903 	case ixgbe_mac_82599EB:
904 	case ixgbe_mac_X540:
905 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
906 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
907 		break;
908 	default:
909 		/* for unknown hardware do not write register */
910 		return;
911 	}
912 
913 	/*
914 	 * We can enable relaxed ordering for reads, but not writes when
915 	 * DCA is enabled.  This is due to a known issue in some chipsets
916 	 * which will cause the DCA tag to be cleared.
917 	 */
918 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
919 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
920 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
921 
922 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
923 }
924 
ixgbe_update_rx_dca(struct ixgbe_adapter * adapter,struct ixgbe_ring * rx_ring,int cpu)925 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
926 				struct ixgbe_ring *rx_ring,
927 				int cpu)
928 {
929 	struct ixgbe_hw *hw = &adapter->hw;
930 	u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
931 	u8 reg_idx = rx_ring->reg_idx;
932 
933 
934 	switch (hw->mac.type) {
935 	case ixgbe_mac_82599EB:
936 	case ixgbe_mac_X540:
937 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
938 		break;
939 	default:
940 		break;
941 	}
942 
943 	/*
944 	 * We can enable relaxed ordering for reads, but not writes when
945 	 * DCA is enabled.  This is due to a known issue in some chipsets
946 	 * which will cause the DCA tag to be cleared.
947 	 */
948 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
949 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
950 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
951 
952 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
953 }
954 
ixgbe_update_dca(struct ixgbe_q_vector * q_vector)955 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
956 {
957 	struct ixgbe_adapter *adapter = q_vector->adapter;
958 	struct ixgbe_ring *ring;
959 	int cpu = get_cpu();
960 
961 	if (q_vector->cpu == cpu)
962 		goto out_no_update;
963 
964 	ixgbe_for_each_ring(ring, q_vector->tx)
965 		ixgbe_update_tx_dca(adapter, ring, cpu);
966 
967 	ixgbe_for_each_ring(ring, q_vector->rx)
968 		ixgbe_update_rx_dca(adapter, ring, cpu);
969 
970 	q_vector->cpu = cpu;
971 out_no_update:
972 	put_cpu();
973 }
974 
ixgbe_setup_dca(struct ixgbe_adapter * adapter)975 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
976 {
977 	int num_q_vectors;
978 	int i;
979 
980 	if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
981 		return;
982 
983 	/* always use CB2 mode, difference is masked in the CB driver */
984 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
985 
986 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
987 		num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
988 	else
989 		num_q_vectors = 1;
990 
991 	for (i = 0; i < num_q_vectors; i++) {
992 		adapter->q_vector[i]->cpu = -1;
993 		ixgbe_update_dca(adapter->q_vector[i]);
994 	}
995 }
996 
__ixgbe_notify_dca(struct device * dev,void * data)997 static int __ixgbe_notify_dca(struct device *dev, void *data)
998 {
999 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1000 	unsigned long event = *(unsigned long *)data;
1001 
1002 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1003 		return 0;
1004 
1005 	switch (event) {
1006 	case DCA_PROVIDER_ADD:
1007 		/* if we're already enabled, don't do it again */
1008 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1009 			break;
1010 		if (dca_add_requester(dev) == 0) {
1011 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1012 			ixgbe_setup_dca(adapter);
1013 			break;
1014 		}
1015 		/* Fall Through since DCA is disabled. */
1016 	case DCA_PROVIDER_REMOVE:
1017 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1018 			dca_remove_requester(dev);
1019 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1020 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1021 		}
1022 		break;
1023 	}
1024 
1025 	return 0;
1026 }
1027 
1028 #endif /* CONFIG_IXGBE_DCA */
ixgbe_rx_hash(struct ixgbe_ring * ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1029 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1030 				 union ixgbe_adv_rx_desc *rx_desc,
1031 				 struct sk_buff *skb)
1032 {
1033 	if (ring->netdev->features & NETIF_F_RXHASH)
1034 		skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1035 }
1036 
1037 #ifdef IXGBE_FCOE
1038 /**
1039  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1040  * @ring: structure containing ring specific data
1041  * @rx_desc: advanced rx descriptor
1042  *
1043  * Returns : true if it is FCoE pkt
1044  */
ixgbe_rx_is_fcoe(struct ixgbe_ring * ring,union ixgbe_adv_rx_desc * rx_desc)1045 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1046 				    union ixgbe_adv_rx_desc *rx_desc)
1047 {
1048 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1049 
1050 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1051 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1052 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1053 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1054 }
1055 
1056 #endif /* IXGBE_FCOE */
1057 /**
1058  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1059  * @ring: structure containing ring specific data
1060  * @rx_desc: current Rx descriptor being processed
1061  * @skb: skb currently being received and modified
1062  **/
ixgbe_rx_checksum(struct ixgbe_ring * ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1063 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1064 				     union ixgbe_adv_rx_desc *rx_desc,
1065 				     struct sk_buff *skb)
1066 {
1067 	skb_checksum_none_assert(skb);
1068 
1069 	/* Rx csum disabled */
1070 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1071 		return;
1072 
1073 	/* if IP and error */
1074 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1075 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1076 		ring->rx_stats.csum_err++;
1077 		return;
1078 	}
1079 
1080 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1081 		return;
1082 
1083 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1084 		__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1085 
1086 		/*
1087 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1088 		 * checksum errors.
1089 		 */
1090 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1091 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1092 			return;
1093 
1094 		ring->rx_stats.csum_err++;
1095 		return;
1096 	}
1097 
1098 	/* It must be a TCP or UDP packet with a valid checksum */
1099 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1100 }
1101 
ixgbe_release_rx_desc(struct ixgbe_ring * rx_ring,u32 val)1102 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1103 {
1104 	rx_ring->next_to_use = val;
1105 
1106 	/* update next to alloc since we have filled the ring */
1107 	rx_ring->next_to_alloc = val;
1108 	/*
1109 	 * Force memory writes to complete before letting h/w
1110 	 * know there are new descriptors to fetch.  (Only
1111 	 * applicable for weak-ordered memory model archs,
1112 	 * such as IA-64).
1113 	 */
1114 	wmb();
1115 	writel(val, rx_ring->tail);
1116 }
1117 
ixgbe_alloc_mapped_page(struct ixgbe_ring * rx_ring,struct ixgbe_rx_buffer * bi)1118 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1119 				    struct ixgbe_rx_buffer *bi)
1120 {
1121 	struct page *page = bi->page;
1122 	dma_addr_t dma = bi->dma;
1123 
1124 	/* since we are recycling buffers we should seldom need to alloc */
1125 	if (likely(dma))
1126 		return true;
1127 
1128 	/* alloc new page for storage */
1129 	if (likely(!page)) {
1130 		page = alloc_pages(GFP_ATOMIC | __GFP_COLD,
1131 				   ixgbe_rx_pg_order(rx_ring));
1132 		if (unlikely(!page)) {
1133 			rx_ring->rx_stats.alloc_rx_page_failed++;
1134 			return false;
1135 		}
1136 		bi->page = page;
1137 	}
1138 
1139 	/* map page for use */
1140 	dma = dma_map_page(rx_ring->dev, page, 0,
1141 			   ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1142 
1143 	/*
1144 	 * if mapping failed free memory back to system since
1145 	 * there isn't much point in holding memory we can't use
1146 	 */
1147 	if (dma_mapping_error(rx_ring->dev, dma)) {
1148 		put_page(page);
1149 		bi->page = NULL;
1150 
1151 		rx_ring->rx_stats.alloc_rx_page_failed++;
1152 		return false;
1153 	}
1154 
1155 	bi->dma = dma;
1156 	bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1157 
1158 	return true;
1159 }
1160 
1161 /**
1162  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1163  * @rx_ring: ring to place buffers on
1164  * @cleaned_count: number of buffers to replace
1165  **/
ixgbe_alloc_rx_buffers(struct ixgbe_ring * rx_ring,u16 cleaned_count)1166 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1167 {
1168 	union ixgbe_adv_rx_desc *rx_desc;
1169 	struct ixgbe_rx_buffer *bi;
1170 	u16 i = rx_ring->next_to_use;
1171 
1172 	/* nothing to do */
1173 	if (!cleaned_count)
1174 		return;
1175 
1176 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1177 	bi = &rx_ring->rx_buffer_info[i];
1178 	i -= rx_ring->count;
1179 
1180 	do {
1181 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1182 			break;
1183 
1184 		/*
1185 		 * Refresh the desc even if buffer_addrs didn't change
1186 		 * because each write-back erases this info.
1187 		 */
1188 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1189 
1190 		rx_desc++;
1191 		bi++;
1192 		i++;
1193 		if (unlikely(!i)) {
1194 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1195 			bi = rx_ring->rx_buffer_info;
1196 			i -= rx_ring->count;
1197 		}
1198 
1199 		/* clear the hdr_addr for the next_to_use descriptor */
1200 		rx_desc->read.hdr_addr = 0;
1201 
1202 		cleaned_count--;
1203 	} while (cleaned_count);
1204 
1205 	i += rx_ring->count;
1206 
1207 	if (rx_ring->next_to_use != i)
1208 		ixgbe_release_rx_desc(rx_ring, i);
1209 }
1210 
1211 /**
1212  * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1213  * @data: pointer to the start of the headers
1214  * @max_len: total length of section to find headers in
1215  *
1216  * This function is meant to determine the length of headers that will
1217  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
1218  * motivation of doing this is to only perform one pull for IPv4 TCP
1219  * packets so that we can do basic things like calculating the gso_size
1220  * based on the average data per packet.
1221  **/
ixgbe_get_headlen(unsigned char * data,unsigned int max_len)1222 static unsigned int ixgbe_get_headlen(unsigned char *data,
1223 				      unsigned int max_len)
1224 {
1225 	union {
1226 		unsigned char *network;
1227 		/* l2 headers */
1228 		struct ethhdr *eth;
1229 		struct vlan_hdr *vlan;
1230 		/* l3 headers */
1231 		struct iphdr *ipv4;
1232 	} hdr;
1233 	__be16 protocol;
1234 	u8 nexthdr = 0;	/* default to not TCP */
1235 	u8 hlen;
1236 
1237 	/* this should never happen, but better safe than sorry */
1238 	if (max_len < ETH_HLEN)
1239 		return max_len;
1240 
1241 	/* initialize network frame pointer */
1242 	hdr.network = data;
1243 
1244 	/* set first protocol and move network header forward */
1245 	protocol = hdr.eth->h_proto;
1246 	hdr.network += ETH_HLEN;
1247 
1248 	/* handle any vlan tag if present */
1249 	if (protocol == __constant_htons(ETH_P_8021Q)) {
1250 		if ((hdr.network - data) > (max_len - VLAN_HLEN))
1251 			return max_len;
1252 
1253 		protocol = hdr.vlan->h_vlan_encapsulated_proto;
1254 		hdr.network += VLAN_HLEN;
1255 	}
1256 
1257 	/* handle L3 protocols */
1258 	if (protocol == __constant_htons(ETH_P_IP)) {
1259 		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1260 			return max_len;
1261 
1262 		/* access ihl as a u8 to avoid unaligned access on ia64 */
1263 		hlen = (hdr.network[0] & 0x0F) << 2;
1264 
1265 		/* verify hlen meets minimum size requirements */
1266 		if (hlen < sizeof(struct iphdr))
1267 			return hdr.network - data;
1268 
1269 		/* record next protocol */
1270 		nexthdr = hdr.ipv4->protocol;
1271 		hdr.network += hlen;
1272 #ifdef IXGBE_FCOE
1273 	} else if (protocol == __constant_htons(ETH_P_FCOE)) {
1274 		if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1275 			return max_len;
1276 		hdr.network += FCOE_HEADER_LEN;
1277 #endif
1278 	} else {
1279 		return hdr.network - data;
1280 	}
1281 
1282 	/* finally sort out TCP */
1283 	if (nexthdr == IPPROTO_TCP) {
1284 		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1285 			return max_len;
1286 
1287 		/* access doff as a u8 to avoid unaligned access on ia64 */
1288 		hlen = (hdr.network[12] & 0xF0) >> 2;
1289 
1290 		/* verify hlen meets minimum size requirements */
1291 		if (hlen < sizeof(struct tcphdr))
1292 			return hdr.network - data;
1293 
1294 		hdr.network += hlen;
1295 	}
1296 
1297 	/*
1298 	 * If everything has gone correctly hdr.network should be the
1299 	 * data section of the packet and will be the end of the header.
1300 	 * If not then it probably represents the end of the last recognized
1301 	 * header.
1302 	 */
1303 	if ((hdr.network - data) < max_len)
1304 		return hdr.network - data;
1305 	else
1306 		return max_len;
1307 }
1308 
ixgbe_get_rsc_cnt(struct ixgbe_ring * rx_ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1309 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1310 			      union ixgbe_adv_rx_desc *rx_desc,
1311 			      struct sk_buff *skb)
1312 {
1313 	__le32 rsc_enabled;
1314 	u32 rsc_cnt;
1315 
1316 	if (!ring_is_rsc_enabled(rx_ring))
1317 		return;
1318 
1319 	rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1320 		      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1321 
1322 	/* If this is an RSC frame rsc_cnt should be non-zero */
1323 	if (!rsc_enabled)
1324 		return;
1325 
1326 	rsc_cnt = le32_to_cpu(rsc_enabled);
1327 	rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1328 
1329 	IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1330 }
1331 
ixgbe_set_rsc_gso_size(struct ixgbe_ring * ring,struct sk_buff * skb)1332 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1333 				   struct sk_buff *skb)
1334 {
1335 	u16 hdr_len = skb_headlen(skb);
1336 
1337 	/* set gso_size to avoid messing up TCP MSS */
1338 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1339 						 IXGBE_CB(skb)->append_cnt);
1340 }
1341 
ixgbe_update_rsc_stats(struct ixgbe_ring * rx_ring,struct sk_buff * skb)1342 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1343 				   struct sk_buff *skb)
1344 {
1345 	/* if append_cnt is 0 then frame is not RSC */
1346 	if (!IXGBE_CB(skb)->append_cnt)
1347 		return;
1348 
1349 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1350 	rx_ring->rx_stats.rsc_flush++;
1351 
1352 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1353 
1354 	/* gso_size is computed using append_cnt so always clear it last */
1355 	IXGBE_CB(skb)->append_cnt = 0;
1356 }
1357 
1358 /**
1359  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1360  * @rx_ring: rx descriptor ring packet is being transacted on
1361  * @rx_desc: pointer to the EOP Rx descriptor
1362  * @skb: pointer to current skb being populated
1363  *
1364  * This function checks the ring, descriptor, and packet information in
1365  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1366  * other fields within the skb.
1367  **/
ixgbe_process_skb_fields(struct ixgbe_ring * rx_ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1368 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1369 				     union ixgbe_adv_rx_desc *rx_desc,
1370 				     struct sk_buff *skb)
1371 {
1372 	ixgbe_update_rsc_stats(rx_ring, skb);
1373 
1374 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1375 
1376 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1377 
1378 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1379 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1380 		__vlan_hwaccel_put_tag(skb, vid);
1381 	}
1382 
1383 	skb_record_rx_queue(skb, rx_ring->queue_index);
1384 
1385 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1386 }
1387 
ixgbe_rx_skb(struct ixgbe_q_vector * q_vector,struct sk_buff * skb)1388 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1389 			 struct sk_buff *skb)
1390 {
1391 	struct ixgbe_adapter *adapter = q_vector->adapter;
1392 
1393 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1394 		napi_gro_receive(&q_vector->napi, skb);
1395 	else
1396 		netif_rx(skb);
1397 }
1398 
1399 /**
1400  * ixgbe_is_non_eop - process handling of non-EOP buffers
1401  * @rx_ring: Rx ring being processed
1402  * @rx_desc: Rx descriptor for current buffer
1403  * @skb: Current socket buffer containing buffer in progress
1404  *
1405  * This function updates next to clean.  If the buffer is an EOP buffer
1406  * this function exits returning false, otherwise it will place the
1407  * sk_buff in the next buffer to be chained and return true indicating
1408  * that this is in fact a non-EOP buffer.
1409  **/
ixgbe_is_non_eop(struct ixgbe_ring * rx_ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1410 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1411 			     union ixgbe_adv_rx_desc *rx_desc,
1412 			     struct sk_buff *skb)
1413 {
1414 	u32 ntc = rx_ring->next_to_clean + 1;
1415 
1416 	/* fetch, update, and store next to clean */
1417 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1418 	rx_ring->next_to_clean = ntc;
1419 
1420 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1421 
1422 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1423 		return false;
1424 
1425 	/* append_cnt indicates packet is RSC, if so fetch nextp */
1426 	if (IXGBE_CB(skb)->append_cnt) {
1427 		ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1428 		ntc &= IXGBE_RXDADV_NEXTP_MASK;
1429 		ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1430 	}
1431 
1432 	/* place skb in next buffer to be received */
1433 	rx_ring->rx_buffer_info[ntc].skb = skb;
1434 	rx_ring->rx_stats.non_eop_descs++;
1435 
1436 	return true;
1437 }
1438 
1439 /**
1440  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1441  * @rx_ring: rx descriptor ring packet is being transacted on
1442  * @rx_desc: pointer to the EOP Rx descriptor
1443  * @skb: pointer to current skb being fixed
1444  *
1445  * Check for corrupted packet headers caused by senders on the local L2
1446  * embedded NIC switch not setting up their Tx Descriptors right.  These
1447  * should be very rare.
1448  *
1449  * Also address the case where we are pulling data in on pages only
1450  * and as such no data is present in the skb header.
1451  *
1452  * In addition if skb is not at least 60 bytes we need to pad it so that
1453  * it is large enough to qualify as a valid Ethernet frame.
1454  *
1455  * Returns true if an error was encountered and skb was freed.
1456  **/
ixgbe_cleanup_headers(struct ixgbe_ring * rx_ring,union ixgbe_adv_rx_desc * rx_desc,struct sk_buff * skb)1457 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1458 				  union ixgbe_adv_rx_desc *rx_desc,
1459 				  struct sk_buff *skb)
1460 {
1461 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1462 	struct net_device *netdev = rx_ring->netdev;
1463 	unsigned char *va;
1464 	unsigned int pull_len;
1465 
1466 	/* if the page was released unmap it, else just sync our portion */
1467 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1468 		dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1469 			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1470 		IXGBE_CB(skb)->page_released = false;
1471 	} else {
1472 		dma_sync_single_range_for_cpu(rx_ring->dev,
1473 					      IXGBE_CB(skb)->dma,
1474 					      frag->page_offset,
1475 					      ixgbe_rx_bufsz(rx_ring),
1476 					      DMA_FROM_DEVICE);
1477 	}
1478 	IXGBE_CB(skb)->dma = 0;
1479 
1480 	/* verify that the packet does not have any known errors */
1481 	if (unlikely(ixgbe_test_staterr(rx_desc,
1482 					IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1483 	    !(netdev->features & NETIF_F_RXALL))) {
1484 		dev_kfree_skb_any(skb);
1485 		return true;
1486 	}
1487 
1488 	/*
1489 	 * it is valid to use page_address instead of kmap since we are
1490 	 * working with pages allocated out of the lomem pool per
1491 	 * alloc_page(GFP_ATOMIC)
1492 	 */
1493 	va = skb_frag_address(frag);
1494 
1495 	/*
1496 	 * we need the header to contain the greater of either ETH_HLEN or
1497 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1498 	 */
1499 	pull_len = skb_frag_size(frag);
1500 	if (pull_len > 256)
1501 		pull_len = ixgbe_get_headlen(va, pull_len);
1502 
1503 	/* align pull length to size of long to optimize memcpy performance */
1504 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1505 
1506 	/* update all of the pointers */
1507 	skb_frag_size_sub(frag, pull_len);
1508 	frag->page_offset += pull_len;
1509 	skb->data_len -= pull_len;
1510 	skb->tail += pull_len;
1511 
1512 	/*
1513 	 * if we sucked the frag empty then we should free it,
1514 	 * if there are other frags here something is screwed up in hardware
1515 	 */
1516 	if (skb_frag_size(frag) == 0) {
1517 		BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1518 		skb_shinfo(skb)->nr_frags = 0;
1519 		__skb_frag_unref(frag);
1520 		skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1521 	}
1522 
1523 #ifdef IXGBE_FCOE
1524 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1525 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1526 		return false;
1527 
1528 #endif
1529 	/* if skb_pad returns an error the skb was freed */
1530 	if (unlikely(skb->len < 60)) {
1531 		int pad_len = 60 - skb->len;
1532 
1533 		if (skb_pad(skb, pad_len))
1534 			return true;
1535 		__skb_put(skb, pad_len);
1536 	}
1537 
1538 	return false;
1539 }
1540 
1541 /**
1542  * ixgbe_can_reuse_page - determine if we can reuse a page
1543  * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1544  *
1545  * Returns true if page can be reused in another Rx buffer
1546  **/
ixgbe_can_reuse_page(struct ixgbe_rx_buffer * rx_buffer)1547 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1548 {
1549 	struct page *page = rx_buffer->page;
1550 
1551 	/* if we are only owner of page and it is local we can reuse it */
1552 	return likely(page_count(page) == 1) &&
1553 	       likely(page_to_nid(page) == numa_node_id());
1554 }
1555 
1556 /**
1557  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1558  * @rx_ring: rx descriptor ring to store buffers on
1559  * @old_buff: donor buffer to have page reused
1560  *
1561  * Syncronizes page for reuse by the adapter
1562  **/
ixgbe_reuse_rx_page(struct ixgbe_ring * rx_ring,struct ixgbe_rx_buffer * old_buff)1563 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1564 				struct ixgbe_rx_buffer *old_buff)
1565 {
1566 	struct ixgbe_rx_buffer *new_buff;
1567 	u16 nta = rx_ring->next_to_alloc;
1568 	u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1569 
1570 	new_buff = &rx_ring->rx_buffer_info[nta];
1571 
1572 	/* update, and store next to alloc */
1573 	nta++;
1574 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1575 
1576 	/* transfer page from old buffer to new buffer */
1577 	new_buff->page = old_buff->page;
1578 	new_buff->dma = old_buff->dma;
1579 
1580 	/* flip page offset to other buffer and store to new_buff */
1581 	new_buff->page_offset = old_buff->page_offset ^ bufsz;
1582 
1583 	/* sync the buffer for use by the device */
1584 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1585 					 new_buff->page_offset, bufsz,
1586 					 DMA_FROM_DEVICE);
1587 
1588 	/* bump ref count on page before it is given to the stack */
1589 	get_page(new_buff->page);
1590 }
1591 
1592 /**
1593  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1594  * @rx_ring: rx descriptor ring to transact packets on
1595  * @rx_buffer: buffer containing page to add
1596  * @rx_desc: descriptor containing length of buffer written by hardware
1597  * @skb: sk_buff to place the data into
1598  *
1599  * This function is based on skb_add_rx_frag.  I would have used that
1600  * function however it doesn't handle the truesize case correctly since we
1601  * are allocating more memory than might be used for a single receive.
1602  **/
ixgbe_add_rx_frag(struct ixgbe_ring * rx_ring,struct ixgbe_rx_buffer * rx_buffer,struct sk_buff * skb,int size)1603 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1604 			      struct ixgbe_rx_buffer *rx_buffer,
1605 			      struct sk_buff *skb, int size)
1606 {
1607 	skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1608 			   rx_buffer->page, rx_buffer->page_offset,
1609 			   size);
1610 	skb->len += size;
1611 	skb->data_len += size;
1612 	skb->truesize += ixgbe_rx_bufsz(rx_ring);
1613 }
1614 
1615 /**
1616  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1617  * @q_vector: structure containing interrupt and ring information
1618  * @rx_ring: rx descriptor ring to transact packets on
1619  * @budget: Total limit on number of packets to process
1620  *
1621  * This function provides a "bounce buffer" approach to Rx interrupt
1622  * processing.  The advantage to this is that on systems that have
1623  * expensive overhead for IOMMU access this provides a means of avoiding
1624  * it by maintaining the mapping of the page to the syste.
