f32f828b7f3dc31490179210f86c72c5968d84f7
[cascardo/linux.git] / drivers / net / ethernet / cisco / enic / enic_main.c
1 /*
2  * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * This program is free software; you may redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; version 2 of the License.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16  * SOFTWARE.
17  *
18  */
19
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/interrupt.h>
27 #include <linux/workqueue.h>
28 #include <linux/pci.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/if.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/prefetch.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ktime.h>
42
43 #include "cq_enet_desc.h"
44 #include "vnic_dev.h"
45 #include "vnic_intr.h"
46 #include "vnic_stats.h"
47 #include "vnic_vic.h"
48 #include "enic_res.h"
49 #include "enic.h"
50 #include "enic_dev.h"
51 #include "enic_pp.h"
52
53 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
54 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
55 #define MAX_TSO                         (1 << 16)
56 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
57
58 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
59 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
60 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
61
62 /* Supported devices */
63 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
64         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
65         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
66         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
67         { 0, }  /* end of table */
68 };
69
70 MODULE_DESCRIPTION(DRV_DESCRIPTION);
71 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
72 MODULE_LICENSE("GPL");
73 MODULE_VERSION(DRV_VERSION);
74 MODULE_DEVICE_TABLE(pci, enic_id_table);
75
76 #define ENIC_LARGE_PKT_THRESHOLD                1000
77 #define ENIC_MAX_COALESCE_TIMERS                10
78 /*  Interrupt moderation table, which will be used to decide the
79  *  coalescing timer values
80  *  {rx_rate in Mbps, mapping percentage of the range}
81  */
82 struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
83         {4000,  0},
84         {4400, 10},
85         {5060, 20},
86         {5230, 30},
87         {5540, 40},
88         {5820, 50},
89         {6120, 60},
90         {6435, 70},
91         {6745, 80},
92         {7000, 90},
93         {0xFFFFFFFF, 100}
94 };
95
96 /* This table helps the driver to pick different ranges for rx coalescing
97  * timer depending on the link speed.
98  */
99 struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
100         {0,  0}, /* 0  - 4  Gbps */
101         {0,  3}, /* 4  - 10 Gbps */
102         {3,  6}, /* 10 - 40 Gbps */
103 };
104
105 int enic_is_dynamic(struct enic *enic)
106 {
107         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
108 }
109
110 int enic_sriov_enabled(struct enic *enic)
111 {
112         return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
113 }
114
115 static int enic_is_sriov_vf(struct enic *enic)
116 {
117         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
118 }
119
120 int enic_is_valid_vf(struct enic *enic, int vf)
121 {
122 #ifdef CONFIG_PCI_IOV
123         return vf >= 0 && vf < enic->num_vfs;
124 #else
125         return 0;
126 #endif
127 }
128
129 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
130 {
131         struct enic *enic = vnic_dev_priv(wq->vdev);
132
133         if (buf->sop)
134                 pci_unmap_single(enic->pdev, buf->dma_addr,
135                         buf->len, PCI_DMA_TODEVICE);
136         else
137                 pci_unmap_page(enic->pdev, buf->dma_addr,
138                         buf->len, PCI_DMA_TODEVICE);
139
140         if (buf->os_buf)
141                 dev_kfree_skb_any(buf->os_buf);
142 }
143
144 static void enic_wq_free_buf(struct vnic_wq *wq,
145         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
146 {
147         enic_free_wq_buf(wq, buf);
148 }
149
150 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
151         u8 type, u16 q_number, u16 completed_index, void *opaque)
152 {
153         struct enic *enic = vnic_dev_priv(vdev);
154
155         spin_lock(&enic->wq_lock[q_number]);
156
157         vnic_wq_service(&enic->wq[q_number], cq_desc,
158                 completed_index, enic_wq_free_buf,
159                 opaque);
160
161         if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
162             vnic_wq_desc_avail(&enic->wq[q_number]) >=
163             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
164                 netif_wake_subqueue(enic->netdev, q_number);
165
166         spin_unlock(&enic->wq_lock[q_number]);
167
168         return 0;
169 }
170
171 static void enic_log_q_error(struct enic *enic)
172 {
173         unsigned int i;
174         u32 error_status;
175
176         for (i = 0; i < enic->wq_count; i++) {
177                 error_status = vnic_wq_error_status(&enic->wq[i]);
178                 if (error_status)
179                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
180                                 i, error_status);
181         }
182
183         for (i = 0; i < enic->rq_count; i++) {
184                 error_status = vnic_rq_error_status(&enic->rq[i]);
185                 if (error_status)
186                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
187                                 i, error_status);
188         }
189 }
190
191 static void enic_msglvl_check(struct enic *enic)
192 {
193         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
194
195         if (msg_enable != enic->msg_enable) {
196                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
197                         enic->msg_enable, msg_enable);
198                 enic->msg_enable = msg_enable;
199         }
200 }
201
202 static void enic_mtu_check(struct enic *enic)
203 {
204         u32 mtu = vnic_dev_mtu(enic->vdev);
205         struct net_device *netdev = enic->netdev;
206
207         if (mtu && mtu != enic->port_mtu) {
208                 enic->port_mtu = mtu;
209                 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
210                         mtu = max_t(int, ENIC_MIN_MTU,
211                                 min_t(int, ENIC_MAX_MTU, mtu));
212                         if (mtu != netdev->mtu)
213                                 schedule_work(&enic->change_mtu_work);
214                 } else {
215                         if (mtu < netdev->mtu)
216                                 netdev_warn(netdev,
217                                         "interface MTU (%d) set higher "
218                                         "than switch port MTU (%d)\n",
219                                         netdev->mtu, mtu);
220                 }
221         }
222 }
223
224 static void enic_link_check(struct enic *enic)
225 {
226         int link_status = vnic_dev_link_status(enic->vdev);
227         int carrier_ok = netif_carrier_ok(enic->netdev);
228
229         if (link_status && !carrier_ok) {
230                 netdev_info(enic->netdev, "Link UP\n");
231                 netif_carrier_on(enic->netdev);
232         } else if (!link_status && carrier_ok) {
233                 netdev_info(enic->netdev, "Link DOWN\n");
234                 netif_carrier_off(enic->netdev);
235         }
236 }
237
238 static void enic_notify_check(struct enic *enic)
239 {
240         enic_msglvl_check(enic);
241         enic_mtu_check(enic);
242         enic_link_check(enic);
243 }
244
245 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
246
247 static irqreturn_t enic_isr_legacy(int irq, void *data)
248 {
249         struct net_device *netdev = data;
250         struct enic *enic = netdev_priv(netdev);
251         unsigned int io_intr = enic_legacy_io_intr();
252         unsigned int err_intr = enic_legacy_err_intr();
253         unsigned int notify_intr = enic_legacy_notify_intr();
254         u32 pba;
255
256         vnic_intr_mask(&enic->intr[io_intr]);
257
258         pba = vnic_intr_legacy_pba(enic->legacy_pba);
259         if (!pba) {
260                 vnic_intr_unmask(&enic->intr[io_intr]);
261                 return IRQ_NONE;        /* not our interrupt */
262         }
263
264         if (ENIC_TEST_INTR(pba, notify_intr)) {
265                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
266                 enic_notify_check(enic);
267         }
268
269         if (ENIC_TEST_INTR(pba, err_intr)) {
270                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
271                 enic_log_q_error(enic);
272                 /* schedule recovery from WQ/RQ error */
273                 schedule_work(&enic->reset);
274                 return IRQ_HANDLED;
275         }
276
277         if (ENIC_TEST_INTR(pba, io_intr)) {
278                 if (napi_schedule_prep(&enic->napi[0]))
279                         __napi_schedule(&enic->napi[0]);
280         } else {
281                 vnic_intr_unmask(&enic->intr[io_intr]);
282         }
283
284         return IRQ_HANDLED;
285 }
286
287 static irqreturn_t enic_isr_msi(int irq, void *data)
288 {
289         struct enic *enic = data;
290
291         /* With MSI, there is no sharing of interrupts, so this is
292          * our interrupt and there is no need to ack it.  The device
293          * is not providing per-vector masking, so the OS will not
294          * write to PCI config space to mask/unmask the interrupt.
295          * We're using mask_on_assertion for MSI, so the device
296          * automatically masks the interrupt when the interrupt is
297          * generated.  Later, when exiting polling, the interrupt
298          * will be unmasked (see enic_poll).
299          *
300          * Also, the device uses the same PCIe Traffic Class (TC)
301          * for Memory Write data and MSI, so there are no ordering
302          * issues; the MSI will always arrive at the Root Complex
303          * _after_ corresponding Memory Writes (i.e. descriptor
304          * writes).
