OSDN Git Service

Merge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
[uclinux-h8/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 #include <linux/numa.h>
43 #ifdef CONFIG_RFS_ACCEL
44 #include <linux/cpu_rmap.h>
45 #endif
46 #include <linux/crash_dump.h>
47 #include <net/busy_poll.h>
48 #include <net/vxlan.h>
49
50 #include "cq_enet_desc.h"
51 #include "vnic_dev.h"
52 #include "vnic_intr.h"
53 #include "vnic_stats.h"
54 #include "vnic_vic.h"
55 #include "enic_res.h"
56 #include "enic.h"
57 #include "enic_dev.h"
58 #include "enic_pp.h"
59 #include "enic_clsf.h"
60
61 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
62 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
63 #define MAX_TSO                         (1 << 16)
64 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
65
66 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF      0x0071  /* enet SRIOV VF */
69
70 #define RX_COPYBREAK_DEFAULT            256
71
72 /* Supported devices */
73 static const struct pci_device_id enic_id_table[] = {
74         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
75         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
76         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
77         { 0, }  /* end of table */
78 };
79
80 MODULE_DESCRIPTION(DRV_DESCRIPTION);
81 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
82 MODULE_LICENSE("GPL");
83 MODULE_VERSION(DRV_VERSION);
84 MODULE_DEVICE_TABLE(pci, enic_id_table);
85
86 #define ENIC_LARGE_PKT_THRESHOLD                1000
87 #define ENIC_MAX_COALESCE_TIMERS                10
88 /*  Interrupt moderation table, which will be used to decide the
89  *  coalescing timer values
90  *  {rx_rate in Mbps, mapping percentage of the range}
91  */
92 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
93         {4000,  0},
94         {4400, 10},
95         {5060, 20},
96         {5230, 30},
97         {5540, 40},
98         {5820, 50},
99         {6120, 60},
100         {6435, 70},
101         {6745, 80},
102         {7000, 90},
103         {0xFFFFFFFF, 100}
104 };
105
106 /* This table helps the driver to pick different ranges for rx coalescing
107  * timer depending on the link speed.
108  */
109 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
110         {0,  0}, /* 0  - 4  Gbps */
111         {0,  3}, /* 4  - 10 Gbps */
112         {3,  6}, /* 10 - 40 Gbps */
113 };
114
115 static void enic_init_affinity_hint(struct enic *enic)
116 {
117         int numa_node = dev_to_node(&enic->pdev->dev);
118         int i;
119
120         for (i = 0; i < enic->intr_count; i++) {
121                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) ||
122                     (enic->msix[i].affinity_mask &&
123                      !cpumask_empty(enic->msix[i].affinity_mask)))
124                         continue;
125                 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask,
126                                        GFP_KERNEL))
127                         cpumask_set_cpu(cpumask_local_spread(i, numa_node),
128                                         enic->msix[i].affinity_mask);
129         }
130 }
131
132 static void enic_free_affinity_hint(struct enic *enic)
133 {
134         int i;
135
136         for (i = 0; i < enic->intr_count; i++) {
137                 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i))
138                         continue;
139                 free_cpumask_var(enic->msix[i].affinity_mask);
140         }
141 }
142
143 static void enic_set_affinity_hint(struct enic *enic)
144 {
145         int i;
146         int err;
147
148         for (i = 0; i < enic->intr_count; i++) {
149                 if (enic_is_err_intr(enic, i)           ||
150                     enic_is_notify_intr(enic, i)        ||
151                     !enic->msix[i].affinity_mask        ||
152                     cpumask_empty(enic->msix[i].affinity_mask))
153                         continue;
154                 err = irq_set_affinity_hint(enic->msix_entry[i].vector,
155                                             enic->msix[i].affinity_mask);
156                 if (err)
157                         netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n",
158                                     err);
159         }
160
161         for (i = 0; i < enic->wq_count; i++) {
162                 int wq_intr = enic_msix_wq_intr(enic, i);
163
164                 if (enic->msix[wq_intr].affinity_mask &&
165                     !cpumask_empty(enic->msix[wq_intr].affinity_mask))
166                         netif_set_xps_queue(enic->netdev,
167                                             enic->msix[wq_intr].affinity_mask,
168                                             i);
169         }
170 }
171
172 static void enic_unset_affinity_hint(struct enic *enic)
173 {
174         int i;
175
176         for (i = 0; i < enic->intr_count; i++)
177                 irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
178 }
179
180 static void enic_udp_tunnel_add(struct net_device *netdev,
181                                 struct udp_tunnel_info *ti)
182 {
183         struct enic *enic = netdev_priv(netdev);
184         __be16 port = ti->port;
185         int err;
186
187         spin_lock_bh(&enic->devcmd_lock);
188
189         if (ti->type != UDP_TUNNEL_TYPE_VXLAN) {
190                 netdev_info(netdev, "udp_tnl: only vxlan tunnel offload supported");
191                 goto error;
192         }
193
194         switch (ti->sa_family) {
195         case AF_INET6:
196                 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)) {
197                         netdev_info(netdev, "vxlan: only IPv4 offload supported");
198                         goto error;
199                 }
200                 /* Fall through */
201         case AF_INET:
202                 break;
203         default:
204                 goto error;
205         }
206
207         if (enic->vxlan.vxlan_udp_port_number) {
208                 if (ntohs(port) == enic->vxlan.vxlan_udp_port_number)
209                         netdev_warn(netdev, "vxlan: udp port already offloaded");
210                 else
211                         netdev_info(netdev, "vxlan: offload supported for only one UDP port");
212
213                 goto error;
214         }
215         if ((vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) != 1) &&
216             !(enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ)) {
217                 netdev_info(netdev, "vxlan: vxlan offload with multi wq not supported on this adapter");
218                 goto error;
219         }
220
221         err = vnic_dev_overlay_offload_cfg(enic->vdev,
222                                            OVERLAY_CFG_VXLAN_PORT_UPDATE,
223                                            ntohs(port));
224         if (err)
225                 goto error;
226
227         err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
228                                             enic->vxlan.patch_level);
229         if (err)
230                 goto error;
231
232         enic->vxlan.vxlan_udp_port_number = ntohs(port);
233
234         netdev_info(netdev, "vxlan fw-vers-%d: offload enabled for udp port: %d, sa_family: %d ",
235                     (int)enic->vxlan.patch_level, ntohs(port), ti->sa_family);
236
237         goto unlock;
238
239 error:
240         netdev_info(netdev, "failed to offload udp port: %d, sa_family: %d, type: %d",
241                     ntohs(port), ti->sa_family, ti->type);
242 unlock:
243         spin_unlock_bh(&enic->devcmd_lock);
244 }
245
246 static void enic_udp_tunnel_del(struct net_device *netdev,
247                                 struct udp_tunnel_info *ti)
248 {
249         struct enic *enic = netdev_priv(netdev);
250         int err;
251
252         spin_lock_bh(&enic->devcmd_lock);
253
254         if ((ntohs(ti->port) != enic->vxlan.vxlan_udp_port_number) ||
255             ti->type != UDP_TUNNEL_TYPE_VXLAN) {
256                 netdev_info(netdev, "udp_tnl: port:%d, sa_family: %d, type: %d not offloaded",
257                             ntohs(ti->port), ti->sa_family, ti->type);
258                 goto unlock;
259         }
260
261         err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
262                                             OVERLAY_OFFLOAD_DISABLE);
263         if (err) {
264                 netdev_err(netdev, "vxlan: del offload udp port: %d failed",
265                            ntohs(ti->port));
266                 goto unlock;
267         }
268
269         enic->vxlan.vxlan_udp_port_number = 0;
270
271         netdev_info(netdev, "vxlan: del offload udp port %d, family %d\n",
272                     ntohs(ti->port), ti->sa_family);
273
274 unlock:
275         spin_unlock_bh(&enic->devcmd_lock);
276 }
277
278 static netdev_features_t enic_features_check(struct sk_buff *skb,
279                                              struct net_device *dev,
280                                              netdev_features_t features)
281 {
282         const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
283         struct enic *enic = netdev_priv(dev);
284         struct udphdr *udph;
285         u16 port = 0;
286         u8 proto;
287
288         if (!skb->encapsulation)
289                 return features;
290
291         features = vxlan_features_check(skb, features);
292
293         switch (vlan_get_protocol(skb)) {
294         case htons(ETH_P_IPV6):
295                 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6))
296                         goto out;
297                 proto = ipv6_hdr(skb)->nexthdr;
298                 break;
299         case htons(ETH_P_IP):
300                 proto = ip_hdr(skb)->protocol;
301                 break;
302         default:
303                 goto out;
304         }
305
306         switch (eth->h_proto) {
307         case ntohs(ETH_P_IPV6):
308                 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6))
309                         goto out;
310                 /* Fall through */
311         case ntohs(ETH_P_IP):
312                 break;
313         default:
314                 goto out;
315         }
316
317
318         if (proto == IPPROTO_UDP) {
319                 udph = udp_hdr(skb);
320                 port = be16_to_cpu(udph->dest);
321         }
322
323         /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
324          * for other UDP port tunnels
325          */
326         if (port  != enic->vxlan.vxlan_udp_port_number)
327                 goto out;
328
329         return features;
330
331 out:
332         return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
333 }
334
335 int enic_is_dynamic(struct enic *enic)
336 {
337         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
338 }
339
340 int enic_sriov_enabled(struct enic *enic)
341 {
342         return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
343 }
344
345 static int enic_is_sriov_vf(struct enic *enic)
346 {
347         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
348 }
349
350 int enic_is_valid_vf(struct enic *enic, int vf)
351 {
352 #ifdef CONFIG_PCI_IOV
353         return vf >= 0 && vf < enic->num_vfs;
354 #else
355         return 0;
356 #endif
357 }
358
359 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
360 {
361         struct enic *enic = vnic_dev_priv(wq->vdev);
362
363         if (buf->sop)
364                 pci_unmap_single(enic->pdev, buf->dma_addr,
365                         buf->len, PCI_DMA_TODEVICE);
366         else
367                 pci_unmap_page(enic->pdev, buf->dma_addr,
368                         buf->len, PCI_DMA_TODEVICE);
369
370         if (buf->os_buf)
371                 dev_kfree_skb_any(buf->os_buf);
372 }
373
374 static void enic_wq_free_buf(struct vnic_wq *wq,
375         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
376 {
377         enic_free_wq_buf(wq, buf);
378 }
379
380 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
381         u8 type, u16 q_number, u16 completed_index, void *opaque)
382 {
383         struct enic *enic = vnic_dev_priv(vdev);
384
385         spin_lock(&enic->wq_lock[q_number]);
386
387         vnic_wq_service(&enic->wq[q_number], cq_desc,
388                 completed_index, enic_wq_free_buf,
389                 opaque);
390
391         if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
392             vnic_wq_desc_avail(&enic->wq[q_number]) >=
393             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
394                 netif_wake_subqueue(enic->netdev, q_number);
395
396         spin_unlock(&enic->wq_lock[q_number]);
397
398         return 0;
399 }
400
401 static bool enic_log_q_error(struct enic *enic)
402 {
403         unsigned int i;
404         u32 error_status;
405         bool err = false;
406
407         for (i = 0; i < enic->wq_count; i++) {
408                 error_status = vnic_wq_error_status(&enic->wq[i]);
409                 err |= error_status;
410                 if (error_status)
411                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
412                                 i, error_status);
413         }
414
415         for (i = 0; i < enic->rq_count; i++) {
416                 error_status = vnic_rq_error_status(&enic->rq[i]);
417                 err |= error_status;
418                 if (error_status)
419                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
420                                 i, error_status);
421         }
422
423         return err;
424 }
425
426 static void enic_msglvl_check(struct enic *enic)
427 {
428         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
429
430         if (msg_enable != enic->msg_enable) {
431                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
432                         enic->msg_enable, msg_enable);
433                 enic->msg_enable = msg_enable;
434         }
435 }
436
437 static void enic_mtu_check(struct enic *enic)
438 {
439         u32 mtu = vnic_dev_mtu(enic->vdev);
440         struct net_device *netdev = enic->netdev;
441
442         if (mtu && mtu != enic->port_mtu) {
443                 enic->port_mtu = mtu;
444                 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
445                         mtu = max_t(int, ENIC_MIN_MTU,
446                                 min_t(int, ENIC_MAX_MTU, mtu));
447                         if (mtu != netdev->mtu)
448                                 schedule_work(&enic->change_mtu_work);
449                 } else {
450                         if (mtu < netdev->mtu)
451                                 netdev_warn(netdev,
452                                         "interface MTU (%d) set higher "
453                                         "than switch port MTU (%d)\n",
454                                         netdev->mtu, mtu);
455                 }
456         }
457 }
458
459 static void enic_link_check(struct enic *enic)
460 {
461         int link_status = vnic_dev_link_status(enic->vdev);
462         int carrier_ok = netif_carrier_ok(enic->netdev);
463
464         if (link_status && !carrier_ok) {
465                 netdev_info(enic->netdev, "Link UP\n");
466                 netif_carrier_on(enic->netdev);
467         } else if (!link_status && carrier_ok) {
468                 netdev_info(enic->netdev, "Link DOWN\n");
469                 netif_carrier_off(enic->netdev);
470         }
471 }
472
473 static void enic_notify_check(struct enic *enic)
474 {
475         enic_msglvl_check(enic);
476         enic_mtu_check(enic);
477         enic_link_check(enic);
478 }
479
480 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
481
482 static irqreturn_t enic_isr_legacy(int irq, void *data)
483 {
484         struct net_device *netdev = data;
485         struct enic *enic = netdev_priv(netdev);
486         unsigned int io_intr = enic_legacy_io_intr();
487         unsigned int err_intr = enic_legacy_err_intr();
488         unsigned int notify_intr = enic_legacy_notify_intr();
489         u32 pba;
490
491         vnic_intr_mask(&enic->intr[io_intr]);
492
493         pba = vnic_intr_legacy_pba(enic->legacy_pba);
494         if (!pba) {
495                 vnic_intr_unmask(&enic->intr[io_intr]);
496                 return IRQ_NONE;        /* not our interrupt */
497         }
498
499         if (ENIC_TEST_INTR(pba, notify_intr)) {
500                 enic_notify_check(enic);
501                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
502         }
503
504         if (ENIC_TEST_INTR(pba, err_intr)) {
505                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
506                 enic_log_q_error(enic);
507                 /* schedule recovery from WQ/RQ error */
508                 schedule_work(&enic->reset);
509                 return IRQ_HANDLED;
510         }
511
512         if (ENIC_TEST_INTR(pba, io_intr))
513                 napi_schedule_irqoff(&enic->napi[0]);
514         else
515                 vnic_intr_unmask(&enic->intr[io_intr]);
516
517         return IRQ_HANDLED;
518 }
519
520 static irqreturn_t enic_isr_msi(int irq, void *data)
521 {
522         struct enic *enic = data;
523
524         /* With MSI, there is no sharing of interrupts, so this is
525          * our interrupt and there is no need to ack it.  The device
526          * is not providing per-vector masking, so the OS will not
527          * write to PCI config space to mask/unmask the interrupt.