1625  *
1626  * Returns true if all work is completed without reaching budget
1627  **/
ixgbe_clean_rx_irq(struct ixgbe_q_vector * q_vector,struct ixgbe_ring * rx_ring,int budget)1628 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1629 			       struct ixgbe_ring *rx_ring,
1630 			       int budget)
1631 {
1632 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1633 #ifdef IXGBE_FCOE
1634 	struct ixgbe_adapter *adapter = q_vector->adapter;
1635 	int ddp_bytes = 0;
1636 #endif /* IXGBE_FCOE */
1637 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1638 
1639 	do {
1640 		struct ixgbe_rx_buffer *rx_buffer;
1641 		union ixgbe_adv_rx_desc *rx_desc;
1642 		struct sk_buff *skb;
1643 		struct page *page;
1644 		u16 ntc;
1645 
1646 		/* return some buffers to hardware, one at a time is too slow */
1647 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1648 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1649 			cleaned_count = 0;
1650 		}
1651 
1652 		ntc = rx_ring->next_to_clean;
1653 		rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1654 		rx_buffer = &rx_ring->rx_buffer_info[ntc];
1655 
1656 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1657 			break;
1658 
1659 		/*
1660 		 * This memory barrier is needed to keep us from reading
1661 		 * any other fields out of the rx_desc until we know the
1662 		 * RXD_STAT_DD bit is set
1663 		 */
1664 		rmb();
1665 
1666 		page = rx_buffer->page;
1667 		prefetchw(page);
1668 
1669 		skb = rx_buffer->skb;
1670 
1671 		if (likely(!skb)) {
1672 			void *page_addr = page_address(page) +
1673 					  rx_buffer->page_offset;
1674 
1675 			/* prefetch first cache line of first page */
1676 			prefetch(page_addr);
1677 #if L1_CACHE_BYTES < 128
1678 			prefetch(page_addr + L1_CACHE_BYTES);
1679 #endif
1680 
1681 			/* allocate a skb to store the frags */
1682 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1683 							IXGBE_RX_HDR_SIZE);
1684 			if (unlikely(!skb)) {
1685 				rx_ring->rx_stats.alloc_rx_buff_failed++;
1686 				break;
1687 			}
1688 
1689 			/*
1690 			 * we will be copying header into skb->data in
1691 			 * pskb_may_pull so it is in our interest to prefetch
1692 			 * it now to avoid a possible cache miss
1693 			 */
1694 			prefetchw(skb->data);
1695 
1696 			/*
1697 			 * Delay unmapping of the first packet. It carries the
1698 			 * header information, HW may still access the header
1699 			 * after the writeback.  Only unmap it when EOP is
1700 			 * reached
1701 			 */
1702 			IXGBE_CB(skb)->dma = rx_buffer->dma;
1703 		} else {
1704 			/* we are reusing so sync this buffer for CPU use */
1705 			dma_sync_single_range_for_cpu(rx_ring->dev,
1706 						      rx_buffer->dma,
1707 						      rx_buffer->page_offset,
1708 						      ixgbe_rx_bufsz(rx_ring),
1709 						      DMA_FROM_DEVICE);
1710 		}
1711 
1712 		/* pull page into skb */
1713 		ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1714 				  le16_to_cpu(rx_desc->wb.upper.length));
1715 
1716 		if (ixgbe_can_reuse_page(rx_buffer)) {
1717 			/* hand second half of page back to the ring */
1718 			ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1719 		} else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1720 			/* the page has been released from the ring */
1721 			IXGBE_CB(skb)->page_released = true;
1722 		} else {
1723 			/* we are not reusing the buffer so unmap it */
1724 			dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1725 				       ixgbe_rx_pg_size(rx_ring),
1726 				       DMA_FROM_DEVICE);
1727 		}
1728 
1729 		/* clear contents of buffer_info */
1730 		rx_buffer->skb = NULL;
1731 		rx_buffer->dma = 0;
1732 		rx_buffer->page = NULL;
1733 
1734 		ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1735 
1736 		cleaned_count++;
1737 
1738 		/* place incomplete frames back on ring for completion */
1739 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1740 			continue;
1741 
1742 		/* verify the packet layout is correct */
1743 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1744 			continue;
1745 
1746 		/* probably a little skewed due to removing CRC */
1747 		total_rx_bytes += skb->len;
1748 		total_rx_packets++;
1749 
1750 		/* populate checksum, timestamp, VLAN, and protocol */
1751 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1752 
1753 #ifdef IXGBE_FCOE
1754 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
1755 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1756 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1757 			if (!ddp_bytes) {
1758 				dev_kfree_skb_any(skb);
1759 				continue;
1760 			}
1761 		}
1762 
1763 #endif /* IXGBE_FCOE */
1764 		ixgbe_rx_skb(q_vector, skb);
1765 
1766 		/* update budget accounting */
1767 		budget--;
1768 	} while (likely(budget));
1769 
1770 #ifdef IXGBE_FCOE
1771 	/* include DDPed FCoE data */
1772 	if (ddp_bytes > 0) {
1773 		unsigned int mss;
1774 
1775 		mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1776 			sizeof(struct fc_frame_header) -
1777 			sizeof(struct fcoe_crc_eof);
1778 		if (mss > 512)
1779 			mss &= ~511;
1780 		total_rx_bytes += ddp_bytes;
1781 		total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1782 	}
1783 
1784 #endif /* IXGBE_FCOE */
1785 	u64_stats_update_begin(&rx_ring->syncp);
1786 	rx_ring->stats.packets += total_rx_packets;
1787 	rx_ring->stats.bytes += total_rx_bytes;
1788 	u64_stats_update_end(&rx_ring->syncp);
1789 	q_vector->rx.total_packets += total_rx_packets;
1790 	q_vector->rx.total_bytes += total_rx_bytes;
1791 
1792 	if (cleaned_count)
1793 		ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1794 
1795 	return !!budget;
1796 }
1797 
1798 /**
1799  * ixgbe_configure_msix - Configure MSI-X hardware
1800  * @adapter: board private structure
1801  *
1802  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1803  * interrupts.
1804  **/
ixgbe_configure_msix(struct ixgbe_adapter * adapter)1805 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1806 {
1807 	struct ixgbe_q_vector *q_vector;
1808 	int q_vectors, v_idx;
1809 	u32 mask;
1810 
1811 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1812 
1813 	/* Populate MSIX to EITR Select */
1814 	if (adapter->num_vfs > 32) {
1815 		u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1816 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1817 	}
1818 
1819 	/*
1820 	 * Populate the IVAR table and set the ITR values to the
1821 	 * corresponding register.
1822 	 */
1823 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1824 		struct ixgbe_ring *ring;
1825 		q_vector = adapter->q_vector[v_idx];
1826 
1827 		ixgbe_for_each_ring(ring, q_vector->rx)
1828 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1829 
1830 		ixgbe_for_each_ring(ring, q_vector->tx)
1831 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1832 
1833 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1834 			/* tx only vector */
1835 			if (adapter->tx_itr_setting == 1)
1836 				q_vector->itr = IXGBE_10K_ITR;
1837 			else
1838 				q_vector->itr = adapter->tx_itr_setting;
1839 		} else {
1840 			/* rx or rx/tx vector */
1841 			if (adapter->rx_itr_setting == 1)
1842 				q_vector->itr = IXGBE_20K_ITR;
1843 			else
1844 				q_vector->itr = adapter->rx_itr_setting;
1845 		}
1846 
1847 		ixgbe_write_eitr(q_vector);
1848 	}
1849 
1850 	switch (adapter->hw.mac.type) {
1851 	case ixgbe_mac_82598EB:
1852 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1853 			       v_idx);
1854 		break;
1855 	case ixgbe_mac_82599EB:
1856 	case ixgbe_mac_X540:
1857 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
1858 		break;
1859 	default:
1860 		break;
1861 	}
1862 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1863 
1864 	/* set up to autoclear timer, and the vectors */
1865 	mask = IXGBE_EIMS_ENABLE_MASK;
1866 	mask &= ~(IXGBE_EIMS_OTHER |
1867 		  IXGBE_EIMS_MAILBOX |
1868 		  IXGBE_EIMS_LSC);
1869 
1870 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1871 }
1872 
1873 enum latency_range {
1874 	lowest_latency = 0,
1875 	low_latency = 1,
1876 	bulk_latency = 2,
1877 	latency_invalid = 255
1878 };
1879 
1880 /**
1881  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1882  * @q_vector: structure containing interrupt and ring information
1883  * @ring_container: structure containing ring performance data
1884  *
1885  *      Stores a new ITR value based on packets and byte
1886  *      counts during the last interrupt.  The advantage of per interrupt
1887  *      computation is faster updates and more accurate ITR for the current
1888  *      traffic pattern.  Constants in this function were computed
1889  *      based on theoretical maximum wire speed and thresholds were set based
1890  *      on testing data as well as attempting to minimize response time
1891  *      while increasing bulk throughput.
1892  *      this functionality is controlled by the InterruptThrottleRate module
1893  *      parameter (see ixgbe_param.c)
1894  **/
ixgbe_update_itr(struct ixgbe_q_vector * q_vector,struct ixgbe_ring_container * ring_container)1895 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1896 			     struct ixgbe_ring_container *ring_container)
1897 {
1898 	int bytes = ring_container->total_bytes;
1899 	int packets = ring_container->total_packets;
1900 	u32 timepassed_us;
1901 	u64 bytes_perint;
1902 	u8 itr_setting = ring_container->itr;
1903 
1904 	if (packets == 0)
1905 		return;
1906 
1907 	/* simple throttlerate management
1908 	 *   0-10MB/s   lowest (100000 ints/s)
1909 	 *  10-20MB/s   low    (20000 ints/s)
1910 	 *  20-1249MB/s bulk   (8000 ints/s)
1911 	 */
1912 	/* what was last interrupt timeslice? */
1913 	timepassed_us = q_vector->itr >> 2;
1914 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1915 
1916 	switch (itr_setting) {
1917 	case lowest_latency:
1918 		if (bytes_perint > 10)
1919 			itr_setting = low_latency;
1920 		break;
1921 	case low_latency:
1922 		if (bytes_perint > 20)
1923 			itr_setting = bulk_latency;
1924 		else if (bytes_perint <= 10)
1925 			itr_setting = lowest_latency;
1926 		break;
1927 	case bulk_latency:
1928 		if (bytes_perint <= 20)
1929 			itr_setting = low_latency;
1930 		break;
1931 	}
1932 
1933 	/* clear work counters since we have the values we need */
1934 	ring_container->total_bytes = 0;
1935 	ring_container->total_packets = 0;
1936 
1937 	/* write updated itr to ring container */
1938 	ring_container->itr = itr_setting;
1939 }
1940 
1941 /**
1942  * ixgbe_write_eitr - write EITR register in hardware specific way
1943  * @q_vector: structure containing interrupt and ring information
1944  *
1945  * This function is made to be called by ethtool and by the driver
1946  * when it needs to update EITR registers at runtime.  Hardware
1947  * specific quirks/differences are taken care of here.
1948  */
ixgbe_write_eitr(struct ixgbe_q_vector * q_vector)1949 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1950 {
1951 	struct ixgbe_adapter *adapter = q_vector->adapter;
1952 	struct ixgbe_hw *hw = &adapter->hw;
1953 	int v_idx = q_vector->v_idx;
1954 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1955 
1956 	switch (adapter->hw.mac.type) {
1957 	case ixgbe_mac_82598EB:
1958 		/* must write high and low 16 bits to reset counter */
1959 		itr_reg |= (itr_reg << 16);
1960 		break;
1961 	case ixgbe_mac_82599EB:
1962 	case ixgbe_mac_X540:
1963 		/*
1964 		 * set the WDIS bit to not clear the timer bits and cause an
1965 		 * immediate assertion of the interrupt
1966 		 */
1967 		itr_reg |= IXGBE_EITR_CNT_WDIS;
1968 		break;
1969 	default:
1970 		break;
1971 	}
1972 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1973 }
1974 
ixgbe_set_itr(struct ixgbe_q_vector * q_vector)1975 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1976 {
1977 	u32 new_itr = q_vector->itr;
1978 	u8 current_itr;
1979 
1980 	ixgbe_update_itr(q_vector, &q_vector->tx);
1981 	ixgbe_update_itr(q_vector, &q_vector->rx);
1982 
1983 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1984 
1985 	switch (current_itr) {
1986 	/* counts and packets in update_itr are dependent on these numbers */
1987 	case lowest_latency:
1988 		new_itr = IXGBE_100K_ITR;
1989 		break;
1990 	case low_latency:
1991 		new_itr = IXGBE_20K_ITR;
1992 		break;
1993 	case bulk_latency:
1994 		new_itr = IXGBE_8K_ITR;
1995 		break;
1996 	default:
1997 		break;
1998 	}
1999 
2000 	if (new_itr != q_vector->itr) {
2001 		/* do an exponential smoothing */
2002 		new_itr = (10 * new_itr * q_vector->itr) /
2003 			  ((9 * new_itr) + q_vector->itr);
2004 
2005 		/* save the algorithm value here */
2006 		q_vector->itr = new_itr;
2007 
2008 		ixgbe_write_eitr(q_vector);
2009 	}
2010 }
2011 
2012 /**
2013  * ixgbe_check_overtemp_subtask - check for over temperature
2014  * @adapter: pointer to adapter
2015  **/
ixgbe_check_overtemp_subtask(struct ixgbe_adapter * adapter)2016 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2017 {
2018 	struct ixgbe_hw *hw = &adapter->hw;
2019 	u32 eicr = adapter->interrupt_event;
2020 
2021 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2022 		return;
2023 
2024 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2025 	    !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2026 		return;
2027 
2028 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2029 
2030 	switch (hw->device_id) {
2031 	case IXGBE_DEV_ID_82599_T3_LOM:
2032 		/*
2033 		 * Since the warning interrupt is for both ports
2034 		 * we don't have to check if:
2035 		 *  - This interrupt wasn't for our port.
2036 		 *  - We may have missed the interrupt so always have to
2037 		 *    check if we  got a LSC
2038 		 */
2039 		if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2040 		    !(eicr & IXGBE_EICR_LSC))
2041 			return;
2042 
2043 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2044 			u32 autoneg;
2045 			bool link_up = false;
2046 
2047 			hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2048 
2049 			if (link_up)
2050 				return;
2051 		}
2052 
2053 		/* Check if this is not due to overtemp */
2054 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2055 			return;
2056 
2057 		break;
2058 	default:
2059 		if (!(eicr & IXGBE_EICR_GPI_SDP0))
2060 			return;
2061 		break;
2062 	}
2063 	e_crit(drv,
2064 	       "Network adapter has been stopped because it has over heated. "
2065 	       "Restart the computer. If the problem persists, "
2066 	       "power off the system and replace the adapter\n");
2067 
2068 	adapter->interrupt_event = 0;
2069 }
2070 
ixgbe_check_fan_failure(struct ixgbe_adapter * adapter,u32 eicr)2071 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2072 {
2073 	struct ixgbe_hw *hw = &adapter->hw;
2074 
2075 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2076 	    (eicr & IXGBE_EICR_GPI_SDP1)) {
2077 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2078 		/* write to clear the interrupt */
2079 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2080 	}
2081 }
2082 
ixgbe_check_overtemp_event(struct ixgbe_adapter * adapter,u32 eicr)2083 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2084 {
2085 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2086 		return;
2087 
2088 	switch (adapter->hw.mac.type) {
2089 	case ixgbe_mac_82599EB:
2090 		/*
2091 		 * Need to check link state so complete overtemp check
2092 		 * on service task
2093 		 */
2094 		if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2095 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2096 			adapter->interrupt_event = eicr;
2097 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2098 			ixgbe_service_event_schedule(adapter);
2099 			return;
2100 		}
2101 		return;
2102 	case ixgbe_mac_X540:
2103 		if (!(eicr & IXGBE_EICR_TS))
2104 			return;
2105 		break;
2106 	default:
2107 		return;
2108 	}
2109 
2110 	e_crit(drv,
2111 	       "Network adapter has been stopped because it has over heated. "
2112 	       "Restart the computer. If the problem persists, "
2113 	       "power off the system and replace the adapter\n");
2114 }
2115 
ixgbe_check_sfp_event(struct ixgbe_adapter * adapter,u32 eicr)2116 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2117 {
2118 	struct ixgbe_hw *hw = &adapter->hw;
2119 
2120 	if (eicr & IXGBE_EICR_GPI_SDP2) {
2121 		/* Clear the interrupt */
2122 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2123 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2124 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2125 			ixgbe_service_event_schedule(adapter);
2126 		}
2127 	}
2128 
2129 	if (eicr & IXGBE_EICR_GPI_SDP1) {
2130 		/* Clear the interrupt */
2131 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2132 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2133 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2134 			ixgbe_service_event_schedule(adapter);
2135 		}
2136 	}
2137 }
2138 
ixgbe_check_lsc(struct ixgbe_adapter * adapter)2139 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2140 {
2141 	struct ixgbe_hw *hw = &adapter->hw;
2142 
2143 	adapter->lsc_int++;
2144 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2145 	adapter->link_check_timeout = jiffies;
2146 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2147 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2148 		IXGBE_WRITE_FLUSH(hw);
2149 		ixgbe_service_event_schedule(adapter);
2150 	}
2151 }
2152 
ixgbe_irq_enable_queues(struct ixgbe_adapter * adapter,u64 qmask)2153 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2154 					   u64 qmask)
2155 {
2156 	u32 mask;
2157 	struct ixgbe_hw *hw = &adapter->hw;
2158 
2159 	switch (hw->mac.type) {
2160 	case ixgbe_mac_82598EB:
2161 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2162 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2163 		break;
2164 	case ixgbe_mac_82599EB:
2165 	case ixgbe_mac_X540:
2166 		mask = (qmask & 0xFFFFFFFF);
2167 		if (mask)
2168 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2169 		mask = (qmask >> 32);
2170 		if (mask)
2171 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2172 		break;
2173 	default:
2174 		break;
2175 	}
2176 	/* skip the flush */
2177 }
2178 
ixgbe_irq_disable_queues(struct ixgbe_adapter * adapter,u64 qmask)2179 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2180 					    u64 qmask)
2181 {
2182 	u32 mask;
2183 	struct ixgbe_hw *hw = &adapter->hw;
2184 
2185 	switch (hw->mac.type) {
2186 	case ixgbe_mac_82598EB:
2187 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2188 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2189 		break;
2190 	case ixgbe_mac_82599EB:
2191 	case ixgbe_mac_X540:
2192 		mask = (qmask & 0xFFFFFFFF);
2193 		if (mask)
2194 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2195 		mask = (qmask >> 32);
2196 		if (mask)
2197 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2198 		break;
2199 	default:
2200 		break;
2201 	}
2202 	/* skip the flush */
2203 }
2204 
2205 /**
2206  * ixgbe_irq_enable - Enable default interrupt generation settings
2207  * @adapter: board private structure
2208  **/
ixgbe_irq_enable(struct ixgbe_adapter * adapter,bool queues,bool flush)2209 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2210 				    bool flush)
2211 {
2212 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2213 
2214 	/* don't reenable LSC while waiting for link */
2215 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2216 		mask &= ~IXGBE_EIMS_LSC;
2217 
2218 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2219 		switch (adapter->hw.mac.type) {
2220 		case ixgbe_mac_82599EB:
2221 			mask |= IXGBE_EIMS_GPI_SDP0;
2222 			break;
2223 		case ixgbe_mac_X540:
2224 			mask |= IXGBE_EIMS_TS;
2225 			break;
2226 		default:
2227 			break;
2228 		}
2229 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2230 		mask |= IXGBE_EIMS_GPI_SDP1;
2231 	switch (adapter->hw.mac.type) {
2232 	case ixgbe_mac_82599EB:
2233 		mask |= IXGBE_EIMS_GPI_SDP1;
2234 		mask |= IXGBE_EIMS_GPI_SDP2;
2235 	case ixgbe_mac_X540:
2236 		mask |= IXGBE_EIMS_ECC;
2237 		mask |= IXGBE_EIMS_MAILBOX;
2238 		break;
2239 	default:
2240 		break;
2241 	}
2242 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2243 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2244 		mask |= IXGBE_EIMS_FLOW_DIR;
2245 
2246 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2247 	if (queues)
2248 		ixgbe_irq_enable_queues(adapter, ~0);
2249 	if (flush)
2250 		IXGBE_WRITE_FLUSH(&adapter->hw);
2251 }
2252 
ixgbe_msix_other(int irq,void * data)2253 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2254 {
2255 	struct ixgbe_adapter *adapter = data;
2256 	struct ixgbe_hw *hw = &adapter->hw;
2257 	u32 eicr;
2258 
2259 	/*
2260 	 * Workaround for Silicon errata.  Use clear-by-write instead
2261 	 * of clear-by-read.  Reading with EICS will return the
2262 	 * interrupt causes without clearing, which later be done
2263 	 * with the write to EICR.
2264 	 */
2265 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2266 
2267 	/* The lower 16bits of the EICR register are for the queue interrupts
2268 	 * which should be masked here in order to not accidently clear them if
2269 	 * the bits are high when ixgbe_msix_other is called. There is a race
2270 	 * condition otherwise which results in possible performance loss
2271 	 * especially if the ixgbe_msix_other interrupt is triggering
2272 	 * consistently (as it would when PPS is turned on for the X540 device)
2273 	 */
2274 	eicr &= 0xFFFF0000;
2275 
2276 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2277 
2278 	if (eicr & IXGBE_EICR_LSC)
2279 		ixgbe_check_lsc(adapter);
2280 
2281 	if (eicr & IXGBE_EICR_MAILBOX)
2282 		ixgbe_msg_task(adapter);
2283 
2284 	switch (hw->mac.type) {
2285 	case ixgbe_mac_82599EB:
2286 	case ixgbe_mac_X540:
2287 		if (eicr & IXGBE_EICR_ECC)
2288 			e_info(link, "Received unrecoverable ECC Err, please "
2289 			       "reboot\n");
2290 		/* Handle Flow Director Full threshold interrupt */
2291 		if (eicr & IXGBE_EICR_FLOW_DIR) {
2292 			int reinit_count = 0;
2293 			int i;
2294 			for (i = 0; i < adapter->num_tx_queues; i++) {
2295 				struct ixgbe_ring *ring = adapter->tx_ring[i];
2296 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2297 						       &ring->state))
2298 					reinit_count++;
2299 			}
2300 			if (reinit_count) {
2301 				/* no more flow director interrupts until after init */
2302 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2303 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2304 				ixgbe_service_event_schedule(adapter);
2305 			}
2306 		}
2307 		ixgbe_check_sfp_event(adapter, eicr);
2308 		ixgbe_check_overtemp_event(adapter, eicr);
2309 		break;
2310 	default:
2311 		break;
2312 	}
2313 
2314 	ixgbe_check_fan_failure(adapter, eicr);
2315 
2316 	/* re-enable the original interrupt state, no lsc, no queues */
2317 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2318 		ixgbe_irq_enable(adapter, false, false);
2319 
2320 	return IRQ_HANDLED;
2321 }
2322 
ixgbe_msix_clean_rings(int irq,void * data)2323 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2324 {
2325 	struct ixgbe_q_vector *q_vector = data;
2326 
2327 	/* EIAM disabled interrupts (on this vector) for us */
2328 
2329 	if (q_vector->rx.ring || q_vector->tx.ring)
2330 		napi_schedule(&q_vector->napi);
2331 
2332 	return IRQ_HANDLED;
2333 }
2334 
2335 /**
2336  * ixgbe_poll - NAPI Rx polling callback
2337  * @napi: structure for representing this polling device
2338  * @budget: how many packets driver is allowed to clean
2339  *
2340  * This function is used for legacy and MSI, NAPI mode
2341  **/
ixgbe_poll(struct napi_struct * napi,int budget)2342 int ixgbe_poll(struct napi_struct *napi, int budget)
2343 {
2344 	struct ixgbe_q_vector *q_vector =
2345 				container_of(napi, struct ixgbe_q_vector, napi);
2346 	struct ixgbe_adapter *adapter = q_vector->adapter;
2347 	struct ixgbe_ring *ring;
2348 	int per_ring_budget;
2349 	bool clean_complete = true;
2350 
2351 #ifdef CONFIG_IXGBE_DCA
2352 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2353 		ixgbe_update_dca(q_vector);
2354 #endif
2355 
2356 	ixgbe_for_each_ring(ring, q_vector->tx)
2357 		clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2358 
2359 	/* attempt to distribute budget to each queue fairly, but don't allow
2360 	 * the budget to go below 1 because we'll exit polling */
2361 	if (q_vector->rx.count > 1)
2362 		per_ring_budget = max(budget/q_vector->rx.count, 1);
2363 	else
2364 		per_ring_budget = budget;
2365 
2366 	ixgbe_for_each_ring(ring, q_vector->rx)
2367 		clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2368 						     per_ring_budget);
2369 
2370 	/* If all work not completed, return budget and keep polling */
2371 	if (!clean_complete)
2372 		return budget;
2373 
2374 	/* all work done, exit the polling mode */
2375 	napi_complete(napi);
2376 	if (adapter->rx_itr_setting & 1)
2377 		ixgbe_set_itr(q_vector);
2378 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2379 		ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2380 
2381 	return 0;
2382 }
2383 
2384 /**
2385  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2386  * @adapter: board private structure
2387  *
2388  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2389  * interrupts from the kernel.
2390  **/
ixgbe_request_msix_irqs(struct ixgbe_adapter * adapter)2391 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2392 {
2393 	struct net_device *netdev = adapter->netdev;
2394 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2395 	int vector, err;
2396 	int ri = 0, ti = 0;
2397 
2398 	for (vector = 0; vector < q_vectors; vector++) {
2399 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2400 		struct msix_entry *entry = &adapter->msix_entries[vector];
2401 
2402 		if (q_vector->tx.ring && q_vector->rx.ring) {
2403 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2404 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
2405 			ti++;
2406 		} else if (q_vector->rx.ring) {
2407 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2408 				 "%s-%s-%d", netdev->name, "rx", ri++);
2409 		} else if (q_vector->tx.ring) {
2410 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2411 				 "%s-%s-%d", netdev->name, "tx", ti++);
2412 		} else {
2413 			/* skip this unused q_vector */
2414 			continue;
2415 		}
2416 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2417 				  q_vector->name, q_vector);
2418 		if (err) {
2419 			e_err(probe, "request_irq failed for MSIX interrupt "
2420 			      "Error: %d\n", err);
2421 			goto free_queue_irqs;
2422 		}
2423 		/* If Flow Director is enabled, set interrupt affinity */
2424 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2425 			/* assign the mask for this irq */
2426 			irq_set_affinity_hint(entry->vector,
2427 					      &q_vector->affinity_mask);
2428 		}
2429 	}
2430 
2431 	err = request_irq(adapter->msix_entries[vector].vector,
2432 			  ixgbe_msix_other, 0, netdev->name, adapter);
2433 	if (err) {
2434 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
2435 		goto free_queue_irqs;
2436 	}
2437 
2438 	return 0;
2439 
2440 free_queue_irqs:
2441 	while (vector) {
2442 		vector--;
2443 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2444 				      NULL);
2445 		free_irq(adapter->msix_entries[vector].vector,
2446 			 adapter->q_vector[vector]);
2447 	}
2448 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2449 	pci_disable_msix(adapter->pdev);
2450 	kfree(adapter->msix_entries);
2451 	adapter->msix_entries = NULL;
2452 	return err;
2453 }
2454 
2455 /**
2456  * ixgbe_intr - legacy mode Interrupt Handler
2457  * @irq: interrupt number
2458  * @data: pointer to a network interface device structure
2459  **/
ixgbe_intr(int irq,void * data)2460 static irqreturn_t ixgbe_intr(int irq, void *data)
2461 {
2462 	struct ixgbe_adapter *adapter = data;
2463 	struct ixgbe_hw *hw = &adapter->hw;
2464 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2465 	u32 eicr;
2466 
2467 	/*
2468 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2469 	 * before the read of EICR.
2470 	 */
2471 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2472 
2473 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2474 	 * therefore no explicit interrupt disable is necessary */
2475 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2476 	if (!eicr) {
2477 		/*
2478 		 * shared interrupt alert!
2479 		 * make sure interrupts are enabled because the read will
2480 		 * have disabled interrupts due to EIAM
2481 		 * finish the workaround of silicon errata on 82598.  Unmask
2482 		 * the interrupt that we masked before the EICR read.
2483 		 */
2484 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
2485 			ixgbe_irq_enable(adapter, true, true);
2486 		return IRQ_NONE;	/* Not our interrupt */
2487 	}
2488 
2489 	if (eicr & IXGBE_EICR_LSC)
2490 		ixgbe_check_lsc(adapter);
2491 
2492 	switch (hw->mac.type) {
2493 	case ixgbe_mac_82599EB:
2494 		ixgbe_check_sfp_event(adapter, eicr);
2495 		/* Fall through */
2496 	case ixgbe_mac_X540:
2497 		if (eicr & IXGBE_EICR_ECC)
2498 			e_info(link, "Received unrecoverable ECC err, please "
2499 				     "reboot\n");
2500 		ixgbe_check_overtemp_event(adapter, eicr);
2501 		break;
2502 	default:
2503 		break;
2504 	}
2505 
2506 	ixgbe_check_fan_failure(adapter, eicr);
2507 
2508 	/* would disable interrupts here but EIAM disabled it */
2509 	napi_schedule(&q_vector->napi);
2510 
2511 	/*
2512 	 * re-enable link(maybe) and non-queue interrupts, no flush.