305          */
306
307         napi_schedule(&enic->napi[0]);
308
309         return IRQ_HANDLED;
310 }
311
312 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
313 {
314         struct napi_struct *napi = data;
315
316         /* schedule NAPI polling for RQ cleanup */
317         napi_schedule(napi);
318
319         return IRQ_HANDLED;
320 }
321
322 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
323 {
324         struct enic *enic = data;
325         unsigned int cq;
326         unsigned int intr;
327         unsigned int wq_work_to_do = -1; /* no limit */
328         unsigned int wq_work_done;
329         unsigned int wq_irq;
330
331         wq_irq = (u32)irq - enic->msix_entry[enic_msix_wq_intr(enic, 0)].vector;
332         cq = enic_cq_wq(enic, wq_irq);
333         intr = enic_msix_wq_intr(enic, wq_irq);
334
335         wq_work_done = vnic_cq_service(&enic->cq[cq],
336                 wq_work_to_do, enic_wq_service, NULL);
337
338         vnic_intr_return_credits(&enic->intr[intr],
339                 wq_work_done,
340                 1 /* unmask intr */,
341                 1 /* reset intr timer */);
342
343         return IRQ_HANDLED;
344 }
345
346 static irqreturn_t enic_isr_msix_err(int irq, void *data)
347 {
348         struct enic *enic = data;
349         unsigned int intr = enic_msix_err_intr(enic);
350
351         vnic_intr_return_all_credits(&enic->intr[intr]);
352
353         enic_log_q_error(enic);
354
355         /* schedule recovery from WQ/RQ error */
356         schedule_work(&enic->reset);
357
358         return IRQ_HANDLED;
359 }
360
361 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
362 {
363         struct enic *enic = data;
364         unsigned int intr = enic_msix_notify_intr(enic);
365
366         vnic_intr_return_all_credits(&enic->intr[intr]);
367         enic_notify_check(enic);
368
369         return IRQ_HANDLED;
370 }
371
372 static inline void enic_queue_wq_skb_cont(struct enic *enic,
373         struct vnic_wq *wq, struct sk_buff *skb,
374         unsigned int len_left, int loopback)
375 {
376         const skb_frag_t *frag;
377
378         /* Queue additional data fragments */
379         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
380                 len_left -= skb_frag_size(frag);
381                 enic_queue_wq_desc_cont(wq, skb,
382                         skb_frag_dma_map(&enic->pdev->dev,
383                                          frag, 0, skb_frag_size(frag),
384                                          DMA_TO_DEVICE),
385                         skb_frag_size(frag),
386                         (len_left == 0),        /* EOP? */
387                         loopback);
388         }
389 }
390
391 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
392         struct vnic_wq *wq, struct sk_buff *skb,
393         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
394 {
395         unsigned int head_len = skb_headlen(skb);
396         unsigned int len_left = skb->len - head_len;
397         int eop = (len_left == 0);
398
399         /* Queue the main skb fragment. The fragments are no larger
400          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
401          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
402          * per fragment is queued.
403          */
404         enic_queue_wq_desc(wq, skb,
405                 pci_map_single(enic->pdev, skb->data,
406                         head_len, PCI_DMA_TODEVICE),
407                 head_len,
408                 vlan_tag_insert, vlan_tag,
409                 eop, loopback);
410
411         if (!eop)
412                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
413 }
414
415 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
416         struct vnic_wq *wq, struct sk_buff *skb,
417         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
418 {
419         unsigned int head_len = skb_headlen(skb);
420         unsigned int len_left = skb->len - head_len;
421         unsigned int hdr_len = skb_checksum_start_offset(skb);
422         unsigned int csum_offset = hdr_len + skb->csum_offset;
423         int eop = (len_left == 0);
424
425         /* Queue the main skb fragment. The fragments are no larger
426          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
427          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
428          * per fragment is queued.
429          */
430         enic_queue_wq_desc_csum_l4(wq, skb,
431                 pci_map_single(enic->pdev, skb->data,
432                         head_len, PCI_DMA_TODEVICE),
433                 head_len,
434                 csum_offset,
435                 hdr_len,
436                 vlan_tag_insert, vlan_tag,
437                 eop, loopback);
438
439         if (!eop)
440                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
441 }
442
443 static inline void enic_queue_wq_skb_tso(struct enic *enic,
444         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
445         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
446 {
447         unsigned int frag_len_left = skb_headlen(skb);
448         unsigned int len_left = skb->len - frag_len_left;
449         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
450         int eop = (len_left == 0);
451         unsigned int len;
452         dma_addr_t dma_addr;
453         unsigned int offset = 0;
454         skb_frag_t *frag;
455
456         /* Preload TCP csum field with IP pseudo hdr calculated
457          * with IP length set to zero.  HW will later add in length
458          * to each TCP segment resulting from the TSO.
459          */
460
461         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
462                 ip_hdr(skb)->check = 0;
463                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
464                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
465         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
466                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
467                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
468         }
469
470         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
471          * for the main skb fragment
472          */
473         while (frag_len_left) {
474                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
475                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
476                                 len, PCI_DMA_TODEVICE);
477                 enic_queue_wq_desc_tso(wq, skb,
478                         dma_addr,
479                         len,
480                         mss, hdr_len,
481                         vlan_tag_insert, vlan_tag,
482                         eop && (len == frag_len_left), loopback);
483                 frag_len_left -= len;
484                 offset += len;
485         }
486
487         if (eop)
488                 return;
489
490         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
491          * for additional data fragments
492          */
493         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
494                 len_left -= skb_frag_size(frag);
495                 frag_len_left = skb_frag_size(frag);
496                 offset = 0;
497
498                 while (frag_len_left) {
499                         len = min(frag_len_left,
500                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
501                         dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
502                                                     offset, len,
503                                                     DMA_TO_DEVICE);
504                         enic_queue_wq_desc_cont(wq, skb,
505                                 dma_addr,
506                                 len,
507                                 (len_left == 0) &&
508                                 (len == frag_len_left),         /* EOP? */
509                                 loopback);
510                         frag_len_left -= len;
511                         offset += len;
512                 }
513         }
514 }
515
516 static inline void enic_queue_wq_skb(struct enic *enic,
517         struct vnic_wq *wq, struct sk_buff *skb)
518 {
519         unsigned int mss = skb_shinfo(skb)->gso_size;
520         unsigned int vlan_tag = 0;
521         int vlan_tag_insert = 0;
522         int loopback = 0;
523
524         if (vlan_tx_tag_present(skb)) {
525                 /* VLAN tag from trunking driver */
526                 vlan_tag_insert = 1;
527                 vlan_tag = vlan_tx_tag_get(skb);
528         } else if (enic->loop_enable) {
529                 vlan_tag = enic->loop_tag;
530                 loopback = 1;
531         }
532
533         if (mss)
534                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
535                         vlan_tag_insert, vlan_tag, loopback);
536         else if (skb->ip_summed == CHECKSUM_PARTIAL)
537                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
538                         vlan_tag_insert, vlan_tag, loopback);
539         else
540                 enic_queue_wq_skb_vlan(enic, wq, skb,
541                         vlan_tag_insert, vlan_tag, loopback);
542 }
543
544 /* netif_tx_lock held, process context with BHs disabled, or BH */
545 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
546         struct net_device *netdev)
547 {
548         struct enic *enic = netdev_priv(netdev);
549         struct vnic_wq *wq;
550         unsigned long flags;
551         unsigned int txq_map;
552
553         if (skb->len <= 0) {
554                 dev_kfree_skb_any(skb);
555                 return NETDEV_TX_OK;
556         }
557
558         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
559         wq = &enic->wq[txq_map];
560
561         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
562          * which is very likely.  In the off chance it's going to take
563          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
564          */
565
566         if (skb_shinfo(skb)->gso_size == 0 &&
567             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
568             skb_linearize(skb)) {
569                 dev_kfree_skb_any(skb);
570                 return NETDEV_TX_OK;
571         }
572
573         spin_lock_irqsave(&enic->wq_lock[txq_map], flags);
574
575         if (vnic_wq_desc_avail(wq) <
576             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
577                 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
578                 /* This is a hard error, log it */
579                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
580                 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
581                 return NETDEV_TX_BUSY;
582         }
583
584         enic_queue_wq_skb(enic, wq, skb);
585
586         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
587                 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
588
589         spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
590
591         return NETDEV_TX_OK;
592 }
593
594 /* dev_base_lock rwlock held, nominally process context */
595 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
596                                                 struct rtnl_link_stats64 *net_stats)
597 {
598         struct enic *enic = netdev_priv(netdev);
599         struct vnic_stats *stats;
600
601         enic_dev_stats_dump(enic, &stats);
602
603         net_stats->tx_packets = stats->tx.tx_frames_ok;