528          * We're using mask_on_assertion for MSI, so the device
529          * automatically masks the interrupt when the interrupt is
530          * generated.  Later, when exiting polling, the interrupt
531          * will be unmasked (see enic_poll).
532          *
533          * Also, the device uses the same PCIe Traffic Class (TC)
534          * for Memory Write data and MSI, so there are no ordering
535          * issues; the MSI will always arrive at the Root Complex
536          * _after_ corresponding Memory Writes (i.e. descriptor
537          * writes).
538          */
539
540         napi_schedule_irqoff(&enic->napi[0]);
541
542         return IRQ_HANDLED;
543 }
544
545 static irqreturn_t enic_isr_msix(int irq, void *data)
546 {
547         struct napi_struct *napi = data;
548
549         napi_schedule_irqoff(napi);
550
551         return IRQ_HANDLED;
552 }
553
554 static irqreturn_t enic_isr_msix_err(int irq, void *data)
555 {
556         struct enic *enic = data;
557         unsigned int intr = enic_msix_err_intr(enic);
558
559         vnic_intr_return_all_credits(&enic->intr[intr]);
560
561         if (enic_log_q_error(enic))
562                 /* schedule recovery from WQ/RQ error */
563                 schedule_work(&enic->reset);
564
565         return IRQ_HANDLED;
566 }
567
568 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
569 {
570         struct enic *enic = data;
571         unsigned int intr = enic_msix_notify_intr(enic);
572
573         enic_notify_check(enic);
574         vnic_intr_return_all_credits(&enic->intr[intr]);
575
576         return IRQ_HANDLED;
577 }
578
579 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
580                                   struct sk_buff *skb, unsigned int len_left,
581                                   int loopback)
582 {
583         const skb_frag_t *frag;
584         dma_addr_t dma_addr;
585
586         /* Queue additional data fragments */
587         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
588                 len_left -= skb_frag_size(frag);
589                 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
590                                             skb_frag_size(frag),
591                                             DMA_TO_DEVICE);
592                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
593                         return -ENOMEM;
594                 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
595                                         (len_left == 0),        /* EOP? */
596                                         loopback);
597         }
598
599         return 0;
600 }
601
602 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
603                                   struct sk_buff *skb, int vlan_tag_insert,
604                                   unsigned int vlan_tag, int loopback)
605 {
606         unsigned int head_len = skb_headlen(skb);
607         unsigned int len_left = skb->len - head_len;
608         int eop = (len_left == 0);
609         dma_addr_t dma_addr;
610         int err = 0;
611
612         dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
613                                   PCI_DMA_TODEVICE);
614         if (unlikely(enic_dma_map_check(enic, dma_addr)))
615                 return -ENOMEM;
616
617         /* Queue the main skb fragment. The fragments are no larger
618          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
619          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
620          * per fragment is queued.
621          */
622         enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
623                            vlan_tag, eop, loopback);
624
625         if (!eop)
626                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
627
628         return err;
629 }
630
631 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
632                                      struct sk_buff *skb, int vlan_tag_insert,
633                                      unsigned int vlan_tag, int loopback)
634 {
635         unsigned int head_len = skb_headlen(skb);
636         unsigned int len_left = skb->len - head_len;
637         unsigned int hdr_len = skb_checksum_start_offset(skb);
638         unsigned int csum_offset = hdr_len + skb->csum_offset;
639         int eop = (len_left == 0);
640         dma_addr_t dma_addr;
641         int err = 0;
642
643         dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
644                                   PCI_DMA_TODEVICE);
645         if (unlikely(enic_dma_map_check(enic, dma_addr)))
646                 return -ENOMEM;
647
648         /* Queue the main skb fragment. The fragments are no larger
649          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
650          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
651          * per fragment is queued.
652          */
653         enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
654                                    hdr_len, vlan_tag_insert, vlan_tag, eop,
655                                    loopback);
656
657         if (!eop)
658                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
659
660         return err;
661 }
662
663 static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
664 {
665         const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
666
667         switch (eth->h_proto) {
668         case ntohs(ETH_P_IP):
669                 inner_ip_hdr(skb)->check = 0;
670                 inner_tcp_hdr(skb)->check =
671                         ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr,
672                                            inner_ip_hdr(skb)->daddr, 0,
673                                            IPPROTO_TCP, 0);
674                 break;
675         case ntohs(ETH_P_IPV6):
676                 inner_tcp_hdr(skb)->check =
677                         ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr,
678                                          &inner_ipv6_hdr(skb)->daddr, 0,
679                                          IPPROTO_TCP, 0);
680                 break;
681         default:
682                 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload");
683                 break;
684         }
685 }
686
687 static void enic_preload_tcp_csum(struct sk_buff *skb)
688 {
689         /* Preload TCP csum field with IP pseudo hdr calculated
690          * with IP length set to zero.  HW will later add in length
691          * to each TCP segment resulting from the TSO.
692          */
693
694         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
695                 ip_hdr(skb)->check = 0;
696                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
697                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
698         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
699                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
700                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
701         }
702 }
703
704 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
705                                  struct sk_buff *skb, unsigned int mss,
706                                  int vlan_tag_insert, unsigned int vlan_tag,
707                                  int loopback)
708 {
709         unsigned int frag_len_left = skb_headlen(skb);
710         unsigned int len_left = skb->len - frag_len_left;
711         int eop = (len_left == 0);
712         unsigned int offset = 0;
713         unsigned int hdr_len;
714         dma_addr_t dma_addr;
715         unsigned int len;
716         skb_frag_t *frag;
717
718         if (skb->encapsulation) {
719                 hdr_len = skb_inner_transport_header(skb) - skb->data;
720                 hdr_len += inner_tcp_hdrlen(skb);
721                 enic_preload_tcp_csum_encap(skb);
722         } else {
723                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
724                 enic_preload_tcp_csum(skb);
725         }
726
727         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
728          * for the main skb fragment
729          */
730         while (frag_len_left) {
731                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
732                 dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
733                                           PCI_DMA_TODEVICE);
734                 if (unlikely(enic_dma_map_check(enic, dma_addr)))
735                         return -ENOMEM;
736                 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
737                                        vlan_tag_insert, vlan_tag,
738                                        eop && (len == frag_len_left), loopback);
739                 frag_len_left -= len;
740                 offset += len;
741         }
742
743         if (eop)
744                 return 0;
745
746         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
747          * for additional data fragments
748          */
749         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
750                 len_left -= skb_frag_size(frag);
751                 frag_len_left = skb_frag_size(frag);
752                 offset = 0;
753
754                 while (frag_len_left) {
755                         len = min(frag_len_left,
756                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
757                         dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
758                                                     offset, len,
759                                                     DMA_TO_DEVICE);
760                         if (unlikely(enic_dma_map_check(enic, dma_addr)))
761                                 return -ENOMEM;
762                         enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
763                                                 (len_left == 0) &&
764                                                  (len == frag_len_left),/*EOP*/
765                                                 loopback);
766                         frag_len_left -= len;
767                         offset += len;
768                 }
769         }
770
771         return 0;
772 }
773
774 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
775                                           struct sk_buff *skb,
776                                           int vlan_tag_insert,
777                                           unsigned int vlan_tag, int loopback)
778 {
779         unsigned int head_len = skb_headlen(skb);
780         unsigned int len_left = skb->len - head_len;
781         /* Hardware will overwrite the checksum fields, calculating from
782          * scratch and ignoring the value placed by software.
