projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
sunvnet: Use one Tx queue per vnet_port
[deliverable/linux.git] / drivers / net / ethernet / sun / sunvnet.c
1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
2 *
3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4 */
5
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8 #include <linux/module.h>
9 #include <linux/kernel.h>
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/netdevice.h>
15 #include <linux/ethtool.h>
16 #include <linux/etherdevice.h>
17 #include <linux/mutex.h>
18 #include <linux/if_vlan.h>
19
20 #if IS_ENABLED(CONFIG_IPV6)
21 #include <linux/icmpv6.h>
22 #endif
23
24 #include <net/icmp.h>
25 #include <net/route.h>
26
27 #include <asm/vio.h>
28 #include <asm/ldc.h>
29
30 #include "sunvnet.h"
31
32 #define DRV_MODULE_NAME "sunvnet"
33 #define DRV_MODULE_VERSION "1.0"
34 #define DRV_MODULE_RELDATE "June 25, 2007"
35
36 static char version[] =
37 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
38 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
39 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
40 MODULE_LICENSE("GPL");
41 MODULE_VERSION(DRV_MODULE_VERSION);
42
43 #define VNET_MAX_TXQS 16
44
45 /* Heuristic for the number of times to exponentially backoff and
46 * retry sending an LDC trigger when EAGAIN is encountered
47 */
48 #define VNET_MAX_RETRIES 10
49
50 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
51
52 /* Ordered from largest major to lowest */
53 static struct vio_version vnet_versions[] = {
54 { .major = 1, .minor = 6 },
55 { .major = 1, .minor = 0 },
56 };
57
58 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
59 {
60 return vio_dring_avail(dr, VNET_TX_RING_SIZE);
61 }
62
63 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
64 {
65 struct vio_msg_tag *pkt = arg;
66
67 pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
68 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
69 pr_err("Resetting connection\n");
70
71 ldc_disconnect(port->vio.lp);
72
73 return -ECONNRESET;
74 }
75
76 static int vnet_send_attr(struct vio_driver_state *vio)
77 {
78 struct vnet_port *port = to_vnet_port(vio);
79 struct net_device *dev = port->vp->dev;
80 struct vio_net_attr_info pkt;
81 int framelen = ETH_FRAME_LEN;
82 int i;
83
84 memset(&pkt, 0, sizeof(pkt));
85 pkt.tag.type = VIO_TYPE_CTRL;
86 pkt.tag.stype = VIO_SUBTYPE_INFO;
87 pkt.tag.stype_env = VIO_ATTR_INFO;
88 pkt.tag.sid = vio_send_sid(vio);
89 if (vio_version_before(vio, 1, 2))
90 pkt.xfer_mode = VIO_DRING_MODE;
91 else
92 pkt.xfer_mode = VIO_NEW_DRING_MODE;
93 pkt.addr_type = VNET_ADDR_ETHERMAC;
94 pkt.ack_freq = 0;
95 for (i = 0; i < 6; i++)
96 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
97 if (vio_version_after(vio, 1, 3)) {
98 if (port->rmtu) {
99 port->rmtu = min(VNET_MAXPACKET, port->rmtu);
100 pkt.mtu = port->rmtu;
101 } else {
102 port->rmtu = VNET_MAXPACKET;
103 pkt.mtu = port->rmtu;
104 }
105 if (vio_version_after_eq(vio, 1, 6))
106 pkt.options = VIO_TX_DRING;
107 } else if (vio_version_before(vio, 1, 3)) {
108 pkt.mtu = framelen;
109 } else { /* v1.3 */
110 pkt.mtu = framelen + VLAN_HLEN;
111 }
112
113 pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
114 pkt.cflags = 0;
115
116 viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
117 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
118 "cflags[0x%04x] lso_max[%u]\n",
119 pkt.xfer_mode, pkt.addr_type,
120 (unsigned long long)pkt.addr,
121 pkt.ack_freq, pkt.plnk_updt, pkt.options,
122 (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
123
124
125 return vio_ldc_send(vio, &pkt, sizeof(pkt));
126 }
127
128 static int handle_attr_info(struct vio_driver_state *vio,
129 struct vio_net_attr_info *pkt)
130 {
131 struct vnet_port *port = to_vnet_port(vio);
132 u64 localmtu;
133 u8 xfer_mode;
134
135 viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
136 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
137 " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
138 pkt->xfer_mode, pkt->addr_type,
139 (unsigned long long)pkt->addr,
140 pkt->ack_freq, pkt->plnk_updt, pkt->options,
141 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
142 pkt->ipv4_lso_maxlen);
143