604         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
605         net_stats->tx_errors = stats->tx.tx_errors;
606         net_stats->tx_dropped = stats->tx.tx_drops;
607
608         net_stats->rx_packets = stats->rx.rx_frames_ok;
609         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
610         net_stats->rx_errors = stats->rx.rx_errors;
611         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
612         net_stats->rx_over_errors = enic->rq_truncated_pkts;
613         net_stats->rx_crc_errors = enic->rq_bad_fcs;
614         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
615
616         return net_stats;
617 }
618
619 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
620 {
621         struct enic *enic = netdev_priv(netdev);
622
623         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
624                 unsigned int mc_count = netdev_mc_count(netdev);
625
626                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
627                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
628
629                 return -ENOSPC;
630         }
631
632         enic_dev_add_addr(enic, mc_addr);
633         enic->mc_count++;
634
635         return 0;
636 }
637
638 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
639 {
640         struct enic *enic = netdev_priv(netdev);
641
642         enic_dev_del_addr(enic, mc_addr);
643         enic->mc_count--;
644
645         return 0;
646 }
647
648 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
649 {
650         struct enic *enic = netdev_priv(netdev);
651
652         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
653                 unsigned int uc_count = netdev_uc_count(netdev);
654
655                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
656                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
657
658                 return -ENOSPC;
659         }
660
661         enic_dev_add_addr(enic, uc_addr);
662         enic->uc_count++;
663
664         return 0;
665 }
666
667 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
668 {
669         struct enic *enic = netdev_priv(netdev);
670
671         enic_dev_del_addr(enic, uc_addr);
672         enic->uc_count--;
673
674         return 0;
675 }
676
677 void enic_reset_addr_lists(struct enic *enic)
678 {
679         struct net_device *netdev = enic->netdev;
680
681         __dev_uc_unsync(netdev, NULL);
682         __dev_mc_unsync(netdev, NULL);
683
684         enic->mc_count = 0;
685         enic->uc_count = 0;
686         enic->flags = 0;
687 }
688
689 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
690 {
691         struct enic *enic = netdev_priv(netdev);
692
693         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
694                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
695                         return -EADDRNOTAVAIL;
696         } else {
697                 if (!is_valid_ether_addr(addr))
698                         return -EADDRNOTAVAIL;
699         }
700
701         memcpy(netdev->dev_addr, addr, netdev->addr_len);
702
703         return 0;
704 }
705
706 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
707 {
708         struct enic *enic = netdev_priv(netdev);
709         struct sockaddr *saddr = p;
710         char *addr = saddr->sa_data;
711         int err;
712
713         if (netif_running(enic->netdev)) {
714                 err = enic_dev_del_station_addr(enic);
715                 if (err)
716                         return err;
717         }
718
719         err = enic_set_mac_addr(netdev, addr);
720         if (err)
721                 return err;
722
723         if (netif_running(enic->netdev)) {
724                 err = enic_dev_add_station_addr(enic);
725                 if (err)
726                         return err;
727         }
728
729         return err;
730 }
731
732 static int enic_set_mac_address(struct net_device *netdev, void *p)
733 {
734         struct sockaddr *saddr = p;
735         char *addr = saddr->sa_data;
736         struct enic *enic = netdev_priv(netdev);
737         int err;
738
739         err = enic_dev_del_station_addr(enic);
740         if (err)
741                 return err;
742
743         err = enic_set_mac_addr(netdev, addr);
744         if (err)
745                 return err;
746
747         return enic_dev_add_station_addr(enic);
748 }
749
750 /* netif_tx_lock held, BHs disabled */
751 static void enic_set_rx_mode(struct net_device *netdev)
752 {
753         struct enic *enic = netdev_priv(netdev);
754         int directed = 1;
755         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
756         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
757         int promisc = (netdev->flags & IFF_PROMISC) ||
758                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
759         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
760                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
761         unsigned int flags = netdev->flags |
762                 (allmulti ? IFF_ALLMULTI : 0) |
763                 (promisc ? IFF_PROMISC : 0);
764
765         if (enic->flags != flags) {
766                 enic->flags = flags;
767                 enic_dev_packet_filter(enic, directed,
768                         multicast, broadcast, promisc, allmulti);
769         }
770
771         if (!promisc) {
772                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
773                 if (!allmulti)
774                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
775         }
776 }
777
778 /* netif_tx_lock held, BHs disabled */
779 static void enic_tx_timeout(struct net_device *netdev)
780 {
781         struct enic *enic = netdev_priv(netdev);
782         schedule_work(&enic->reset);
783 }
784
785 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
786 {
787         struct enic *enic = netdev_priv(netdev);
788         struct enic_port_profile *pp;
789         int err;
790
791         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
792         if (err)
793                 return err;
794
795         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
796                 if (vf == PORT_SELF_VF) {
797                         memcpy(pp->vf_mac, mac, ETH_ALEN);
798                         return 0;
799                 } else {
800                         /*
801                          * For sriov vf's set the mac in hw
802                          */
803                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
804                                 vnic_dev_set_mac_addr, mac);
805                         return enic_dev_status_to_errno(err);
806                 }
807         } else
808                 return -EINVAL;
809 }
810
811 static int enic_set_vf_port(struct net_device *netdev, int vf,
812         struct nlattr *port[])
813 {
814         struct enic *enic = netdev_priv(netdev);
815         struct enic_port_profile prev_pp;
816         struct enic_port_profile *pp;
817         int err = 0, restore_pp = 1;
818
819         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
820         if (err)
821                 return err;
822
823         if (!port[IFLA_PORT_REQUEST])
824                 return -EOPNOTSUPP;
825
826         memcpy(&prev_pp, pp, sizeof(*enic->pp));
827         memset(pp, 0, sizeof(*enic->pp));
828
829         pp->set |= ENIC_SET_REQUEST;
830         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
831
832         if (port[IFLA_PORT_PROFILE]) {
833                 pp->set |= ENIC_SET_NAME;
834                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
835                         PORT_PROFILE_MAX);
836         }
837
838         if (port[IFLA_PORT_INSTANCE_UUID]) {
839                 pp->set |= ENIC_SET_INSTANCE;
840                 memcpy(pp->instance_uuid,
841                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
842         }
843
844         if (port[IFLA_PORT_HOST_UUID]) {
845                 pp->set |= ENIC_SET_HOST;
846                 memcpy(pp->host_uuid,
847                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
848         }
849
850         if (vf == PORT_SELF_VF) {
851                 /* Special case handling: mac came from IFLA_VF_MAC */
852                 if (!is_zero_ether_addr(prev_pp.vf_mac))
853                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
854
855                 if (is_zero_ether_addr(netdev->dev_addr))
856                         eth_hw_addr_random(netdev);
857         } else {
858                 /* SR-IOV VF: get mac from adapter */
859                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
860                         vnic_dev_get_mac_addr, pp->mac_addr);
861                 if (err) {
862                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
863                         memcpy(pp, &prev_pp, sizeof(*pp));
864                         return enic_dev_status_to_errno(err);
865                 }
866         }
867
868         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
869         if (err) {
870                 if (restore_pp) {
871                         /* Things are still the way they were: Implicit
872                          * DISASSOCIATE failed
873                          */
874                         memcpy(pp, &prev_pp, sizeof(*pp));
875                 } else {
876                         memset(pp, 0, sizeof(*pp));
877                         if (vf == PORT_SELF_VF)
878                                 memset(netdev->dev_addr, 0, ETH_ALEN);
879                 }
880         } else {
881                 /* Set flag to indicate that the port assoc/disassoc
882                  * request has been sent out to fw
883                  */
884                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
885
886                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
887                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
888                         memset(pp->mac_addr, 0, ETH_ALEN);
889                         if (vf == PORT_SELF_VF)
890                                 memset(netdev->dev_addr, 0, ETH_ALEN);
891                 }
892         }
893
894         if (vf == PORT_SELF_VF)
895                 memset(pp->vf_mac, 0, ETH_ALEN);
896
897         return err;
898 }
899
900 static int enic_get_vf_port(struct net_device *netdev, int vf,
901         struct sk_buff *skb)
902 {
903         struct enic *enic = netdev_priv(netdev);
904         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
905         struct enic_port_profile *pp;
906         int err;
907
908         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
909         if (err)
910                 return err;
911
912         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
913                 return -ENODATA;
914
915         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
916         if (err)
917                 return err;
918
919         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
920             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
921             ((pp->set & ENIC_SET_NAME) &&
922              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
923             ((pp->set & ENIC_SET_INSTANCE) &&
924              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
925                      pp->instance_uuid)) ||
926             ((pp->set & ENIC_SET_HOST) &&