783          * Offload mode = 00
784          * mss[2], mss[1], mss[0] bits are set
785          */
786         unsigned int mss_or_csum = 7;
787         int eop = (len_left == 0);
788         dma_addr_t dma_addr;
789         int err = 0;
790
791         dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
792                                   PCI_DMA_TODEVICE);
793         if (unlikely(enic_dma_map_check(enic, dma_addr)))
794                 return -ENOMEM;
795
796         enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
797                               vlan_tag_insert, vlan_tag,
798                               WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
799                               loopback);
800         if (!eop)
801                 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
802
803         return err;
804 }
805
806 static inline void enic_queue_wq_skb(struct enic *enic,
807         struct vnic_wq *wq, struct sk_buff *skb)
808 {
809         unsigned int mss = skb_shinfo(skb)->gso_size;
810         unsigned int vlan_tag = 0;
811         int vlan_tag_insert = 0;
812         int loopback = 0;
813         int err;
814
815         if (skb_vlan_tag_present(skb)) {
816                 /* VLAN tag from trunking driver */
817                 vlan_tag_insert = 1;
818                 vlan_tag = skb_vlan_tag_get(skb);
819         } else if (enic->loop_enable) {
820                 vlan_tag = enic->loop_tag;
821                 loopback = 1;
822         }
823
824         if (mss)
825                 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
826                                             vlan_tag_insert, vlan_tag,
827                                             loopback);
828         else if (skb->encapsulation)
829                 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
830                                               vlan_tag, loopback);
831         else if (skb->ip_summed == CHECKSUM_PARTIAL)
832                 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
833                                                 vlan_tag, loopback);
834         else
835                 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
836                                              vlan_tag, loopback);
837         if (unlikely(err)) {
838                 struct vnic_wq_buf *buf;
839
840                 buf = wq->to_use->prev;
841                 /* while not EOP of previous pkt && queue not empty.
842                  * For all non EOP bufs, os_buf is NULL.
843                  */
844                 while (!buf->os_buf && (buf->next != wq->to_clean)) {
845                         enic_free_wq_buf(wq, buf);
846                         wq->ring.desc_avail++;
847                         buf = buf->prev;
848                 }
849                 wq->to_use = buf->next;
850                 dev_kfree_skb(skb);
851         }
852 }
853
854 /* netif_tx_lock held, process context with BHs disabled, or BH */
855 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
856         struct net_device *netdev)
857 {
858         struct enic *enic = netdev_priv(netdev);
859         struct vnic_wq *wq;
860         unsigned int txq_map;
861         struct netdev_queue *txq;
862
863         if (skb->len <= 0) {
864                 dev_kfree_skb_any(skb);
865                 return NETDEV_TX_OK;
866         }
867
868         txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
869         wq = &enic->wq[txq_map];
870         txq = netdev_get_tx_queue(netdev, txq_map);
871
872         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
873          * which is very likely.  In the off chance it's going to take
874          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
875          */
876
877         if (skb_shinfo(skb)->gso_size == 0 &&
878             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
879             skb_linearize(skb)) {
880                 dev_kfree_skb_any(skb);
881                 return NETDEV_TX_OK;
882         }
883
884         spin_lock(&enic->wq_lock[txq_map]);
885
886         if (vnic_wq_desc_avail(wq) <
887             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
888                 netif_tx_stop_queue(txq);
889                 /* This is a hard error, log it */
890                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
891                 spin_unlock(&enic->wq_lock[txq_map]);
892                 return NETDEV_TX_BUSY;
893         }
894
895         enic_queue_wq_skb(enic, wq, skb);
896
897         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
898                 netif_tx_stop_queue(txq);
899         skb_tx_timestamp(skb);
900         if (!skb->xmit_more || netif_xmit_stopped(txq))
901                 vnic_wq_doorbell(wq);
902
903         spin_unlock(&enic->wq_lock[txq_map]);
904
905         return NETDEV_TX_OK;
906 }
907
908 /* dev_base_lock rwlock held, nominally process context */
909 static void enic_get_stats(struct net_device *netdev,
910                            struct rtnl_link_stats64 *net_stats)
911 {
912         struct enic *enic = netdev_priv(netdev);
913         struct vnic_stats *stats;
914         int err;
915
916         err = enic_dev_stats_dump(enic, &stats);
917         /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
918          * For other failures, like devcmd failure, we return previously
919          * recorded stats.
920          */
921         if (err == -ENOMEM)
922                 return;
923
924         net_stats->tx_packets = stats->tx.tx_frames_ok;
925         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
926         net_stats->tx_errors = stats->tx.tx_errors;
927         net_stats->tx_dropped = stats->tx.tx_drops;
928
929         net_stats->rx_packets = stats->rx.rx_frames_ok;
930         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
931         net_stats->rx_errors = stats->rx.rx_errors;
932         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
933         net_stats->rx_over_errors = enic->rq_truncated_pkts;
934         net_stats->rx_crc_errors = enic->rq_bad_fcs;
935         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
936 }
937
938 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
939 {
940         struct enic *enic = netdev_priv(netdev);
941
942         if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
943                 unsigned int mc_count = netdev_mc_count(netdev);
944
945                 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
946                             ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
947
948                 return -ENOSPC;
949         }
950
951         enic_dev_add_addr(enic, mc_addr);
952         enic->mc_count++;
953
954         return 0;
955 }
956
957 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
958 {
959         struct enic *enic = netdev_priv(netdev);
960
961         enic_dev_del_addr(enic, mc_addr);
962         enic->mc_count--;
963
964         return 0;
965 }
966
967 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
968 {
969         struct enic *enic = netdev_priv(netdev);
970
971         if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
972                 unsigned int uc_count = netdev_uc_count(netdev);
973
974                 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
975                             ENIC_UNICAST_PERFECT_FILTERS, uc_count);
976
977                 return -ENOSPC;
978         }
979
980         enic_dev_add_addr(enic, uc_addr);
981         enic->uc_count++;
982
983         return 0;
984 }
985
986 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
987 {
988         struct enic *enic = netdev_priv(netdev);
989
990         enic_dev_del_addr(enic, uc_addr);
991         enic->uc_count--;
992
993         return 0;
994 }
995
996 void enic_reset_addr_lists(struct enic *enic)
997 {
998         struct net_device *netdev = enic->netdev;
999
1000         __dev_uc_unsync(netdev, NULL);
1001         __dev_mc_unsync(netdev, NULL);
1002
1003         enic->mc_count = 0;
1004         enic->uc_count = 0;
1005         enic->flags = 0;
1006 }
1007
1008 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
1009 {
1010         struct enic *enic = netdev_priv(netdev);
1011
1012         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
1013                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
1014                         return -EADDRNOTAVAIL;
1015         } else {
1016                 if (!is_valid_ether_addr(addr))
1017                         return -EADDRNOTAVAIL;
1018         }
1019
1020         memcpy(netdev->dev_addr, addr, netdev->addr_len);
1021
1022         return 0;
1023 }
1024
1025 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
1026 {
1027         struct enic *enic = netdev_priv(netdev);
1028         struct sockaddr *saddr = p;
1029         char *addr = saddr->sa_data;
1030         int err;
1031
1032         if (netif_running(enic->netdev)) {
1033                 err = enic_dev_del_station_addr(enic);
1034                 if (err)
1035                         return err;
1036         }
1037
1038         err = enic_set_mac_addr(netdev, addr);
1039         if (err)
1040                 return err;
1041
1042         if (netif_running(enic->netdev)) {
1043                 err = enic_dev_add_station_addr(enic);
1044                 if (err)
1045                         return err;
1046         }
1047
1048         return err;
1049 }
1050
1051 static int enic_set_mac_address(struct net_device *netdev, void *p)
1052 {
1053         struct sockaddr *saddr = p;
1054         char *addr = saddr->sa_data;
1055         struct enic *enic = netdev_priv(netdev);
1056         int err;
1057
1058         err = enic_dev_del_station_addr(enic);
1059         if (err)
1060                 return err;
1061
1062         err = enic_set_mac_addr(netdev, addr);
1063         if (err)
1064                 return err;
1065
1066         return enic_dev_add_station_addr(enic);
1067 }
1068
1069 /* netif_tx_lock held, BHs disabled */
1070 static void enic_set_rx_mode(struct net_device *netdev)
1071 {
1072         struct enic *enic = netdev_priv(netdev);
1073         int directed = 1;
1074         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1075         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1076         int promisc = (netdev->flags & IFF_PROMISC) ||
1077                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1078         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1079                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1080         unsigned int flags = netdev->flags |
1081                 (allmulti ? IFF_ALLMULTI : 0) |
1082                 (promisc ? IFF_PROMISC : 0);
1083
1084         if (enic->flags != flags) {
1085                 enic->flags = flags;
1086                 enic_dev_packet_filter(enic, directed,
1087                         multicast, broadcast, promisc, allmulti);
1088         }
1089
1090         if (!promisc) {
1091                 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1092                 if (!allmulti)
1093                         __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1094         }
1095 }
1096
1097 /* netif_tx_lock held, BHs disabled */
1098 static void enic_tx_timeout(struct net_device *netdev)
1099 {
1100         struct enic *enic = netdev_priv(netdev);
1101         schedule_work(&enic->tx_hang_reset);
1102 }
1103
1104 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1105 {
1106         struct enic *enic = netdev_priv(netdev);
1107         struct enic_port_profile *pp;
1108         int err;
1109
1110         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1111         if (err)
1112                 return err;
1113
1114         if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1115                 if (vf == PORT_SELF_VF) {
1116                         memcpy(pp->vf_mac, mac, ETH_ALEN);
1117                         return 0;
1118                 } else {
1119                         /*
1120                          * For sriov vf's set the mac in hw
1121                          */
1122                         ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1123                                 vnic_dev_set_mac_addr, mac);
1124                         return enic_dev_status_to_errno(err);
1125                 }
1126         } else
1127                 return -EINVAL;
1128 }
1129
1130 static int enic_set_vf_port(struct net_device *netdev, int vf,
1131         struct nlattr *port[])
1132 {
1133         struct enic *enic = netdev_priv(netdev);
1134         struct enic_port_profile prev_pp;
1135         struct enic_port_profile *pp;
1136         int err = 0, restore_pp = 1;
1137
1138         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1139         if (err)
1140                 return err;
1141
1142         if (!port[IFLA_PORT_REQUEST])
1143                 return -EOPNOTSUPP;
1144
1145         memcpy(&prev_pp, pp, sizeof(*enic->pp));
1146         memset(pp, 0, sizeof(*enic->pp));
1147
1148         pp->set |= ENIC_SET_REQUEST;
1149         pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1150
1151         if (port[IFLA_PORT_PROFILE]) {
1152                 pp->set |= ENIC_SET_NAME;
1153                 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1154                         PORT_PROFILE_MAX);
1155         }
1156
1157         if (port[IFLA_PORT_INSTANCE_UUID]) {
1158                 pp->set |= ENIC_SET_INSTANCE;
1159                 memcpy(pp->instance_uuid,
1160                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1161         }
1162
1163         if (port[IFLA_PORT_HOST_UUID]) {
1164                 pp->set |= ENIC_SET_HOST;
1165                 memcpy(pp->host_uuid,
1166                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1167         }
1168
1169         if (vf == PORT_SELF_VF) {
1170                 /* Special case handling: mac came from IFLA_VF_MAC */
1171                 if (!is_zero_ether_addr(prev_pp.vf_mac))
1172                         memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1173
1174                 if (is_zero_ether_addr(netdev->dev_addr))
1175                         eth_hw_addr_random(netdev);
1176         } else {
1177                 /* SR-IOV VF: get mac from adapter */
1178                 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1179                         vnic_dev_get_mac_addr, pp->mac_addr);
1180                 if (err) {
1181                         netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1182                         memcpy(pp, &prev_pp, sizeof(*pp));
1183                         return enic_dev_status_to_errno(err);
1184                 }
1185         }
1186
1187         err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1188         if (err) {
1189                 if (restore_pp) {
1190                         /* Things are still the way they were: Implicit
1191                          * DISASSOCIATE failed
1192                          */
1193                         memcpy(pp, &prev_pp, sizeof(*pp));
1194                 } else {
1195                         memset(pp, 0, sizeof(*pp));
1196                         if (vf == PORT_SELF_VF)
1197                                 eth_zero_addr(netdev->dev_addr);
1198                 }
1199         } else {
1200                 /* Set flag to indicate that the port assoc/disassoc
1201                  * request has been sent out to fw
1202                  */