2513 	 * ixgbe_poll will re-enable the queue interrupts
2514 	 */
2515 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
2516 		ixgbe_irq_enable(adapter, false, false);
2517 
2518 	return IRQ_HANDLED;
2519 }
2520 
2521 /**
2522  * ixgbe_request_irq - initialize interrupts
2523  * @adapter: board private structure
2524  *
2525  * Attempts to configure interrupts using the best available
2526  * capabilities of the hardware and kernel.
2527  **/
ixgbe_request_irq(struct ixgbe_adapter * adapter)2528 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2529 {
2530 	struct net_device *netdev = adapter->netdev;
2531 	int err;
2532 
2533 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2534 		err = ixgbe_request_msix_irqs(adapter);
2535 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2536 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2537 				  netdev->name, adapter);
2538 	else
2539 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2540 				  netdev->name, adapter);
2541 
2542 	if (err)
2543 		e_err(probe, "request_irq failed, Error %d\n", err);
2544 
2545 	return err;
2546 }
2547 
ixgbe_free_irq(struct ixgbe_adapter * adapter)2548 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2549 {
2550 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2551 		int i, q_vectors;
2552 
2553 		q_vectors = adapter->num_msix_vectors;
2554 		i = q_vectors - 1;
2555 		free_irq(adapter->msix_entries[i].vector, adapter);
2556 		i--;
2557 
2558 		for (; i >= 0; i--) {
2559 			/* free only the irqs that were actually requested */
2560 			if (!adapter->q_vector[i]->rx.ring &&
2561 			    !adapter->q_vector[i]->tx.ring)
2562 				continue;
2563 
2564 			/* clear the affinity_mask in the IRQ descriptor */
2565 			irq_set_affinity_hint(adapter->msix_entries[i].vector,
2566 					      NULL);
2567 
2568 			free_irq(adapter->msix_entries[i].vector,
2569 				 adapter->q_vector[i]);
2570 		}
2571 	} else {
2572 		free_irq(adapter->pdev->irq, adapter);
2573 	}
2574 }
2575 
2576 /**
2577  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2578  * @adapter: board private structure
2579  **/
ixgbe_irq_disable(struct ixgbe_adapter * adapter)2580 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2581 {
2582 	switch (adapter->hw.mac.type) {
2583 	case ixgbe_mac_82598EB:
2584 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2585 		break;
2586 	case ixgbe_mac_82599EB:
2587 	case ixgbe_mac_X540:
2588 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2589 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2590 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2591 		break;
2592 	default:
2593 		break;
2594 	}
2595 	IXGBE_WRITE_FLUSH(&adapter->hw);
2596 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2597 		int i;
2598 		for (i = 0; i < adapter->num_msix_vectors; i++)
2599 			synchronize_irq(adapter->msix_entries[i].vector);
2600 	} else {
2601 		synchronize_irq(adapter->pdev->irq);
2602 	}
2603 }
2604 
2605 /**
2606  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2607  *
2608  **/
ixgbe_configure_msi_and_legacy(struct ixgbe_adapter * adapter)2609 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2610 {
2611 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2612 
2613 	/* rx/tx vector */
2614 	if (adapter->rx_itr_setting == 1)
2615 		q_vector->itr = IXGBE_20K_ITR;
2616 	else
2617 		q_vector->itr = adapter->rx_itr_setting;
2618 
2619 	ixgbe_write_eitr(q_vector);
2620 
2621 	ixgbe_set_ivar(adapter, 0, 0, 0);
2622 	ixgbe_set_ivar(adapter, 1, 0, 0);
2623 
2624 	e_info(hw, "Legacy interrupt IVAR setup done\n");
2625 }
2626 
2627 /**
2628  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2629  * @adapter: board private structure
2630  * @ring: structure containing ring specific data
2631  *
2632  * Configure the Tx descriptor ring after a reset.
2633  **/
ixgbe_configure_tx_ring(struct ixgbe_adapter * adapter,struct ixgbe_ring * ring)2634 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2635 			     struct ixgbe_ring *ring)
2636 {
2637 	struct ixgbe_hw *hw = &adapter->hw;
2638 	u64 tdba = ring->dma;
2639 	int wait_loop = 10;
2640 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
2641 	u8 reg_idx = ring->reg_idx;
2642 
2643 	/* disable queue to avoid issues while updating state */
2644 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2645 	IXGBE_WRITE_FLUSH(hw);
2646 
2647 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2648 			(tdba & DMA_BIT_MASK(32)));
2649 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2650 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2651 			ring->count * sizeof(union ixgbe_adv_tx_desc));
2652 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2653 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2654 	ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2655 
2656 	/*
2657 	 * set WTHRESH to encourage burst writeback, it should not be set
2658 	 * higher than 1 when ITR is 0 as it could cause false TX hangs
2659 	 *
2660 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2661 	 * to or less than the number of on chip descriptors, which is
2662 	 * currently 40.
2663 	 */
2664 	if (!ring->q_vector || (ring->q_vector->itr < 8))
2665 		txdctl |= (1 << 16);	/* WTHRESH = 1 */
2666 	else
2667 		txdctl |= (8 << 16);	/* WTHRESH = 8 */
2668 
2669 	/*
2670 	 * Setting PTHRESH to 32 both improves performance
2671 	 * and avoids a TX hang with DFP enabled
2672 	 */
2673 	txdctl |= (1 << 8) |	/* HTHRESH = 1 */
2674 		   32;		/* PTHRESH = 32 */
2675 
2676 	/* reinitialize flowdirector state */
2677 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2678 	    adapter->atr_sample_rate) {
2679 		ring->atr_sample_rate = adapter->atr_sample_rate;
2680 		ring->atr_count = 0;
2681 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2682 	} else {
2683 		ring->atr_sample_rate = 0;
2684 	}
2685 
2686 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2687 
2688 	/* enable queue */
2689 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2690 
2691 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2692 	if (hw->mac.type == ixgbe_mac_82598EB &&
2693 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2694 		return;
2695 
2696 	/* poll to verify queue is enabled */
2697 	do {
2698 		usleep_range(1000, 2000);
2699 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2700 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2701 	if (!wait_loop)
2702 		e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2703 }
2704 
ixgbe_setup_mtqc(struct ixgbe_adapter * adapter)2705 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2706 {
2707 	struct ixgbe_hw *hw = &adapter->hw;
2708 	u32 rttdcs;
2709 	u32 reg;
2710 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2711 
2712 	if (hw->mac.type == ixgbe_mac_82598EB)
2713 		return;
2714 
2715 	/* disable the arbiter while setting MTQC */
2716 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2717 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
2718 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2719 
2720 	/* set transmit pool layout */
2721 	switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2722 	case (IXGBE_FLAG_SRIOV_ENABLED):
2723 		IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2724 				(IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2725 		break;
2726 	default:
2727 		if (!tcs)
2728 			reg = IXGBE_MTQC_64Q_1PB;
2729 		else if (tcs <= 4)
2730 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2731 		else
2732 			reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2733 
2734 		IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2735 
2736 		/* Enable Security TX Buffer IFG for multiple pb */
2737 		if (tcs) {
2738 			reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2739 			reg |= IXGBE_SECTX_DCB;
2740 			IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2741 		}
2742 		break;
2743 	}
2744 
2745 	/* re-enable the arbiter */
2746 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2747 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2748 }
2749 
2750 /**
2751  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2752  * @adapter: board private structure
2753  *
2754  * Configure the Tx unit of the MAC after a reset.
2755  **/
ixgbe_configure_tx(struct ixgbe_adapter * adapter)2756 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2757 {
2758 	struct ixgbe_hw *hw = &adapter->hw;
2759 	u32 dmatxctl;
2760 	u32 i;
2761 
2762 	ixgbe_setup_mtqc(adapter);
2763 
2764 	if (hw->mac.type != ixgbe_mac_82598EB) {
2765 		/* DMATXCTL.EN must be before Tx queues are enabled */
2766 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2767 		dmatxctl |= IXGBE_DMATXCTL_TE;
2768 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2769 	}
2770 
2771 	/* Setup the HW Tx Head and Tail descriptor pointers */
2772 	for (i = 0; i < adapter->num_tx_queues; i++)
2773 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2774 }
2775 
2776 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2777 
ixgbe_configure_srrctl(struct ixgbe_adapter * adapter,struct ixgbe_ring * rx_ring)2778 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2779 				   struct ixgbe_ring *rx_ring)
2780 {
2781 	u32 srrctl;
2782 	u8 reg_idx = rx_ring->reg_idx;
2783 
2784 	switch (adapter->hw.mac.type) {
2785 	case ixgbe_mac_82598EB: {
2786 		struct ixgbe_ring_feature *feature = adapter->ring_feature;
2787 		const int mask = feature[RING_F_RSS].mask;
2788 		reg_idx = reg_idx & mask;
2789 	}
2790 		break;
2791 	case ixgbe_mac_82599EB:
2792 	case ixgbe_mac_X540:
2793 	default:
2794 		break;
2795 	}
2796 
2797 	srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2798 
2799 	srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2800 	srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2801 	if (adapter->num_vfs)
2802 		srrctl |= IXGBE_SRRCTL_DROP_EN;
2803 
2804 	srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2805 		  IXGBE_SRRCTL_BSIZEHDR_MASK;
2806 
2807 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2808 	srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2809 #else
2810 	srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2811 #endif
2812 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2813 
2814 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2815 }
2816 
ixgbe_setup_mrqc(struct ixgbe_adapter * adapter)2817 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2818 {
2819 	struct ixgbe_hw *hw = &adapter->hw;
2820 	static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2821 			  0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2822 			  0x6A3E67EA, 0x14364D17, 0x3BED200D};
2823 	u32 mrqc = 0, reta = 0;
2824 	u32 rxcsum;
2825 	int i, j;
2826 	u8 tcs = netdev_get_num_tc(adapter->netdev);
2827 	int maxq = adapter->ring_feature[RING_F_RSS].indices;
2828 
2829 	if (tcs)
2830 		maxq = min(maxq, adapter->num_tx_queues / tcs);
2831 
2832 	/* Fill out hash function seeds */
2833 	for (i = 0; i < 10; i++)
2834 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2835 
2836 	/* Fill out redirection table */
2837 	for (i = 0, j = 0; i < 128; i++, j++) {
2838 		if (j == maxq)
2839 			j = 0;
2840 		/* reta = 4-byte sliding window of
2841 		 * 0x00..(indices-1)(indices-1)00..etc. */
2842 		reta = (reta << 8) | (j * 0x11);
2843 		if ((i & 3) == 3)
2844 			IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2845 	}
2846 
2847 	/* Disable indicating checksum in descriptor, enables RSS hash */
2848 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2849 	rxcsum |= IXGBE_RXCSUM_PCSD;
2850 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2851 
2852 	if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2853 	    (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2854 		mrqc = IXGBE_MRQC_RSSEN;
2855 	} else {
2856 		int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2857 					     | IXGBE_FLAG_SRIOV_ENABLED);
2858 
2859 		switch (mask) {
2860 		case (IXGBE_FLAG_RSS_ENABLED):
2861 			if (!tcs)
2862 				mrqc = IXGBE_MRQC_RSSEN;
2863 			else if (tcs <= 4)
2864 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
2865 			else
2866 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
2867 			break;
2868 		case (IXGBE_FLAG_SRIOV_ENABLED):
2869 			mrqc = IXGBE_MRQC_VMDQEN;
2870 			break;
2871 		default:
2872 			break;
2873 		}
2874 	}
2875 
2876 	/* Perform hash on these packet types */
2877 	mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2878 	      | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2879 	      | IXGBE_MRQC_RSS_FIELD_IPV6
2880 	      | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2881 
2882 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2883 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2884 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2885 		mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2886 
2887 	IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2888 }
2889 
2890 /**
2891  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2892  * @adapter:    address of board private structure
2893  * @index:      index of ring to set
2894  **/
ixgbe_configure_rscctl(struct ixgbe_adapter * adapter,struct ixgbe_ring * ring)2895 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2896 				   struct ixgbe_ring *ring)
2897 {
2898 	struct ixgbe_hw *hw = &adapter->hw;
2899 	u32 rscctrl;
2900 	u8 reg_idx = ring->reg_idx;
2901 
2902 	if (!ring_is_rsc_enabled(ring))
2903 		return;
2904 
2905 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2906 	rscctrl |= IXGBE_RSCCTL_RSCEN;
2907 	/*
2908 	 * we must limit the number of descriptors so that the
2909 	 * total size of max desc * buf_len is not greater
2910 	 * than 65536
2911 	 */
2912 #if (PAGE_SIZE <= 8192)
2913 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2914 #elif (PAGE_SIZE <= 16384)
2915 	rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2916 #else
2917 	rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2918 #endif
2919 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2920 }
2921 
2922 /**
2923  *  ixgbe_set_uta - Set unicast filter table address
2924  *  @adapter: board private structure
2925  *
2926  *  The unicast table address is a register array of 32-bit registers.
2927  *  The table is meant to be used in a way similar to how the MTA is used
2928  *  however due to certain limitations in the hardware it is necessary to
2929  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2930  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2931  **/
ixgbe_set_uta(struct ixgbe_adapter * adapter)2932 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2933 {
2934 	struct ixgbe_hw *hw = &adapter->hw;
2935 	int i;
2936 
2937 	/* The UTA table only exists on 82599 hardware and newer */
2938 	if (hw->mac.type < ixgbe_mac_82599EB)
2939 		return;
2940 
2941 	/* we only need to do this if VMDq is enabled */
2942 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2943 		return;
2944 
2945 	for (i = 0; i < 128; i++)
2946 		IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2947 }
2948 
2949 #define IXGBE_MAX_RX_DESC_POLL 10
ixgbe_rx_desc_queue_enable(struct ixgbe_adapter * adapter,struct ixgbe_ring * ring)2950 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2951 				       struct ixgbe_ring *ring)
2952 {
2953 	struct ixgbe_hw *hw = &adapter->hw;
2954 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2955 	u32 rxdctl;
2956 	u8 reg_idx = ring->reg_idx;
2957 
2958 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2959 	if (hw->mac.type == ixgbe_mac_82598EB &&
2960 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2961 		return;
2962 
2963 	do {
2964 		usleep_range(1000, 2000);
2965 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2966 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2967 
2968 	if (!wait_loop) {
2969 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2970 		      "the polling period\n", reg_idx);
2971 	}
2972 }
2973 
ixgbe_disable_rx_queue(struct ixgbe_adapter * adapter,struct ixgbe_ring * ring)2974 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
2975 			    struct ixgbe_ring *ring)
2976 {
2977 	struct ixgbe_hw *hw = &adapter->hw;
2978 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2979 	u32 rxdctl;
2980 	u8 reg_idx = ring->reg_idx;
2981 
2982 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2983 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
2984 
2985 	/* write value back with RXDCTL.ENABLE bit cleared */
2986 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2987 
2988 	if (hw->mac.type == ixgbe_mac_82598EB &&
2989 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2990 		return;
2991 
2992 	/* the hardware may take up to 100us to really disable the rx queue */
2993 	do {
2994 		udelay(10);
2995 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2996 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
2997 
2998 	if (!wait_loop) {
2999 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3000 		      "the polling period\n", reg_idx);
3001 	}
3002 }
3003 
ixgbe_configure_rx_ring(struct ixgbe_adapter * adapter,struct ixgbe_ring * ring)3004 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3005 			     struct ixgbe_ring *ring)
3006 {
3007 	struct ixgbe_hw *hw = &adapter->hw;
3008 	u64 rdba = ring->dma;
3009 	u32 rxdctl;
3010 	u8 reg_idx = ring->reg_idx;
3011 
3012 	/* disable queue to avoid issues while updating state */
3013 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3014 	ixgbe_disable_rx_queue(adapter, ring);
3015 
3016 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3017 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3018 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3019 			ring->count * sizeof(union ixgbe_adv_rx_desc));
3020 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3021 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3022 	ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3023 
3024 	ixgbe_configure_srrctl(adapter, ring);
3025 	ixgbe_configure_rscctl(adapter, ring);
3026 
3027 	/* If operating in IOV mode set RLPML for X540 */
3028 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3029 	    hw->mac.type == ixgbe_mac_X540) {
3030 		rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3031 		rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3032 			    ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3033 	}
3034 
3035 	if (hw->mac.type == ixgbe_mac_82598EB) {
3036 		/*
3037 		 * enable cache line friendly hardware writes:
3038 		 * PTHRESH=32 descriptors (half the internal cache),
3039 		 * this also removes ugly rx_no_buffer_count increment
3040 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
3041 		 * WTHRESH=8 burst writeback up to two cache lines
3042 		 */
3043 		rxdctl &= ~0x3FFFFF;
3044 		rxdctl |=  0x080420;
3045 	}
3046 
3047 	/* enable receive descriptor ring */
3048 	rxdctl |= IXGBE_RXDCTL_ENABLE;
3049 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3050 
3051 	ixgbe_rx_desc_queue_enable(adapter, ring);
3052 	ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3053 }
3054 
ixgbe_setup_psrtype(struct ixgbe_adapter * adapter)3055 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3056 {
3057 	struct ixgbe_hw *hw = &adapter->hw;
3058 	int p;
3059 
3060 	/* PSRTYPE must be initialized in non 82598 adapters */
3061 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3062 		      IXGBE_PSRTYPE_UDPHDR |
3063 		      IXGBE_PSRTYPE_IPV4HDR |
3064 		      IXGBE_PSRTYPE_L2HDR |
3065 		      IXGBE_PSRTYPE_IPV6HDR;
3066 
3067 	if (hw->mac.type == ixgbe_mac_82598EB)
3068 		return;
3069 
3070 	if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3071 		psrtype |= (adapter->num_rx_queues_per_pool << 29);
3072 
3073 	for (p = 0; p < adapter->num_rx_pools; p++)
3074 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3075 				psrtype);
3076 }
3077 
ixgbe_configure_virtualization(struct ixgbe_adapter * adapter)3078 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3079 {
3080 	struct ixgbe_hw *hw = &adapter->hw;
3081 	u32 gcr_ext;
3082 	u32 vt_reg_bits;
3083 	u32 reg_offset, vf_shift;
3084 	u32 vmdctl;
3085 	int i;
3086 
3087 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3088 		return;
3089 
3090 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3091 	vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3092 	vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3093 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3094 
3095 	vf_shift = adapter->num_vfs % 32;
3096 	reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3097 
3098 	/* Enable only the PF's pool for Tx/Rx */
3099 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3100 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3101 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3102 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3103 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3104 
3105 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3106 	hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3107 
3108 	/*
3109 	 * Set up VF register offsets for selected VT Mode,
3110 	 * i.e. 32 or 64 VFs for SR-IOV
3111 	 */
3112 	gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3113 	gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3114 	gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3115 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3116 
3117 	/* enable Tx loopback for VF/PF communication */
3118 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3119 	/* Enable MAC Anti-Spoofing */
3120 	hw->mac.ops.set_mac_anti_spoofing(hw,
3121 					   (adapter->num_vfs != 0),
3122 					  adapter->num_vfs);
3123 	/* For VFs that have spoof checking turned off */
3124 	for (i = 0; i < adapter->num_vfs; i++) {
3125 		if (!adapter->vfinfo[i].spoofchk_enabled)
3126 			ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3127 	}
3128 }
3129 
ixgbe_set_rx_buffer_len(struct ixgbe_adapter * adapter)3130 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3131 {
3132 	struct ixgbe_hw *hw = &adapter->hw;
3133 	struct net_device *netdev = adapter->netdev;
3134 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3135 	struct ixgbe_ring *rx_ring;
3136 	int i;
3137 	u32 mhadd, hlreg0;
3138 
3139 #ifdef IXGBE_FCOE
3140 	/* adjust max frame to be able to do baby jumbo for FCoE */
3141 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3142 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3143 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3144 
3145 #endif /* IXGBE_FCOE */
3146 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3147 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3148 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
3149 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3150 
3151 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3152 	}
3153 
3154 	/* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3155 	max_frame += VLAN_HLEN;
3156 
3157 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3158 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3159 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3160 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3161 
3162 	/*
3163 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3164 	 * the Base and Length of the Rx Descriptor Ring
3165 	 */
3166 	for (i = 0; i < adapter->num_rx_queues; i++) {
3167 		rx_ring = adapter->rx_ring[i];
3168 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3169 			set_ring_rsc_enabled(rx_ring);
3170 		else
3171 			clear_ring_rsc_enabled(rx_ring);
3172 	}
3173 }
3174 
ixgbe_setup_rdrxctl(struct ixgbe_adapter * adapter)3175 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3176 {
3177 	struct ixgbe_hw *hw = &adapter->hw;
3178 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3179 
3180 	switch (hw->mac.type) {
3181 	case ixgbe_mac_82598EB:
3182 		/*
3183 		 * For VMDq support of different descriptor types or
3184 		 * buffer sizes through the use of multiple SRRCTL
3185 		 * registers, RDRXCTL.MVMEN must be set to 1
3186 		 *
3187 		 * also, the manual doesn't mention it clearly but DCA hints
3188 		 * will only use queue 0's tags unless this bit is set.  Side
3189 		 * effects of setting this bit are only that SRRCTL must be
3190 		 * fully programmed [0..15]
3191 		 */
3192 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3193 		break;
3194 	case ixgbe_mac_82599EB:
3195 	case ixgbe_mac_X540:
3196 		/* Disable RSC for ACK packets */
3197 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3198 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3199 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3200 		/* hardware requires some bits to be set by default */
3201 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3202 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3203 		break;
3204 	default:
3205 		/* We should do nothing since we don't know this hardware */
3206 		return;
3207 	}
3208 
3209 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3210 }
3211 
3212 /**
3213  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3214  * @adapter: board private structure
3215  *
3216  * Configure the Rx unit of the MAC after a reset.
3217  **/
ixgbe_configure_rx(struct ixgbe_adapter * adapter)3218 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3219 {
3220 	struct ixgbe_hw *hw = &adapter->hw;
3221 	int i;
3222 	u32 rxctrl;
3223 
3224 	/* disable receives while setting up the descriptors */
3225 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3226 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3227 
3228 	ixgbe_setup_psrtype(adapter);
3229 	ixgbe_setup_rdrxctl(adapter);
3230 
3231 	/* Program registers for the distribution of queues */
3232 	ixgbe_setup_mrqc(adapter);
3233 
3234 	ixgbe_set_uta(adapter);
3235 
3236 	/* set_rx_buffer_len must be called before ring initialization */
3237 	ixgbe_set_rx_buffer_len(adapter);
3238 
3239 	/*
3240 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
3241 	 * the Base and Length of the Rx Descriptor Ring
3242 	 */
3243 	for (i = 0; i < adapter->num_rx_queues; i++)
3244 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3245 
3246 	/* disable drop enable for 82598 parts */
3247 	if (hw->mac.type == ixgbe_mac_82598EB)
3248 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
3249 
3250 	/* enable all receives */
3251 	rxctrl |= IXGBE_RXCTRL_RXEN;
3252 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
3253 }
3254 
ixgbe_vlan_rx_add_vid(struct net_device * netdev,u16 vid)3255 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3256 {
3257 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3258 	struct ixgbe_hw *hw = &adapter->hw;
3259 	int pool_ndx = adapter->num_vfs;
3260 
3261 	/* add VID to filter table */
3262 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3263 	set_bit(vid, adapter->active_vlans);
3264 
3265 	return 0;
3266 }
3267 
ixgbe_vlan_rx_kill_vid(struct net_device * netdev,u16 vid)3268 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3269 {
3270 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3271 	struct ixgbe_hw *hw = &adapter->hw;
3272 	int pool_ndx = adapter->num_vfs;
3273 
3274 	/* remove VID from filter table */
3275 	hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3276 	clear_bit(vid, adapter->active_vlans);
3277 
3278 	return 0;
3279 }
3280 
3281 /**
3282  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3283  * @adapter: driver data
3284  */
ixgbe_vlan_filter_disable(struct ixgbe_adapter * adapter)3285 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3286 {
3287 	struct ixgbe_hw *hw = &adapter->hw;
3288 	u32 vlnctrl;
3289 
3290 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3291 	vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3292 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3293 }
3294 
3295 /**
3296  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3297  * @adapter: driver data
3298  */
ixgbe_vlan_filter_enable(struct ixgbe_adapter * adapter)3299 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3300 {
3301 	struct ixgbe_hw *hw = &adapter->hw;
3302 	u32 vlnctrl;
3303 
3304 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3305 	vlnctrl |= IXGBE_VLNCTRL_VFE;
3306 	vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3307 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3308 }
3309 
3310 /**
3311  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3312  * @adapter: driver data
3313  */
ixgbe_vlan_strip_disable(struct ixgbe_adapter * adapter)3314 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3315 {
3316 	struct ixgbe_hw *hw = &adapter->hw;
3317 	u32 vlnctrl;
3318 	int i, j;
3319 
3320 	switch (hw->mac.type) {
3321 	case ixgbe_mac_82598EB:
3322 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3323 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
3324 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3325 		break;
3326 	case ixgbe_mac_82599EB:
3327 	case ixgbe_mac_X540:
3328 		for (i = 0; i < adapter->num_rx_queues; i++) {
3329 			j = adapter->rx_ring[i]->reg_idx;
3330 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3331 			vlnctrl &= ~IXGBE_RXDCTL_VME;
3332 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3333 		}
3334 		break;
3335 	default:
3336 		break;
3337 	}
3338 }
3339 
3340 /**
3341  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3342  * @adapter: driver data
3343  */
ixgbe_vlan_strip_enable(struct ixgbe_adapter * adapter)3344 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3345 {
3346 	struct ixgbe_hw *hw = &adapter->hw;
3347 	u32 vlnctrl;
3348 	int i, j;
3349 
3350 	switch (hw->mac.type) {
3351 	case ixgbe_mac_82598EB:
3352 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3353 		vlnctrl |= IXGBE_VLNCTRL_VME;
3354 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3355 		break;
3356 	case ixgbe_mac_82599EB:
3357 	case ixgbe_mac_X540:
3358 		for (i = 0; i < adapter->num_rx_queues; i++) {
3359 			j = adapter->rx_ring[i]->reg_idx;
3360 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3361 			vlnctrl |= IXGBE_RXDCTL_VME;
3362 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3363 		}
3364 		break;
3365 	default:
3366 		break;
3367 	}
3368 }
3369 
ixgbe_restore_vlan(struct ixgbe_adapter * adapter)3370 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3371 {
3372 	u16 vid;
3373 
3374 	ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3375 
3376 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3377 		ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3378 }
3379 
3380 /**
3381  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3382  * @netdev: network interface device structure
3383  *
3384  * Writes unicast address list to the RAR table.