927              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
928                 goto nla_put_failure;
929         return 0;
930
931 nla_put_failure:
932         return -EMSGSIZE;
933 }
934
935 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
936 {
937         struct enic *enic = vnic_dev_priv(rq->vdev);
938
939         if (!buf->os_buf)
940                 return;
941
942         pci_unmap_single(enic->pdev, buf->dma_addr,
943                 buf->len, PCI_DMA_FROMDEVICE);
944         dev_kfree_skb_any(buf->os_buf);
945 }
946
947 static int enic_rq_alloc_buf(struct vnic_rq *rq)
948 {
949         struct enic *enic = vnic_dev_priv(rq->vdev);
950         struct net_device *netdev = enic->netdev;
951         struct sk_buff *skb;
952         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
953         unsigned int os_buf_index = 0;
954         dma_addr_t dma_addr;
955
956         skb = netdev_alloc_skb_ip_align(netdev, len);
957         if (!skb)
958                 return -ENOMEM;
959
960         dma_addr = pci_map_single(enic->pdev, skb->data,
961                 len, PCI_DMA_FROMDEVICE);
962
963         enic_queue_rq_desc(rq, skb, os_buf_index,
964                 dma_addr, len);
965
966         return 0;
967 }
968
969 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
970                                       u32 pkt_len)
971 {
972         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
973                 pkt_size->large_pkt_bytes_cnt += pkt_len;
974         else
975                 pkt_size->small_pkt_bytes_cnt += pkt_len;
976 }
977
978 static void enic_rq_indicate_buf(struct vnic_rq *rq,
979         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
980         int skipped, void *opaque)
981 {
982         struct enic *enic = vnic_dev_priv(rq->vdev);
983         struct net_device *netdev = enic->netdev;
984         struct sk_buff *skb;
985         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
986
987         u8 type, color, eop, sop, ingress_port, vlan_stripped;
988         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
989         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
990         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
991         u8 packet_error;
992         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
993         u32 rss_hash;
994
995         if (skipped)
996                 return;
997
998         skb = buf->os_buf;
999         prefetch(skb->data - NET_IP_ALIGN);
1000         pci_unmap_single(enic->pdev, buf->dma_addr,
1001                 buf->len, PCI_DMA_FROMDEVICE);
1002
1003         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1004                 &type, &color, &q_number, &completed_index,
1005                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1006                 &csum_not_calc, &rss_hash, &bytes_written,
1007                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1008                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1009                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1010                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1011                 &fcs_ok);
1012
1013         if (packet_error) {
1014
1015                 if (!fcs_ok) {
1016                         if (bytes_written > 0)
1017                                 enic->rq_bad_fcs++;
1018                         else if (bytes_written == 0)
1019                                 enic->rq_truncated_pkts++;
1020                 }
1021
1022                 dev_kfree_skb_any(skb);
1023
1024                 return;
1025         }
1026
1027         if (eop && bytes_written > 0) {
1028
1029                 /* Good receive
1030                  */
1031
1032                 skb_put(skb, bytes_written);
1033                 skb->protocol = eth_type_trans(skb, netdev);
1034                 skb_record_rx_queue(skb, q_number);
1035                 if (netdev->features & NETIF_F_RXHASH) {
1036                         skb_set_hash(skb, rss_hash,
1037                                      (rss_type &
1038                                       (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1039                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1040                                        NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1041                                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1042                 }
1043
1044                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1045                         skb->csum = htons(checksum);
1046                         skb->ip_summed = CHECKSUM_COMPLETE;
1047                 }
1048
1049                 if (vlan_stripped)
1050                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1051
1052                 if (netdev->features & NETIF_F_GRO)
1053                         napi_gro_receive(&enic->napi[q_number], skb);
1054                 else
1055                         netif_receive_skb(skb);
1056                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1057                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1058                                                   bytes_written);
1059         } else {
1060
1061                 /* Buffer overflow
1062                  */
1063
1064                 dev_kfree_skb_any(skb);
1065         }
1066 }
1067
1068 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1069         u8 type, u16 q_number, u16 completed_index, void *opaque)
1070 {
1071         struct enic *enic = vnic_dev_priv(vdev);
1072
1073         vnic_rq_service(&enic->rq[q_number], cq_desc,
1074                 completed_index, VNIC_RQ_RETURN_DESC,
1075                 enic_rq_indicate_buf, opaque);
1076
1077         return 0;
1078 }
1079
1080 static int enic_poll(struct napi_struct *napi, int budget)
1081 {
1082         struct net_device *netdev = napi->dev;
1083         struct enic *enic = netdev_priv(netdev);
1084         unsigned int cq_rq = enic_cq_rq(enic, 0);
1085         unsigned int cq_wq = enic_cq_wq(enic, 0);
1086         unsigned int intr = enic_legacy_io_intr();
1087         unsigned int rq_work_to_do = budget;
1088         unsigned int wq_work_to_do = -1; /* no limit */
1089         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1090         int err;
1091
1092         /* Service RQ (first) and WQ
1093          */
1094
1095         if (budget > 0)
1096                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1097                         rq_work_to_do, enic_rq_service, NULL);
1098
1099         wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1100                 wq_work_to_do, enic_wq_service, NULL);
1101
1102         /* Accumulate intr event credits for this polling
1103          * cycle.  An intr event is the completion of a
1104          * a WQ or RQ packet.
1105          */
1106
1107         work_done = rq_work_done + wq_work_done;
1108
1109         if (work_done > 0)
1110                 vnic_intr_return_credits(&enic->intr[intr],
1111                         work_done,
1112                         0 /* don't unmask intr */,
1113                         0 /* don't reset intr timer */);
1114
1115         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1116
1117         /* Buffer allocation failed. Stay in polling
1118          * mode so we can try to fill the ring again.
1119          */
1120
1121         if (err)
1122                 rq_work_done = rq_work_to_do;
1123
1124         if (rq_work_done < rq_work_to_do) {
1125
1126                 /* Some work done, but not enough to stay in polling,
1127                  * exit polling
1128                  */
1129
1130                 napi_complete(napi);
1131                 vnic_intr_unmask(&enic->intr[intr]);
1132         }
1133
1134         return rq_work_done;
1135 }
1136
1137 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1138 {
1139         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1140         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1141         u32 timer = cq->tobe_rx_coal_timeval;
1142
1143         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1144                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1145                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1146         }
1147 }
1148
1149 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1150 {
1151         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1152         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1153         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1154         int index;
1155         u32 timer;
1156         u32 range_start;
1157         u32 traffic;
1158         u64 delta;
1159         ktime_t now = ktime_get();
1160
1161         delta = ktime_us_delta(now, cq->prev_ts);
1162         if (delta < ENIC_AIC_TS_BREAK)
1163                 return;
1164         cq->prev_ts = now;
1165
1166         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1167                   pkt_size_counter->small_pkt_bytes_cnt;
1168         /* The table takes Mbps
1169          * traffic *= 8    => bits
1170          * traffic *= (10^6 / delta)    => bps
1171          * traffic /= 10^6     => Mbps
1172          *
1173          * Combining, traffic *= (8 / delta)
1174          */
1175
1176         traffic <<= 3;
1177         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1178
1179         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1180                 if (traffic < mod_table[index].rx_rate)
1181                         break;
1182         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1183                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1184                       rx_coal->small_pkt_range_start :
1185                       rx_coal->large_pkt_range_start;
1186         timer = range_start + ((rx_coal->range_end - range_start) *
1187                                mod_table[index].range_percent / 100);
1188         /* Damping */
1189         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1190
1191         pkt_size_counter->large_pkt_bytes_cnt = 0;
1192         pkt_size_counter->small_pkt_bytes_cnt = 0;
1193 }
1194
1195 static int enic_poll_msix(struct napi_struct *napi, int budget)
1196 {
1197         struct net_device *netdev = napi->dev;
1198         struct enic *enic = netdev_priv(netdev);
1199         unsigned int rq = (napi - &enic->napi[0]);
1200         unsigned int cq = enic_cq_rq(enic, rq);
1201         unsigned int intr = enic_msix_rq_intr(enic, rq);
1202         unsigned int work_to_do = budget;
1203         unsigned int work_done = 0;
1204         int err;
1205
1206         /* Service RQ
1207          */
1208
1209         if (budget > 0)
1210                 work_done = vnic_cq_service(&enic->cq[cq],
1211                         work_to_do, enic_rq_service, NULL);
1212
1213         /* Return intr event credits for this polling
1214          * cycle.  An intr event is the completion of a
1215          * RQ packet.
1216          */
1217
1218         if (work_done > 0)
1219                 vnic_intr_return_credits(&enic->intr[intr],
1220                         work_done,
1221                         0 /* don't unmask intr */,
1222                         0 /* don't reset intr timer */);
1223
1224         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1225
1226         /* Buffer allocation failed. Stay in polling mode
1227          * so we can try to fill the ring again.