1203                 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1204
1205                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1206                 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1207                         eth_zero_addr(pp->mac_addr);
1208                         if (vf == PORT_SELF_VF)
1209                                 eth_zero_addr(netdev->dev_addr);
1210                 }
1211         }
1212
1213         if (vf == PORT_SELF_VF)
1214                 eth_zero_addr(pp->vf_mac);
1215
1216         return err;
1217 }
1218
1219 static int enic_get_vf_port(struct net_device *netdev, int vf,
1220         struct sk_buff *skb)
1221 {
1222         struct enic *enic = netdev_priv(netdev);
1223         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1224         struct enic_port_profile *pp;
1225         int err;
1226
1227         ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1228         if (err)
1229                 return err;
1230
1231         if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1232                 return -ENODATA;
1233
1234         err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1235         if (err)
1236                 return err;
1237
1238         if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1239             nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1240             ((pp->set & ENIC_SET_NAME) &&
1241              nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1242             ((pp->set & ENIC_SET_INSTANCE) &&
1243              nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1244                      pp->instance_uuid)) ||
1245             ((pp->set & ENIC_SET_HOST) &&
1246              nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1247                 goto nla_put_failure;
1248         return 0;
1249
1250 nla_put_failure:
1251         return -EMSGSIZE;
1252 }
1253
1254 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1255 {
1256         struct enic *enic = vnic_dev_priv(rq->vdev);
1257
1258         if (!buf->os_buf)
1259                 return;
1260
1261         pci_unmap_single(enic->pdev, buf->dma_addr,
1262                 buf->len, PCI_DMA_FROMDEVICE);
1263         dev_kfree_skb_any(buf->os_buf);
1264         buf->os_buf = NULL;
1265 }
1266
1267 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1268 {
1269         struct enic *enic = vnic_dev_priv(rq->vdev);
1270         struct net_device *netdev = enic->netdev;
1271         struct sk_buff *skb;
1272         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1273         unsigned int os_buf_index = 0;
1274         dma_addr_t dma_addr;
1275         struct vnic_rq_buf *buf = rq->to_use;
1276
1277         if (buf->os_buf) {
1278                 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1279                                    buf->len);
1280
1281                 return 0;
1282         }
1283         skb = netdev_alloc_skb_ip_align(netdev, len);
1284         if (!skb)
1285                 return -ENOMEM;
1286
1287         dma_addr = pci_map_single(enic->pdev, skb->data, len,
1288                                   PCI_DMA_FROMDEVICE);
1289         if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1290                 dev_kfree_skb(skb);
1291                 return -ENOMEM;
1292         }
1293
1294         enic_queue_rq_desc(rq, skb, os_buf_index,
1295                 dma_addr, len);
1296
1297         return 0;
1298 }
1299
1300 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1301                                       u32 pkt_len)
1302 {
1303         if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1304                 pkt_size->large_pkt_bytes_cnt += pkt_len;
1305         else
1306                 pkt_size->small_pkt_bytes_cnt += pkt_len;
1307 }
1308
1309 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1310                              struct vnic_rq_buf *buf, u16 len)
1311 {
1312         struct enic *enic = netdev_priv(netdev);
1313         struct sk_buff *new_skb;
1314
1315         if (len > enic->rx_copybreak)
1316                 return false;
1317         new_skb = netdev_alloc_skb_ip_align(netdev, len);
1318         if (!new_skb)
1319                 return false;
1320         pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1321                                     DMA_FROM_DEVICE);
1322         memcpy(new_skb->data, (*skb)->data, len);
1323         *skb = new_skb;
1324
1325         return true;
1326 }
1327
1328 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1329         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1330         int skipped, void *opaque)
1331 {
1332         struct enic *enic = vnic_dev_priv(rq->vdev);
1333         struct net_device *netdev = enic->netdev;
1334         struct sk_buff *skb;
1335         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1336
1337         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1338         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1339         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1340         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1341         u8 packet_error;
1342         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1343         u32 rss_hash;
1344         bool outer_csum_ok = true, encap = false;
1345
1346         if (skipped)
1347                 return;
1348
1349         skb = buf->os_buf;
1350
1351         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1352                 &type, &color, &q_number, &completed_index,
1353                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1354                 &csum_not_calc, &rss_hash, &bytes_written,
1355                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1356                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1357                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1358                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1359                 &fcs_ok);
1360
1361         if (packet_error) {
1362
1363                 if (!fcs_ok) {
1364                         if (bytes_written > 0)
1365                                 enic->rq_bad_fcs++;
1366                         else if (bytes_written == 0)
1367                                 enic->rq_truncated_pkts++;
1368                 }
1369
1370                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1371                                  PCI_DMA_FROMDEVICE);
1372                 dev_kfree_skb_any(skb);
1373                 buf->os_buf = NULL;
1374
1375                 return;
1376         }
1377
1378         if (eop && bytes_written > 0) {
1379
1380                 /* Good receive
1381                  */
1382
1383                 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1384                         buf->os_buf = NULL;
1385                         pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1386                                          PCI_DMA_FROMDEVICE);
1387                 }
1388                 prefetch(skb->data - NET_IP_ALIGN);
1389
1390                 skb_put(skb, bytes_written);
1391                 skb->protocol = eth_type_trans(skb, netdev);
1392                 skb_record_rx_queue(skb, q_number);
1393                 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1394                     (type == 3)) {
1395                         switch (rss_type) {
1396                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1397                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1398                         case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1399                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1400                                 break;
1401                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1402                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1403                         case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1404                                 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1405                                 break;
1406                         }
1407                 }
1408                 if (enic->vxlan.vxlan_udp_port_number) {
1409                         switch (enic->vxlan.patch_level) {
1410                         case 0:
1411                                 if (fcoe) {
1412                                         encap = true;
1413                                         outer_csum_ok = fcoe_fc_crc_ok;
1414                                 }
1415                                 break;
1416                         case 2:
1417                                 if ((type == 7) &&
1418                                     (rss_hash & BIT(0))) {
1419                                         encap = true;
1420                                         outer_csum_ok = (rss_hash & BIT(1)) &&
1421                                                         (rss_hash & BIT(2));
1422                                 }
1423                                 break;
1424                         }
1425                 }
1426
1427                 /* Hardware does not provide whole packet checksum. It only
1428                  * provides pseudo checksum. Since hw validates the packet
1429                  * checksum but not provide us the checksum value. use
1430                  * CHECSUM_UNNECESSARY.
1431                  *
1432                  * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1433                  * inner csum_ok. outer_csum_ok is set by hw when outer udp
1434                  * csum is correct or is zero.
1435                  */
1436                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1437                     tcp_udp_csum_ok && outer_csum_ok &&
1438                     (ipv4_csum_ok || ipv6)) {
1439                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1440                         skb->csum_level = encap;
1441                 }
1442
1443                 if (vlan_stripped)
1444                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1445
1446                 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1447                 if (!(netdev->features & NETIF_F_GRO))
1448                         netif_receive_skb(skb);
1449                 else
1450                         napi_gro_receive(&enic->napi[q_number], skb);
1451                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1452                         enic_intr_update_pkt_size(&cq->pkt_size_counter,
1453                                                   bytes_written);
1454         } else {
1455
1456                 /* Buffer overflow
1457                  */
1458
1459                 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1460                                  PCI_DMA_FROMDEVICE);
1461                 dev_kfree_skb_any(skb);
1462                 buf->os_buf = NULL;
1463         }
1464 }
1465
1466 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1467         u8 type, u16 q_number, u16 completed_index, void *opaque)
1468 {
1469         struct enic *enic = vnic_dev_priv(vdev);
1470
1471         vnic_rq_service(&enic->rq[q_number], cq_desc,
1472                 completed_index, VNIC_RQ_RETURN_DESC,
1473                 enic_rq_indicate_buf, opaque);
1474
1475         return 0;
1476 }
1477
1478 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1479 {
1480         unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1481         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1482         u32 timer = cq->tobe_rx_coal_timeval;
1483
1484         if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1485                 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1486                 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1487         }
1488 }
1489
1490 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1491 {
1492         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1493         struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1494         struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1495         int index;
1496         u32 timer;
1497         u32 range_start;
1498         u32 traffic;
1499         u64 delta;
1500         ktime_t now = ktime_get();
1501
1502         delta = ktime_us_delta(now, cq->prev_ts);
1503         if (delta < ENIC_AIC_TS_BREAK)
1504                 return;
1505         cq->prev_ts = now;
1506
1507         traffic = pkt_size_counter->large_pkt_bytes_cnt +
1508                   pkt_size_counter->small_pkt_bytes_cnt;
1509         /* The table takes Mbps
1510          * traffic *= 8    => bits
1511          * traffic *= (10^6 / delta)    => bps
1512          * traffic /= 10^6     => Mbps
1513          *
1514          * Combining, traffic *= (8 / delta)
1515          */
1516
1517         traffic <<= 3;
1518         traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1519
1520         for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1521                 if (traffic < mod_table[index].rx_rate)
1522                         break;
1523         range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1524                        pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1525                       rx_coal->small_pkt_range_start :
1526                       rx_coal->large_pkt_range_start;
1527         timer = range_start + ((rx_coal->range_end - range_start) *
1528                                mod_table[index].range_percent / 100);
1529         /* Damping */
1530         cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1531
1532         pkt_size_counter->large_pkt_bytes_cnt = 0;
1533         pkt_size_counter->small_pkt_bytes_cnt = 0;
1534 }
1535
1536 static int enic_poll(struct napi_struct *napi, int budget)
1537 {
1538         struct net_device *netdev = napi->dev;
1539         struct enic *enic = netdev_priv(netdev);
1540         unsigned int cq_rq = enic_cq_rq(enic, 0);
1541         unsigned int cq_wq = enic_cq_wq(enic, 0);
1542         unsigned int intr = enic_legacy_io_intr();
1543         unsigned int rq_work_to_do = budget;
1544         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1545         unsigned int  work_done, rq_work_done = 0, wq_work_done;
1546         int err;
1547
1548         wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1549                                        enic_wq_service, NULL);
1550
1551         if (budget > 0)
1552                 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1553                         rq_work_to_do, enic_rq_service, NULL);
1554
1555         /* Accumulate intr event credits for this polling
1556          * cycle.  An intr event is the completion of a
1557          * a WQ or RQ packet.
1558          */
1559
1560         work_done = rq_work_done + wq_work_done;
1561
1562         if (work_done > 0)
1563                 vnic_intr_return_credits(&enic->intr[intr],
1564                         work_done,
1565                         0 /* don't unmask intr */,
1566                         0 /* don't reset intr timer */);
1567
1568         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1569
1570         /* Buffer allocation failed. Stay in polling
1571          * mode so we can try to fill the ring again.
1572          */
1573
1574         if (err)
1575                 rq_work_done = rq_work_to_do;
1576         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1577                 /* Call the function which refreshes the intr coalescing timer
1578                  * value based on the traffic.