3385  * Returns: -ENOMEM on failure/insufficient address space
3386  *                0 on no addresses written
3387  *                X on writing X addresses to the RAR table
3388  **/
ixgbe_write_uc_addr_list(struct net_device * netdev)3389 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3390 {
3391 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3392 	struct ixgbe_hw *hw = &adapter->hw;
3393 	unsigned int vfn = adapter->num_vfs;
3394 	unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3395 	int count = 0;
3396 
3397 	/* return ENOMEM indicating insufficient memory for addresses */
3398 	if (netdev_uc_count(netdev) > rar_entries)
3399 		return -ENOMEM;
3400 
3401 	if (!netdev_uc_empty(netdev) && rar_entries) {
3402 		struct netdev_hw_addr *ha;
3403 		/* return error if we do not support writing to RAR table */
3404 		if (!hw->mac.ops.set_rar)
3405 			return -ENOMEM;
3406 
3407 		netdev_for_each_uc_addr(ha, netdev) {
3408 			if (!rar_entries)
3409 				break;
3410 			hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3411 					    vfn, IXGBE_RAH_AV);
3412 			count++;
3413 		}
3414 	}
3415 	/* write the addresses in reverse order to avoid write combining */
3416 	for (; rar_entries > 0 ; rar_entries--)
3417 		hw->mac.ops.clear_rar(hw, rar_entries);
3418 
3419 	return count;
3420 }
3421 
3422 /**
3423  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3424  * @netdev: network interface device structure
3425  *
3426  * The set_rx_method entry point is called whenever the unicast/multicast
3427  * address list or the network interface flags are updated.  This routine is
3428  * responsible for configuring the hardware for proper unicast, multicast and
3429  * promiscuous mode.
3430  **/
ixgbe_set_rx_mode(struct net_device * netdev)3431 void ixgbe_set_rx_mode(struct net_device *netdev)
3432 {
3433 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
3434 	struct ixgbe_hw *hw = &adapter->hw;
3435 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3436 	int count;
3437 
3438 	/* Check for Promiscuous and All Multicast modes */
3439 
3440 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3441 
3442 	/* set all bits that we expect to always be set */
3443 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3444 	fctrl |= IXGBE_FCTRL_BAM;
3445 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3446 	fctrl |= IXGBE_FCTRL_PMCF;
3447 
3448 	/* clear the bits we are changing the status of */
3449 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3450 
3451 	if (netdev->flags & IFF_PROMISC) {
3452 		hw->addr_ctrl.user_set_promisc = true;
3453 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3454 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3455 		/* don't hardware filter vlans in promisc mode */
3456 		ixgbe_vlan_filter_disable(adapter);
3457 	} else {
3458 		if (netdev->flags & IFF_ALLMULTI) {
3459 			fctrl |= IXGBE_FCTRL_MPE;
3460 			vmolr |= IXGBE_VMOLR_MPE;
3461 		} else {
3462 			/*
3463 			 * Write addresses to the MTA, if the attempt fails
3464 			 * then we should just turn on promiscuous mode so
3465 			 * that we can at least receive multicast traffic
3466 			 */
3467 			hw->mac.ops.update_mc_addr_list(hw, netdev);
3468 			vmolr |= IXGBE_VMOLR_ROMPE;
3469 		}
3470 		ixgbe_vlan_filter_enable(adapter);
3471 		hw->addr_ctrl.user_set_promisc = false;
3472 		/*
3473 		 * Write addresses to available RAR registers, if there is not
3474 		 * sufficient space to store all the addresses then enable
3475 		 * unicast promiscuous mode
3476 		 */
3477 		count = ixgbe_write_uc_addr_list(netdev);
3478 		if (count < 0) {
3479 			fctrl |= IXGBE_FCTRL_UPE;
3480 			vmolr |= IXGBE_VMOLR_ROPE;
3481 		}
3482 	}
3483 
3484 	if (adapter->num_vfs) {
3485 		ixgbe_restore_vf_multicasts(adapter);
3486 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3487 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3488 			   IXGBE_VMOLR_ROPE);
3489 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3490 	}
3491 
3492 	/* This is useful for sniffing bad packets. */
3493 	if (adapter->netdev->features & NETIF_F_RXALL) {
3494 		/* UPE and MPE will be handled by normal PROMISC logic
3495 		 * in e1000e_set_rx_mode */
3496 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3497 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3498 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3499 
3500 		fctrl &= ~(IXGBE_FCTRL_DPF);
3501 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
3502 	}
3503 
3504 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3505 
3506 	if (netdev->features & NETIF_F_HW_VLAN_RX)
3507 		ixgbe_vlan_strip_enable(adapter);
3508 	else
3509 		ixgbe_vlan_strip_disable(adapter);
3510 }
3511 
ixgbe_napi_enable_all(struct ixgbe_adapter * adapter)3512 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3513 {
3514 	int q_idx;
3515 	struct ixgbe_q_vector *q_vector;
3516 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3517 
3518 	/* legacy and MSI only use one vector */
3519 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3520 		q_vectors = 1;
3521 
3522 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3523 		q_vector = adapter->q_vector[q_idx];
3524 		napi_enable(&q_vector->napi);
3525 	}
3526 }
3527 
ixgbe_napi_disable_all(struct ixgbe_adapter * adapter)3528 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3529 {
3530 	int q_idx;
3531 	struct ixgbe_q_vector *q_vector;
3532 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3533 
3534 	/* legacy and MSI only use one vector */
3535 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3536 		q_vectors = 1;
3537 
3538 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3539 		q_vector = adapter->q_vector[q_idx];
3540 		napi_disable(&q_vector->napi);
3541 	}
3542 }
3543 
3544 #ifdef CONFIG_IXGBE_DCB
3545 /*
3546  * ixgbe_configure_dcb - Configure DCB hardware
3547  * @adapter: ixgbe adapter struct
3548  *
3549  * This is called by the driver on open to configure the DCB hardware.
3550  * This is also called by the gennetlink interface when reconfiguring
3551  * the DCB state.
3552  */
ixgbe_configure_dcb(struct ixgbe_adapter * adapter)3553 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3554 {
3555 	struct ixgbe_hw *hw = &adapter->hw;
3556 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3557 
3558 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3559 		if (hw->mac.type == ixgbe_mac_82598EB)
3560 			netif_set_gso_max_size(adapter->netdev, 65536);
3561 		return;
3562 	}
3563 
3564 	if (hw->mac.type == ixgbe_mac_82598EB)
3565 		netif_set_gso_max_size(adapter->netdev, 32768);
3566 
3567 
3568 	/* Enable VLAN tag insert/strip */
3569 	adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3570 
3571 	hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3572 
3573 #ifdef IXGBE_FCOE
3574 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3575 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3576 #endif
3577 
3578 	/* reconfigure the hardware */
3579 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3580 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3581 						DCB_TX_CONFIG);
3582 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3583 						DCB_RX_CONFIG);
3584 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3585 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3586 		ixgbe_dcb_hw_ets(&adapter->hw,
3587 				 adapter->ixgbe_ieee_ets,
3588 				 max_frame);
3589 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
3590 					adapter->ixgbe_ieee_pfc->pfc_en,
3591 					adapter->ixgbe_ieee_ets->prio_tc);
3592 	}
3593 
3594 	/* Enable RSS Hash per TC */
3595 	if (hw->mac.type != ixgbe_mac_82598EB) {
3596 		int i;
3597 		u32 reg = 0;
3598 
3599 		for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3600 			u8 msb = 0;
3601 			u8 cnt = adapter->netdev->tc_to_txq[i].count;
3602 
3603 			while (cnt >>= 1)
3604 				msb++;
3605 
3606 			reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3607 		}
3608 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3609 	}
3610 }
3611 #endif
3612 
3613 /* Additional bittime to account for IXGBE framing */
3614 #define IXGBE_ETH_FRAMING 20
3615 
3616 /*
3617  * ixgbe_hpbthresh - calculate high water mark for flow control
3618  *
3619  * @adapter: board private structure to calculate for
3620  * @pb - packet buffer to calculate
3621  */
ixgbe_hpbthresh(struct ixgbe_adapter * adapter,int pb)3622 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3623 {
3624 	struct ixgbe_hw *hw = &adapter->hw;
3625 	struct net_device *dev = adapter->netdev;
3626 	int link, tc, kb, marker;
3627 	u32 dv_id, rx_pba;
3628 
3629 	/* Calculate max LAN frame size */
3630 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3631 
3632 #ifdef IXGBE_FCOE
3633 	/* FCoE traffic class uses FCOE jumbo frames */
3634 	if (dev->features & NETIF_F_FCOE_MTU) {
3635 		int fcoe_pb = 0;
3636 
3637 #ifdef CONFIG_IXGBE_DCB
3638 		fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3639 
3640 #endif
3641 		if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3642 			tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3643 	}
3644 #endif
3645 
3646 	/* Calculate delay value for device */
3647 	switch (hw->mac.type) {
3648 	case ixgbe_mac_X540:
3649 		dv_id = IXGBE_DV_X540(link, tc);
3650 		break;
3651 	default:
3652 		dv_id = IXGBE_DV(link, tc);
3653 		break;
3654 	}
3655 
3656 	/* Loopback switch introduces additional latency */
3657 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3658 		dv_id += IXGBE_B2BT(tc);
3659 
3660 	/* Delay value is calculated in bit times convert to KB */
3661 	kb = IXGBE_BT2KB(dv_id);
3662 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3663 
3664 	marker = rx_pba - kb;
3665 
3666 	/* It is possible that the packet buffer is not large enough
3667 	 * to provide required headroom. In this case throw an error
3668 	 * to user and a do the best we can.
3669 	 */
3670 	if (marker < 0) {
3671 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
3672 			    "headroom to support flow control."
3673 			    "Decrease MTU or number of traffic classes\n", pb);
3674 		marker = tc + 1;
3675 	}
3676 
3677 	return marker;
3678 }
3679 
3680 /*
3681  * ixgbe_lpbthresh - calculate low water mark for for flow control
3682  *
3683  * @adapter: board private structure to calculate for
3684  * @pb - packet buffer to calculate
3685  */
ixgbe_lpbthresh(struct ixgbe_adapter * adapter)3686 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3687 {
3688 	struct ixgbe_hw *hw = &adapter->hw;
3689 	struct net_device *dev = adapter->netdev;
3690 	int tc;
3691 	u32 dv_id;
3692 
3693 	/* Calculate max LAN frame size */
3694 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3695 
3696 	/* Calculate delay value for device */
3697 	switch (hw->mac.type) {
3698 	case ixgbe_mac_X540:
3699 		dv_id = IXGBE_LOW_DV_X540(tc);
3700 		break;
3701 	default:
3702 		dv_id = IXGBE_LOW_DV(tc);
3703 		break;
3704 	}
3705 
3706 	/* Delay value is calculated in bit times convert to KB */
3707 	return IXGBE_BT2KB(dv_id);
3708 }
3709 
3710 /*
3711  * ixgbe_pbthresh_setup - calculate and setup high low water marks
3712  */
ixgbe_pbthresh_setup(struct ixgbe_adapter * adapter)3713 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3714 {
3715 	struct ixgbe_hw *hw = &adapter->hw;
3716 	int num_tc = netdev_get_num_tc(adapter->netdev);
3717 	int i;
3718 
3719 	if (!num_tc)
3720 		num_tc = 1;
3721 
3722 	hw->fc.low_water = ixgbe_lpbthresh(adapter);
3723 
3724 	for (i = 0; i < num_tc; i++) {
3725 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3726 
3727 		/* Low water marks must not be larger than high water marks */
3728 		if (hw->fc.low_water > hw->fc.high_water[i])
3729 			hw->fc.low_water = 0;
3730 	}
3731 }
3732 
ixgbe_configure_pb(struct ixgbe_adapter * adapter)3733 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3734 {
3735 	struct ixgbe_hw *hw = &adapter->hw;
3736 	int hdrm;
3737 	u8 tc = netdev_get_num_tc(adapter->netdev);
3738 
3739 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3740 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3741 		hdrm = 32 << adapter->fdir_pballoc;
3742 	else
3743 		hdrm = 0;
3744 
3745 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3746 	ixgbe_pbthresh_setup(adapter);
3747 }
3748 
ixgbe_fdir_filter_restore(struct ixgbe_adapter * adapter)3749 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3750 {
3751 	struct ixgbe_hw *hw = &adapter->hw;
3752 	struct hlist_node *node, *node2;
3753 	struct ixgbe_fdir_filter *filter;
3754 
3755 	spin_lock(&adapter->fdir_perfect_lock);
3756 
3757 	if (!hlist_empty(&adapter->fdir_filter_list))
3758 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3759 
3760 	hlist_for_each_entry_safe(filter, node, node2,
3761 				  &adapter->fdir_filter_list, fdir_node) {
3762 		ixgbe_fdir_write_perfect_filter_82599(hw,
3763 				&filter->filter,
3764 				filter->sw_idx,
3765 				(filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3766 				IXGBE_FDIR_DROP_QUEUE :
3767 				adapter->rx_ring[filter->action]->reg_idx);
3768 	}
3769 
3770 	spin_unlock(&adapter->fdir_perfect_lock);
3771 }
3772 
ixgbe_configure(struct ixgbe_adapter * adapter)3773 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3774 {
3775 	struct ixgbe_hw *hw = &adapter->hw;
3776 
3777 	ixgbe_configure_pb(adapter);
3778 #ifdef CONFIG_IXGBE_DCB
3779 	ixgbe_configure_dcb(adapter);
3780 #endif
3781 
3782 	ixgbe_set_rx_mode(adapter->netdev);
3783 	ixgbe_restore_vlan(adapter);
3784 
3785 #ifdef IXGBE_FCOE
3786 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3787 		ixgbe_configure_fcoe(adapter);
3788 
3789 #endif /* IXGBE_FCOE */
3790 
3791 	switch (hw->mac.type) {
3792 	case ixgbe_mac_82599EB:
3793 	case ixgbe_mac_X540:
3794 		hw->mac.ops.disable_rx_buff(hw);
3795 		break;
3796 	default:
3797 		break;
3798 	}
3799 
3800 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3801 		ixgbe_init_fdir_signature_82599(&adapter->hw,
3802 						adapter->fdir_pballoc);
3803 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3804 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
3805 					      adapter->fdir_pballoc);
3806 		ixgbe_fdir_filter_restore(adapter);
3807 	}
3808 
3809 	switch (hw->mac.type) {
3810 	case ixgbe_mac_82599EB:
3811 	case ixgbe_mac_X540:
3812 		hw->mac.ops.enable_rx_buff(hw);
3813 		break;
3814 	default:
3815 		break;
3816 	}
3817 
3818 	ixgbe_configure_virtualization(adapter);
3819 
3820 	ixgbe_configure_tx(adapter);
3821 	ixgbe_configure_rx(adapter);
3822 }
3823 
ixgbe_is_sfp(struct ixgbe_hw * hw)3824 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3825 {
3826 	switch (hw->phy.type) {
3827 	case ixgbe_phy_sfp_avago:
3828 	case ixgbe_phy_sfp_ftl:
3829 	case ixgbe_phy_sfp_intel:
3830 	case ixgbe_phy_sfp_unknown:
3831 	case ixgbe_phy_sfp_passive_tyco:
3832 	case ixgbe_phy_sfp_passive_unknown:
3833 	case ixgbe_phy_sfp_active_unknown:
3834 	case ixgbe_phy_sfp_ftl_active:
3835 		return true;
3836 	case ixgbe_phy_nl:
3837 		if (hw->mac.type == ixgbe_mac_82598EB)
3838 			return true;
3839 	default:
3840 		return false;
3841 	}
3842 }
3843 
3844 /**
3845  * ixgbe_sfp_link_config - set up SFP+ link
3846  * @adapter: pointer to private adapter struct
3847  **/
ixgbe_sfp_link_config(struct ixgbe_adapter * adapter)3848 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3849 {
3850 	/*
3851 	 * We are assuming the worst case scenario here, and that
3852 	 * is that an SFP was inserted/removed after the reset
3853 	 * but before SFP detection was enabled.  As such the best
3854 	 * solution is to just start searching as soon as we start
3855 	 */
3856 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3857 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3858 
3859 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3860 }
3861 
3862 /**
3863  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3864  * @hw: pointer to private hardware struct
3865  *
3866  * Returns 0 on success, negative on failure
3867  **/
ixgbe_non_sfp_link_config(struct ixgbe_hw * hw)3868 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3869 {
3870 	u32 autoneg;
3871 	bool negotiation, link_up = false;
3872 	u32 ret = IXGBE_ERR_LINK_SETUP;
3873 
3874 	if (hw->mac.ops.check_link)
3875 		ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3876 
3877 	if (ret)
3878 		goto link_cfg_out;
3879 
3880 	autoneg = hw->phy.autoneg_advertised;
3881 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3882 		ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3883 							&negotiation);
3884 	if (ret)
3885 		goto link_cfg_out;
3886 
3887 	if (hw->mac.ops.setup_link)
3888 		ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3889 link_cfg_out:
3890 	return ret;
3891 }
3892 
ixgbe_setup_gpie(struct ixgbe_adapter * adapter)3893 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3894 {
3895 	struct ixgbe_hw *hw = &adapter->hw;
3896 	u32 gpie = 0;
3897 
3898 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3899 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3900 		       IXGBE_GPIE_OCD;
3901 		gpie |= IXGBE_GPIE_EIAME;
3902 		/*
3903 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
3904 		 * this saves a register write for every interrupt
3905 		 */
3906 		switch (hw->mac.type) {
3907 		case ixgbe_mac_82598EB:
3908 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3909 			break;
3910 		case ixgbe_mac_82599EB:
3911 		case ixgbe_mac_X540:
3912 		default:
3913 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3914 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3915 			break;
3916 		}
3917 	} else {
3918 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
3919 		 * specifically only auto mask tx and rx interrupts */
3920 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3921 	}
3922 
3923 	/* XXX: to interrupt immediately for EICS writes, enable this */
3924 	/* gpie |= IXGBE_GPIE_EIMEN; */
3925 
3926 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3927 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3928 		gpie |= IXGBE_GPIE_VTMODE_64;
3929 	}
3930 
3931 	/* Enable Thermal over heat sensor interrupt */
3932 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3933 		switch (adapter->hw.mac.type) {
3934 		case ixgbe_mac_82599EB:
3935 			gpie |= IXGBE_SDP0_GPIEN;
3936 			break;
3937 		case ixgbe_mac_X540:
3938 			gpie |= IXGBE_EIMS_TS;
3939 			break;
3940 		default:
3941 			break;
3942 		}
3943 	}
3944 
3945 	/* Enable fan failure interrupt */
3946 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3947 		gpie |= IXGBE_SDP1_GPIEN;
3948 
3949 	if (hw->mac.type == ixgbe_mac_82599EB) {
3950 		gpie |= IXGBE_SDP1_GPIEN;
3951 		gpie |= IXGBE_SDP2_GPIEN;
3952 	}
3953 
3954 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3955 }
3956 
ixgbe_up_complete(struct ixgbe_adapter * adapter)3957 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3958 {
3959 	struct ixgbe_hw *hw = &adapter->hw;
3960 	int err;
3961 	u32 ctrl_ext;
3962 
3963 	ixgbe_get_hw_control(adapter);
3964 	ixgbe_setup_gpie(adapter);
3965 
3966 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3967 		ixgbe_configure_msix(adapter);
3968 	else
3969 		ixgbe_configure_msi_and_legacy(adapter);
3970 
3971 	/* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3972 	if (hw->mac.ops.enable_tx_laser &&
3973 	    ((hw->phy.multispeed_fiber) ||
3974 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3975 	      (hw->mac.type == ixgbe_mac_82599EB))))
3976 		hw->mac.ops.enable_tx_laser(hw);
3977 
3978 	clear_bit(__IXGBE_DOWN, &adapter->state);
3979 	ixgbe_napi_enable_all(adapter);
3980 
3981 	if (ixgbe_is_sfp(hw)) {
3982 		ixgbe_sfp_link_config(adapter);
3983 	} else {
3984 		err = ixgbe_non_sfp_link_config(hw);
3985 		if (err)
3986 			e_err(probe, "link_config FAILED %d\n", err);
3987 	}
3988 
3989 	/* clear any pending interrupts, may auto mask */
3990 	IXGBE_READ_REG(hw, IXGBE_EICR);
3991 	ixgbe_irq_enable(adapter, true, true);
3992 
3993 	/*
3994 	 * If this adapter has a fan, check to see if we had a failure
3995 	 * before we enabled the interrupt.
3996 	 */
3997 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3998 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3999 		if (esdp & IXGBE_ESDP_SDP1)
4000 			e_crit(drv, "Fan has stopped, replace the adapter\n");
4001 	}
4002 
4003 	/* enable transmits */
4004 	netif_tx_start_all_queues(adapter->netdev);
4005 
4006 	/* bring the link up in the watchdog, this could race with our first
4007 	 * link up interrupt but shouldn't be a problem */
4008 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4009 	adapter->link_check_timeout = jiffies;
4010 	mod_timer(&adapter->service_timer, jiffies);
4011 
4012 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
4013 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4014 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4015 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4016 }
4017 
ixgbe_reinit_locked(struct ixgbe_adapter * adapter)4018 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4019 {
4020 	WARN_ON(in_interrupt());
4021 	/* put off any impending NetWatchDogTimeout */
4022 	adapter->netdev->trans_start = jiffies;
4023 
4024 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4025 		usleep_range(1000, 2000);
4026 	ixgbe_down(adapter);
4027 	/*
4028 	 * If SR-IOV enabled then wait a bit before bringing the adapter
4029 	 * back up to give the VFs time to respond to the reset.  The
4030 	 * two second wait is based upon the watchdog timer cycle in
4031 	 * the VF driver.
4032 	 */
4033 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4034 		msleep(2000);
4035 	ixgbe_up(adapter);
4036 	clear_bit(__IXGBE_RESETTING, &adapter->state);
4037 }
4038 
ixgbe_up(struct ixgbe_adapter * adapter)4039 void ixgbe_up(struct ixgbe_adapter *adapter)
4040 {
4041 	/* hardware has been reset, we need to reload some things */
4042 	ixgbe_configure(adapter);
4043 
4044 	ixgbe_up_complete(adapter);
4045 }
4046 
ixgbe_reset(struct ixgbe_adapter * adapter)4047 void ixgbe_reset(struct ixgbe_adapter *adapter)
4048 {
4049 	struct ixgbe_hw *hw = &adapter->hw;
4050 	int err;
4051 
4052 	/* lock SFP init bit to prevent race conditions with the watchdog */
4053 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4054 		usleep_range(1000, 2000);
4055 
4056 	/* clear all SFP and link config related flags while holding SFP_INIT */
4057 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4058 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
4059 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4060 
4061 	err = hw->mac.ops.init_hw(hw);
4062 	switch (err) {
4063 	case 0:
4064 	case IXGBE_ERR_SFP_NOT_PRESENT:
4065 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
4066 		break;
4067 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4068 		e_dev_err("master disable timed out\n");
4069 		break;
4070 	case IXGBE_ERR_EEPROM_VERSION:
4071 		/* We are running on a pre-production device, log a warning */
4072 		e_dev_warn("This device is a pre-production adapter/LOM. "
4073 			   "Please be aware there may be issues associated with "
4074 			   "your hardware.  If you are experiencing problems "
4075 			   "please contact your Intel or hardware "
4076 			   "representative who provided you with this "
4077 			   "hardware.\n");
4078 		break;
4079 	default:
4080 		e_dev_err("Hardware Error: %d\n", err);
4081 	}
4082 
4083 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4084 
4085 	/* reprogram the RAR[0] in case user changed it. */
4086 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4087 			    IXGBE_RAH_AV);
4088 }
4089 
4090 /**
4091  * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4092  * @rx_ring: ring to setup
4093  *
4094  * On many IA platforms the L1 cache has a critical stride of 4K, this
4095  * results in each receive buffer starting in the same cache set.  To help
4096  * reduce the pressure on this cache set we can interleave the offsets so
4097  * that only every other buffer will be in the same cache set.