1228          */
1229
1230         if (err)
1231                 work_done = work_to_do;
1232         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1233                 /* Call the function which refreshes
1234                  * the intr coalescing timer value based on
1235                  * the traffic.  This is supported only in
1236                  * the case of MSI-x mode
1237                  */
1238                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1239
1240         if (work_done < work_to_do) {
1241
1242                 /* Some work done, but not enough to stay in polling,
1243                  * exit polling
1244                  */
1245
1246                 napi_complete(napi);
1247                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1248                         enic_set_int_moderation(enic, &enic->rq[rq]);
1249                 vnic_intr_unmask(&enic->intr[intr]);
1250         }
1251
1252         return work_done;
1253 }
1254
1255 static void enic_notify_timer(unsigned long data)
1256 {
1257         struct enic *enic = (struct enic *)data;
1258
1259         enic_notify_check(enic);
1260
1261         mod_timer(&enic->notify_timer,
1262                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1263 }
1264
1265 static void enic_free_intr(struct enic *enic)
1266 {
1267         struct net_device *netdev = enic->netdev;
1268         unsigned int i;
1269
1270         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1271         case VNIC_DEV_INTR_MODE_INTX:
1272                 free_irq(enic->pdev->irq, netdev);
1273                 break;
1274         case VNIC_DEV_INTR_MODE_MSI:
1275                 free_irq(enic->pdev->irq, enic);
1276                 break;
1277         case VNIC_DEV_INTR_MODE_MSIX:
1278                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1279                         if (enic->msix[i].requested)
1280                                 free_irq(enic->msix_entry[i].vector,
1281                                         enic->msix[i].devid);
1282                 break;
1283         default:
1284                 break;
1285         }
1286 }
1287
1288 static int enic_request_intr(struct enic *enic)
1289 {
1290         struct net_device *netdev = enic->netdev;
1291         unsigned int i, intr;
1292         int err = 0;
1293
1294         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1295
1296         case VNIC_DEV_INTR_MODE_INTX:
1297
1298                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1299                         IRQF_SHARED, netdev->name, netdev);
1300                 break;
1301
1302         case VNIC_DEV_INTR_MODE_MSI:
1303
1304                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1305                         0, netdev->name, enic);
1306                 break;
1307
1308         case VNIC_DEV_INTR_MODE_MSIX:
1309
1310                 for (i = 0; i < enic->rq_count; i++) {
1311                         intr = enic_msix_rq_intr(enic, i);
1312                         snprintf(enic->msix[intr].devname,
1313                                 sizeof(enic->msix[intr].devname),
1314                                 "%.11s-rx-%d", netdev->name, i);
1315                         enic->msix[intr].isr = enic_isr_msix_rq;
1316                         enic->msix[intr].devid = &enic->napi[i];
1317                 }
1318
1319                 for (i = 0; i < enic->wq_count; i++) {
1320                         intr = enic_msix_wq_intr(enic, i);
1321                         snprintf(enic->msix[intr].devname,
1322                                 sizeof(enic->msix[intr].devname),
1323                                 "%.11s-tx-%d", netdev->name, i);
1324                         enic->msix[intr].isr = enic_isr_msix_wq;
1325                         enic->msix[intr].devid = enic;
1326                 }
1327
1328                 intr = enic_msix_err_intr(enic);
1329                 snprintf(enic->msix[intr].devname,
1330                         sizeof(enic->msix[intr].devname),
1331                         "%.11s-err", netdev->name);
1332                 enic->msix[intr].isr = enic_isr_msix_err;
1333                 enic->msix[intr].devid = enic;
1334
1335                 intr = enic_msix_notify_intr(enic);
1336                 snprintf(enic->msix[intr].devname,
1337                         sizeof(enic->msix[intr].devname),
1338                         "%.11s-notify", netdev->name);
1339                 enic->msix[intr].isr = enic_isr_msix_notify;
1340                 enic->msix[intr].devid = enic;
1341
1342                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1343                         enic->msix[i].requested = 0;
1344
1345                 for (i = 0; i < enic->intr_count; i++) {
1346                         err = request_irq(enic->msix_entry[i].vector,
1347                                 enic->msix[i].isr, 0,
1348                                 enic->msix[i].devname,
1349                                 enic->msix[i].devid);
1350                         if (err) {
1351                                 enic_free_intr(enic);
1352                                 break;
1353                         }
1354                         enic->msix[i].requested = 1;
1355                 }
1356
1357                 break;
1358
1359         default:
1360                 break;
1361         }
1362
1363         return err;
1364 }
1365
1366 static void enic_synchronize_irqs(struct enic *enic)
1367 {
1368         unsigned int i;
1369
1370         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1371         case VNIC_DEV_INTR_MODE_INTX:
1372         case VNIC_DEV_INTR_MODE_MSI:
1373                 synchronize_irq(enic->pdev->irq);
1374                 break;
1375         case VNIC_DEV_INTR_MODE_MSIX:
1376                 for (i = 0; i < enic->intr_count; i++)
1377                         synchronize_irq(enic->msix_entry[i].vector);
1378                 break;
1379         default:
1380                 break;
1381         }
1382 }
1383
1384 static void enic_set_rx_coal_setting(struct enic *enic)
1385 {
1386         unsigned int speed;
1387         int index = -1;
1388         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1389
1390         /* If intr mode is not MSIX, do not do adaptive coalescing */
1391         if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1392                 netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1393                 return;
1394         }
1395
1396         /* 1. Read the link speed from fw
1397          * 2. Pick the default range for the speed
1398          * 3. Update it in enic->rx_coalesce_setting
1399          */
1400         speed = vnic_dev_port_speed(enic->vdev);
1401         if (ENIC_LINK_SPEED_10G < speed)
1402                 index = ENIC_LINK_40G_INDEX;
1403         else if (ENIC_LINK_SPEED_4G < speed)
1404                 index = ENIC_LINK_10G_INDEX;
1405         else
1406                 index = ENIC_LINK_4G_INDEX;
1407
1408         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1409         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1410         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1411
1412         /* Start with the value provided by UCSM */
1413         for (index = 0; index < enic->rq_count; index++)
1414                 enic->cq[index].cur_rx_coal_timeval =
1415                                 enic->config.intr_timer_usec;
1416
1417         rx_coal->use_adaptive_rx_coalesce = 1;
1418 }
1419
1420 static int enic_dev_notify_set(struct enic *enic)
1421 {
1422         int err;
1423
1424         spin_lock(&enic->devcmd_lock);
1425         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1426         case VNIC_DEV_INTR_MODE_INTX:
1427                 err = vnic_dev_notify_set(enic->vdev,
1428                         enic_legacy_notify_intr());
1429                 break;
1430         case VNIC_DEV_INTR_MODE_MSIX:
1431                 err = vnic_dev_notify_set(enic->vdev,
1432                         enic_msix_notify_intr(enic));
1433                 break;
1434         default:
1435                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1436                 break;
1437         }
1438         spin_unlock(&enic->devcmd_lock);
1439
1440         return err;
1441 }
1442
1443 static void enic_notify_timer_start(struct enic *enic)
1444 {
1445         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1446         case VNIC_DEV_INTR_MODE_MSI:
1447                 mod_timer(&enic->notify_timer, jiffies);
1448                 break;
1449         default:
1450                 /* Using intr for notification for INTx/MSI-X */
1451                 break;
1452         }
1453 }
1454
1455 /* rtnl lock is held, process context */
1456 static int enic_open(struct net_device *netdev)
1457 {
1458         struct enic *enic = netdev_priv(netdev);
1459         unsigned int i;
1460         int err;
1461
1462         err = enic_request_intr(enic);
1463         if (err) {
1464                 netdev_err(netdev, "Unable to request irq.\n");
1465                 return err;
1466         }
1467
1468         err = enic_dev_notify_set(enic);
1469         if (err) {
1470                 netdev_err(netdev,
1471                         "Failed to alloc notify buffer, aborting.\n");
1472                 goto err_out_free_intr;
1473         }
1474
1475         for (i = 0; i < enic->rq_count; i++) {
1476                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1477                 /* Need at least one buffer on ring to get going */
1478                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1479                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1480                         err = -ENOMEM;
1481                         goto err_out_notify_unset;
1482                 }
1483         }
1484
1485         for (i = 0; i < enic->wq_count; i++)
1486                 vnic_wq_enable(&enic->wq[i]);
1487         for (i = 0; i < enic->rq_count; i++)
1488                 vnic_rq_enable(&enic->rq[i]);
1489
1490         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1491                 enic_dev_add_station_addr(enic);
1492
1493         enic_set_rx_mode(netdev);
1494
1495         netif_tx_wake_all_queues(netdev);
1496
1497         for (i = 0; i < enic->rq_count; i++)
1498                 napi_enable(&enic->napi[i]);
1499
1500         enic_dev_enable(enic);
1501
1502         for (i = 0; i < enic->intr_count; i++)
1503                 vnic_intr_unmask(&enic->intr[i]);
1504
1505         enic_notify_timer_start(enic);
1506
1507         return 0;
1508
1509 err_out_notify_unset:
1510         enic_dev_notify_unset(enic);
1511 err_out_free_intr:
1512         enic_free_intr(enic);
1513
1514         return err;
1515 }
1516
1517 /* rtnl lock is held, process context */
1518 static int enic_stop(struct net_device *netdev)
1519 {
1520         struct enic *enic = netdev_priv(netdev);
1521         unsigned int i;
1522         int err;
1523
1524         for (i = 0; i < enic->intr_count; i++) {
1525                 vnic_intr_mask(&enic->intr[i]);
1526                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1527         }
1528
1529         enic_synchronize_irqs(enic);
1530
1531         del_timer_sync(&enic->notify_timer);
1532
1533         enic_dev_disable(enic);
1534
1535         for (i = 0; i < enic->rq_count; i++)
1536                 napi_disable(&enic->napi[i]);
1537
1538         netif_carrier_off(netdev);
1539         netif_tx_disable(netdev);
1540
1541         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1542                 enic_dev_del_station_addr(enic);
1543
1544         for (i = 0; i < enic->wq_count; i++) {
1545                 err = vnic_wq_disable(&enic->wq[i]);