1579                  */
1580                 enic_calc_int_moderation(enic, &enic->rq[0]);
1581
1582         if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1583
1584                 /* Some work done, but not enough to stay in polling,
1585                  * exit polling
1586                  */
1587
1588                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1589                         enic_set_int_moderation(enic, &enic->rq[0]);
1590                 vnic_intr_unmask(&enic->intr[intr]);
1591         }
1592
1593         return rq_work_done;
1594 }
1595
1596 #ifdef CONFIG_RFS_ACCEL
1597 static void enic_free_rx_cpu_rmap(struct enic *enic)
1598 {
1599         free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1600         enic->netdev->rx_cpu_rmap = NULL;
1601 }
1602
1603 static void enic_set_rx_cpu_rmap(struct enic *enic)
1604 {
1605         int i, res;
1606
1607         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1608                 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1609                 if (unlikely(!enic->netdev->rx_cpu_rmap))
1610                         return;
1611                 for (i = 0; i < enic->rq_count; i++) {
1612                         res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1613                                                enic->msix_entry[i].vector);
1614                         if (unlikely(res)) {
1615                                 enic_free_rx_cpu_rmap(enic);
1616                                 return;
1617                         }
1618                 }
1619         }
1620 }
1621
1622 #else
1623
1624 static void enic_free_rx_cpu_rmap(struct enic *enic)
1625 {
1626 }
1627
1628 static void enic_set_rx_cpu_rmap(struct enic *enic)
1629 {
1630 }
1631
1632 #endif /* CONFIG_RFS_ACCEL */
1633
1634 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1635 {
1636         struct net_device *netdev = napi->dev;
1637         struct enic *enic = netdev_priv(netdev);
1638         unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1639         struct vnic_wq *wq = &enic->wq[wq_index];
1640         unsigned int cq;
1641         unsigned int intr;
1642         unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1643         unsigned int wq_work_done;
1644         unsigned int wq_irq;
1645
1646         wq_irq = wq->index;
1647         cq = enic_cq_wq(enic, wq_irq);
1648         intr = enic_msix_wq_intr(enic, wq_irq);
1649         wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1650                                        enic_wq_service, NULL);
1651
1652         vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1653                                  0 /* don't unmask intr */,
1654                                  1 /* reset intr timer */);
1655         if (!wq_work_done) {
1656                 napi_complete(napi);
1657                 vnic_intr_unmask(&enic->intr[intr]);
1658                 return 0;
1659         }
1660
1661         return budget;
1662 }
1663
1664 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1665 {
1666         struct net_device *netdev = napi->dev;
1667         struct enic *enic = netdev_priv(netdev);
1668         unsigned int rq = (napi - &enic->napi[0]);
1669         unsigned int cq = enic_cq_rq(enic, rq);
1670         unsigned int intr = enic_msix_rq_intr(enic, rq);
1671         unsigned int work_to_do = budget;
1672         unsigned int work_done = 0;
1673         int err;
1674
1675         /* Service RQ
1676          */
1677
1678         if (budget > 0)
1679                 work_done = vnic_cq_service(&enic->cq[cq],
1680                         work_to_do, enic_rq_service, NULL);
1681
1682         /* Return intr event credits for this polling
1683          * cycle.  An intr event is the completion of a
1684          * RQ packet.
1685          */
1686
1687         if (work_done > 0)
1688                 vnic_intr_return_credits(&enic->intr[intr],
1689                         work_done,
1690                         0 /* don't unmask intr */,
1691                         0 /* don't reset intr timer */);
1692
1693         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1694
1695         /* Buffer allocation failed. Stay in polling mode
1696          * so we can try to fill the ring again.
1697          */
1698
1699         if (err)
1700                 work_done = work_to_do;
1701         if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1702                 /* Call the function which refreshes the intr coalescing timer
1703                  * value based on the traffic.
1704                  */
1705                 enic_calc_int_moderation(enic, &enic->rq[rq]);
1706
1707         if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1708
1709                 /* Some work done, but not enough to stay in polling,
1710                  * exit polling
1711                  */
1712
1713                 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1714                         enic_set_int_moderation(enic, &enic->rq[rq]);
1715                 vnic_intr_unmask(&enic->intr[intr]);
1716         }
1717
1718         return work_done;
1719 }
1720
1721 static void enic_notify_timer(struct timer_list *t)
1722 {
1723         struct enic *enic = from_timer(enic, t, notify_timer);
1724
1725         enic_notify_check(enic);
1726
1727         mod_timer(&enic->notify_timer,
1728                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1729 }
1730
1731 static void enic_free_intr(struct enic *enic)
1732 {
1733         struct net_device *netdev = enic->netdev;
1734         unsigned int i;
1735
1736         enic_free_rx_cpu_rmap(enic);
1737         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1738         case VNIC_DEV_INTR_MODE_INTX:
1739                 free_irq(enic->pdev->irq, netdev);
1740                 break;
1741         case VNIC_DEV_INTR_MODE_MSI:
1742                 free_irq(enic->pdev->irq, enic);
1743                 break;
1744         case VNIC_DEV_INTR_MODE_MSIX:
1745                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1746                         if (enic->msix[i].requested)
1747                                 free_irq(enic->msix_entry[i].vector,
1748                                         enic->msix[i].devid);
1749                 break;
1750         default:
1751                 break;
1752         }
1753 }
1754
1755 static int enic_request_intr(struct enic *enic)
1756 {
1757         struct net_device *netdev = enic->netdev;
1758         unsigned int i, intr;
1759         int err = 0;
1760
1761         enic_set_rx_cpu_rmap(enic);
1762         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1763
1764         case VNIC_DEV_INTR_MODE_INTX:
1765
1766                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1767                         IRQF_SHARED, netdev->name, netdev);
1768                 break;
1769
1770         case VNIC_DEV_INTR_MODE_MSI:
1771
1772                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1773                         0, netdev->name, enic);
1774                 break;
1775
1776         case VNIC_DEV_INTR_MODE_MSIX:
1777
1778                 for (i = 0; i < enic->rq_count; i++) {
1779                         intr = enic_msix_rq_intr(enic, i);
1780                         snprintf(enic->msix[intr].devname,
1781                                 sizeof(enic->msix[intr].devname),
1782                                 "%s-rx-%u", netdev->name, i);
1783                         enic->msix[intr].isr = enic_isr_msix;
1784                         enic->msix[intr].devid = &enic->napi[i];
1785                 }
1786
1787                 for (i = 0; i < enic->wq_count; i++) {
1788                         int wq = enic_cq_wq(enic, i);
1789
1790                         intr = enic_msix_wq_intr(enic, i);
1791                         snprintf(enic->msix[intr].devname,
1792                                 sizeof(enic->msix[intr].devname),
1793                                 "%s-tx-%u", netdev->name, i);
1794                         enic->msix[intr].isr = enic_isr_msix;
1795                         enic->msix[intr].devid = &enic->napi[wq];
1796                 }
1797
1798                 intr = enic_msix_err_intr(enic);
1799                 snprintf(enic->msix[intr].devname,
1800                         sizeof(enic->msix[intr].devname),
1801                         "%s-err", netdev->name);
1802                 enic->msix[intr].isr = enic_isr_msix_err;
1803                 enic->msix[intr].devid = enic;
1804
1805                 intr = enic_msix_notify_intr(enic);
1806                 snprintf(enic->msix[intr].devname,
1807                         sizeof(enic->msix[intr].devname),
1808                         "%s-notify", netdev->name);
1809                 enic->msix[intr].isr = enic_isr_msix_notify;
1810                 enic->msix[intr].devid = enic;
1811
1812                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1813                         enic->msix[i].requested = 0;
1814
1815                 for (i = 0; i < enic->intr_count; i++) {
1816                         err = request_irq(enic->msix_entry[i].vector,
1817                                 enic->msix[i].isr, 0,
1818                                 enic->msix[i].devname,
1819                                 enic->msix[i].devid);
1820                         if (err) {
1821                                 enic_free_intr(enic);
1822                                 break;
1823                         }
1824                         enic->msix[i].requested = 1;
1825                 }
1826
1827                 break;
1828
1829         default:
1830                 break;
1831         }
1832
1833         return err;
1834 }
1835
1836 static void enic_synchronize_irqs(struct enic *enic)
1837 {
1838         unsigned int i;
1839
1840         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1841         case VNIC_DEV_INTR_MODE_INTX:
1842         case VNIC_DEV_INTR_MODE_MSI:
1843                 synchronize_irq(enic->pdev->irq);
1844                 break;
1845         case VNIC_DEV_INTR_MODE_MSIX:
1846                 for (i = 0; i < enic->intr_count; i++)
1847                         synchronize_irq(enic->msix_entry[i].vector);
1848                 break;
1849         default:
1850                 break;
1851         }
1852 }
1853
1854 static void enic_set_rx_coal_setting(struct enic *enic)
1855 {
1856         unsigned int speed;
1857         int index = -1;
1858         struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1859
1860         /* 1. Read the link speed from fw
1861          * 2. Pick the default range for the speed
1862          * 3. Update it in enic->rx_coalesce_setting
1863          */
1864         speed = vnic_dev_port_speed(enic->vdev);
1865         if (ENIC_LINK_SPEED_10G < speed)
1866                 index = ENIC_LINK_40G_INDEX;
1867         else if (ENIC_LINK_SPEED_4G < speed)
1868                 index = ENIC_LINK_10G_INDEX;
1869         else
1870                 index = ENIC_LINK_4G_INDEX;
1871
1872         rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1873         rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1874         rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1875
1876         /* Start with the value provided by UCSM */
1877         for (index = 0; index < enic->rq_count; index++)
1878                 enic->cq[index].cur_rx_coal_timeval =
1879                                 enic->config.intr_timer_usec;
1880
1881         rx_coal->use_adaptive_rx_coalesce = 1;
1882 }
1883
1884 static int enic_dev_notify_set(struct enic *enic)
1885 {
1886         int err;
1887
1888         spin_lock_bh(&enic->devcmd_lock);
1889         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1890         case VNIC_DEV_INTR_MODE_INTX:
1891                 err = vnic_dev_notify_set(enic->vdev,
1892                         enic_legacy_notify_intr());
1893                 break;
1894         case VNIC_DEV_INTR_MODE_MSIX:
1895                 err = vnic_dev_notify_set(enic->vdev,
1896                         enic_msix_notify_intr(enic));
1897                 break;
1898         default:
1899                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1900                 break;
1901         }
1902         spin_unlock_bh(&enic->devcmd_lock);
1903
1904         return err;
1905 }
1906
1907 static void enic_notify_timer_start(struct enic *enic)
1908 {
1909         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1910         case VNIC_DEV_INTR_MODE_MSI:
1911                 mod_timer(&enic->notify_timer, jiffies);
1912                 break;
1913         default:
1914                 /* Using intr for notification for INTx/MSI-X */
1915                 break;
1916         }
1917 }
1918
1919 /* rtnl lock is held, process context */
1920 static int enic_open(struct net_device *netdev)
1921 {
1922         struct enic *enic = netdev_priv(netdev);
1923         unsigned int i;
1924         int err, ret;
1925
1926         err = enic_request_intr(enic);
1927         if (err) {
1928                 netdev_err(netdev, "Unable to request irq.\n");
1929                 return err;
1930         }
1931         enic_init_affinity_hint(enic);
1932         enic_set_affinity_hint(enic);
1933
1934         err = enic_dev_notify_set(enic);
1935         if (err) {
1936                 netdev_err(netdev,
1937                         "Failed to alloc notify buffer, aborting.\n");
1938                 goto err_out_free_intr;
1939         }
1940
1941         for (i = 0; i < enic->rq_count; i++) {
1942                 /* enable rq before updating rq desc */
1943                 vnic_rq_enable(&enic->rq[i]);
1944                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1945                 /* Need at least one buffer on ring to get going */
1946                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1947                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1948                         err = -ENOMEM;
1949                         goto err_out_free_rq;
1950                 }
1951         }
1952
1953         for (i = 0; i < enic->wq_count; i++)
1954                 vnic_wq_enable(&enic->wq[i]);
1955
1956         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1957                 enic_dev_add_station_addr(enic);
1958
1959         enic_set_rx_mode(netdev);
1960
1961         netif_tx_wake_all_queues(netdev);
1962
1963         for (i = 0; i < enic->rq_count; i++)
1964                 napi_enable(&enic->napi[i]);