4098  **/
ixgbe_init_rx_page_offset(struct ixgbe_ring * rx_ring)4099 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4100 {
4101 	struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4102 	u16 i;
4103 
4104 	for (i = 0; i < rx_ring->count; i += 2) {
4105 		rx_buffer[0].page_offset = 0;
4106 		rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4107 		rx_buffer = &rx_buffer[2];
4108 	}
4109 }
4110 
4111 /**
4112  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4113  * @rx_ring: ring to free buffers from
4114  **/
ixgbe_clean_rx_ring(struct ixgbe_ring * rx_ring)4115 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4116 {
4117 	struct device *dev = rx_ring->dev;
4118 	unsigned long size;
4119 	u16 i;
4120 
4121 	/* ring already cleared, nothing to do */
4122 	if (!rx_ring->rx_buffer_info)
4123 		return;
4124 
4125 	/* Free all the Rx ring sk_buffs */
4126 	for (i = 0; i < rx_ring->count; i++) {
4127 		struct ixgbe_rx_buffer *rx_buffer;
4128 
4129 		rx_buffer = &rx_ring->rx_buffer_info[i];
4130 		if (rx_buffer->skb) {
4131 			struct sk_buff *skb = rx_buffer->skb;
4132 			if (IXGBE_CB(skb)->page_released) {
4133 				dma_unmap_page(dev,
4134 					       IXGBE_CB(skb)->dma,
4135 					       ixgbe_rx_bufsz(rx_ring),
4136 					       DMA_FROM_DEVICE);
4137 				IXGBE_CB(skb)->page_released = false;
4138 			}
4139 			dev_kfree_skb(skb);
4140 		}
4141 		rx_buffer->skb = NULL;
4142 		if (rx_buffer->dma)
4143 			dma_unmap_page(dev, rx_buffer->dma,
4144 				       ixgbe_rx_pg_size(rx_ring),
4145 				       DMA_FROM_DEVICE);
4146 		rx_buffer->dma = 0;
4147 		if (rx_buffer->page)
4148 			put_page(rx_buffer->page);
4149 		rx_buffer->page = NULL;
4150 	}
4151 
4152 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4153 	memset(rx_ring->rx_buffer_info, 0, size);
4154 
4155 	ixgbe_init_rx_page_offset(rx_ring);
4156 
4157 	/* Zero out the descriptor ring */
4158 	memset(rx_ring->desc, 0, rx_ring->size);
4159 
4160 	rx_ring->next_to_alloc = 0;
4161 	rx_ring->next_to_clean = 0;
4162 	rx_ring->next_to_use = 0;
4163 }
4164 
4165 /**
4166  * ixgbe_clean_tx_ring - Free Tx Buffers
4167  * @tx_ring: ring to be cleaned
4168  **/
ixgbe_clean_tx_ring(struct ixgbe_ring * tx_ring)4169 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4170 {
4171 	struct ixgbe_tx_buffer *tx_buffer_info;
4172 	unsigned long size;
4173 	u16 i;
4174 
4175 	/* ring already cleared, nothing to do */
4176 	if (!tx_ring->tx_buffer_info)
4177 		return;
4178 
4179 	/* Free all the Tx ring sk_buffs */
4180 	for (i = 0; i < tx_ring->count; i++) {
4181 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
4182 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4183 	}
4184 
4185 	netdev_tx_reset_queue(txring_txq(tx_ring));
4186 
4187 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4188 	memset(tx_ring->tx_buffer_info, 0, size);
4189 
4190 	/* Zero out the descriptor ring */
4191 	memset(tx_ring->desc, 0, tx_ring->size);
4192 
4193 	tx_ring->next_to_use = 0;
4194 	tx_ring->next_to_clean = 0;
4195 }
4196 
4197 /**
4198  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4199  * @adapter: board private structure
4200  **/
ixgbe_clean_all_rx_rings(struct ixgbe_adapter * adapter)4201 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4202 {
4203 	int i;
4204 
4205 	for (i = 0; i < adapter->num_rx_queues; i++)
4206 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4207 }
4208 
4209 /**
4210  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4211  * @adapter: board private structure
4212  **/
ixgbe_clean_all_tx_rings(struct ixgbe_adapter * adapter)4213 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4214 {
4215 	int i;
4216 
4217 	for (i = 0; i < adapter->num_tx_queues; i++)
4218 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4219 }
4220 
ixgbe_fdir_filter_exit(struct ixgbe_adapter * adapter)4221 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4222 {
4223 	struct hlist_node *node, *node2;
4224 	struct ixgbe_fdir_filter *filter;
4225 
4226 	spin_lock(&adapter->fdir_perfect_lock);
4227 
4228 	hlist_for_each_entry_safe(filter, node, node2,
4229 				  &adapter->fdir_filter_list, fdir_node) {
4230 		hlist_del(&filter->fdir_node);
4231 		kfree(filter);
4232 	}
4233 	adapter->fdir_filter_count = 0;
4234 
4235 	spin_unlock(&adapter->fdir_perfect_lock);
4236 }
4237 
ixgbe_down(struct ixgbe_adapter * adapter)4238 void ixgbe_down(struct ixgbe_adapter *adapter)
4239 {
4240 	struct net_device *netdev = adapter->netdev;
4241 	struct ixgbe_hw *hw = &adapter->hw;
4242 	u32 rxctrl;
4243 	int i;
4244 
4245 	/* signal that we are down to the interrupt handler */
4246 	set_bit(__IXGBE_DOWN, &adapter->state);
4247 
4248 	/* disable receives */
4249 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4250 	IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4251 
4252 	/* disable all enabled rx queues */
4253 	for (i = 0; i < adapter->num_rx_queues; i++)
4254 		/* this call also flushes the previous write */
4255 		ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4256 
4257 	usleep_range(10000, 20000);
4258 
4259 	netif_tx_stop_all_queues(netdev);
4260 
4261 	/* call carrier off first to avoid false dev_watchdog timeouts */
4262 	netif_carrier_off(netdev);
4263 	netif_tx_disable(netdev);
4264 
4265 	ixgbe_irq_disable(adapter);
4266 
4267 	ixgbe_napi_disable_all(adapter);
4268 
4269 	adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4270 			     IXGBE_FLAG2_RESET_REQUESTED);
4271 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4272 
4273 	del_timer_sync(&adapter->service_timer);
4274 
4275 	if (adapter->num_vfs) {
4276 		/* Clear EITR Select mapping */
4277 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4278 
4279 		/* Mark all the VFs as inactive */
4280 		for (i = 0 ; i < adapter->num_vfs; i++)
4281 			adapter->vfinfo[i].clear_to_send = false;
4282 
4283 		/* ping all the active vfs to let them know we are going down */
4284 		ixgbe_ping_all_vfs(adapter);
4285 
4286 		/* Disable all VFTE/VFRE TX/RX */
4287 		ixgbe_disable_tx_rx(adapter);
4288 	}
4289 
4290 	/* disable transmits in the hardware now that interrupts are off */
4291 	for (i = 0; i < adapter->num_tx_queues; i++) {
4292 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4293 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4294 	}
4295 
4296 	/* Disable the Tx DMA engine on 82599 and X540 */
4297 	switch (hw->mac.type) {
4298 	case ixgbe_mac_82599EB:
4299 	case ixgbe_mac_X540:
4300 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4301 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4302 				 ~IXGBE_DMATXCTL_TE));
4303 		break;
4304 	default:
4305 		break;
4306 	}
4307 
4308 	if (!pci_channel_offline(adapter->pdev))
4309 		ixgbe_reset(adapter);
4310 
4311 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4312 	if (hw->mac.ops.disable_tx_laser &&
4313 	    ((hw->phy.multispeed_fiber) ||
4314 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4315 	      (hw->mac.type == ixgbe_mac_82599EB))))
4316 		hw->mac.ops.disable_tx_laser(hw);
4317 
4318 	ixgbe_clean_all_tx_rings(adapter);
4319 	ixgbe_clean_all_rx_rings(adapter);
4320 
4321 #ifdef CONFIG_IXGBE_DCA
4322 	/* since we reset the hardware DCA settings were cleared */
4323 	ixgbe_setup_dca(adapter);
4324 #endif
4325 }
4326 
4327 /**
4328  * ixgbe_tx_timeout - Respond to a Tx Hang
4329  * @netdev: network interface device structure
4330  **/
ixgbe_tx_timeout(struct net_device * netdev)4331 static void ixgbe_tx_timeout(struct net_device *netdev)
4332 {
4333 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4334 
4335 	/* Do the reset outside of interrupt context */
4336 	ixgbe_tx_timeout_reset(adapter);
4337 }
4338 
4339 /**
4340  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4341  * @adapter: board private structure to initialize
4342  *
4343  * ixgbe_sw_init initializes the Adapter private data structure.
4344  * Fields are initialized based on PCI device information and
4345  * OS network device settings (MTU size).
4346  **/
ixgbe_sw_init(struct ixgbe_adapter * adapter)4347 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4348 {
4349 	struct ixgbe_hw *hw = &adapter->hw;
4350 	struct pci_dev *pdev = adapter->pdev;
4351 	unsigned int rss;
4352 #ifdef CONFIG_IXGBE_DCB
4353 	int j;
4354 	struct tc_configuration *tc;
4355 #endif
4356 
4357 	/* PCI config space info */
4358 
4359 	hw->vendor_id = pdev->vendor;
4360 	hw->device_id = pdev->device;
4361 	hw->revision_id = pdev->revision;
4362 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4363 	hw->subsystem_device_id = pdev->subsystem_device;
4364 
4365 	/* Set capability flags */
4366 	rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4367 	adapter->ring_feature[RING_F_RSS].indices = rss;
4368 	adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4369 	switch (hw->mac.type) {
4370 	case ixgbe_mac_82598EB:
4371 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
4372 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4373 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4374 		break;
4375 	case ixgbe_mac_X540:
4376 		adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4377 	case ixgbe_mac_82599EB:
4378 		adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4379 		adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4380 		adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4381 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4382 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4383 		/* Flow Director hash filters enabled */
4384 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4385 		adapter->atr_sample_rate = 20;
4386 		adapter->ring_feature[RING_F_FDIR].indices =
4387 							 IXGBE_MAX_FDIR_INDICES;
4388 		adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4389 #ifdef IXGBE_FCOE
4390 		adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4391 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4392 		adapter->ring_feature[RING_F_FCOE].indices = 0;
4393 #ifdef CONFIG_IXGBE_DCB
4394 		/* Default traffic class to use for FCoE */
4395 		adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4396 #endif
4397 #endif /* IXGBE_FCOE */
4398 		break;
4399 	default:
4400 		break;
4401 	}
4402 
4403 	/* n-tuple support exists, always init our spinlock */
4404 	spin_lock_init(&adapter->fdir_perfect_lock);
4405 
4406 #ifdef CONFIG_IXGBE_DCB
4407 	switch (hw->mac.type) {
4408 	case ixgbe_mac_X540:
4409 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4410 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4411 		break;
4412 	default:
4413 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4414 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4415 		break;
4416 	}
4417 
4418 	/* Configure DCB traffic classes */
4419 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4420 		tc = &adapter->dcb_cfg.tc_config[j];
4421 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
4422 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4423 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
4424 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4425 		tc->dcb_pfc = pfc_disabled;
4426 	}
4427 
4428 	/* Initialize default user to priority mapping, UPx->TC0 */
4429 	tc = &adapter->dcb_cfg.tc_config[0];
4430 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4431 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4432 
4433 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4434 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4435 	adapter->dcb_cfg.pfc_mode_enable = false;
4436 	adapter->dcb_set_bitmap = 0x00;
4437 	adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4438 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4439 	       sizeof(adapter->temp_dcb_cfg));
4440 
4441 #endif
4442 
4443 	/* default flow control settings */
4444 	hw->fc.requested_mode = ixgbe_fc_full;
4445 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
4446 #ifdef CONFIG_DCB
4447 	adapter->last_lfc_mode = hw->fc.current_mode;
4448 #endif
4449 	ixgbe_pbthresh_setup(adapter);
4450 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4451 	hw->fc.send_xon = true;
4452 	hw->fc.disable_fc_autoneg = false;
4453 
4454 	/* enable itr by default in dynamic mode */
4455 	adapter->rx_itr_setting = 1;
4456 	adapter->tx_itr_setting = 1;
4457 
4458 	/* set default ring sizes */
4459 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4460 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4461 
4462 	/* set default work limits */
4463 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4464 
4465 	/* initialize eeprom parameters */
4466 	if (ixgbe_init_eeprom_params_generic(hw)) {
4467 		e_dev_err("EEPROM initialization failed\n");
4468 		return -EIO;
4469 	}
4470 
4471 	set_bit(__IXGBE_DOWN, &adapter->state);
4472 
4473 	return 0;
4474 }
4475 
4476 /**
4477  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4478  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4479  *
4480  * Return 0 on success, negative on failure
4481  **/
ixgbe_setup_tx_resources(struct ixgbe_ring * tx_ring)4482 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4483 {
4484 	struct device *dev = tx_ring->dev;
4485 	int orig_node = dev_to_node(dev);
4486 	int numa_node = -1;
4487 	int size;
4488 
4489 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4490 
4491 	if (tx_ring->q_vector)
4492 		numa_node = tx_ring->q_vector->numa_node;
4493 
4494 	tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4495 	if (!tx_ring->tx_buffer_info)
4496 		tx_ring->tx_buffer_info = vzalloc(size);
4497 	if (!tx_ring->tx_buffer_info)
4498 		goto err;
4499 
4500 	/* round up to nearest 4K */
4501 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4502 	tx_ring->size = ALIGN(tx_ring->size, 4096);
4503 
4504 	set_dev_node(dev, numa_node);
4505 	tx_ring->desc = dma_alloc_coherent(dev,
4506 					   tx_ring->size,
4507 					   &tx_ring->dma,
4508 					   GFP_KERNEL);
4509 	set_dev_node(dev, orig_node);
4510 	if (!tx_ring->desc)
4511 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4512 						   &tx_ring->dma, GFP_KERNEL);
4513 	if (!tx_ring->desc)
4514 		goto err;
4515 
4516 	tx_ring->next_to_use = 0;
4517 	tx_ring->next_to_clean = 0;
4518 	return 0;
4519 
4520 err:
4521 	vfree(tx_ring->tx_buffer_info);
4522 	tx_ring->tx_buffer_info = NULL;
4523 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4524 	return -ENOMEM;
4525 }
4526 
4527 /**
4528  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4529  * @adapter: board private structure
4530  *
4531  * If this function returns with an error, then it's possible one or
4532  * more of the rings is populated (while the rest are not).  It is the
4533  * callers duty to clean those orphaned rings.
4534  *
4535  * Return 0 on success, negative on failure
4536  **/
ixgbe_setup_all_tx_resources(struct ixgbe_adapter * adapter)4537 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4538 {
4539 	int i, err = 0;
4540 
4541 	for (i = 0; i < adapter->num_tx_queues; i++) {
4542 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4543 		if (!err)
4544 			continue;
4545 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4546 		break;
4547 	}
4548 
4549 	return err;
4550 }
4551 
4552 /**
4553  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4554  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4555  *
4556  * Returns 0 on success, negative on failure
4557  **/
ixgbe_setup_rx_resources(struct ixgbe_ring * rx_ring)4558 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4559 {
4560 	struct device *dev = rx_ring->dev;
4561 	int orig_node = dev_to_node(dev);
4562 	int numa_node = -1;
4563 	int size;
4564 
4565 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4566 
4567 	if (rx_ring->q_vector)
4568 		numa_node = rx_ring->q_vector->numa_node;
4569 
4570 	rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4571 	if (!rx_ring->rx_buffer_info)
4572 		rx_ring->rx_buffer_info = vzalloc(size);
4573 	if (!rx_ring->rx_buffer_info)
4574 		goto err;
4575 
4576 	/* Round up to nearest 4K */
4577 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4578 	rx_ring->size = ALIGN(rx_ring->size, 4096);
4579 
4580 	set_dev_node(dev, numa_node);
4581 	rx_ring->desc = dma_alloc_coherent(dev,
4582 					   rx_ring->size,
4583 					   &rx_ring->dma,
4584 					   GFP_KERNEL);
4585 	set_dev_node(dev, orig_node);
4586 	if (!rx_ring->desc)
4587 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4588 						   &rx_ring->dma, GFP_KERNEL);
4589 	if (!rx_ring->desc)
4590 		goto err;
4591 
4592 	rx_ring->next_to_clean = 0;
4593 	rx_ring->next_to_use = 0;
4594 
4595 	ixgbe_init_rx_page_offset(rx_ring);
4596 
4597 	return 0;
4598 err:
4599 	vfree(rx_ring->rx_buffer_info);
4600 	rx_ring->rx_buffer_info = NULL;
4601 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4602 	return -ENOMEM;
4603 }
4604 
4605 /**
4606  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4607  * @adapter: board private structure
4608  *
4609  * If this function returns with an error, then it's possible one or
4610  * more of the rings is populated (while the rest are not).  It is the
4611  * callers duty to clean those orphaned rings.
4612  *
4613  * Return 0 on success, negative on failure
4614  **/
ixgbe_setup_all_rx_resources(struct ixgbe_adapter * adapter)4615 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4616 {
4617 	int i, err = 0;
4618 
4619 	for (i = 0; i < adapter->num_rx_queues; i++) {
4620 		err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4621 		if (!err)
4622 			continue;
4623 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4624 		break;
4625 	}
4626 
4627 	return err;
4628 }
4629 
4630 /**
4631  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4632  * @tx_ring: Tx descriptor ring for a specific queue
4633  *
4634  * Free all transmit software resources
4635  **/
ixgbe_free_tx_resources(struct ixgbe_ring * tx_ring)4636 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4637 {
4638 	ixgbe_clean_tx_ring(tx_ring);
4639 
4640 	vfree(tx_ring->tx_buffer_info);
4641 	tx_ring->tx_buffer_info = NULL;
4642 
4643 	/* if not set, then don't free */
4644 	if (!tx_ring->desc)
4645 		return;
4646 
4647 	dma_free_coherent(tx_ring->dev, tx_ring->size,
4648 			  tx_ring->desc, tx_ring->dma);
4649 
4650 	tx_ring->desc = NULL;
4651 }
4652 
4653 /**
4654  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4655  * @adapter: board private structure
4656  *
4657  * Free all transmit software resources
4658  **/
ixgbe_free_all_tx_resources(struct ixgbe_adapter * adapter)4659 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4660 {
4661 	int i;
4662 
4663 	for (i = 0; i < adapter->num_tx_queues; i++)
4664 		if (adapter->tx_ring[i]->desc)
4665 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
4666 }
4667 
4668 /**
4669  * ixgbe_free_rx_resources - Free Rx Resources
4670  * @rx_ring: ring to clean the resources from
4671  *
4672  * Free all receive software resources
4673  **/
ixgbe_free_rx_resources(struct ixgbe_ring * rx_ring)4674 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4675 {
4676 	ixgbe_clean_rx_ring(rx_ring);
4677 
4678 	vfree(rx_ring->rx_buffer_info);
4679 	rx_ring->rx_buffer_info = NULL;
4680 
4681 	/* if not set, then don't free */
4682 	if (!rx_ring->desc)
4683 		return;
4684 
4685 	dma_free_coherent(rx_ring->dev, rx_ring->size,
4686 			  rx_ring->desc, rx_ring->dma);
4687 
4688 	rx_ring->desc = NULL;
4689 }
4690 
4691 /**
4692  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4693  * @adapter: board private structure
4694  *
4695  * Free all receive software resources
4696  **/
ixgbe_free_all_rx_resources(struct ixgbe_adapter * adapter)4697 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4698 {
4699 	int i;
4700 
4701 	for (i = 0; i < adapter->num_rx_queues; i++)
4702 		if (adapter->rx_ring[i]->desc)
4703 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
4704 }
4705 
4706 /**
4707  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4708  * @netdev: network interface device structure
4709  * @new_mtu: new value for maximum frame size
4710  *
4711  * Returns 0 on success, negative on failure
4712  **/
ixgbe_change_mtu(struct net_device * netdev,int new_mtu)4713 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4714 {
4715 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4716 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4717 
4718 	/* MTU < 68 is an error and causes problems on some kernels */
4719 	if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4720 		return -EINVAL;
4721 
4722 	/*
4723 	 * For 82599EB we cannot allow PF to change MTU greater than 1500
4724 	 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4725 	 * don't allocate and chain buffers correctly.
4726 	 */
4727 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4728 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4729 	    (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4730 			return -EINVAL;
4731 
4732 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4733 
4734 	/* must set new MTU before calling down or up */
4735 	netdev->mtu = new_mtu;
4736 
4737 	if (netif_running(netdev))
4738 		ixgbe_reinit_locked(adapter);
4739 
4740 	return 0;
4741 }
4742 
4743 /**
4744  * ixgbe_open - Called when a network interface is made active
4745  * @netdev: network interface device structure
4746  *
4747  * Returns 0 on success, negative value on failure
4748  *
4749  * The open entry point is called when a network interface is made
4750  * active by the system (IFF_UP).  At this point all resources needed
4751  * for transmit and receive operations are allocated, the interrupt
4752  * handler is registered with the OS, the watchdog timer is started,
4753  * and the stack is notified that the interface is ready.
4754  **/
ixgbe_open(struct net_device * netdev)4755 static int ixgbe_open(struct net_device *netdev)
4756 {
4757 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4758 	int err;
4759 
4760 	/* disallow open during test */
4761 	if (test_bit(__IXGBE_TESTING, &adapter->state))
4762 		return -EBUSY;
4763 
4764 	netif_carrier_off(netdev);
4765 
4766 	/* allocate transmit descriptors */
4767 	err = ixgbe_setup_all_tx_resources(adapter);
4768 	if (err)
4769 		goto err_setup_tx;
4770 
4771 	/* allocate receive descriptors */
4772 	err = ixgbe_setup_all_rx_resources(adapter);
4773 	if (err)
4774 		goto err_setup_rx;
4775 
4776 	ixgbe_configure(adapter);
4777 
4778 	err = ixgbe_request_irq(adapter);
4779 	if (err)
4780 		goto err_req_irq;
4781 
4782 	ixgbe_up_complete(adapter);
4783 
4784 	return 0;
4785 
4786 err_req_irq:
4787 err_setup_rx:
4788 	ixgbe_free_all_rx_resources(adapter);
4789 err_setup_tx:
4790 	ixgbe_free_all_tx_resources(adapter);
4791 	ixgbe_reset(adapter);
4792 
4793 	return err;
4794 }
4795 
4796 /**
4797  * ixgbe_close - Disables a network interface
4798  * @netdev: network interface device structure
4799  *
4800  * Returns 0, this is not allowed to fail
4801  *
4802  * The close entry point is called when an interface is de-activated
4803  * by the OS.  The hardware is still under the drivers control, but
4804  * needs to be disabled.  A global MAC reset is issued to stop the
4805  * hardware, and all transmit and receive resources are freed.
4806  **/
ixgbe_close(struct net_device * netdev)4807 static int ixgbe_close(struct net_device *netdev)
4808 {
4809 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4810 
4811 	ixgbe_down(adapter);
4812 	ixgbe_free_irq(adapter);
4813 
4814 	ixgbe_fdir_filter_exit(adapter);
4815 
4816 	ixgbe_free_all_tx_resources(adapter);
4817 	ixgbe_free_all_rx_resources(adapter);
4818 
4819 	ixgbe_release_hw_control(adapter);
4820 
4821 	return 0;
4822 }
4823 
4824 #ifdef CONFIG_PM
ixgbe_resume(struct pci_dev * pdev)4825 static int ixgbe_resume(struct pci_dev *pdev)
4826 {
4827 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4828 	struct net_device *netdev = adapter->netdev;
4829 	u32 err;
4830 
4831 	pci_set_power_state(pdev, PCI_D0);
4832 	pci_restore_state(pdev);
4833 	/*
4834 	 * pci_restore_state clears dev->state_saved so call
4835 	 * pci_save_state to restore it.
4836 	 */
4837 	pci_save_state(pdev);
4838 
4839 	err = pci_enable_device_mem(pdev);
4840 	if (err) {
4841 		e_dev_err("Cannot enable PCI device from suspend\n");
4842 		return err;
4843 	}
4844 	pci_set_master(pdev);
4845 
4846 	pci_wake_from_d3(pdev, false);
4847 
4848 	rtnl_lock();
4849 	err = ixgbe_init_interrupt_scheme(adapter);
4850 	rtnl_unlock();
4851 	if (err) {
4852 		e_dev_err("Cannot initialize interrupts for device\n");
4853 		return err;
4854 	}
4855 
4856 	ixgbe_reset(adapter);
4857 
4858 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4859 
4860 	if (netif_running(netdev)) {
4861 		err = ixgbe_open(netdev);
4862 		if (err)
4863 			return err;
4864 	}
4865 
4866 	netif_device_attach(netdev);
4867 
4868 	return 0;
4869 }
4870 #endif /* CONFIG_PM */
4871 
__ixgbe_shutdown(struct pci_dev * pdev,bool * enable_wake)4872 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4873 {
4874 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4875 	struct net_device *netdev = adapter->netdev;
4876 	struct ixgbe_hw *hw = &adapter->hw;
4877 	u32 ctrl, fctrl;
4878 	u32 wufc = adapter->wol;
4879 #ifdef CONFIG_PM
4880 	int retval = 0;
4881 #endif
4882 
4883 	netif_device_detach(netdev);
4884 
4885 	if (netif_running(netdev)) {
4886 		rtnl_lock();
4887 		ixgbe_down(adapter);
4888 		ixgbe_free_irq(adapter);
4889 		ixgbe_free_all_tx_resources(adapter);
4890 		ixgbe_free_all_rx_resources(adapter);
4891 		rtnl_unlock();
4892 	}
4893 
4894 	ixgbe_clear_interrupt_scheme(adapter);
4895 
4896 #ifdef CONFIG_PM
4897 	retval = pci_save_state(pdev);
4898 	if (retval)
4899 		return retval;
4900 
4901 #endif
4902 	if (wufc) {
4903 		ixgbe_set_rx_mode(netdev);
4904 
4905 		/*
4906 		 * enable the optics for both mult-speed fiber and
4907 		 * 82599 SFP+ fiber as we can WoL.
4908 		 */
4909 		if (hw->mac.ops.enable_tx_laser &&
4910 		    (hw->phy.multispeed_fiber ||
4911 		    (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4912 		     hw->mac.type == ixgbe_mac_82599EB)))
4913 			hw->mac.ops.enable_tx_laser(hw);
4914 
4915 		/* turn on all-multi mode if wake on multicast is enabled */
4916 		if (wufc & IXGBE_WUFC_MC) {
4917 			fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4918 			fctrl |= IXGBE_FCTRL_MPE;
4919 			IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4920 		}
4921 
4922 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4923 		ctrl |= IXGBE_CTRL_GIO_DIS;
4924 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4925 
4926 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4927 	} else {
4928 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4929 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4930 	}
4931 
4932 	switch (hw->mac.type) {
4933 	case ixgbe_mac_82598EB:
4934 		pci_wake_from_d3(pdev, false);
4935 		break;
4936 	case ixgbe_mac_82599EB:
4937 	case ixgbe_mac_X540:
4938 		pci_wake_from_d3(pdev, !!wufc);
4939 		break;
4940 	default:
4941 		break;
4942 	}
4943 
4944 	*enable_wake = !!wufc;
4945 
4946 	ixgbe_release_hw_control(adapter);
4947 
4948 	pci_disable_device(pdev);
4949 
4950 	return 0;
4951 }
4952 
4953 #ifdef CONFIG_PM
ixgbe_suspend(struct pci_dev * pdev,pm_message_t state)4954 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4955 {
4956 	int retval;
4957 	bool wake;
4958 
4959 	retval = __ixgbe_shutdown(pdev, &wake);
4960 	if (retval)
4961 		return retval;
4962 
4963 	if (wake) {
4964 		pci_prepare_to_sleep(pdev);
4965 	} else {
4966 		pci_wake_from_d3(pdev, false);
4967 		pci_set_power_state(pdev, PCI_D3hot);
4968 	}
4969 
4970 	return 0;
4971 }
4972 #endif /* CONFIG_PM */
4973 
ixgbe_shutdown(struct pci_dev * pdev)4974 static void ixgbe_shutdown(struct pci_dev *pdev)
4975 {
4976 	bool wake;
4977 
4978 	__ixgbe_shutdown(pdev, &wake);
4979 
4980 	if (system_state == SYSTEM_POWER_OFF) {
4981 		pci_wake_from_d3(pdev, wake);
4982 		pci_set_power_state(pdev, PCI_D3hot);
4983 	}
4984 }
4985 
4986 /**
4987  * ixgbe_update_stats - Update the board statistics counters.