1546                 if (err)
1547                         return err;
1548         }
1549         for (i = 0; i < enic->rq_count; i++) {
1550                 err = vnic_rq_disable(&enic->rq[i]);
1551                 if (err)
1552                         return err;
1553         }
1554
1555         enic_dev_notify_unset(enic);
1556         enic_free_intr(enic);
1557
1558         for (i = 0; i < enic->wq_count; i++)
1559                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1560         for (i = 0; i < enic->rq_count; i++)
1561                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1562         for (i = 0; i < enic->cq_count; i++)
1563                 vnic_cq_clean(&enic->cq[i]);
1564         for (i = 0; i < enic->intr_count; i++)
1565                 vnic_intr_clean(&enic->intr[i]);
1566
1567         return 0;
1568 }
1569
1570 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1571 {
1572         struct enic *enic = netdev_priv(netdev);
1573         int running = netif_running(netdev);
1574
1575         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1576                 return -EINVAL;
1577
1578         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1579                 return -EOPNOTSUPP;
1580
1581         if (running)
1582                 enic_stop(netdev);
1583
1584         netdev->mtu = new_mtu;
1585
1586         if (netdev->mtu > enic->port_mtu)
1587                 netdev_warn(netdev,
1588                         "interface MTU (%d) set higher than port MTU (%d)\n",
1589                         netdev->mtu, enic->port_mtu);
1590
1591         if (running)
1592                 enic_open(netdev);
1593
1594         return 0;
1595 }
1596
1597 static void enic_change_mtu_work(struct work_struct *work)
1598 {
1599         struct enic *enic = container_of(work, struct enic, change_mtu_work);
1600         struct net_device *netdev = enic->netdev;
1601         int new_mtu = vnic_dev_mtu(enic->vdev);
1602         int err;
1603         unsigned int i;
1604
1605         new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1606
1607         rtnl_lock();
1608
1609         /* Stop RQ */
1610         del_timer_sync(&enic->notify_timer);
1611
1612         for (i = 0; i < enic->rq_count; i++)
1613                 napi_disable(&enic->napi[i]);
1614
1615         vnic_intr_mask(&enic->intr[0]);
1616         enic_synchronize_irqs(enic);
1617         err = vnic_rq_disable(&enic->rq[0]);
1618         if (err) {
1619                 rtnl_unlock();
1620                 netdev_err(netdev, "Unable to disable RQ.\n");
1621                 return;
1622         }
1623         vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1624         vnic_cq_clean(&enic->cq[0]);
1625         vnic_intr_clean(&enic->intr[0]);
1626
1627         /* Fill RQ with new_mtu-sized buffers */
1628         netdev->mtu = new_mtu;
1629         vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1630         /* Need at least one buffer on ring to get going */
1631         if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1632                 rtnl_unlock();
1633                 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1634                 return;
1635         }
1636
1637         /* Start RQ */
1638         vnic_rq_enable(&enic->rq[0]);
1639         napi_enable(&enic->napi[0]);
1640         vnic_intr_unmask(&enic->intr[0]);
1641         enic_notify_timer_start(enic);
1642
1643         rtnl_unlock();
1644
1645         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1646 }
1647
1648 #ifdef CONFIG_NET_POLL_CONTROLLER
1649 static void enic_poll_controller(struct net_device *netdev)
1650 {
1651         struct enic *enic = netdev_priv(netdev);
1652         struct vnic_dev *vdev = enic->vdev;
1653         unsigned int i, intr;
1654
1655         switch (vnic_dev_get_intr_mode(vdev)) {
1656         case VNIC_DEV_INTR_MODE_MSIX:
1657                 for (i = 0; i < enic->rq_count; i++) {
1658                         intr = enic_msix_rq_intr(enic, i);
1659                         enic_isr_msix_rq(enic->msix_entry[intr].vector,
1660                                 &enic->napi[i]);
1661                 }
1662
1663                 for (i = 0; i < enic->wq_count; i++) {
1664                         intr = enic_msix_wq_intr(enic, i);
1665                         enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
1666                 }
1667
1668                 break;
1669         case VNIC_DEV_INTR_MODE_MSI:
1670                 enic_isr_msi(enic->pdev->irq, enic);
1671                 break;
1672         case VNIC_DEV_INTR_MODE_INTX:
1673                 enic_isr_legacy(enic->pdev->irq, netdev);
1674                 break;
1675         default:
1676                 break;
1677         }
1678 }
1679 #endif
1680
1681 static int enic_dev_wait(struct vnic_dev *vdev,
1682         int (*start)(struct vnic_dev *, int),
1683         int (*finished)(struct vnic_dev *, int *),
1684         int arg)
1685 {
1686         unsigned long time;
1687         int done;
1688         int err;
1689
1690         BUG_ON(in_interrupt());
1691
1692         err = start(vdev, arg);
1693         if (err)
1694                 return err;
1695
1696         /* Wait for func to complete...2 seconds max
1697          */
1698
1699         time = jiffies + (HZ * 2);
1700         do {
1701
1702                 err = finished(vdev, &done);
1703                 if (err)
1704                         return err;
1705
1706                 if (done)
1707                         return 0;
1708
1709                 schedule_timeout_uninterruptible(HZ / 10);
1710
1711         } while (time_after(time, jiffies));
1712
1713         return -ETIMEDOUT;
1714 }
1715
1716 static int enic_dev_open(struct enic *enic)
1717 {
1718         int err;
1719
1720         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1721                 vnic_dev_open_done, 0);
1722         if (err)
1723                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1724                         err);
1725
1726         return err;
1727 }
1728
1729 static int enic_dev_hang_reset(struct enic *enic)
1730 {
1731         int err;
1732
1733         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1734                 vnic_dev_hang_reset_done, 0);
1735         if (err)
1736                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1737                         err);
1738
1739         return err;
1740 }
1741
1742 static int enic_set_rsskey(struct enic *enic)
1743 {
1744         dma_addr_t rss_key_buf_pa;
1745         union vnic_rss_key *rss_key_buf_va = NULL;
1746         union vnic_rss_key rss_key = {
1747                 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1748                 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1749                 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1750                 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1751         };
1752         int err;
1753
1754         rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1755                 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1756         if (!rss_key_buf_va)
1757                 return -ENOMEM;
1758
1759         memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1760
1761         spin_lock(&enic->devcmd_lock);
1762         err = enic_set_rss_key(enic,
1763                 rss_key_buf_pa,
1764                 sizeof(union vnic_rss_key));
1765         spin_unlock(&enic->devcmd_lock);
1766
1767         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1768                 rss_key_buf_va, rss_key_buf_pa);
1769
1770         return err;
1771 }
1772
1773 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1774 {
1775         dma_addr_t rss_cpu_buf_pa;
1776         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1777         unsigned int i;
1778         int err;
1779
1780         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1781                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1782         if (!rss_cpu_buf_va)
1783                 return -ENOMEM;
1784
1785         for (i = 0; i < (1 << rss_hash_bits); i++)
1786                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1787
1788         spin_lock(&enic->devcmd_lock);
1789         err = enic_set_rss_cpu(enic,
1790                 rss_cpu_buf_pa,
1791                 sizeof(union vnic_rss_cpu));
1792         spin_unlock(&enic->devcmd_lock);
1793
1794         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1795                 rss_cpu_buf_va, rss_cpu_buf_pa);
1796
1797         return err;
1798 }
1799
1800 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1801         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1802 {
1803         const u8 tso_ipid_split_en = 0;
1804         const u8 ig_vlan_strip_en = 1;
1805         int err;
1806
1807         /* Enable VLAN tag stripping.
1808         */
1809
1810         spin_lock(&enic->devcmd_lock);
1811         err = enic_set_nic_cfg(enic,
1812                 rss_default_cpu, rss_hash_type,
1813                 rss_hash_bits, rss_base_cpu,
1814                 rss_enable, tso_ipid_split_en,
1815                 ig_vlan_strip_en);
1816         spin_unlock(&enic->devcmd_lock);
1817
1818         return err;
1819 }
1820
1821 static int enic_set_rss_nic_cfg(struct enic *enic)
1822 {
1823         struct device *dev = enic_get_dev(enic);
1824         const u8 rss_default_cpu = 0;
1825         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1826                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1827                 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1828                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1829         const u8 rss_hash_bits = 7;
1830         const u8 rss_base_cpu = 0;
1831         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1832
1833         if (rss_enable) {
1834                 if (!enic_set_rsskey(enic)) {
1835                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1836                                 rss_enable = 0;
1837                                 dev_warn(dev, "RSS disabled, "
1838                                         "Failed to set RSS cpu indirection table.");
1839                         }
1840                 } else {
1841                         rss_enable = 0;
1842                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1843                 }
1844         }
1845
1846         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1847                 rss_hash_bits, rss_base_cpu, rss_enable);
1848 }
1849
1850 static void enic_reset(struct work_struct *work)
1851 {
1852         struct enic *enic = container_of(work, struct enic, reset);
1853
1854         if (!netif_running(enic->netdev))
1855                 return;
1856
1857         rtnl_lock();
1858
1859         spin_lock(&enic->enic_api_lock);
1860         enic_dev_hang_notify(enic);
1861         enic_stop(enic->netdev);
1862         enic_dev_hang_reset(enic);
1863         enic_reset_addr_lists(enic);
1864         enic_init_vnic_resources(enic);
1865         enic_set_rss_nic_cfg(enic);
1866         enic_dev_set_ig_vlan_rewrite_mode(enic);
1867         enic_open(enic->netdev);
1868         spin_unlock(&enic->enic_api_lock);
1869         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
1870
1871         rtnl_unlock();
1872 }
1873
1874 static int enic_set_intr_mode(struct enic *enic)
1875 {
1876         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
1877         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
1878         unsigned int i;
1879
1880         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1881          * on system capabilities.