1965
1966         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1967                 for (i = 0; i < enic->wq_count; i++)
1968                         napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1969         enic_dev_enable(enic);
1970
1971         for (i = 0; i < enic->intr_count; i++)
1972                 vnic_intr_unmask(&enic->intr[i]);
1973
1974         enic_notify_timer_start(enic);
1975         enic_rfs_timer_start(enic);
1976
1977         return 0;
1978
1979 err_out_free_rq:
1980         for (i = 0; i < enic->rq_count; i++) {
1981                 ret = vnic_rq_disable(&enic->rq[i]);
1982                 if (!ret)
1983                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1984         }
1985         enic_dev_notify_unset(enic);
1986 err_out_free_intr:
1987         enic_unset_affinity_hint(enic);
1988         enic_free_intr(enic);
1989
1990         return err;
1991 }
1992
1993 /* rtnl lock is held, process context */
1994 static int enic_stop(struct net_device *netdev)
1995 {
1996         struct enic *enic = netdev_priv(netdev);
1997         unsigned int i;
1998         int err;
1999
2000         for (i = 0; i < enic->intr_count; i++) {
2001                 vnic_intr_mask(&enic->intr[i]);
2002                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
2003         }
2004
2005         enic_synchronize_irqs(enic);
2006
2007         del_timer_sync(&enic->notify_timer);
2008         enic_rfs_flw_tbl_free(enic);
2009
2010         enic_dev_disable(enic);
2011
2012         for (i = 0; i < enic->rq_count; i++)
2013                 napi_disable(&enic->napi[i]);
2014
2015         netif_carrier_off(netdev);
2016         netif_tx_disable(netdev);
2017         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2018                 for (i = 0; i < enic->wq_count; i++)
2019                         napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
2020
2021         if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
2022                 enic_dev_del_station_addr(enic);
2023
2024         for (i = 0; i < enic->wq_count; i++) {
2025                 err = vnic_wq_disable(&enic->wq[i]);
2026                 if (err)
2027                         return err;
2028         }
2029         for (i = 0; i < enic->rq_count; i++) {
2030                 err = vnic_rq_disable(&enic->rq[i]);
2031                 if (err)
2032                         return err;
2033         }
2034
2035         enic_dev_notify_unset(enic);
2036         enic_unset_affinity_hint(enic);
2037         enic_free_intr(enic);
2038
2039         for (i = 0; i < enic->wq_count; i++)
2040                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
2041         for (i = 0; i < enic->rq_count; i++)
2042                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
2043         for (i = 0; i < enic->cq_count; i++)
2044                 vnic_cq_clean(&enic->cq[i]);
2045         for (i = 0; i < enic->intr_count; i++)
2046                 vnic_intr_clean(&enic->intr[i]);
2047
2048         return 0;
2049 }
2050
2051 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
2052 {
2053         bool running = netif_running(netdev);
2054         int err = 0;
2055
2056         ASSERT_RTNL();
2057         if (running) {
2058                 err = enic_stop(netdev);
2059                 if (err)
2060                         return err;
2061         }
2062
2063         netdev->mtu = new_mtu;
2064
2065         if (running) {
2066                 err = enic_open(netdev);
2067                 if (err)
2068                         return err;
2069         }
2070
2071         return 0;
2072 }
2073
2074 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2075 {
2076         struct enic *enic = netdev_priv(netdev);
2077
2078         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2079                 return -EOPNOTSUPP;
2080
2081         if (netdev->mtu > enic->port_mtu)
2082                 netdev_warn(netdev,
2083                             "interface MTU (%d) set higher than port MTU (%d)\n",
2084                             netdev->mtu, enic->port_mtu);
2085
2086         return _enic_change_mtu(netdev, new_mtu);
2087 }
2088
2089 static void enic_change_mtu_work(struct work_struct *work)
2090 {
2091         struct enic *enic = container_of(work, struct enic, change_mtu_work);
2092         struct net_device *netdev = enic->netdev;
2093         int new_mtu = vnic_dev_mtu(enic->vdev);
2094
2095         rtnl_lock();
2096         (void)_enic_change_mtu(netdev, new_mtu);
2097         rtnl_unlock();
2098
2099         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2100 }
2101
2102 #ifdef CONFIG_NET_POLL_CONTROLLER
2103 static void enic_poll_controller(struct net_device *netdev)
2104 {
2105         struct enic *enic = netdev_priv(netdev);
2106         struct vnic_dev *vdev = enic->vdev;
2107         unsigned int i, intr;
2108
2109         switch (vnic_dev_get_intr_mode(vdev)) {
2110         case VNIC_DEV_INTR_MODE_MSIX:
2111                 for (i = 0; i < enic->rq_count; i++) {
2112                         intr = enic_msix_rq_intr(enic, i);
2113                         enic_isr_msix(enic->msix_entry[intr].vector,
2114                                       &enic->napi[i]);
2115                 }
2116
2117                 for (i = 0; i < enic->wq_count; i++) {
2118                         intr = enic_msix_wq_intr(enic, i);
2119                         enic_isr_msix(enic->msix_entry[intr].vector,
2120                                       &enic->napi[enic_cq_wq(enic, i)]);
2121                 }
2122
2123                 break;
2124         case VNIC_DEV_INTR_MODE_MSI:
2125                 enic_isr_msi(enic->pdev->irq, enic);
2126                 break;
2127         case VNIC_DEV_INTR_MODE_INTX:
2128                 enic_isr_legacy(enic->pdev->irq, netdev);
2129                 break;
2130         default:
2131                 break;
2132         }
2133 }
2134 #endif
2135
2136 static int enic_dev_wait(struct vnic_dev *vdev,
2137         int (*start)(struct vnic_dev *, int),
2138         int (*finished)(struct vnic_dev *, int *),
2139         int arg)
2140 {
2141         unsigned long time;
2142         int done;
2143         int err;
2144
2145         BUG_ON(in_interrupt());
2146
2147         err = start(vdev, arg);
2148         if (err)
2149                 return err;
2150
2151         /* Wait for func to complete...2 seconds max
2152          */
2153
2154         time = jiffies + (HZ * 2);
2155         do {
2156
2157                 err = finished(vdev, &done);
2158                 if (err)
2159                         return err;
2160
2161                 if (done)
2162                         return 0;
2163
2164                 schedule_timeout_uninterruptible(HZ / 10);
2165
2166         } while (time_after(time, jiffies));
2167
2168         return -ETIMEDOUT;
2169 }
2170
2171 static int enic_dev_open(struct enic *enic)
2172 {
2173         int err;
2174         u32 flags = CMD_OPENF_IG_DESCCACHE;
2175
2176         err = enic_dev_wait(enic->vdev, vnic_dev_open,
2177                 vnic_dev_open_done, flags);
2178         if (err)
2179                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2180                         err);
2181
2182         return err;
2183 }
2184
2185 static int enic_dev_soft_reset(struct enic *enic)
2186 {
2187         int err;
2188
2189         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2190                             vnic_dev_soft_reset_done, 0);
2191         if (err)
2192                 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2193                            err);
2194
2195         return err;
2196 }
2197
2198 static int enic_dev_hang_reset(struct enic *enic)
2199 {
2200         int err;
2201
2202         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2203                 vnic_dev_hang_reset_done, 0);
2204         if (err)
2205                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2206                         err);
2207
2208         return err;
2209 }
2210
2211 int __enic_set_rsskey(struct enic *enic)
2212 {
2213         union vnic_rss_key *rss_key_buf_va;
2214         dma_addr_t rss_key_buf_pa;
2215         int i, kidx, bidx, err;
2216
2217         rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
2218                                                sizeof(union vnic_rss_key),
2219                                                &rss_key_buf_pa);
2220         if (!rss_key_buf_va)
2221                 return -ENOMEM;
2222
2223         for (i = 0; i < ENIC_RSS_LEN; i++) {
2224                 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2225                 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2226                 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2227         }
2228         spin_lock_bh(&enic->devcmd_lock);
2229         err = enic_set_rss_key(enic,
2230                 rss_key_buf_pa,
2231                 sizeof(union vnic_rss_key));
2232         spin_unlock_bh(&enic->devcmd_lock);
2233
2234         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2235                 rss_key_buf_va, rss_key_buf_pa);
2236
2237         return err;
2238 }
2239
2240 static int enic_set_rsskey(struct enic *enic)
2241 {
2242         netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2243
2244         return __enic_set_rsskey(enic);
2245 }
2246
2247 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2248 {
2249         dma_addr_t rss_cpu_buf_pa;
2250         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2251         unsigned int i;
2252         int err;
2253
2254         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2255                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2256         if (!rss_cpu_buf_va)
2257                 return -ENOMEM;
2258
2259         for (i = 0; i < (1 << rss_hash_bits); i++)
2260                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2261
2262         spin_lock_bh(&enic->devcmd_lock);
2263         err = enic_set_rss_cpu(enic,
2264                 rss_cpu_buf_pa,
2265                 sizeof(union vnic_rss_cpu));
2266         spin_unlock_bh(&enic->devcmd_lock);
2267
2268         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2269                 rss_cpu_buf_va, rss_cpu_buf_pa);
2270
2271         return err;
2272 }
2273
2274 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2275         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2276 {
2277         const u8 tso_ipid_split_en = 0;
2278         const u8 ig_vlan_strip_en = 1;
2279         int err;
2280
2281         /* Enable VLAN tag stripping.
2282         */
2283
2284         spin_lock_bh(&enic->devcmd_lock);
2285         err = enic_set_nic_cfg(enic,
2286                 rss_default_cpu, rss_hash_type,
2287                 rss_hash_bits, rss_base_cpu,
2288                 rss_enable, tso_ipid_split_en,
2289                 ig_vlan_strip_en);
2290         spin_unlock_bh(&enic->devcmd_lock);
2291
2292         return err;
2293 }
2294
2295 static int enic_set_rss_nic_cfg(struct enic *enic)
2296 {
2297         struct device *dev = enic_get_dev(enic);
2298         const u8 rss_default_cpu = 0;
2299         const u8 rss_hash_bits = 7;
2300         const u8 rss_base_cpu = 0;
2301         u8 rss_hash_type;
2302         int res;
2303         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2304
2305         spin_lock_bh(&enic->devcmd_lock);
2306         res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type);
2307         spin_unlock_bh(&enic->devcmd_lock);
2308         if (res) {
2309                 /* defaults for old adapters
2310                  */
2311                 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4      |
2312                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4  |
2313                                 NIC_CFG_RSS_HASH_TYPE_IPV6      |
2314                                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2315         }
2316
2317         if (rss_enable) {
2318                 if (!enic_set_rsskey(enic)) {
2319                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
2320                                 rss_enable = 0;
2321                                 dev_warn(dev, "RSS disabled, "
2322                                         "Failed to set RSS cpu indirection table.");
2323                         }
2324                 } else {
2325                         rss_enable = 0;
2326                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2327                 }
2328         }
2329
2330         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2331                 rss_hash_bits, rss_base_cpu, rss_enable);
2332 }
2333
2334 static void enic_reset(struct work_struct *work)
2335 {
2336         struct enic *enic = container_of(work, struct enic, reset);
2337
2338         if (!netif_running(enic->netdev))
2339                 return;
2340
2341         rtnl_lock();
2342
2343         spin_lock(&enic->enic_api_lock);
2344         enic_stop(enic->netdev);
2345         enic_dev_soft_reset(enic);
2346         enic_reset_addr_lists(enic);
2347         enic_init_vnic_resources(enic);
2348         enic_set_rss_nic_cfg(enic);
2349         enic_dev_set_ig_vlan_rewrite_mode(enic);
2350         enic_open(enic->netdev);
2351         spin_unlock(&enic->enic_api_lock);
2352         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2353
2354         rtnl_unlock();
2355 }
2356
2357 static void enic_tx_hang_reset(struct work_struct *work)
2358 {
2359         struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2360
2361         rtnl_lock();
2362
2363         spin_lock(&enic->enic_api_lock);
2364         enic_dev_hang_notify(enic);
2365         enic_stop(enic->netdev);
2366         enic_dev_hang_reset(enic);
2367         enic_reset_addr_lists(enic);
2368         enic_init_vnic_resources(enic);
2369         enic_set_rss_nic_cfg(enic);
2370         enic_dev_set_ig_vlan_rewrite_mode(enic);
2371         enic_open(enic->netdev);
2372         spin_unlock(&enic->enic_api_lock);
2373         call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2374
2375         rtnl_unlock();
2376 }
2377
2378 static int enic_set_intr_mode(struct enic *enic)
2379 {
2380         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2381         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2382         unsigned int i;
2383
2384         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2385          * on system capabilities.