4988  * @adapter: board private structure
4989  **/
ixgbe_update_stats(struct ixgbe_adapter * adapter)4990 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4991 {
4992 	struct net_device *netdev = adapter->netdev;
4993 	struct ixgbe_hw *hw = &adapter->hw;
4994 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
4995 	u64 total_mpc = 0;
4996 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4997 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
4998 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
4999 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5000 #ifdef IXGBE_FCOE
5001 	struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5002 	unsigned int cpu;
5003 	u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5004 #endif /* IXGBE_FCOE */
5005 
5006 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5007 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5008 		return;
5009 
5010 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5011 		u64 rsc_count = 0;
5012 		u64 rsc_flush = 0;
5013 		for (i = 0; i < 16; i++)
5014 			adapter->hw_rx_no_dma_resources +=
5015 				IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5016 		for (i = 0; i < adapter->num_rx_queues; i++) {
5017 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5018 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5019 		}
5020 		adapter->rsc_total_count = rsc_count;
5021 		adapter->rsc_total_flush = rsc_flush;
5022 	}
5023 
5024 	for (i = 0; i < adapter->num_rx_queues; i++) {
5025 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5026 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5027 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5028 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5029 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5030 		bytes += rx_ring->stats.bytes;
5031 		packets += rx_ring->stats.packets;
5032 	}
5033 	adapter->non_eop_descs = non_eop_descs;
5034 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5035 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5036 	adapter->hw_csum_rx_error = hw_csum_rx_error;
5037 	netdev->stats.rx_bytes = bytes;
5038 	netdev->stats.rx_packets = packets;
5039 
5040 	bytes = 0;
5041 	packets = 0;
5042 	/* gather some stats to the adapter struct that are per queue */
5043 	for (i = 0; i < adapter->num_tx_queues; i++) {
5044 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5045 		restart_queue += tx_ring->tx_stats.restart_queue;
5046 		tx_busy += tx_ring->tx_stats.tx_busy;
5047 		bytes += tx_ring->stats.bytes;
5048 		packets += tx_ring->stats.packets;
5049 	}
5050 	adapter->restart_queue = restart_queue;
5051 	adapter->tx_busy = tx_busy;
5052 	netdev->stats.tx_bytes = bytes;
5053 	netdev->stats.tx_packets = packets;
5054 
5055 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5056 
5057 	/* 8 register reads */
5058 	for (i = 0; i < 8; i++) {
5059 		/* for packet buffers not used, the register should read 0 */
5060 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5061 		missed_rx += mpc;
5062 		hwstats->mpc[i] += mpc;
5063 		total_mpc += hwstats->mpc[i];
5064 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5065 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5066 		switch (hw->mac.type) {
5067 		case ixgbe_mac_82598EB:
5068 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5069 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5070 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5071 			hwstats->pxonrxc[i] +=
5072 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5073 			break;
5074 		case ixgbe_mac_82599EB:
5075 		case ixgbe_mac_X540:
5076 			hwstats->pxonrxc[i] +=
5077 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5078 			break;
5079 		default:
5080 			break;
5081 		}
5082 	}
5083 
5084 	/*16 register reads */
5085 	for (i = 0; i < 16; i++) {
5086 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5087 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5088 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
5089 		    (hw->mac.type == ixgbe_mac_X540)) {
5090 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5091 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5092 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5093 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5094 		}
5095 	}
5096 
5097 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5098 	/* work around hardware counting issue */
5099 	hwstats->gprc -= missed_rx;
5100 
5101 	ixgbe_update_xoff_received(adapter);
5102 
5103 	/* 82598 hardware only has a 32 bit counter in the high register */
5104 	switch (hw->mac.type) {
5105 	case ixgbe_mac_82598EB:
5106 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5107 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5108 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5109 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5110 		break;
5111 	case ixgbe_mac_X540:
5112 		/* OS2BMC stats are X540 only*/
5113 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5114 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5115 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5116 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5117 	case ixgbe_mac_82599EB:
5118 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5119 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5120 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5121 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5122 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5123 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5124 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5125 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5126 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5127 #ifdef IXGBE_FCOE
5128 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5129 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5130 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5131 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5132 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5133 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5134 		/* Add up per cpu counters for total ddp aloc fail */
5135 		if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5136 			for_each_possible_cpu(cpu) {
5137 				fcoe_noddp_counts_sum +=
5138 					*per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5139 				fcoe_noddp_ext_buff_counts_sum +=
5140 					*per_cpu_ptr(fcoe->
5141 						pcpu_noddp_ext_buff, cpu);
5142 			}
5143 		}
5144 		hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5145 		hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5146 #endif /* IXGBE_FCOE */
5147 		break;
5148 	default:
5149 		break;
5150 	}
5151 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5152 	hwstats->bprc += bprc;
5153 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5154 	if (hw->mac.type == ixgbe_mac_82598EB)
5155 		hwstats->mprc -= bprc;
5156 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5157 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5158 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5159 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5160 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5161 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5162 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5163 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5164 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5165 	hwstats->lxontxc += lxon;
5166 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5167 	hwstats->lxofftxc += lxoff;
5168 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5169 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5170 	/*
5171 	 * 82598 errata - tx of flow control packets is included in tx counters
5172 	 */
5173 	xon_off_tot = lxon + lxoff;
5174 	hwstats->gptc -= xon_off_tot;
5175 	hwstats->mptc -= xon_off_tot;
5176 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5177 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5178 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5179 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5180 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5181 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5182 	hwstats->ptc64 -= xon_off_tot;
5183 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5184 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5185 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5186 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5187 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5188 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5189 
5190 	/* Fill out the OS statistics structure */
5191 	netdev->stats.multicast = hwstats->mprc;
5192 
5193 	/* Rx Errors */
5194 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5195 	netdev->stats.rx_dropped = 0;
5196 	netdev->stats.rx_length_errors = hwstats->rlec;
5197 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
5198 	netdev->stats.rx_missed_errors = total_mpc;
5199 }
5200 
5201 /**
5202  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5203  * @adapter - pointer to the device adapter structure
5204  **/
ixgbe_fdir_reinit_subtask(struct ixgbe_adapter * adapter)5205 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5206 {
5207 	struct ixgbe_hw *hw = &adapter->hw;
5208 	int i;
5209 
5210 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5211 		return;
5212 
5213 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5214 
5215 	/* if interface is down do nothing */
5216 	if (test_bit(__IXGBE_DOWN, &adapter->state))
5217 		return;
5218 
5219 	/* do nothing if we are not using signature filters */
5220 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5221 		return;
5222 
5223 	adapter->fdir_overflow++;
5224 
5225 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5226 		for (i = 0; i < adapter->num_tx_queues; i++)
5227 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5228 			        &(adapter->tx_ring[i]->state));
5229 		/* re-enable flow director interrupts */
5230 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5231 	} else {
5232 		e_err(probe, "failed to finish FDIR re-initialization, "
5233 		      "ignored adding FDIR ATR filters\n");
5234 	}
5235 }
5236 
5237 /**
5238  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5239  * @adapter - pointer to the device adapter structure
5240  *
5241  * This function serves two purposes.  First it strobes the interrupt lines
5242  * in order to make certain interrupts are occurring.  Secondly it sets the
5243  * bits needed to check for TX hangs.  As a result we should immediately
5244  * determine if a hang has occurred.
5245  */
ixgbe_check_hang_subtask(struct ixgbe_adapter * adapter)5246 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5247 {
5248 	struct ixgbe_hw *hw = &adapter->hw;
5249 	u64 eics = 0;
5250 	int i;
5251 
5252 	/* If we're down or resetting, just bail */
5253 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5254 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5255 		return;
5256 
5257 	/* Force detection of hung controller */
5258 	if (netif_carrier_ok(adapter->netdev)) {
5259 		for (i = 0; i < adapter->num_tx_queues; i++)
5260 			set_check_for_tx_hang(adapter->tx_ring[i]);
5261 	}
5262 
5263 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5264 		/*
5265 		 * for legacy and MSI interrupts don't set any bits
5266 		 * that are enabled for EIAM, because this operation
5267 		 * would set *both* EIMS and EICS for any bit in EIAM
5268 		 */
5269 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
5270 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5271 	} else {
5272 		/* get one bit for every active tx/rx interrupt vector */
5273 		for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5274 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
5275 			if (qv->rx.ring || qv->tx.ring)
5276 				eics |= ((u64)1 << i);
5277 		}
5278 	}
5279 
5280 	/* Cause software interrupt to ensure rings are cleaned */
5281 	ixgbe_irq_rearm_queues(adapter, eics);
5282 
5283 }
5284 
5285 /**
5286  * ixgbe_watchdog_update_link - update the link status
5287  * @adapter - pointer to the device adapter structure
5288  * @link_speed - pointer to a u32 to store the link_speed
5289  **/
ixgbe_watchdog_update_link(struct ixgbe_adapter * adapter)5290 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5291 {
5292 	struct ixgbe_hw *hw = &adapter->hw;
5293 	u32 link_speed = adapter->link_speed;
5294 	bool link_up = adapter->link_up;
5295 	int i;
5296 
5297 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5298 		return;
5299 
5300 	if (hw->mac.ops.check_link) {
5301 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5302 	} else {
5303 		/* always assume link is up, if no check link function */
5304 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5305 		link_up = true;
5306 	}
5307 	if (link_up) {
5308 		if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5309 			for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5310 				hw->mac.ops.fc_enable(hw, i);
5311 		} else {
5312 			hw->mac.ops.fc_enable(hw, 0);
5313 		}
5314 	}
5315 
5316 	if (link_up ||
5317 	    time_after(jiffies, (adapter->link_check_timeout +
5318 				 IXGBE_TRY_LINK_TIMEOUT))) {
5319 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5320 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5321 		IXGBE_WRITE_FLUSH(hw);
5322 	}
5323 
5324 	adapter->link_up = link_up;
5325 	adapter->link_speed = link_speed;
5326 }
5327 
5328 /**
5329  * ixgbe_watchdog_link_is_up - update netif_carrier status and
5330  *                             print link up message
5331  * @adapter - pointer to the device adapter structure
5332  **/
ixgbe_watchdog_link_is_up(struct ixgbe_adapter * adapter)5333 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5334 {
5335 	struct net_device *netdev = adapter->netdev;
5336 	struct ixgbe_hw *hw = &adapter->hw;
5337 	u32 link_speed = adapter->link_speed;
5338 	bool flow_rx, flow_tx;
5339 
5340 	/* only continue if link was previously down */
5341 	if (netif_carrier_ok(netdev))
5342 		return;
5343 
5344 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5345 
5346 	switch (hw->mac.type) {
5347 	case ixgbe_mac_82598EB: {
5348 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5349 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5350 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5351 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5352 	}
5353 		break;
5354 	case ixgbe_mac_X540:
5355 	case ixgbe_mac_82599EB: {
5356 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5357 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5358 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5359 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5360 	}
5361 		break;
5362 	default:
5363 		flow_tx = false;
5364 		flow_rx = false;
5365 		break;
5366 	}
5367 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5368 	       (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5369 	       "10 Gbps" :
5370 	       (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5371 	       "1 Gbps" :
5372 	       (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5373 	       "100 Mbps" :
5374 	       "unknown speed"))),
5375 	       ((flow_rx && flow_tx) ? "RX/TX" :
5376 	       (flow_rx ? "RX" :
5377 	       (flow_tx ? "TX" : "None"))));
5378 
5379 	netif_carrier_on(netdev);
5380 	ixgbe_check_vf_rate_limit(adapter);
5381 }
5382 
5383 /**
5384  * ixgbe_watchdog_link_is_down - update netif_carrier status and
5385  *                               print link down message
5386  * @adapter - pointer to the adapter structure
5387  **/
ixgbe_watchdog_link_is_down(struct ixgbe_adapter * adapter)5388 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5389 {
5390 	struct net_device *netdev = adapter->netdev;
5391 	struct ixgbe_hw *hw = &adapter->hw;
5392 
5393 	adapter->link_up = false;
5394 	adapter->link_speed = 0;
5395 
5396 	/* only continue if link was up previously */
5397 	if (!netif_carrier_ok(netdev))
5398 		return;
5399 
5400 	/* poll for SFP+ cable when link is down */
5401 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5402 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5403 
5404 	e_info(drv, "NIC Link is Down\n");
5405 	netif_carrier_off(netdev);
5406 }
5407 
5408 /**
5409  * ixgbe_watchdog_flush_tx - flush queues on link down
5410  * @adapter - pointer to the device adapter structure
5411  **/
ixgbe_watchdog_flush_tx(struct ixgbe_adapter * adapter)5412 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5413 {
5414 	int i;
5415 	int some_tx_pending = 0;
5416 
5417 	if (!netif_carrier_ok(adapter->netdev)) {
5418 		for (i = 0; i < adapter->num_tx_queues; i++) {
5419 			struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5420 			if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5421 				some_tx_pending = 1;
5422 				break;
5423 			}
5424 		}
5425 
5426 		if (some_tx_pending) {
5427 			/* We've lost link, so the controller stops DMA,
5428 			 * but we've got queued Tx work that's never going
5429 			 * to get done, so reset controller to flush Tx.
5430 			 * (Do the reset outside of interrupt context).
5431 			 */
5432 			adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5433 		}
5434 	}
5435 }
5436 
ixgbe_spoof_check(struct ixgbe_adapter * adapter)5437 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5438 {
5439 	u32 ssvpc;
5440 
5441 	/* Do not perform spoof check for 82598 */
5442 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5443 		return;
5444 
5445 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5446 
5447 	/*
5448 	 * ssvpc register is cleared on read, if zero then no
5449 	 * spoofed packets in the last interval.
5450 	 */
5451 	if (!ssvpc)
5452 		return;
5453 
5454 	e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5455 }
5456 
5457 /**
5458  * ixgbe_watchdog_subtask - check and bring link up
5459  * @adapter - pointer to the device adapter structure
5460  **/
ixgbe_watchdog_subtask(struct ixgbe_adapter * adapter)5461 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5462 {
5463 	/* if interface is down do nothing */
5464 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5465 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5466 		return;
5467 
5468 	ixgbe_watchdog_update_link(adapter);
5469 
5470 	if (adapter->link_up)
5471 		ixgbe_watchdog_link_is_up(adapter);
5472 	else
5473 		ixgbe_watchdog_link_is_down(adapter);
5474 
5475 	ixgbe_spoof_check(adapter);
5476 	ixgbe_update_stats(adapter);
5477 
5478 	ixgbe_watchdog_flush_tx(adapter);
5479 }
5480 
5481 /**
5482  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5483  * @adapter - the ixgbe adapter structure
5484  **/
ixgbe_sfp_detection_subtask(struct ixgbe_adapter * adapter)5485 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5486 {
5487 	struct ixgbe_hw *hw = &adapter->hw;
5488 	s32 err;
5489 
5490 	/* not searching for SFP so there is nothing to do here */
5491 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5492 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5493 		return;
5494 
5495 	/* someone else is in init, wait until next service event */
5496 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5497 		return;
5498 
5499 	err = hw->phy.ops.identify_sfp(hw);
5500 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5501 		goto sfp_out;
5502 
5503 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5504 		/* If no cable is present, then we need to reset
5505 		 * the next time we find a good cable. */
5506 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5507 	}
5508 
5509 	/* exit on error */
5510 	if (err)
5511 		goto sfp_out;
5512 
5513 	/* exit if reset not needed */
5514 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5515 		goto sfp_out;
5516 
5517 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5518 
5519 	/*
5520 	 * A module may be identified correctly, but the EEPROM may not have
5521 	 * support for that module.  setup_sfp() will fail in that case, so
5522 	 * we should not allow that module to load.
5523 	 */
5524 	if (hw->mac.type == ixgbe_mac_82598EB)
5525 		err = hw->phy.ops.reset(hw);
5526 	else
5527 		err = hw->mac.ops.setup_sfp(hw);
5528 
5529 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5530 		goto sfp_out;
5531 
5532 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5533 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5534 
5535 sfp_out:
5536 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5537 
5538 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5539 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5540 		e_dev_err("failed to initialize because an unsupported "
5541 			  "SFP+ module type was detected.\n");
5542 		e_dev_err("Reload the driver after installing a "
5543 			  "supported module.\n");
5544 		unregister_netdev(adapter->netdev);
5545 	}
5546 }
5547 
5548 /**
5549  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5550  * @adapter - the ixgbe adapter structure
5551  **/
ixgbe_sfp_link_config_subtask(struct ixgbe_adapter * adapter)5552 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5553 {
5554 	struct ixgbe_hw *hw = &adapter->hw;
5555 	u32 autoneg;
5556 	bool negotiation;
5557 
5558 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5559 		return;
5560 
5561 	/* someone else is in init, wait until next service event */
5562 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5563 		return;
5564 
5565 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5566 
5567 	autoneg = hw->phy.autoneg_advertised;
5568 	if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5569 		hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5570 	if (hw->mac.ops.setup_link)
5571 		hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5572 
5573 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5574 	adapter->link_check_timeout = jiffies;
5575 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5576 }
5577 
5578 #ifdef CONFIG_PCI_IOV
ixgbe_check_for_bad_vf(struct ixgbe_adapter * adapter)5579 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5580 {
5581 	int vf;
5582 	struct ixgbe_hw *hw = &adapter->hw;
5583 	struct net_device *netdev = adapter->netdev;
5584 	u32 gpc;
5585 	u32 ciaa, ciad;
5586 
5587 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5588 	if (gpc) /* If incrementing then no need for the check below */
5589 		return;
5590 	/*
5591 	 * Check to see if a bad DMA write target from an errant or
5592 	 * malicious VF has caused a PCIe error.  If so then we can
5593 	 * issue a VFLR to the offending VF(s) and then resume without
5594 	 * requesting a full slot reset.
5595 	 */
5596 
5597 	for (vf = 0; vf < adapter->num_vfs; vf++) {
5598 		ciaa = (vf << 16) | 0x80000000;
5599 		/* 32 bit read so align, we really want status at offset 6 */
5600 		ciaa |= PCI_COMMAND;
5601 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5602 		ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5603 		ciaa &= 0x7FFFFFFF;
5604 		/* disable debug mode asap after reading data */
5605 		IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5606 		/* Get the upper 16 bits which will be the PCI status reg */
5607 		ciad >>= 16;
5608 		if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5609 			netdev_err(netdev, "VF %d Hung DMA\n", vf);
5610 			/* Issue VFLR */
5611 			ciaa = (vf << 16) | 0x80000000;
5612 			ciaa |= 0xA8;
5613 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5614 			ciad = 0x00008000;  /* VFLR */
5615 			IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5616 			ciaa &= 0x7FFFFFFF;
5617 			IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5618 		}
5619 	}
5620 }
5621 
5622 #endif
5623 /**
5624  * ixgbe_service_timer - Timer Call-back
5625  * @data: pointer to adapter cast into an unsigned long
5626  **/
ixgbe_service_timer(unsigned long data)5627 static void ixgbe_service_timer(unsigned long data)
5628 {
5629 	struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5630 	unsigned long next_event_offset;
5631 	bool ready = true;
5632 
5633 	/* poll faster when waiting for link */
5634 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5635 		next_event_offset = HZ / 10;
5636 	else
5637 		next_event_offset = HZ * 2;
5638 
5639 #ifdef CONFIG_PCI_IOV
5640 	/*
5641 	 * don't bother with SR-IOV VF DMA hang check if there are
5642 	 * no VFs or the link is down
5643 	 */
5644 	if (!adapter->num_vfs ||
5645 	    (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5646 		goto normal_timer_service;
5647 
5648 	/* If we have VFs allocated then we must check for DMA hangs */
5649 	ixgbe_check_for_bad_vf(adapter);
5650 	next_event_offset = HZ / 50;
5651 	adapter->timer_event_accumulator++;
5652 
5653 	if (adapter->timer_event_accumulator >= 100)
5654 		adapter->timer_event_accumulator = 0;
5655 	else
5656 		ready = false;
5657 
5658 normal_timer_service:
5659 #endif
5660 	/* Reset the timer */
5661 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5662 
5663 	if (ready)
5664 		ixgbe_service_event_schedule(adapter);
5665 }
5666 
ixgbe_reset_subtask(struct ixgbe_adapter * adapter)5667 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5668 {
5669 	if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5670 		return;
5671 
5672 	adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5673 
5674 	/* If we're already down or resetting, just bail */
5675 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5676 	    test_bit(__IXGBE_RESETTING, &adapter->state))
5677 		return;
5678 
5679 	ixgbe_dump(adapter);
5680 	netdev_err(adapter->netdev, "Reset adapter\n");
5681 	adapter->tx_timeout_count++;
5682 
5683 	ixgbe_reinit_locked(adapter);
5684 }
5685 
5686 /**
5687  * ixgbe_service_task - manages and runs subtasks
5688  * @work: pointer to work_struct containing our data
5689  **/
ixgbe_service_task(struct work_struct * work)5690 static void ixgbe_service_task(struct work_struct *work)
5691 {
5692 	struct ixgbe_adapter *adapter = container_of(work,
5693 						     struct ixgbe_adapter,
5694 						     service_task);
5695 
5696 	ixgbe_reset_subtask(adapter);
5697 	ixgbe_sfp_detection_subtask(adapter);
5698 	ixgbe_sfp_link_config_subtask(adapter);
5699 	ixgbe_check_overtemp_subtask(adapter);
5700 	ixgbe_watchdog_subtask(adapter);
5701 	ixgbe_fdir_reinit_subtask(adapter);
5702 	ixgbe_check_hang_subtask(adapter);
5703 
5704 	ixgbe_service_event_complete(adapter);
5705 }
5706 
ixgbe_tso(struct ixgbe_ring * tx_ring,struct ixgbe_tx_buffer * first,u8 * hdr_len)5707 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5708 		     struct ixgbe_tx_buffer *first,
5709 		     u8 *hdr_len)
5710 {
5711 	struct sk_buff *skb = first->skb;
5712 	u32 vlan_macip_lens, type_tucmd;
5713 	u32 mss_l4len_idx, l4len;
5714 
5715 	if (!skb_is_gso(skb))
5716 		return 0;
5717 
5718 	if (skb_header_cloned(skb)) {
5719 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5720 		if (err)
5721 			return err;
5722 	}
5723 
5724 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5725 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5726 
5727 	if (first->protocol == __constant_htons(ETH_P_IP)) {
5728 		struct iphdr *iph = ip_hdr(skb);
5729 		iph->tot_len = 0;
5730 		iph->check = 0;
5731 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5732 							 iph->daddr, 0,
5733 							 IPPROTO_TCP,
5734 							 0);
5735 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5736 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5737 				   IXGBE_TX_FLAGS_CSUM |
5738 				   IXGBE_TX_FLAGS_IPV4;
5739 	} else if (skb_is_gso_v6(skb)) {
5740 		ipv6_hdr(skb)->payload_len = 0;
5741 		tcp_hdr(skb)->check =
5742 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5743 				     &ipv6_hdr(skb)->daddr,
5744 				     0, IPPROTO_TCP, 0);
5745 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5746 				   IXGBE_TX_FLAGS_CSUM;
5747 	}
5748 
5749 	/* compute header lengths */
5750 	l4len = tcp_hdrlen(skb);
5751 	*hdr_len = skb_transport_offset(skb) + l4len;
5752 
5753 	/* update gso size and bytecount with header size */
5754 	first->gso_segs = skb_shinfo(skb)->gso_segs;
5755 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
5756 
5757 	/* mss_l4len_id: use 1 as index for TSO */
5758 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5759 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5760 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5761 
5762 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5763 	vlan_macip_lens = skb_network_header_len(skb);
5764 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5765 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5766 
5767 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5768 			  mss_l4len_idx);
5769 
5770 	return 1;
5771 }
5772 
ixgbe_tx_csum(struct ixgbe_ring * tx_ring,struct ixgbe_tx_buffer * first)5773 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5774 			  struct ixgbe_tx_buffer *first)
5775 {
5776 	struct sk_buff *skb = first->skb;
5777 	u32 vlan_macip_lens = 0;
5778 	u32 mss_l4len_idx = 0;
5779 	u32 type_tucmd = 0;
5780 
5781 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
5782 		if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5783 		    !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5784 			return;
5785 	} else {
5786 		u8 l4_hdr = 0;
5787 		switch (first->protocol) {
5788 		case __constant_htons(ETH_P_IP):
5789 			vlan_macip_lens |= skb_network_header_len(skb);
5790 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5791 			l4_hdr = ip_hdr(skb)->protocol;
5792 			break;
5793 		case __constant_htons(ETH_P_IPV6):
5794 			vlan_macip_lens |= skb_network_header_len(skb);
5795 			l4_hdr = ipv6_hdr(skb)->nexthdr;
5796 			break;
5797 		default:
5798 			if (unlikely(net_ratelimit())) {
5799 				dev_warn(tx_ring->dev,
5800 				 "partial checksum but proto=%x!\n",
5801 				 first->protocol);
5802 			}
5803 			break;
5804 		}
5805 
5806 		switch (l4_hdr) {
5807 		case IPPROTO_TCP:
5808 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5809 			mss_l4len_idx = tcp_hdrlen(skb) <<
5810 					IXGBE_ADVTXD_L4LEN_SHIFT;
5811 			break;
5812 		case IPPROTO_SCTP:
5813 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5814 			mss_l4len_idx = sizeof(struct sctphdr) <<
5815 					IXGBE_ADVTXD_L4LEN_SHIFT;
5816 			break;
5817 		case IPPROTO_UDP:
5818 			mss_l4len_idx = sizeof(struct udphdr) <<
5819 					IXGBE_ADVTXD_L4LEN_SHIFT;
5820 			break;
5821 		default:
5822 			if (unlikely(net_ratelimit())) {
5823 				dev_warn(tx_ring->dev,
5824 				 "partial checksum but l4 proto=%x!\n",
5825 				 l4_hdr);
5826 			}
5827 			break;
5828 		}
5829 
5830 		/* update TX checksum flag */
5831 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5832 	}
5833 
5834 	/* vlan_macip_lens: MACLEN, VLAN tag */
5835 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5836 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5837 
5838 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5839 			  type_tucmd, mss_l4len_idx);
5840 }
5841 
ixgbe_tx_cmd_type(u32 tx_flags)5842 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5843 {
5844 	/* set type for advanced descriptor with frame checksum insertion */
5845 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5846 				      IXGBE_ADVTXD_DCMD_IFCS |
5847 				      IXGBE_ADVTXD_DCMD_DEXT);
5848 
5849 	/* set HW vlan bit if vlan is present */
5850 	if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5851 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5852 
5853 	/* set segmentation enable bits for TSO/FSO */
5854 #ifdef IXGBE_FCOE
5855 	if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5856 #else
5857 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
5858 #endif
5859 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5860 
5861 	return cmd_type;
5862 }
5863 
ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc * tx_desc,u32 tx_flags,unsigned int paylen)5864 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5865 				   u32 tx_flags, unsigned int paylen)
5866 {
5867 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5868 
5869 	/* enable L4 checksum for TSO and TX checksum offload */
5870 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5871 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5872 
5873 	/* enble IPv4 checksum for TSO */
5874 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5875 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5876 
5877 	/* use index 1 context for TSO/FSO/FCOE */
5878 #ifdef IXGBE_FCOE
5879 	if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5880 #else
5881 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
5882 #endif
5883 		olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5884 
5885 	/*
5886 	 * Check Context must be set if Tx switch is enabled, which it
5887 	 * always is for case where virtual functions are running
5888 	 */
5889 #ifdef IXGBE_FCOE
5890 	if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5891 #else
5892 	if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5893 #endif
5894 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5895 
5896 	tx_desc->read.olinfo_status = olinfo_status;
5897 }
5898 
5899 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5900 		       IXGBE_TXD_CMD_RS)
5901 
ixgbe_tx_map(struct ixgbe_ring * tx_ring,struct ixgbe_tx_buffer * first,const u8 hdr_len)5902 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5903 			 struct ixgbe_tx_buffer *first,
5904 			 const u8 hdr_len)
5905 {
5906 	dma_addr_t dma;
5907 	struct sk_buff *skb = first->skb;
5908 	struct ixgbe_tx_buffer *tx_buffer;
5909 	union ixgbe_adv_tx_desc *tx_desc;
5910 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5911 	unsigned int data_len = skb->data_len;
5912 	unsigned int size = skb_headlen(skb);
5913 	unsigned int paylen = skb->len - hdr_len;
5914 	u32 tx_flags = first->tx_flags;
5915 	__le32 cmd_type;
5916 	u16 i = tx_ring->next_to_use;
5917 
5918 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
5919 
5920 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5921 	cmd_type = ixgbe_tx_cmd_type(tx_flags);
5922 
5923 #ifdef IXGBE_FCOE
5924 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5925 		if (data_len < sizeof(struct fcoe_crc_eof)) {
5926 			size -= sizeof(struct fcoe_crc_eof) - data_len;
5927 			data_len = 0;
5928 		} else {
5929 			data_len -= sizeof(struct fcoe_crc_eof);
5930 		}
5931 	}
5932 
5933 #endif
5934 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5935 	if (dma_mapping_error(tx_ring->dev, dma))
5936 		goto dma_error;
5937 
5938 	/* record length, and DMA address */
5939 	dma_unmap_len_set(first, len, size);
5940 	dma_unmap_addr_set(first, dma, dma);
5941 
5942 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
5943 
5944 	for (;;) {
5945 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5946 			tx_desc->read.cmd_type_len =
5947 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5948 
5949 			i++;
5950 			tx_desc++;
5951 			if (i == tx_ring->count) {
5952 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5953 				i = 0;
5954 			}
5955 
5956 			dma += IXGBE_MAX_DATA_PER_TXD;
5957 			size -= IXGBE_MAX_DATA_PER_TXD;
5958 
5959 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
5960 			tx_desc->read.olinfo_status = 0;
5961 		}
5962 
5963 		if (likely(!data_len))
5964 			break;
5965 
5966 		if (unlikely(skb->no_fcs))
5967 			cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
5968 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
5969 
5970 		i++;
5971 		tx_desc++;
5972 		if (i == tx_ring->count) {
5973 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5974 			i = 0;
5975 		}
5976 
5977 #ifdef IXGBE_FCOE
5978 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
5979 #else
5980 		size = skb_frag_size(frag);
5981 #endif
5982 		data_len -= size;
5983 
5984 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
5985 				       DMA_TO_DEVICE);
5986 		if (dma_mapping_error(tx_ring->dev, dma))
5987 			goto dma_error;
5988 
5989 		tx_buffer = &tx_ring->tx_buffer_info[i];
5990 		dma_unmap_len_set(tx_buffer, len, size);
5991 		dma_unmap_addr_set(tx_buffer, dma, dma);
5992 
5993 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
5994 		tx_desc->read.olinfo_status = 0;
5995 
5996 		frag++;
5997 	}
5998 
5999 	/* write last descriptor with RS and EOP bits */
6000 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6001 	tx_desc->read.cmd_type_len = cmd_type;
6002 
6003 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6004 
6005 	/* set the timestamp */
6006 	first->time_stamp = jiffies;
6007 
6008 	/*
6009 	 * Force memory writes to complete before letting h/w know there
6010 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
6011 	 * memory model archs, such as IA-64).