1882          *
1883          * Try MSI-X first
1884          *
1885          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1886          * (the second to last INTR is used for WQ/RQ errors)
1887          * (the last INTR is used for notifications)
1888          */
1889
1890         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1891         for (i = 0; i < n + m + 2; i++)
1892                 enic->msix_entry[i].entry = i;
1893
1894         /* Use multiple RQs if RSS is enabled
1895          */
1896
1897         if (ENIC_SETTING(enic, RSS) &&
1898             enic->config.intr_mode < 1 &&
1899             enic->rq_count >= n &&
1900             enic->wq_count >= m &&
1901             enic->cq_count >= n + m &&
1902             enic->intr_count >= n + m + 2) {
1903
1904                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1905                                           n + m + 2, n + m + 2) > 0) {
1906
1907                         enic->rq_count = n;
1908                         enic->wq_count = m;
1909                         enic->cq_count = n + m;
1910                         enic->intr_count = n + m + 2;
1911
1912                         vnic_dev_set_intr_mode(enic->vdev,
1913                                 VNIC_DEV_INTR_MODE_MSIX);
1914
1915                         return 0;
1916                 }
1917         }
1918
1919         if (enic->config.intr_mode < 1 &&
1920             enic->rq_count >= 1 &&
1921             enic->wq_count >= m &&
1922             enic->cq_count >= 1 + m &&
1923             enic->intr_count >= 1 + m + 2) {
1924                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1925                                           1 + m + 2, 1 + m + 2) > 0) {
1926
1927                         enic->rq_count = 1;
1928                         enic->wq_count = m;
1929                         enic->cq_count = 1 + m;
1930                         enic->intr_count = 1 + m + 2;
1931
1932                         vnic_dev_set_intr_mode(enic->vdev,
1933                                 VNIC_DEV_INTR_MODE_MSIX);
1934
1935                         return 0;
1936                 }
1937         }
1938
1939         /* Next try MSI
1940          *
1941          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1942          */
1943
1944         if (enic->config.intr_mode < 2 &&
1945             enic->rq_count >= 1 &&
1946             enic->wq_count >= 1 &&
1947             enic->cq_count >= 2 &&
1948             enic->intr_count >= 1 &&
1949             !pci_enable_msi(enic->pdev)) {
1950
1951                 enic->rq_count = 1;
1952                 enic->wq_count = 1;
1953                 enic->cq_count = 2;
1954                 enic->intr_count = 1;
1955
1956                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1957
1958                 return 0;
1959         }
1960
1961         /* Next try INTx
1962          *
1963          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1964          * (the first INTR is used for WQ/RQ)
1965          * (the second INTR is used for WQ/RQ errors)
1966          * (the last INTR is used for notifications)
1967          */
1968
1969         if (enic->config.intr_mode < 3 &&
1970             enic->rq_count >= 1 &&
1971             enic->wq_count >= 1 &&
1972             enic->cq_count >= 2 &&
1973             enic->intr_count >= 3) {
1974
1975                 enic->rq_count = 1;
1976                 enic->wq_count = 1;
1977                 enic->cq_count = 2;
1978                 enic->intr_count = 3;
1979
1980                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1981
1982                 return 0;
1983         }
1984
1985         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1986
1987         return -EINVAL;
1988 }
1989
1990 static void enic_clear_intr_mode(struct enic *enic)
1991 {
1992         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1993         case VNIC_DEV_INTR_MODE_MSIX:
1994                 pci_disable_msix(enic->pdev);
1995                 break;
1996         case VNIC_DEV_INTR_MODE_MSI:
1997                 pci_disable_msi(enic->pdev);
1998                 break;
1999         default:
2000                 break;
2001         }
2002
2003         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2004 }
2005
2006 static const struct net_device_ops enic_netdev_dynamic_ops = {
2007         .ndo_open               = enic_open,
2008         .ndo_stop               = enic_stop,
2009         .ndo_start_xmit         = enic_hard_start_xmit,
2010         .ndo_get_stats64        = enic_get_stats,
2011         .ndo_validate_addr      = eth_validate_addr,
2012         .ndo_set_rx_mode        = enic_set_rx_mode,
2013         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2014         .ndo_change_mtu         = enic_change_mtu,
2015         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2016         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2017         .ndo_tx_timeout         = enic_tx_timeout,
2018         .ndo_set_vf_port        = enic_set_vf_port,
2019         .ndo_get_vf_port        = enic_get_vf_port,
2020         .ndo_set_vf_mac         = enic_set_vf_mac,
2021 #ifdef CONFIG_NET_POLL_CONTROLLER
2022         .ndo_poll_controller    = enic_poll_controller,
2023 #endif
2024 };
2025
2026 static const struct net_device_ops enic_netdev_ops = {
2027         .ndo_open               = enic_open,
2028         .ndo_stop               = enic_stop,
2029         .ndo_start_xmit         = enic_hard_start_xmit,
2030         .ndo_get_stats64        = enic_get_stats,
2031         .ndo_validate_addr      = eth_validate_addr,
2032         .ndo_set_mac_address    = enic_set_mac_address,
2033         .ndo_set_rx_mode        = enic_set_rx_mode,
2034         .ndo_change_mtu         = enic_change_mtu,
2035         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2036         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2037         .ndo_tx_timeout         = enic_tx_timeout,
2038         .ndo_set_vf_port        = enic_set_vf_port,
2039         .ndo_get_vf_port        = enic_get_vf_port,
2040         .ndo_set_vf_mac         = enic_set_vf_mac,
2041 #ifdef CONFIG_NET_POLL_CONTROLLER
2042         .ndo_poll_controller    = enic_poll_controller,
2043 #endif
2044 };
2045
2046 static void enic_dev_deinit(struct enic *enic)
2047 {
2048         unsigned int i;
2049
2050         for (i = 0; i < enic->rq_count; i++)
2051                 netif_napi_del(&enic->napi[i]);
2052
2053         enic_free_vnic_resources(enic);
2054         enic_clear_intr_mode(enic);
2055 }
2056
2057 static int enic_dev_init(struct enic *enic)
2058 {
2059         struct device *dev = enic_get_dev(enic);
2060         struct net_device *netdev = enic->netdev;
2061         unsigned int i;
2062         int err;
2063
2064         /* Get interrupt coalesce timer info */
2065         err = enic_dev_intr_coal_timer_info(enic);
2066         if (err) {
2067                 dev_warn(dev, "Using default conversion factor for "
2068                         "interrupt coalesce timer\n");
2069                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2070         }
2071
2072         /* Get vNIC configuration
2073          */
2074
2075         err = enic_get_vnic_config(enic);
2076         if (err) {
2077                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2078                 return err;
2079         }
2080
2081         /* Get available resource counts
2082          */
2083
2084         enic_get_res_counts(enic);
2085
2086         /* Set interrupt mode based on resource counts and system
2087          * capabilities
2088          */
2089
2090         err = enic_set_intr_mode(enic);
2091         if (err) {
2092                 dev_err(dev, "Failed to set intr mode based on resource "
2093                         "counts and system capabilities, aborting\n");
2094                 return err;
2095         }
2096
2097         /* Allocate and configure vNIC resources
2098          */
2099
2100         err = enic_alloc_vnic_resources(enic);
2101         if (err) {
2102                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2103                 goto err_out_free_vnic_resources;
2104         }
2105
2106         enic_init_vnic_resources(enic);
2107
2108         err = enic_set_rss_nic_cfg(enic);
2109         if (err) {
2110                 dev_err(dev, "Failed to config nic, aborting\n");
2111                 goto err_out_free_vnic_resources;
2112         }
2113
2114         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2115         default:
2116                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2117                 break;
2118         case VNIC_DEV_INTR_MODE_MSIX:
2119                 for (i = 0; i < enic->rq_count; i++)
2120                         netif_napi_add(netdev, &enic->napi[i],
2121                                 enic_poll_msix, 64);
2122                 break;
2123         }
2124
2125         return 0;
2126
2127 err_out_free_vnic_resources:
2128         enic_clear_intr_mode(enic);
2129         enic_free_vnic_resources(enic);
2130
2131         return err;
2132 }
2133
2134 static void enic_iounmap(struct enic *enic)
2135 {
2136         unsigned int i;
2137
2138         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2139                 if (enic->bar[i].vaddr)
2140                         iounmap(enic->bar[i].vaddr);
2141 }
2142
2143 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2144 {
2145         struct device *dev = &pdev->dev;
2146         struct net_device *netdev;
2147         struct enic *enic;
2148         int using_dac = 0;
2149         unsigned int i;
2150         int err;
2151 #ifdef CONFIG_PCI_IOV
2152         int pos = 0;
2153 #endif
2154         int num_pps = 1;
2155
2156         /* Allocate net device structure and initialize.  Private
2157          * instance data is initialized to zero.
2158          */
2159
2160         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2161                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2162         if (!netdev)
2163                 return -ENOMEM;
2164
2165         pci_set_drvdata(pdev, netdev);
2166
2167         SET_NETDEV_DEV(netdev, &pdev->dev);
2168
2169         enic = netdev_priv(netdev);
2170         enic->netdev = netdev;
2171         enic->pdev = pdev;
2172
2173         /* Setup PCI resources
2174          */
2175
2176         err = pci_enable_device_mem(pdev);
2177         if (err) {
2178                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2179                 goto err_out_free_netdev;
2180         }
2181
2182         err = pci_request_regions(pdev, DRV_NAME);
2183         if (err) {
2184                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2185                 goto err_out_disable_device;
2186         }
2187
2188         pci_set_master(pdev);
2189
2190         /* Query PCI controller on system for DMA addressing
2191          * limitation for the device.  Try 64-bit first, and
2192          * fail to 32-bit.