2386          *
2387          * Try MSI-X first
2388          *
2389          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2390          * (the second to last INTR is used for WQ/RQ errors)
2391          * (the last INTR is used for notifications)
2392          */
2393
2394         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2395         for (i = 0; i < n + m + 2; i++)
2396                 enic->msix_entry[i].entry = i;
2397
2398         /* Use multiple RQs if RSS is enabled
2399          */
2400
2401         if (ENIC_SETTING(enic, RSS) &&
2402             enic->config.intr_mode < 1 &&
2403             enic->rq_count >= n &&
2404             enic->wq_count >= m &&
2405             enic->cq_count >= n + m &&
2406             enic->intr_count >= n + m + 2) {
2407
2408                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2409                                           n + m + 2, n + m + 2) > 0) {
2410
2411                         enic->rq_count = n;
2412                         enic->wq_count = m;
2413                         enic->cq_count = n + m;
2414                         enic->intr_count = n + m + 2;
2415
2416                         vnic_dev_set_intr_mode(enic->vdev,
2417                                 VNIC_DEV_INTR_MODE_MSIX);
2418
2419                         return 0;
2420                 }
2421         }
2422
2423         if (enic->config.intr_mode < 1 &&
2424             enic->rq_count >= 1 &&
2425             enic->wq_count >= m &&
2426             enic->cq_count >= 1 + m &&
2427             enic->intr_count >= 1 + m + 2) {
2428                 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2429                                           1 + m + 2, 1 + m + 2) > 0) {
2430
2431                         enic->rq_count = 1;
2432                         enic->wq_count = m;
2433                         enic->cq_count = 1 + m;
2434                         enic->intr_count = 1 + m + 2;
2435
2436                         vnic_dev_set_intr_mode(enic->vdev,
2437                                 VNIC_DEV_INTR_MODE_MSIX);
2438
2439                         return 0;
2440                 }
2441         }
2442
2443         /* Next try MSI
2444          *
2445          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2446          */
2447
2448         if (enic->config.intr_mode < 2 &&
2449             enic->rq_count >= 1 &&
2450             enic->wq_count >= 1 &&
2451             enic->cq_count >= 2 &&
2452             enic->intr_count >= 1 &&
2453             !pci_enable_msi(enic->pdev)) {
2454
2455                 enic->rq_count = 1;
2456                 enic->wq_count = 1;
2457                 enic->cq_count = 2;
2458                 enic->intr_count = 1;
2459
2460                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2461
2462                 return 0;
2463         }
2464
2465         /* Next try INTx
2466          *
2467          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2468          * (the first INTR is used for WQ/RQ)
2469          * (the second INTR is used for WQ/RQ errors)
2470          * (the last INTR is used for notifications)
2471          */
2472
2473         if (enic->config.intr_mode < 3 &&
2474             enic->rq_count >= 1 &&
2475             enic->wq_count >= 1 &&
2476             enic->cq_count >= 2 &&
2477             enic->intr_count >= 3) {
2478
2479                 enic->rq_count = 1;
2480                 enic->wq_count = 1;
2481                 enic->cq_count = 2;
2482                 enic->intr_count = 3;
2483
2484                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2485
2486                 return 0;
2487         }
2488
2489         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2490
2491         return -EINVAL;
2492 }
2493
2494 static void enic_clear_intr_mode(struct enic *enic)
2495 {
2496         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2497         case VNIC_DEV_INTR_MODE_MSIX:
2498                 pci_disable_msix(enic->pdev);
2499                 break;
2500         case VNIC_DEV_INTR_MODE_MSI:
2501                 pci_disable_msi(enic->pdev);
2502                 break;
2503         default:
2504                 break;
2505         }
2506
2507         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2508 }
2509
2510 static const struct net_device_ops enic_netdev_dynamic_ops = {
2511         .ndo_open               = enic_open,
2512         .ndo_stop               = enic_stop,
2513         .ndo_start_xmit         = enic_hard_start_xmit,
2514         .ndo_get_stats64        = enic_get_stats,
2515         .ndo_validate_addr      = eth_validate_addr,
2516         .ndo_set_rx_mode        = enic_set_rx_mode,
2517         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2518         .ndo_change_mtu         = enic_change_mtu,
2519         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2520         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2521         .ndo_tx_timeout         = enic_tx_timeout,
2522         .ndo_set_vf_port        = enic_set_vf_port,
2523         .ndo_get_vf_port        = enic_get_vf_port,
2524         .ndo_set_vf_mac         = enic_set_vf_mac,
2525 #ifdef CONFIG_NET_POLL_CONTROLLER
2526         .ndo_poll_controller    = enic_poll_controller,
2527 #endif
2528 #ifdef CONFIG_RFS_ACCEL
2529         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2530 #endif
2531         .ndo_udp_tunnel_add     = enic_udp_tunnel_add,
2532         .ndo_udp_tunnel_del     = enic_udp_tunnel_del,
2533         .ndo_features_check     = enic_features_check,
2534 };
2535
2536 static const struct net_device_ops enic_netdev_ops = {
2537         .ndo_open               = enic_open,
2538         .ndo_stop               = enic_stop,
2539         .ndo_start_xmit         = enic_hard_start_xmit,
2540         .ndo_get_stats64        = enic_get_stats,
2541         .ndo_validate_addr      = eth_validate_addr,
2542         .ndo_set_mac_address    = enic_set_mac_address,
2543         .ndo_set_rx_mode        = enic_set_rx_mode,
2544         .ndo_change_mtu         = enic_change_mtu,
2545         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2546         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2547         .ndo_tx_timeout         = enic_tx_timeout,
2548         .ndo_set_vf_port        = enic_set_vf_port,
2549         .ndo_get_vf_port        = enic_get_vf_port,
2550         .ndo_set_vf_mac         = enic_set_vf_mac,
2551 #ifdef CONFIG_NET_POLL_CONTROLLER
2552         .ndo_poll_controller    = enic_poll_controller,
2553 #endif
2554 #ifdef CONFIG_RFS_ACCEL
2555         .ndo_rx_flow_steer      = enic_rx_flow_steer,
2556 #endif
2557         .ndo_udp_tunnel_add     = enic_udp_tunnel_add,
2558         .ndo_udp_tunnel_del     = enic_udp_tunnel_del,
2559         .ndo_features_check     = enic_features_check,
2560 };
2561
2562 static void enic_dev_deinit(struct enic *enic)
2563 {
2564         unsigned int i;
2565
2566         for (i = 0; i < enic->rq_count; i++) {
2567                 napi_hash_del(&enic->napi[i]);
2568                 netif_napi_del(&enic->napi[i]);
2569         }
2570         if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2571                 for (i = 0; i < enic->wq_count; i++)
2572                         netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2573
2574         enic_free_vnic_resources(enic);
2575         enic_clear_intr_mode(enic);
2576         enic_free_affinity_hint(enic);
2577 }
2578
2579 static void enic_kdump_kernel_config(struct enic *enic)
2580 {
2581         if (is_kdump_kernel()) {
2582                 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2583                 enic->rq_count = 1;
2584                 enic->wq_count = 1;
2585                 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2586                 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2587                 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2588         }
2589 }
2590
2591 static int enic_dev_init(struct enic *enic)
2592 {
2593         struct device *dev = enic_get_dev(enic);
2594         struct net_device *netdev = enic->netdev;
2595         unsigned int i;
2596         int err;
2597
2598         /* Get interrupt coalesce timer info */
2599         err = enic_dev_intr_coal_timer_info(enic);
2600         if (err) {
2601                 dev_warn(dev, "Using default conversion factor for "
2602                         "interrupt coalesce timer\n");
2603                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2604         }
2605
2606         /* Get vNIC configuration
2607          */
2608
2609         err = enic_get_vnic_config(enic);
2610         if (err) {
2611                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2612                 return err;
2613         }
2614
2615         /* Get available resource counts
2616          */
2617
2618         enic_get_res_counts(enic);
2619
2620         /* modify resource count if we are in kdump_kernel
2621          */
2622         enic_kdump_kernel_config(enic);
2623
2624         /* Set interrupt mode based on resource counts and system
2625          * capabilities
2626          */
2627
2628         err = enic_set_intr_mode(enic);
2629         if (err) {
2630                 dev_err(dev, "Failed to set intr mode based on resource "
2631                         "counts and system capabilities, aborting\n");
2632                 return err;
2633         }
2634
2635         /* Allocate and configure vNIC resources
2636          */
2637
2638         err = enic_alloc_vnic_resources(enic);
2639         if (err) {
2640                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2641                 goto err_out_free_vnic_resources;
2642         }
2643
2644         enic_init_vnic_resources(enic);
2645
2646         err = enic_set_rss_nic_cfg(enic);
2647         if (err) {
2648                 dev_err(dev, "Failed to config nic, aborting\n");
2649                 goto err_out_free_vnic_resources;
2650         }
2651
2652         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2653         default:
2654                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2655                 break;
2656         case VNIC_DEV_INTR_MODE_MSIX:
2657                 for (i = 0; i < enic->rq_count; i++) {
2658                         netif_napi_add(netdev, &enic->napi[i],
2659                                 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2660                 }
2661                 for (i = 0; i < enic->wq_count; i++)
2662                         netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2663                                        enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2664                 break;
2665         }
2666
2667         return 0;
2668
2669 err_out_free_vnic_resources:
2670         enic_free_affinity_hint(enic);
2671         enic_clear_intr_mode(enic);
2672         enic_free_vnic_resources(enic);
2673
2674         return err;
2675 }
2676
2677 static void enic_iounmap(struct enic *enic)
2678 {
2679         unsigned int i;
2680
2681         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2682                 if (enic->bar[i].vaddr)
2683                         iounmap(enic->bar[i].vaddr);
2684 }
2685
2686 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2687 {
2688         struct device *dev = &pdev->dev;
2689         struct net_device *netdev;
2690         struct enic *enic;
2691         int using_dac = 0;
2692         unsigned int i;
2693         int err;
2694 #ifdef CONFIG_PCI_IOV
2695         int pos = 0;
2696 #endif
2697         int num_pps = 1;
2698
2699         /* Allocate net device structure and initialize.  Private
2700          * instance data is initialized to zero.
2701          */
2702
2703         netdev = alloc_etherdev_mqs(sizeof(struct enic),
2704                                     ENIC_RQ_MAX, ENIC_WQ_MAX);
2705         if (!netdev)
2706                 return -ENOMEM;
2707
2708         pci_set_drvdata(pdev, netdev);
2709
2710         SET_NETDEV_DEV(netdev, &pdev->dev);
2711
2712         enic = netdev_priv(netdev);
2713         enic->netdev = netdev;
2714         enic->pdev = pdev;
2715
2716         /* Setup PCI resources
2717          */
2718
2719         err = pci_enable_device_mem(pdev);
2720         if (err) {
2721                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2722                 goto err_out_free_netdev;
2723         }
2724
2725         err = pci_request_regions(pdev, DRV_NAME);
2726         if (err) {
2727                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2728                 goto err_out_disable_device;
2729         }
2730
2731         pci_set_master(pdev);
2732
2733         /* Query PCI controller on system for DMA addressing
2734          * limitation for the device.  Try 47-bit first, and
2735          * fail to 32-bit.