6012 	 *
6013 	 * We also need this memory barrier to make certain all of the
6014 	 * status bits have been updated before next_to_watch is written.
6015 	 */
6016 	wmb();
6017 
6018 	/* set next_to_watch value indicating a packet is present */
6019 	first->next_to_watch = tx_desc;
6020 
6021 	i++;
6022 	if (i == tx_ring->count)
6023 		i = 0;
6024 
6025 	tx_ring->next_to_use = i;
6026 
6027 	/* notify HW of packet */
6028 	writel(i, tx_ring->tail);
6029 
6030 	return;
6031 dma_error:
6032 	dev_err(tx_ring->dev, "TX DMA map failed\n");
6033 
6034 	/* clear dma mappings for failed tx_buffer_info map */
6035 	for (;;) {
6036 		tx_buffer = &tx_ring->tx_buffer_info[i];
6037 		ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6038 		if (tx_buffer == first)
6039 			break;
6040 		if (i == 0)
6041 			i = tx_ring->count;
6042 		i--;
6043 	}
6044 
6045 	tx_ring->next_to_use = i;
6046 }
6047 
ixgbe_atr(struct ixgbe_ring * ring,struct ixgbe_tx_buffer * first)6048 static void ixgbe_atr(struct ixgbe_ring *ring,
6049 		      struct ixgbe_tx_buffer *first)
6050 {
6051 	struct ixgbe_q_vector *q_vector = ring->q_vector;
6052 	union ixgbe_atr_hash_dword input = { .dword = 0 };
6053 	union ixgbe_atr_hash_dword common = { .dword = 0 };
6054 	union {
6055 		unsigned char *network;
6056 		struct iphdr *ipv4;
6057 		struct ipv6hdr *ipv6;
6058 	} hdr;
6059 	struct tcphdr *th;
6060 	__be16 vlan_id;
6061 
6062 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
6063 	if (!q_vector)
6064 		return;
6065 
6066 	/* do nothing if sampling is disabled */
6067 	if (!ring->atr_sample_rate)
6068 		return;
6069 
6070 	ring->atr_count++;
6071 
6072 	/* snag network header to get L4 type and address */
6073 	hdr.network = skb_network_header(first->skb);
6074 
6075 	/* Currently only IPv4/IPv6 with TCP is supported */
6076 	if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6077 	     hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6078 	    (first->protocol != __constant_htons(ETH_P_IP) ||
6079 	     hdr.ipv4->protocol != IPPROTO_TCP))
6080 		return;
6081 
6082 	th = tcp_hdr(first->skb);
6083 
6084 	/* skip this packet since it is invalid or the socket is closing */
6085 	if (!th || th->fin)
6086 		return;
6087 
6088 	/* sample on all syn packets or once every atr sample count */
6089 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6090 		return;
6091 
6092 	/* reset sample count */
6093 	ring->atr_count = 0;
6094 
6095 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6096 
6097 	/*
6098 	 * src and dst are inverted, think how the receiver sees them
6099 	 *
6100 	 * The input is broken into two sections, a non-compressed section
6101 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
6102 	 * is XORed together and stored in the compressed dword.
6103 	 */
6104 	input.formatted.vlan_id = vlan_id;
6105 
6106 	/*
6107 	 * since src port and flex bytes occupy the same word XOR them together
6108 	 * and write the value to source port portion of compressed dword
6109 	 */
6110 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6111 		common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6112 	else
6113 		common.port.src ^= th->dest ^ first->protocol;
6114 	common.port.dst ^= th->source;
6115 
6116 	if (first->protocol == __constant_htons(ETH_P_IP)) {
6117 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6118 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6119 	} else {
6120 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6121 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6122 			     hdr.ipv6->saddr.s6_addr32[1] ^
6123 			     hdr.ipv6->saddr.s6_addr32[2] ^
6124 			     hdr.ipv6->saddr.s6_addr32[3] ^
6125 			     hdr.ipv6->daddr.s6_addr32[0] ^
6126 			     hdr.ipv6->daddr.s6_addr32[1] ^
6127 			     hdr.ipv6->daddr.s6_addr32[2] ^
6128 			     hdr.ipv6->daddr.s6_addr32[3];
6129 	}
6130 
6131 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
6132 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6133 					      input, common, ring->queue_index);
6134 }
6135 
__ixgbe_maybe_stop_tx(struct ixgbe_ring * tx_ring,u16 size)6136 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6137 {
6138 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6139 	/* Herbert's original patch had:
6140 	 *  smp_mb__after_netif_stop_queue();
6141 	 * but since that doesn't exist yet, just open code it. */
6142 	smp_mb();
6143 
6144 	/* We need to check again in a case another CPU has just
6145 	 * made room available. */
6146 	if (likely(ixgbe_desc_unused(tx_ring) < size))
6147 		return -EBUSY;
6148 
6149 	/* A reprieve! - use start_queue because it doesn't call schedule */
6150 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6151 	++tx_ring->tx_stats.restart_queue;
6152 	return 0;
6153 }
6154 
ixgbe_maybe_stop_tx(struct ixgbe_ring * tx_ring,u16 size)6155 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6156 {
6157 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
6158 		return 0;
6159 	return __ixgbe_maybe_stop_tx(tx_ring, size);
6160 }
6161 
ixgbe_select_queue(struct net_device * dev,struct sk_buff * skb)6162 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6163 {
6164 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6165 	int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6166 					       smp_processor_id();
6167 #ifdef IXGBE_FCOE
6168 	__be16 protocol = vlan_get_protocol(skb);
6169 
6170 	if (((protocol == htons(ETH_P_FCOE)) ||
6171 	    (protocol == htons(ETH_P_FIP))) &&
6172 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6173 		txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6174 		txq += adapter->ring_feature[RING_F_FCOE].mask;
6175 		return txq;
6176 	}
6177 #endif
6178 
6179 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6180 		while (unlikely(txq >= dev->real_num_tx_queues))
6181 			txq -= dev->real_num_tx_queues;
6182 		return txq;
6183 	}
6184 
6185 	return skb_tx_hash(dev, skb);
6186 }
6187 
ixgbe_xmit_frame_ring(struct sk_buff * skb,struct ixgbe_adapter * adapter,struct ixgbe_ring * tx_ring)6188 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6189 			  struct ixgbe_adapter *adapter,
6190 			  struct ixgbe_ring *tx_ring)
6191 {
6192 	struct ixgbe_tx_buffer *first;
6193 	int tso;
6194 	u32 tx_flags = 0;
6195 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6196 	unsigned short f;
6197 #endif
6198 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
6199 	__be16 protocol = skb->protocol;
6200 	u8 hdr_len = 0;
6201 
6202 	/*
6203 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6204 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6205 	 *       + 2 desc gap to keep tail from touching head,
6206 	 *       + 1 desc for context descriptor,
6207 	 * otherwise try next time
6208 	 */
6209 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6210 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6211 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6212 #else
6213 	count += skb_shinfo(skb)->nr_frags;
6214 #endif
6215 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6216 		tx_ring->tx_stats.tx_busy++;
6217 		return NETDEV_TX_BUSY;
6218 	}
6219 
6220 	/* record the location of the first descriptor for this packet */
6221 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6222 	first->skb = skb;
6223 	first->bytecount = skb->len;
6224 	first->gso_segs = 1;
6225 
6226 	/* if we have a HW VLAN tag being added default to the HW one */
6227 	if (vlan_tx_tag_present(skb)) {
6228 		tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6229 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6230 	/* else if it is a SW VLAN check the next protocol and store the tag */
6231 	} else if (protocol == __constant_htons(ETH_P_8021Q)) {
6232 		struct vlan_hdr *vhdr, _vhdr;
6233 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6234 		if (!vhdr)
6235 			goto out_drop;
6236 
6237 		protocol = vhdr->h_vlan_encapsulated_proto;
6238 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6239 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
6240 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6241 	}
6242 
6243 #ifdef CONFIG_PCI_IOV
6244 	/*
6245 	 * Use the l2switch_enable flag - would be false if the DMA
6246 	 * Tx switch had been disabled.
6247 	 */
6248 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6249 		tx_flags |= IXGBE_TX_FLAGS_TXSW;
6250 
6251 #endif
6252 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6253 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6254 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6255 	     (skb->priority != TC_PRIO_CONTROL))) {
6256 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6257 		tx_flags |= (skb->priority & 0x7) <<
6258 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6259 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6260 			struct vlan_ethhdr *vhdr;
6261 			if (skb_header_cloned(skb) &&
6262 			    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6263 				goto out_drop;
6264 			vhdr = (struct vlan_ethhdr *)skb->data;
6265 			vhdr->h_vlan_TCI = htons(tx_flags >>
6266 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
6267 		} else {
6268 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6269 		}
6270 	}
6271 
6272 	/* record initial flags and protocol */
6273 	first->tx_flags = tx_flags;
6274 	first->protocol = protocol;
6275 
6276 #ifdef IXGBE_FCOE
6277 	/* setup tx offload for FCoE */
6278 	if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6279 	    (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6280 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
6281 		if (tso < 0)
6282 			goto out_drop;
6283 
6284 		goto xmit_fcoe;
6285 	}
6286 
6287 #endif /* IXGBE_FCOE */
6288 	tso = ixgbe_tso(tx_ring, first, &hdr_len);
6289 	if (tso < 0)
6290 		goto out_drop;
6291 	else if (!tso)
6292 		ixgbe_tx_csum(tx_ring, first);
6293 
6294 	/* add the ATR filter if ATR is on */
6295 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6296 		ixgbe_atr(tx_ring, first);
6297 
6298 #ifdef IXGBE_FCOE
6299 xmit_fcoe:
6300 #endif /* IXGBE_FCOE */
6301 	ixgbe_tx_map(tx_ring, first, hdr_len);
6302 
6303 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6304 
6305 	return NETDEV_TX_OK;
6306 
6307 out_drop:
6308 	dev_kfree_skb_any(first->skb);
6309 	first->skb = NULL;
6310 
6311 	return NETDEV_TX_OK;
6312 }
6313 
ixgbe_xmit_frame(struct sk_buff * skb,struct net_device * netdev)6314 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6315 				    struct net_device *netdev)
6316 {
6317 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6318 	struct ixgbe_ring *tx_ring;
6319 
6320 	if (skb->len <= 0) {
6321 		dev_kfree_skb_any(skb);
6322 		return NETDEV_TX_OK;
6323 	}
6324 
6325 	/*
6326 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
6327 	 * in order to meet this minimum size requirement.
6328 	 */
6329 	if (skb->len < 17) {
6330 		if (skb_padto(skb, 17))
6331 			return NETDEV_TX_OK;
6332 		skb->len = 17;
6333 	}
6334 
6335 	tx_ring = adapter->tx_ring[skb->queue_mapping];
6336 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6337 }
6338 
6339 /**
6340  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6341  * @netdev: network interface device structure
6342  * @p: pointer to an address structure
6343  *
6344  * Returns 0 on success, negative on failure
6345  **/
ixgbe_set_mac(struct net_device * netdev,void * p)6346 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6347 {
6348 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6349 	struct ixgbe_hw *hw = &adapter->hw;
6350 	struct sockaddr *addr = p;
6351 
6352 	if (!is_valid_ether_addr(addr->sa_data))
6353 		return -EADDRNOTAVAIL;
6354 
6355 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6356 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6357 
6358 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6359 			    IXGBE_RAH_AV);
6360 
6361 	return 0;
6362 }
6363 
6364 static int
ixgbe_mdio_read(struct net_device * netdev,int prtad,int devad,u16 addr)6365 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6366 {
6367 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6368 	struct ixgbe_hw *hw = &adapter->hw;
6369 	u16 value;
6370 	int rc;
6371 
6372 	if (prtad != hw->phy.mdio.prtad)
6373 		return -EINVAL;
6374 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6375 	if (!rc)
6376 		rc = value;
6377 	return rc;
6378 }
6379 
ixgbe_mdio_write(struct net_device * netdev,int prtad,int devad,u16 addr,u16 value)6380 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6381 			    u16 addr, u16 value)
6382 {
6383 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6384 	struct ixgbe_hw *hw = &adapter->hw;
6385 
6386 	if (prtad != hw->phy.mdio.prtad)
6387 		return -EINVAL;
6388 	return hw->phy.ops.write_reg(hw, addr, devad, value);
6389 }
6390 
ixgbe_ioctl(struct net_device * netdev,struct ifreq * req,int cmd)6391 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6392 {
6393 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6394 
6395 	return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6396 }
6397 
6398 /**
6399  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6400  * netdev->dev_addrs
6401  * @netdev: network interface device structure
6402  *
6403  * Returns non-zero on failure
6404  **/
ixgbe_add_sanmac_netdev(struct net_device * dev)6405 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6406 {
6407 	int err = 0;
6408 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6409 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6410 
6411 	if (is_valid_ether_addr(mac->san_addr)) {
6412 		rtnl_lock();
6413 		err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6414 		rtnl_unlock();
6415 	}
6416 	return err;
6417 }
6418 
6419 /**
6420  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6421  * netdev->dev_addrs
6422  * @netdev: network interface device structure
6423  *
6424  * Returns non-zero on failure
6425  **/
ixgbe_del_sanmac_netdev(struct net_device * dev)6426 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6427 {
6428 	int err = 0;
6429 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6430 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
6431 
6432 	if (is_valid_ether_addr(mac->san_addr)) {
6433 		rtnl_lock();
6434 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6435 		rtnl_unlock();
6436 	}
6437 	return err;
6438 }
6439 
6440 #ifdef CONFIG_NET_POLL_CONTROLLER
6441 /*
6442  * Polling 'interrupt' - used by things like netconsole to send skbs
6443  * without having to re-enable interrupts. It's not called while
6444  * the interrupt routine is executing.
6445  */
ixgbe_netpoll(struct net_device * netdev)6446 static void ixgbe_netpoll(struct net_device *netdev)
6447 {
6448 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6449 	int i;
6450 
6451 	/* if interface is down do nothing */
6452 	if (test_bit(__IXGBE_DOWN, &adapter->state))
6453 		return;
6454 
6455 	adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6456 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6457 		int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6458 		for (i = 0; i < num_q_vectors; i++) {
6459 			struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6460 			ixgbe_msix_clean_rings(0, q_vector);
6461 		}
6462 	} else {
6463 		ixgbe_intr(adapter->pdev->irq, netdev);
6464 	}
6465 	adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6466 }
6467 
6468 #endif
ixgbe_get_stats64(struct net_device * netdev,struct rtnl_link_stats64 * stats)6469 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6470 						   struct rtnl_link_stats64 *stats)
6471 {
6472 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6473 	int i;
6474 
6475 	rcu_read_lock();
6476 	for (i = 0; i < adapter->num_rx_queues; i++) {
6477 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6478 		u64 bytes, packets;
6479 		unsigned int start;
6480 
6481 		if (ring) {
6482 			do {
6483 				start = u64_stats_fetch_begin_bh(&ring->syncp);
6484 				packets = ring->stats.packets;
6485 				bytes   = ring->stats.bytes;
6486 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6487 			stats->rx_packets += packets;
6488 			stats->rx_bytes   += bytes;
6489 		}
6490 	}
6491 
6492 	for (i = 0; i < adapter->num_tx_queues; i++) {
6493 		struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6494 		u64 bytes, packets;
6495 		unsigned int start;
6496 
6497 		if (ring) {
6498 			do {
6499 				start = u64_stats_fetch_begin_bh(&ring->syncp);
6500 				packets = ring->stats.packets;
6501 				bytes   = ring->stats.bytes;
6502 			} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6503 			stats->tx_packets += packets;
6504 			stats->tx_bytes   += bytes;
6505 		}
6506 	}
6507 	rcu_read_unlock();
6508 	/* following stats updated by ixgbe_watchdog_task() */
6509 	stats->multicast	= netdev->stats.multicast;
6510 	stats->rx_errors	= netdev->stats.rx_errors;
6511 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
6512 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
6513 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
6514 	return stats;
6515 }
6516 
6517 #ifdef CONFIG_IXGBE_DCB
6518 /* ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6519  * #adapter: pointer to ixgbe_adapter
6520  * @tc: number of traffic classes currently enabled
6521  *
6522  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6523  * 802.1Q priority maps to a packet buffer that exists.
6524  */
ixgbe_validate_rtr(struct ixgbe_adapter * adapter,u8 tc)6525 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6526 {
6527 	struct ixgbe_hw *hw = &adapter->hw;
6528 	u32 reg, rsave;
6529 	int i;
6530 
6531 	/* 82598 have a static priority to TC mapping that can not
6532 	 * be changed so no validation is needed.
6533 	 */
6534 	if (hw->mac.type == ixgbe_mac_82598EB)
6535 		return;
6536 
6537 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6538 	rsave = reg;
6539 
6540 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6541 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6542 
6543 		/* If up2tc is out of bounds default to zero */
6544 		if (up2tc > tc)
6545 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6546 	}
6547 
6548 	if (reg != rsave)
6549 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6550 
6551 	return;
6552 }
6553 
6554 /* ixgbe_setup_tc - routine to configure net_device for multiple traffic
6555  * classes.
6556  *
6557  * @netdev: net device to configure
6558  * @tc: number of traffic classes to enable
6559  */
ixgbe_setup_tc(struct net_device * dev,u8 tc)6560 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6561 {
6562 	struct ixgbe_adapter *adapter = netdev_priv(dev);
6563 	struct ixgbe_hw *hw = &adapter->hw;
6564 
6565 	/* Multiple traffic classes requires multiple queues */
6566 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6567 		e_err(drv, "Enable failed, needs MSI-X\n");
6568 		return -EINVAL;
6569 	}
6570 
6571 	/* Hardware supports up to 8 traffic classes */
6572 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6573 	    (hw->mac.type == ixgbe_mac_82598EB &&
6574 	     tc < MAX_TRAFFIC_CLASS))
6575 		return -EINVAL;
6576 
6577 	/* Hardware has to reinitialize queues and interrupts to
6578 	 * match packet buffer alignment. Unfortunately, the
6579 	 * hardware is not flexible enough to do this dynamically.
6580 	 */
6581 	if (netif_running(dev))
6582 		ixgbe_close(dev);
6583 	ixgbe_clear_interrupt_scheme(adapter);
6584 
6585 	if (tc) {
6586 		netdev_set_num_tc(dev, tc);
6587 		adapter->last_lfc_mode = adapter->hw.fc.current_mode;
6588 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6589 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6590 
6591 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6592 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
6593 	} else {
6594 		netdev_reset_tc(dev);
6595 		adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6596 
6597 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6598 		adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6599 
6600 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
6601 		adapter->dcb_cfg.pfc_mode_enable = false;
6602 	}
6603 
6604 	ixgbe_init_interrupt_scheme(adapter);
6605 	ixgbe_validate_rtr(adapter, tc);
6606 	if (netif_running(dev))
6607 		ixgbe_open(dev);
6608 
6609 	return 0;
6610 }
6611 
6612 #endif /* CONFIG_IXGBE_DCB */
ixgbe_do_reset(struct net_device * netdev)6613 void ixgbe_do_reset(struct net_device *netdev)
6614 {
6615 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6616 
6617 	if (netif_running(netdev))
6618 		ixgbe_reinit_locked(adapter);
6619 	else
6620 		ixgbe_reset(adapter);
6621 }
6622 
ixgbe_fix_features(struct net_device * netdev,netdev_features_t features)6623 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6624 					    netdev_features_t features)
6625 {
6626 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6627 
6628 #ifdef CONFIG_DCB
6629 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6630 		features &= ~NETIF_F_HW_VLAN_RX;
6631 #endif
6632 
6633 	/* return error if RXHASH is being enabled when RSS is not supported */
6634 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6635 		features &= ~NETIF_F_RXHASH;
6636 
6637 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6638 	if (!(features & NETIF_F_RXCSUM))
6639 		features &= ~NETIF_F_LRO;
6640 
6641 	/* Turn off LRO if not RSC capable */
6642 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6643 		features &= ~NETIF_F_LRO;
6644 
6645 
6646 	return features;
6647 }
6648 
ixgbe_set_features(struct net_device * netdev,netdev_features_t features)6649 static int ixgbe_set_features(struct net_device *netdev,
6650 			      netdev_features_t features)
6651 {
6652 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6653 	netdev_features_t changed = netdev->features ^ features;
6654 	bool need_reset = false;
6655 
6656 	/* Make sure RSC matches LRO, reset if change */
6657 	if (!(features & NETIF_F_LRO)) {
6658 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6659 			need_reset = true;
6660 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6661 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6662 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6663 		if (adapter->rx_itr_setting == 1 ||
6664 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6665 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6666 			need_reset = true;
6667 		} else if ((changed ^ features) & NETIF_F_LRO) {
6668 			e_info(probe, "rx-usecs set too low, "
6669 			       "disabling RSC\n");
6670 		}
6671 	}
6672 
6673 	/*
6674 	 * Check if Flow Director n-tuple support was enabled or disabled.  If
6675 	 * the state changed, we need to reset.