2193          */
2194
2195         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2196         if (err) {
2197                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2198                 if (err) {
2199                         dev_err(dev, "No usable DMA configuration, aborting\n");
2200                         goto err_out_release_regions;
2201                 }
2202                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2203                 if (err) {
2204                         dev_err(dev, "Unable to obtain %u-bit DMA "
2205                                 "for consistent allocations, aborting\n", 32);
2206                         goto err_out_release_regions;
2207                 }
2208         } else {
2209                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2210                 if (err) {
2211                         dev_err(dev, "Unable to obtain %u-bit DMA "
2212                                 "for consistent allocations, aborting\n", 64);
2213                         goto err_out_release_regions;
2214                 }
2215                 using_dac = 1;
2216         }
2217
2218         /* Map vNIC resources from BAR0-5
2219          */
2220
2221         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2222                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2223                         continue;
2224                 enic->bar[i].len = pci_resource_len(pdev, i);
2225                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2226                 if (!enic->bar[i].vaddr) {
2227                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2228                         err = -ENODEV;
2229                         goto err_out_iounmap;
2230                 }
2231                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2232         }
2233
2234         /* Register vNIC device
2235          */
2236
2237         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2238                 ARRAY_SIZE(enic->bar));
2239         if (!enic->vdev) {
2240                 dev_err(dev, "vNIC registration failed, aborting\n");
2241                 err = -ENODEV;
2242                 goto err_out_iounmap;
2243         }
2244
2245 #ifdef CONFIG_PCI_IOV
2246         /* Get number of subvnics */
2247         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2248         if (pos) {
2249                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2250                         &enic->num_vfs);
2251                 if (enic->num_vfs) {
2252                         err = pci_enable_sriov(pdev, enic->num_vfs);
2253                         if (err) {
2254                                 dev_err(dev, "SRIOV enable failed, aborting."
2255                                         " pci_enable_sriov() returned %d\n",
2256                                         err);
2257                                 goto err_out_vnic_unregister;
2258                         }
2259                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2260                         num_pps = enic->num_vfs;
2261                 }
2262         }
2263 #endif
2264
2265         /* Allocate structure for port profiles */
2266         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2267         if (!enic->pp) {
2268                 err = -ENOMEM;
2269                 goto err_out_disable_sriov_pp;
2270         }
2271
2272         /* Issue device open to get device in known state
2273          */
2274
2275         err = enic_dev_open(enic);
2276         if (err) {
2277                 dev_err(dev, "vNIC dev open failed, aborting\n");
2278                 goto err_out_disable_sriov;
2279         }
2280
2281         /* Setup devcmd lock
2282          */
2283
2284         spin_lock_init(&enic->devcmd_lock);
2285         spin_lock_init(&enic->enic_api_lock);
2286
2287         /*
2288          * Set ingress vlan rewrite mode before vnic initialization
2289          */
2290
2291         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2292         if (err) {
2293                 dev_err(dev,
2294                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2295                 goto err_out_dev_close;
2296         }
2297
2298         /* Issue device init to initialize the vnic-to-switch link.
2299          * We'll start with carrier off and wait for link UP
2300          * notification later to turn on carrier.  We don't need
2301          * to wait here for the vnic-to-switch link initialization
2302          * to complete; link UP notification is the indication that
2303          * the process is complete.
2304          */
2305
2306         netif_carrier_off(netdev);
2307
2308         /* Do not call dev_init for a dynamic vnic.
2309          * For a dynamic vnic, init_prov_info will be
2310          * called later by an upper layer.
2311          */
2312
2313         if (!enic_is_dynamic(enic)) {
2314                 err = vnic_dev_init(enic->vdev, 0);
2315                 if (err) {
2316                         dev_err(dev, "vNIC dev init failed, aborting\n");
2317                         goto err_out_dev_close;
2318                 }
2319         }
2320
2321         err = enic_dev_init(enic);
2322         if (err) {
2323                 dev_err(dev, "Device initialization failed, aborting\n");
2324                 goto err_out_dev_close;
2325         }
2326
2327         netif_set_real_num_tx_queues(netdev, enic->wq_count);
2328         netif_set_real_num_rx_queues(netdev, enic->rq_count);
2329
2330         /* Setup notification timer, HW reset task, and wq locks
2331          */
2332
2333         init_timer(&enic->notify_timer);
2334         enic->notify_timer.function = enic_notify_timer;
2335         enic->notify_timer.data = (unsigned long)enic;
2336
2337         enic_set_rx_coal_setting(enic);
2338         INIT_WORK(&enic->reset, enic_reset);
2339         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2340
2341         for (i = 0; i < enic->wq_count; i++)
2342                 spin_lock_init(&enic->wq_lock[i]);
2343
2344         /* Register net device
2345          */
2346
2347         enic->port_mtu = enic->config.mtu;
2348         (void)enic_change_mtu(netdev, enic->port_mtu);
2349
2350         err = enic_set_mac_addr(netdev, enic->mac_addr);
2351         if (err) {
2352                 dev_err(dev, "Invalid MAC address, aborting\n");
2353                 goto err_out_dev_deinit;
2354         }
2355
2356         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2357         /* rx coalesce time already got initialized. This gets used
2358          * if adaptive coal is turned off
2359          */
2360         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2361
2362         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2363                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2364         else
2365                 netdev->netdev_ops = &enic_netdev_ops;
2366
2367         netdev->watchdog_timeo = 2 * HZ;
2368         enic_set_ethtool_ops(netdev);
2369
2370         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2371         if (ENIC_SETTING(enic, LOOP)) {
2372                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2373                 enic->loop_enable = 1;
2374                 enic->loop_tag = enic->config.loop_tag;
2375                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2376         }
2377         if (ENIC_SETTING(enic, TXCSUM))
2378                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2379         if (ENIC_SETTING(enic, TSO))
2380                 netdev->hw_features |= NETIF_F_TSO |
2381                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2382         if (ENIC_SETTING(enic, RSS))
2383                 netdev->hw_features |= NETIF_F_RXHASH;
2384         if (ENIC_SETTING(enic, RXCSUM))
2385                 netdev->hw_features |= NETIF_F_RXCSUM;
2386
2387         netdev->features |= netdev->hw_features;
2388
2389         if (using_dac)
2390                 netdev->features |= NETIF_F_HIGHDMA;
2391
2392         netdev->priv_flags |= IFF_UNICAST_FLT;
2393
2394         err = register_netdev(netdev);
2395         if (err) {
2396                 dev_err(dev, "Cannot register net device, aborting\n");
2397                 goto err_out_dev_deinit;
2398         }
2399
2400         return 0;
2401
2402 err_out_dev_deinit:
2403         enic_dev_deinit(enic);
2404 err_out_dev_close:
2405         vnic_dev_close(enic->vdev);
2406 err_out_disable_sriov:
2407         kfree(enic->pp);
2408 err_out_disable_sriov_pp:
2409 #ifdef CONFIG_PCI_IOV
2410         if (enic_sriov_enabled(enic)) {
2411                 pci_disable_sriov(pdev);
2412                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2413         }
2414 err_out_vnic_unregister:
2415 #endif
2416         vnic_dev_unregister(enic->vdev);
2417 err_out_iounmap:
2418         enic_iounmap(enic);
2419 err_out_release_regions:
2420         pci_release_regions(pdev);
2421 err_out_disable_device:
2422         pci_disable_device(pdev);
2423 err_out_free_netdev:
2424         free_netdev(netdev);
2425
2426         return err;
2427 }
2428
2429 static void enic_remove(struct pci_dev *pdev)
2430 {
2431         struct net_device *netdev = pci_get_drvdata(pdev);
2432
2433         if (netdev) {
2434                 struct enic *enic = netdev_priv(netdev);
2435
2436                 cancel_work_sync(&enic->reset);
2437                 cancel_work_sync(&enic->change_mtu_work);
2438                 unregister_netdev(netdev);
2439                 enic_dev_deinit(enic);
2440                 vnic_dev_close(enic->vdev);
2441 #ifdef CONFIG_PCI_IOV
2442                 if (enic_sriov_enabled(enic)) {
2443                         pci_disable_sriov(pdev);
2444                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2445                 }
2446 #endif
2447                 kfree(enic->pp);
2448                 vnic_dev_unregister(enic->vdev);
2449                 enic_iounmap(enic);
2450                 pci_release_regions(pdev);
2451                 pci_disable_device(pdev);
2452                 free_netdev(netdev);
2453         }
2454 }
2455
2456 static struct pci_driver enic_driver = {
2457         .name = DRV_NAME,
2458         .id_table = enic_id_table,
2459         .probe = enic_probe,
2460         .remove = enic_remove,
2461 };
2462
2463 static int __init enic_init_module(void)
2464 {
2465         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2466
2467         return pci_register_driver(&enic_driver);
2468 }
2469
2470 static void __exit enic_cleanup_module(void)
2471 {
2472         pci_unregister_driver(&enic_driver);
2473 }
2474
2475 module_init(enic_init_module);
2476 module_exit(enic_cleanup_module);