2736          */
2737
2738         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(47));
2739         if (err) {
2740                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2741                 if (err) {
2742                         dev_err(dev, "No usable DMA configuration, aborting\n");
2743                         goto err_out_release_regions;
2744                 }
2745                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2746                 if (err) {
2747                         dev_err(dev, "Unable to obtain %u-bit DMA "
2748                                 "for consistent allocations, aborting\n", 32);
2749                         goto err_out_release_regions;
2750                 }
2751         } else {
2752                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(47));
2753                 if (err) {
2754                         dev_err(dev, "Unable to obtain %u-bit DMA "
2755                                 "for consistent allocations, aborting\n", 47);
2756                         goto err_out_release_regions;
2757                 }
2758                 using_dac = 1;
2759         }
2760
2761         /* Map vNIC resources from BAR0-5
2762          */
2763
2764         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2765                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2766                         continue;
2767                 enic->bar[i].len = pci_resource_len(pdev, i);
2768                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2769                 if (!enic->bar[i].vaddr) {
2770                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2771                         err = -ENODEV;
2772                         goto err_out_iounmap;
2773                 }
2774                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2775         }
2776
2777         /* Register vNIC device
2778          */
2779
2780         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2781                 ARRAY_SIZE(enic->bar));
2782         if (!enic->vdev) {
2783                 dev_err(dev, "vNIC registration failed, aborting\n");
2784                 err = -ENODEV;
2785                 goto err_out_iounmap;
2786         }
2787
2788         err = vnic_devcmd_init(enic->vdev);
2789
2790         if (err)
2791                 goto err_out_vnic_unregister;
2792
2793 #ifdef CONFIG_PCI_IOV
2794         /* Get number of subvnics */
2795         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2796         if (pos) {
2797                 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2798                         &enic->num_vfs);
2799                 if (enic->num_vfs) {
2800                         err = pci_enable_sriov(pdev, enic->num_vfs);
2801                         if (err) {
2802                                 dev_err(dev, "SRIOV enable failed, aborting."
2803                                         " pci_enable_sriov() returned %d\n",
2804                                         err);
2805                                 goto err_out_vnic_unregister;
2806                         }
2807                         enic->priv_flags |= ENIC_SRIOV_ENABLED;
2808                         num_pps = enic->num_vfs;
2809                 }
2810         }
2811 #endif
2812
2813         /* Allocate structure for port profiles */
2814         enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2815         if (!enic->pp) {
2816                 err = -ENOMEM;
2817                 goto err_out_disable_sriov_pp;
2818         }
2819
2820         /* Issue device open to get device in known state
2821          */
2822
2823         err = enic_dev_open(enic);
2824         if (err) {
2825                 dev_err(dev, "vNIC dev open failed, aborting\n");
2826                 goto err_out_disable_sriov;
2827         }
2828
2829         /* Setup devcmd lock
2830          */
2831
2832         spin_lock_init(&enic->devcmd_lock);
2833         spin_lock_init(&enic->enic_api_lock);
2834
2835         /*
2836          * Set ingress vlan rewrite mode before vnic initialization
2837          */
2838
2839         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2840         if (err) {
2841                 dev_err(dev,
2842                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2843                 goto err_out_dev_close;
2844         }
2845
2846         /* Issue device init to initialize the vnic-to-switch link.
2847          * We'll start with carrier off and wait for link UP
2848          * notification later to turn on carrier.  We don't need
2849          * to wait here for the vnic-to-switch link initialization
2850          * to complete; link UP notification is the indication that
2851          * the process is complete.
2852          */
2853
2854         netif_carrier_off(netdev);
2855
2856         /* Do not call dev_init for a dynamic vnic.
2857          * For a dynamic vnic, init_prov_info will be
2858          * called later by an upper layer.
2859          */
2860
2861         if (!enic_is_dynamic(enic)) {
2862                 err = vnic_dev_init(enic->vdev, 0);
2863                 if (err) {
2864                         dev_err(dev, "vNIC dev init failed, aborting\n");
2865                         goto err_out_dev_close;
2866                 }
2867         }
2868
2869         err = enic_dev_init(enic);
2870         if (err) {
2871                 dev_err(dev, "Device initialization failed, aborting\n");
2872                 goto err_out_dev_close;
2873         }
2874
2875         netif_set_real_num_tx_queues(netdev, enic->wq_count);
2876         netif_set_real_num_rx_queues(netdev, enic->rq_count);
2877
2878         /* Setup notification timer, HW reset task, and wq locks
2879          */
2880
2881         timer_setup(&enic->notify_timer, enic_notify_timer, 0);
2882
2883         enic_rfs_flw_tbl_init(enic);
2884         enic_set_rx_coal_setting(enic);
2885         INIT_WORK(&enic->reset, enic_reset);
2886         INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
2887         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2888
2889         for (i = 0; i < enic->wq_count; i++)
2890                 spin_lock_init(&enic->wq_lock[i]);
2891
2892         /* Register net device
2893          */
2894
2895         enic->port_mtu = enic->config.mtu;
2896
2897         err = enic_set_mac_addr(netdev, enic->mac_addr);
2898         if (err) {
2899                 dev_err(dev, "Invalid MAC address, aborting\n");
2900                 goto err_out_dev_deinit;
2901         }
2902
2903         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2904         /* rx coalesce time already got initialized. This gets used
2905          * if adaptive coal is turned off
2906          */
2907         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2908
2909         if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2910                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2911         else
2912                 netdev->netdev_ops = &enic_netdev_ops;
2913
2914         netdev->watchdog_timeo = 2 * HZ;
2915         enic_set_ethtool_ops(netdev);
2916
2917         netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2918         if (ENIC_SETTING(enic, LOOP)) {
2919                 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2920                 enic->loop_enable = 1;
2921                 enic->loop_tag = enic->config.loop_tag;
2922                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2923         }
2924         if (ENIC_SETTING(enic, TXCSUM))
2925                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2926         if (ENIC_SETTING(enic, TSO))
2927                 netdev->hw_features |= NETIF_F_TSO |
2928                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2929         if (ENIC_SETTING(enic, RSS))
2930                 netdev->hw_features |= NETIF_F_RXHASH;
2931         if (ENIC_SETTING(enic, RXCSUM))
2932                 netdev->hw_features |= NETIF_F_RXCSUM;
2933         if (ENIC_SETTING(enic, VXLAN)) {
2934                 u64 patch_level;
2935                 u64 a1 = 0;
2936
2937                 netdev->hw_enc_features |= NETIF_F_RXCSUM               |
2938                                            NETIF_F_TSO                  |
2939                                            NETIF_F_TSO6                 |
2940                                            NETIF_F_TSO_ECN              |
2941                                            NETIF_F_GSO_UDP_TUNNEL       |
2942                                            NETIF_F_HW_CSUM              |
2943                                            NETIF_F_GSO_UDP_TUNNEL_CSUM;
2944                 netdev->hw_features |= netdev->hw_enc_features;
2945                 /* get bit mask from hw about supported offload bit level
2946                  * BIT(0) = fw supports patch_level 0
2947                  *          fcoe bit = encap
2948                  *          fcoe_fc_crc_ok = outer csum ok
2949                  * BIT(1) = always set by fw
2950                  * BIT(2) = fw supports patch_level 2
2951                  *          BIT(0) in rss_hash = encap
2952                  *          BIT(1,2) in rss_hash = outer_ip_csum_ok/
2953                  *                                 outer_tcp_csum_ok
2954                  * used in enic_rq_indicate_buf
2955                  */
2956                 err = vnic_dev_get_supported_feature_ver(enic->vdev,
2957                                                          VIC_FEATURE_VXLAN,
2958                                                          &patch_level, &a1);
2959                 if (err)
2960                         patch_level = 0;
2961                 enic->vxlan.flags = (u8)a1;
2962                 /* mask bits that are supported by driver
2963                  */
2964                 patch_level &= BIT_ULL(0) | BIT_ULL(2);
2965                 patch_level = fls(patch_level);
2966                 patch_level = patch_level ? patch_level - 1 : 0;
2967                 enic->vxlan.patch_level = patch_level;
2968         }
2969
2970         netdev->features |= netdev->hw_features;
2971         netdev->vlan_features |= netdev->features;
2972
2973 #ifdef CONFIG_RFS_ACCEL
2974         netdev->hw_features |= NETIF_F_NTUPLE;
2975 #endif
2976
2977         if (using_dac)
2978                 netdev->features |= NETIF_F_HIGHDMA;
2979
2980         netdev->priv_flags |= IFF_UNICAST_FLT;
2981
2982         /* MTU range: 68 - 9000 */
2983         netdev->min_mtu = ENIC_MIN_MTU;
2984         netdev->max_mtu = ENIC_MAX_MTU;
2985         netdev->mtu     = enic->port_mtu;
2986
2987         err = register_netdev(netdev);
2988         if (err) {
2989                 dev_err(dev, "Cannot register net device, aborting\n");
2990                 goto err_out_dev_deinit;
2991         }
2992         enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2993
2994         return 0;
2995
2996 err_out_dev_deinit:
2997         enic_dev_deinit(enic);
2998 err_out_dev_close:
2999         vnic_dev_close(enic->vdev);
3000 err_out_disable_sriov:
3001         kfree(enic->pp);
3002 err_out_disable_sriov_pp:
3003 #ifdef CONFIG_PCI_IOV
3004         if (enic_sriov_enabled(enic)) {
3005                 pci_disable_sriov(pdev);
3006                 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3007         }
3008 #endif
3009 err_out_vnic_unregister:
3010         vnic_dev_unregister(enic->vdev);
3011 err_out_iounmap:
3012         enic_iounmap(enic);
3013 err_out_release_regions:
3014         pci_release_regions(pdev);
3015 err_out_disable_device:
3016         pci_disable_device(pdev);
3017 err_out_free_netdev:
3018         free_netdev(netdev);
3019
3020         return err;
3021 }
3022
3023 static void enic_remove(struct pci_dev *pdev)
3024 {
3025         struct net_device *netdev = pci_get_drvdata(pdev);
3026
3027         if (netdev) {
3028                 struct enic *enic = netdev_priv(netdev);
3029
3030                 cancel_work_sync(&enic->reset);
3031                 cancel_work_sync(&enic->change_mtu_work);
3032                 unregister_netdev(netdev);
3033                 enic_dev_deinit(enic);
3034                 vnic_dev_close(enic->vdev);
3035 #ifdef CONFIG_PCI_IOV
3036                 if (enic_sriov_enabled(enic)) {
3037                         pci_disable_sriov(pdev);
3038                         enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3039                 }
3040 #endif
3041                 kfree(enic->pp);
3042                 vnic_dev_unregister(enic->vdev);
3043                 enic_iounmap(enic);
3044                 pci_release_regions(pdev);
3045                 pci_disable_device(pdev);
3046                 free_netdev(netdev);
3047         }
3048 }
3049
3050 static struct pci_driver enic_driver = {
3051         .name = DRV_NAME,
3052         .id_table = enic_id_table,
3053         .probe = enic_probe,
3054         .remove = enic_remove,
3055 };
3056
3057 static int __init enic_init_module(void)
3058 {
3059         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
3060
3061         return pci_register_driver(&enic_driver);
3062 }
3063
3064 static void __exit enic_cleanup_module(void)
3065 {
3066         pci_unregister_driver(&enic_driver);
3067 }
3068
3069 module_init(enic_init_module);
3070 module_exit(enic_cleanup_module);