6676 	 */
6677 	if (!(features & NETIF_F_NTUPLE)) {
6678 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6679 			/* turn off Flow Director, set ATR and reset */
6680 			if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6681 			    !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6682 				adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6683 			need_reset = true;
6684 		}
6685 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6686 	} else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6687 		/* turn off ATR, enable perfect filters and reset */
6688 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6689 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6690 		need_reset = true;
6691 	}
6692 
6693 	if (changed & NETIF_F_RXALL)
6694 		need_reset = true;
6695 
6696 	netdev->features = features;
6697 	if (need_reset)
6698 		ixgbe_do_reset(netdev);
6699 
6700 	return 0;
6701 }
6702 
6703 static const struct net_device_ops ixgbe_netdev_ops = {
6704 	.ndo_open		= ixgbe_open,
6705 	.ndo_stop		= ixgbe_close,
6706 	.ndo_start_xmit		= ixgbe_xmit_frame,
6707 	.ndo_select_queue	= ixgbe_select_queue,
6708 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
6709 	.ndo_validate_addr	= eth_validate_addr,
6710 	.ndo_set_mac_address	= ixgbe_set_mac,
6711 	.ndo_change_mtu		= ixgbe_change_mtu,
6712 	.ndo_tx_timeout		= ixgbe_tx_timeout,
6713 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
6714 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
6715 	.ndo_do_ioctl		= ixgbe_ioctl,
6716 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
6717 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
6718 	.ndo_set_vf_tx_rate	= ixgbe_ndo_set_vf_bw,
6719 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
6720 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
6721 	.ndo_get_stats64	= ixgbe_get_stats64,
6722 #ifdef CONFIG_IXGBE_DCB
6723 	.ndo_setup_tc		= ixgbe_setup_tc,
6724 #endif
6725 #ifdef CONFIG_NET_POLL_CONTROLLER
6726 	.ndo_poll_controller	= ixgbe_netpoll,
6727 #endif
6728 #ifdef IXGBE_FCOE
6729 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6730 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6731 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6732 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
6733 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
6734 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6735 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6736 #endif /* IXGBE_FCOE */
6737 	.ndo_set_features = ixgbe_set_features,
6738 	.ndo_fix_features = ixgbe_fix_features,
6739 };
6740 
ixgbe_probe_vf(struct ixgbe_adapter * adapter,const struct ixgbe_info * ii)6741 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6742 				     const struct ixgbe_info *ii)
6743 {
6744 #ifdef CONFIG_PCI_IOV
6745 	struct ixgbe_hw *hw = &adapter->hw;
6746 
6747 	if (hw->mac.type == ixgbe_mac_82598EB)
6748 		return;
6749 
6750 	/* The 82599 supports up to 64 VFs per physical function
6751 	 * but this implementation limits allocation to 63 so that
6752 	 * basic networking resources are still available to the
6753 	 * physical function
6754 	 */
6755 	adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6756 	ixgbe_enable_sriov(adapter, ii);
6757 #endif /* CONFIG_PCI_IOV */
6758 }
6759 
6760 /**
6761  * ixgbe_probe - Device Initialization Routine
6762  * @pdev: PCI device information struct
6763  * @ent: entry in ixgbe_pci_tbl
6764  *
6765  * Returns 0 on success, negative on failure
6766  *
6767  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6768  * The OS initialization, configuring of the adapter private structure,
6769  * and a hardware reset occur.
6770  **/
ixgbe_probe(struct pci_dev * pdev,const struct pci_device_id * ent)6771 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6772 				 const struct pci_device_id *ent)
6773 {
6774 	struct net_device *netdev;
6775 	struct ixgbe_adapter *adapter = NULL;
6776 	struct ixgbe_hw *hw;
6777 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6778 	static int cards_found;
6779 	int i, err, pci_using_dac;
6780 	u8 part_str[IXGBE_PBANUM_LENGTH];
6781 	unsigned int indices = num_possible_cpus();
6782 #ifdef IXGBE_FCOE
6783 	u16 device_caps;
6784 #endif
6785 	u32 eec;
6786 	u16 wol_cap;
6787 
6788 	/* Catch broken hardware that put the wrong VF device ID in
6789 	 * the PCIe SR-IOV capability.
6790 	 */
6791 	if (pdev->is_virtfn) {
6792 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6793 		     pci_name(pdev), pdev->vendor, pdev->device);
6794 		return -EINVAL;
6795 	}
6796 
6797 	err = pci_enable_device_mem(pdev);
6798 	if (err)
6799 		return err;
6800 
6801 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6802 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6803 		pci_using_dac = 1;
6804 	} else {
6805 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6806 		if (err) {
6807 			err = dma_set_coherent_mask(&pdev->dev,
6808 						    DMA_BIT_MASK(32));
6809 			if (err) {
6810 				dev_err(&pdev->dev,
6811 					"No usable DMA configuration, aborting\n");
6812 				goto err_dma;
6813 			}
6814 		}
6815 		pci_using_dac = 0;
6816 	}
6817 
6818 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6819 					   IORESOURCE_MEM), ixgbe_driver_name);
6820 	if (err) {
6821 		dev_err(&pdev->dev,
6822 			"pci_request_selected_regions failed 0x%x\n", err);
6823 		goto err_pci_reg;
6824 	}
6825 
6826 	pci_enable_pcie_error_reporting(pdev);
6827 
6828 	pci_set_master(pdev);
6829 	pci_save_state(pdev);
6830 
6831 #ifdef CONFIG_IXGBE_DCB
6832 	indices *= MAX_TRAFFIC_CLASS;
6833 #endif
6834 
6835 	if (ii->mac == ixgbe_mac_82598EB)
6836 		indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6837 	else
6838 		indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6839 
6840 #ifdef IXGBE_FCOE
6841 	indices += min_t(unsigned int, num_possible_cpus(),
6842 			 IXGBE_MAX_FCOE_INDICES);
6843 #endif
6844 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6845 	if (!netdev) {
6846 		err = -ENOMEM;
6847 		goto err_alloc_etherdev;
6848 	}
6849 
6850 	SET_NETDEV_DEV(netdev, &pdev->dev);
6851 
6852 	adapter = netdev_priv(netdev);
6853 	pci_set_drvdata(pdev, adapter);
6854 
6855 	adapter->netdev = netdev;
6856 	adapter->pdev = pdev;
6857 	hw = &adapter->hw;
6858 	hw->back = adapter;
6859 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6860 
6861 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6862 			      pci_resource_len(pdev, 0));
6863 	if (!hw->hw_addr) {
6864 		err = -EIO;
6865 		goto err_ioremap;
6866 	}
6867 
6868 	for (i = 1; i <= 5; i++) {
6869 		if (pci_resource_len(pdev, i) == 0)
6870 			continue;
6871 	}
6872 
6873 	netdev->netdev_ops = &ixgbe_netdev_ops;
6874 	ixgbe_set_ethtool_ops(netdev);
6875 	netdev->watchdog_timeo = 5 * HZ;
6876 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6877 
6878 	adapter->bd_number = cards_found;
6879 
6880 	/* Setup hw api */
6881 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6882 	hw->mac.type  = ii->mac;
6883 
6884 	/* EEPROM */
6885 	memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6886 	eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6887 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6888 	if (!(eec & (1 << 8)))
6889 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6890 
6891 	/* PHY */
6892 	memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6893 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6894 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
6895 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6896 	hw->phy.mdio.mmds = 0;
6897 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6898 	hw->phy.mdio.dev = netdev;
6899 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6900 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6901 
6902 	ii->get_invariants(hw);
6903 
6904 	/* setup the private structure */
6905 	err = ixgbe_sw_init(adapter);
6906 	if (err)
6907 		goto err_sw_init;
6908 
6909 	/* Make it possible the adapter to be woken up via WOL */
6910 	switch (adapter->hw.mac.type) {
6911 	case ixgbe_mac_82599EB:
6912 	case ixgbe_mac_X540:
6913 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6914 		break;
6915 	default:
6916 		break;
6917 	}
6918 
6919 	/*
6920 	 * If there is a fan on this device and it has failed log the
6921 	 * failure.
6922 	 */
6923 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6924 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6925 		if (esdp & IXGBE_ESDP_SDP1)
6926 			e_crit(probe, "Fan has stopped, replace the adapter\n");
6927 	}
6928 
6929 	if (allow_unsupported_sfp)
6930 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
6931 
6932 	/* reset_hw fills in the perm_addr as well */
6933 	hw->phy.reset_if_overtemp = true;
6934 	err = hw->mac.ops.reset_hw(hw);
6935 	hw->phy.reset_if_overtemp = false;
6936 	if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6937 	    hw->mac.type == ixgbe_mac_82598EB) {
6938 		err = 0;
6939 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6940 		e_dev_err("failed to load because an unsupported SFP+ "
6941 			  "module type was detected.\n");
6942 		e_dev_err("Reload the driver after installing a supported "
6943 			  "module.\n");
6944 		goto err_sw_init;
6945 	} else if (err) {
6946 		e_dev_err("HW Init failed: %d\n", err);
6947 		goto err_sw_init;
6948 	}
6949 
6950 	ixgbe_probe_vf(adapter, ii);
6951 
6952 	netdev->features = NETIF_F_SG |
6953 			   NETIF_F_IP_CSUM |
6954 			   NETIF_F_IPV6_CSUM |
6955 			   NETIF_F_HW_VLAN_TX |
6956 			   NETIF_F_HW_VLAN_RX |
6957 			   NETIF_F_HW_VLAN_FILTER |
6958 			   NETIF_F_TSO |
6959 			   NETIF_F_TSO6 |
6960 			   NETIF_F_RXHASH |
6961 			   NETIF_F_RXCSUM;
6962 
6963 	netdev->hw_features = netdev->features;
6964 
6965 	switch (adapter->hw.mac.type) {
6966 	case ixgbe_mac_82599EB:
6967 	case ixgbe_mac_X540:
6968 		netdev->features |= NETIF_F_SCTP_CSUM;
6969 		netdev->hw_features |= NETIF_F_SCTP_CSUM |
6970 				       NETIF_F_NTUPLE;
6971 		break;
6972 	default:
6973 		break;
6974 	}
6975 
6976 	netdev->hw_features |= NETIF_F_RXALL;
6977 
6978 	netdev->vlan_features |= NETIF_F_TSO;
6979 	netdev->vlan_features |= NETIF_F_TSO6;
6980 	netdev->vlan_features |= NETIF_F_IP_CSUM;
6981 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6982 	netdev->vlan_features |= NETIF_F_SG;
6983 
6984 	netdev->priv_flags |= IFF_UNICAST_FLT;
6985 	netdev->priv_flags |= IFF_SUPP_NOFCS;
6986 
6987 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6988 		adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6989 				    IXGBE_FLAG_DCB_ENABLED);
6990 
6991 #ifdef CONFIG_IXGBE_DCB
6992 	netdev->dcbnl_ops = &dcbnl_ops;
6993 #endif
6994 
6995 #ifdef IXGBE_FCOE
6996 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6997 		if (hw->mac.ops.get_device_caps) {
6998 			hw->mac.ops.get_device_caps(hw, &device_caps);
6999 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7000 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7001 		}
7002 	}
7003 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7004 		netdev->vlan_features |= NETIF_F_FCOE_CRC;
7005 		netdev->vlan_features |= NETIF_F_FSO;
7006 		netdev->vlan_features |= NETIF_F_FCOE_MTU;
7007 	}
7008 #endif /* IXGBE_FCOE */
7009 	if (pci_using_dac) {
7010 		netdev->features |= NETIF_F_HIGHDMA;
7011 		netdev->vlan_features |= NETIF_F_HIGHDMA;
7012 	}
7013 
7014 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7015 		netdev->hw_features |= NETIF_F_LRO;
7016 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7017 		netdev->features |= NETIF_F_LRO;
7018 
7019 	/* make sure the EEPROM is good */
7020 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7021 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
7022 		err = -EIO;
7023 		goto err_sw_init;
7024 	}
7025 
7026 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7027 	memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7028 
7029 	if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7030 		e_dev_err("invalid MAC address\n");
7031 		err = -EIO;
7032 		goto err_sw_init;
7033 	}
7034 
7035 	setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7036 		    (unsigned long) adapter);
7037 
7038 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
7039 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7040 
7041 	err = ixgbe_init_interrupt_scheme(adapter);
7042 	if (err)
7043 		goto err_sw_init;
7044 
7045 	if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7046 		netdev->hw_features &= ~NETIF_F_RXHASH;
7047 		netdev->features &= ~NETIF_F_RXHASH;
7048 	}
7049 
7050 	/* WOL not supported for all but the following */
7051 	adapter->wol = 0;
7052 	switch (pdev->device) {
7053 	case IXGBE_DEV_ID_82599_SFP:
7054 		/* Only these subdevice supports WOL */
7055 		switch (pdev->subsystem_device) {
7056 		case IXGBE_SUBDEV_ID_82599_560FLR:
7057 			/* only support first port */
7058 			if (hw->bus.func != 0)
7059 				break;
7060 		case IXGBE_SUBDEV_ID_82599_SFP:
7061 			adapter->wol = IXGBE_WUFC_MAG;
7062 			break;
7063 		}
7064 		break;
7065 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7066 		/* All except this subdevice support WOL */
7067 		if (pdev->subsystem_device != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
7068 			adapter->wol = IXGBE_WUFC_MAG;
7069 		break;
7070 	case IXGBE_DEV_ID_82599_KX4:
7071 		adapter->wol = IXGBE_WUFC_MAG;
7072 		break;
7073 	case IXGBE_DEV_ID_X540T:
7074 	case IXGBE_DEV_ID_X540T1:
7075 		/* Check eeprom to see if it is enabled */
7076 		hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7077 		wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
7078 
7079 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
7080 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
7081 		     (hw->bus.func == 0)))
7082 			adapter->wol = IXGBE_WUFC_MAG;
7083 		break;
7084 	}
7085 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7086 
7087 	/* save off EEPROM version number */
7088 	hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7089 	hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7090 
7091 	/* pick up the PCI bus settings for reporting later */
7092 	hw->mac.ops.get_bus_info(hw);
7093 
7094 	/* print bus type/speed/width info */
7095 	e_dev_info("(PCI Express:%s:%s) %pM\n",
7096 		   (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7097 		    hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7098 		    "Unknown"),
7099 		   (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7100 		    hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7101 		    hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7102 		    "Unknown"),
7103 		   netdev->dev_addr);
7104 
7105 	err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7106 	if (err)
7107 		strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7108 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7109 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7110 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7111 		           part_str);
7112 	else
7113 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7114 			   hw->mac.type, hw->phy.type, part_str);
7115 
7116 	if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7117 		e_dev_warn("PCI-Express bandwidth available for this card is "
7118 			   "not sufficient for optimal performance.\n");
7119 		e_dev_warn("For optimal performance a x8 PCI-Express slot "
7120 			   "is required.\n");
7121 	}
7122 
7123 	/* reset the hardware with the new settings */
7124 	err = hw->mac.ops.start_hw(hw);
7125 	if (err == IXGBE_ERR_EEPROM_VERSION) {
7126 		/* We are running on a pre-production device, log a warning */
7127 		e_dev_warn("This device is a pre-production adapter/LOM. "
7128 			   "Please be aware there may be issues associated "
7129 			   "with your hardware.  If you are experiencing "
7130 			   "problems please contact your Intel or hardware "
7131 			   "representative who provided you with this "
7132 			   "hardware.\n");
7133 	}
7134 	strcpy(netdev->name, "eth%d");
7135 	err = register_netdev(netdev);
7136 	if (err)
7137 		goto err_register;
7138 
7139 	/* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7140 	if (hw->mac.ops.disable_tx_laser &&
7141 	    ((hw->phy.multispeed_fiber) ||
7142 	     ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7143 	      (hw->mac.type == ixgbe_mac_82599EB))))
7144 		hw->mac.ops.disable_tx_laser(hw);
7145 
7146 	/* carrier off reporting is important to ethtool even BEFORE open */
7147 	netif_carrier_off(netdev);
7148 
7149 #ifdef CONFIG_IXGBE_DCA
7150 	if (dca_add_requester(&pdev->dev) == 0) {
7151 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7152 		ixgbe_setup_dca(adapter);
7153 	}
7154 #endif
7155 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7156 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7157 		for (i = 0; i < adapter->num_vfs; i++)
7158 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
7159 	}
7160 
7161 	/* firmware requires driver version to be 0xFFFFFFFF
7162 	 * since os does not support feature
7163 	 */
7164 	if (hw->mac.ops.set_fw_drv_ver)
7165 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7166 					   0xFF);
7167 
7168 	/* add san mac addr to netdev */
7169 	ixgbe_add_sanmac_netdev(netdev);
7170 
7171 	e_dev_info("%s\n", ixgbe_default_device_descr);
7172 	cards_found++;
7173 	return 0;
7174 
7175 err_register:
7176 	ixgbe_release_hw_control(adapter);
7177 	ixgbe_clear_interrupt_scheme(adapter);
7178 err_sw_init:
7179 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7180 		ixgbe_disable_sriov(adapter);
7181 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7182 	iounmap(hw->hw_addr);
7183 err_ioremap:
7184 	free_netdev(netdev);
7185 err_alloc_etherdev:
7186 	pci_release_selected_regions(pdev,
7187 				     pci_select_bars(pdev, IORESOURCE_MEM));
7188 err_pci_reg:
7189 err_dma:
7190 	pci_disable_device(pdev);
7191 	return err;
7192 }
7193 
7194 /**
7195  * ixgbe_remove - Device Removal Routine
7196  * @pdev: PCI device information struct
7197  *
7198  * ixgbe_remove is called by the PCI subsystem to alert the driver
7199  * that it should release a PCI device.  The could be caused by a
7200  * Hot-Plug event, or because the driver is going to be removed from
7201  * memory.
7202  **/
ixgbe_remove(struct pci_dev * pdev)7203 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7204 {
7205 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7206 	struct net_device *netdev = adapter->netdev;
7207 
7208 	set_bit(__IXGBE_DOWN, &adapter->state);
7209 	cancel_work_sync(&adapter->service_task);
7210 
7211 #ifdef CONFIG_IXGBE_DCA
7212 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7213 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7214 		dca_remove_requester(&pdev->dev);
7215 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7216 	}
7217 
7218 #endif
7219 #ifdef IXGBE_FCOE
7220 	if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7221 		ixgbe_cleanup_fcoe(adapter);
7222 
7223 #endif /* IXGBE_FCOE */
7224 
7225 	/* remove the added san mac */
7226 	ixgbe_del_sanmac_netdev(netdev);
7227 
7228 	if (netdev->reg_state == NETREG_REGISTERED)
7229 		unregister_netdev(netdev);
7230 
7231 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7232 		if (!(ixgbe_check_vf_assignment(adapter)))
7233 			ixgbe_disable_sriov(adapter);
7234 		else
7235 			e_dev_warn("Unloading driver while VFs are assigned "
7236 				   "- VFs will not be deallocated\n");
7237 	}
7238 
7239 	ixgbe_clear_interrupt_scheme(adapter);
7240 
7241 	ixgbe_release_hw_control(adapter);
7242 
7243 #ifdef CONFIG_DCB
7244 	kfree(adapter->ixgbe_ieee_pfc);
7245 	kfree(adapter->ixgbe_ieee_ets);
7246 
7247 #endif
7248 	iounmap(adapter->hw.hw_addr);
7249 	pci_release_selected_regions(pdev, pci_select_bars(pdev,
7250 				     IORESOURCE_MEM));
7251 
7252 	e_dev_info("complete\n");
7253 
7254 	free_netdev(netdev);
7255 
7256 	pci_disable_pcie_error_reporting(pdev);
7257 
7258 	pci_disable_device(pdev);
7259 }
7260 
7261 /**
7262  * ixgbe_io_error_detected - called when PCI error is detected
7263  * @pdev: Pointer to PCI device
7264  * @state: The current pci connection state
7265  *
7266  * This function is called after a PCI bus error affecting
7267  * this device has been detected.
7268  */
ixgbe_io_error_detected(struct pci_dev * pdev,pci_channel_state_t state)7269 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7270 						pci_channel_state_t state)
7271 {
7272 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7273 	struct net_device *netdev = adapter->netdev;
7274 
7275 #ifdef CONFIG_PCI_IOV
7276 	struct pci_dev *bdev, *vfdev;
7277 	u32 dw0, dw1, dw2, dw3;
7278 	int vf, pos;
7279 	u16 req_id, pf_func;
7280 
7281 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7282 	    adapter->num_vfs == 0)
7283 		goto skip_bad_vf_detection;
7284 
7285 	bdev = pdev->bus->self;
7286 	while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7287 		bdev = bdev->bus->self;
7288 
7289 	if (!bdev)
7290 		goto skip_bad_vf_detection;
7291 
7292 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7293 	if (!pos)
7294 		goto skip_bad_vf_detection;
7295 
7296 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7297 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7298 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7299 	pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7300 
7301 	req_id = dw1 >> 16;
7302 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7303 	if (!(req_id & 0x0080))
7304 		goto skip_bad_vf_detection;
7305 
7306 	pf_func = req_id & 0x01;
7307 	if ((pf_func & 1) == (pdev->devfn & 1)) {
7308 		unsigned int device_id;
7309 
7310 		vf = (req_id & 0x7F) >> 1;
7311 		e_dev_err("VF %d has caused a PCIe error\n", vf);
7312 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7313 				"%8.8x\tdw3: %8.8x\n",
7314 		dw0, dw1, dw2, dw3);
7315 		switch (adapter->hw.mac.type) {
7316 		case ixgbe_mac_82599EB:
7317 			device_id = IXGBE_82599_VF_DEVICE_ID;
7318 			break;
7319 		case ixgbe_mac_X540:
7320 			device_id = IXGBE_X540_VF_DEVICE_ID;
7321 			break;
7322 		default:
7323 			device_id = 0;
7324 			break;
7325 		}
7326 
7327 		/* Find the pci device of the offending VF */
7328 		vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7329 		while (vfdev) {
7330 			if (vfdev->devfn == (req_id & 0xFF))
7331 				break;
7332 			vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7333 					       device_id, vfdev);
7334 		}
7335 		/*
7336 		 * There's a slim chance the VF could have been hot plugged,
7337 		 * so if it is no longer present we don't need to issue the
7338 		 * VFLR.  Just clean up the AER in that case.
7339 		 */
7340 		if (vfdev) {
7341 			e_dev_err("Issuing VFLR to VF %d\n", vf);
7342 			pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7343 		}
7344 
7345 		pci_cleanup_aer_uncorrect_error_status(pdev);
7346 	}
7347 
7348 	/*
7349 	 * Even though the error may have occurred on the other port
7350 	 * we still need to increment the vf error reference count for
7351 	 * both ports because the I/O resume function will be called
7352 	 * for both of them.
7353 	 */
7354 	adapter->vferr_refcount++;
7355 
7356 	return PCI_ERS_RESULT_RECOVERED;
7357 
7358 skip_bad_vf_detection:
7359 #endif /* CONFIG_PCI_IOV */
7360 	netif_device_detach(netdev);
7361 
7362 	if (state == pci_channel_io_perm_failure)
7363 		return PCI_ERS_RESULT_DISCONNECT;
7364 
7365 	if (netif_running(netdev))
7366 		ixgbe_down(adapter);
7367 	pci_disable_device(pdev);
7368 
7369 	/* Request a slot reset. */
7370 	return PCI_ERS_RESULT_NEED_RESET;
7371 }
7372 
7373 /**
7374  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7375  * @pdev: Pointer to PCI device
7376  *
7377  * Restart the card from scratch, as if from a cold-boot.
7378  */
ixgbe_io_slot_reset(struct pci_dev * pdev)7379 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7380 {
7381 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7382 	pci_ers_result_t result;
7383 	int err;
7384 
7385 	if (pci_enable_device_mem(pdev)) {
7386 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
7387 		result = PCI_ERS_RESULT_DISCONNECT;
7388 	} else {
7389 		pci_set_master(pdev);
7390 		pci_restore_state(pdev);
7391 		pci_save_state(pdev);
7392 
7393 		pci_wake_from_d3(pdev, false);
7394 
7395 		ixgbe_reset(adapter);
7396 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7397 		result = PCI_ERS_RESULT_RECOVERED;
7398 	}
7399 
7400 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7401 	if (err) {
7402 		e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7403 			  "failed 0x%0x\n", err);
7404 		/* non-fatal, continue */
7405 	}
7406 
7407 	return result;
7408 }
7409 
7410 /**
7411  * ixgbe_io_resume - called when traffic can start flowing again.
7412  * @pdev: Pointer to PCI device
7413  *
7414  * This callback is called when the error recovery driver tells us that
7415  * its OK to resume normal operation.
7416  */
ixgbe_io_resume(struct pci_dev * pdev)7417 static void ixgbe_io_resume(struct pci_dev *pdev)
7418 {
7419 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7420 	struct net_device *netdev = adapter->netdev;
7421 
7422 #ifdef CONFIG_PCI_IOV
7423 	if (adapter->vferr_refcount) {
7424 		e_info(drv, "Resuming after VF err\n");
7425 		adapter->vferr_refcount--;
7426 		return;
7427 	}
7428 
7429 #endif
7430 	if (netif_running(netdev))
7431 		ixgbe_up(adapter);
7432 
7433 	netif_device_attach(netdev);
7434 }
7435 
7436 static struct pci_error_handlers ixgbe_err_handler = {
7437 	.error_detected = ixgbe_io_error_detected,
7438 	.slot_reset = ixgbe_io_slot_reset,
7439 	.resume = ixgbe_io_resume,
7440 };
7441 
7442 static struct pci_driver ixgbe_driver = {
7443 	.name     = ixgbe_driver_name,
7444 	.id_table = ixgbe_pci_tbl,
7445 	.probe    = ixgbe_probe,
7446 	.remove   = __devexit_p(ixgbe_remove),
7447 #ifdef CONFIG_PM
7448 	.suspend  = ixgbe_suspend,
7449 	.resume   = ixgbe_resume,
7450 #endif
7451 	.shutdown = ixgbe_shutdown,
7452 	.err_handler = &ixgbe_err_handler
7453 };
7454 
7455 /**
7456  * ixgbe_init_module - Driver Registration Routine
7457  *
7458  * ixgbe_init_module is the first routine called when the driver is
7459  * loaded. All it does is register with the PCI subsystem.
7460  **/
ixgbe_init_module(void)7461 static int __init ixgbe_init_module(void)
7462 {
7463 	int ret;
7464 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7465 	pr_info("%s\n", ixgbe_copyright);
7466 
7467 #ifdef CONFIG_IXGBE_DCA
7468 	dca_register_notify(&dca_notifier);
7469 #endif
7470 
7471 	ret = pci_register_driver(&ixgbe_driver);
7472 	return ret;
7473 }
7474 
7475 module_init(ixgbe_init_module);
7476 
7477 /**
7478  * ixgbe_exit_module - Driver Exit Cleanup Routine
7479  *
7480  * ixgbe_exit_module is called just before the driver is removed
7481  * from memory.
7482  **/
ixgbe_exit_module(void)7483 static void __exit ixgbe_exit_module(void)
7484 {
7485 #ifdef CONFIG_IXGBE_DCA
7486 	dca_unregister_notify(&dca_notifier);
7487 #endif
7488 	pci_unregister_driver(&ixgbe_driver);
7489 	rcu_barrier(); /* Wait for completion of call_rcu()'s */
7490 }
7491 
7492 #ifdef CONFIG_IXGBE_DCA
ixgbe_notify_dca(struct notifier_block * nb,unsigned long event,void * p)7493 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7494 			    void *p)
7495 {
7496 	int ret_val;
7497 
7498 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7499 					 __ixgbe_notify_dca);
7500 
7501 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7502 }
7503 
7504 #endif /* CONFIG_IXGBE_DCA */
7505 
7506 module_exit(ixgbe_exit_module);
7507 
7508 /* ixgbe_main.c */
7509