144 pkt->tag.sid = vio_send_sid(vio);
145
146 xfer_mode = pkt->xfer_mode;
147 /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
148 if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
149 xfer_mode = VIO_NEW_DRING_MODE;
150
151 /* MTU negotiation:
152 * < v1.3 - ETH_FRAME_LEN exactly
153 * > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
154 * pkt->mtu for ACK
155 * = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
156 */
157 if (vio_version_before(vio, 1, 3)) {
158 localmtu = ETH_FRAME_LEN;
159 } else if (vio_version_after(vio, 1, 3)) {
160 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
161 localmtu = min(pkt->mtu, localmtu);
162 pkt->mtu = localmtu;
163 } else { /* v1.3 */
164 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
165 }
166 port->rmtu = localmtu;
167
168 /* for version >= 1.6, ACK packet mode we support */
169 if (vio_version_after_eq(vio, 1, 6)) {
170 pkt->xfer_mode = VIO_NEW_DRING_MODE;
171 pkt->options = VIO_TX_DRING;
172 }
173
174 if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
175 pkt->addr_type != VNET_ADDR_ETHERMAC ||
176 pkt->mtu != localmtu) {
177 viodbg(HS, "SEND NET ATTR NACK\n");
178
179 pkt->tag.stype = VIO_SUBTYPE_NACK;
180
181 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
182
183 return -ECONNRESET;
184 } else {
185 viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
186 "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
187 "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
188 pkt->xfer_mode, pkt->addr_type,
189 (unsigned long long)pkt->addr,
190 pkt->ack_freq, pkt->plnk_updt, pkt->options,
191 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
192 pkt->ipv4_lso_maxlen);
193
194 pkt->tag.stype = VIO_SUBTYPE_ACK;
195
196 return vio_ldc_send(vio, pkt, sizeof(*pkt));
197 }
198
199 }
200
201 static int handle_attr_ack(struct vio_driver_state *vio,
202 struct vio_net_attr_info *pkt)
203 {
204 viodbg(HS, "GOT NET ATTR ACK\n");
205
206 return 0;
207 }
208
209 static int handle_attr_nack(struct vio_driver_state *vio,
210 struct vio_net_attr_info *pkt)
211 {
212 viodbg(HS, "GOT NET ATTR NACK\n");
213
214 return -ECONNRESET;
215 }
216
217 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
218 {
219 struct vio_net_attr_info *pkt = arg;
220
221 switch (pkt->tag.stype) {
222 case VIO_SUBTYPE_INFO:
223 return handle_attr_info(vio, pkt);
224
225 case VIO_SUBTYPE_ACK:
226 return handle_attr_ack(vio, pkt);
227
228 case VIO_SUBTYPE_NACK:
229 return handle_attr_nack(vio, pkt);
230
231 default:
232 return -ECONNRESET;
233 }
234 }
235
236 static void vnet_handshake_complete(struct vio_driver_state *vio)
237 {
238 struct vio_dring_state *dr;
239
240 dr = &vio->drings[VIO_DRIVER_RX_RING];
241 dr->snd_nxt = dr->rcv_nxt = 1;
242
243 dr = &vio->drings[VIO_DRIVER_TX_RING];
244 dr->snd_nxt = dr->rcv_nxt = 1;
245 }
246
247 /* The hypervisor interface that implements copying to/from imported
248 * memory from another domain requires that copies are done to 8-byte
249 * aligned buffers, and that the lengths of such copies are also 8-byte
250 * multiples.
251 *
252 * So we align skb->data to an 8-byte multiple and pad-out the data
253 * area so we can round the copy length up to the next multiple of
254 * 8 for the copy.
255 *
256 * The transmitter puts the actual start of the packet 6 bytes into
257 * the buffer it sends over, so that the IP headers after the ethernet
258 * header are aligned properly. These 6 bytes are not in the descriptor
259 * length, they are simply implied. This offset is represented using
260 * the VNET_PACKET_SKIP macro.
261 */
262 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
263 unsigned int len)
264 {
265 struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
266 unsigned long addr, off;
267
268 if (unlikely(!skb))
269 return NULL;
270
271 addr = (unsigned long) skb->data;
272 off = ((addr + 7UL) & ~7UL) - addr;
273 if (off)
274 skb_reserve(skb, off);
275
276 return skb;
277 }
278
279 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
280 struct ldc_trans_cookie *cookies, int ncookies)
281 {
282 struct net_device *dev = port->vp->dev;
283 unsigned int copy_len;
284 struct sk_buff *skb;
285 int err;
286
287 err = -EMSGSIZE;
288 if (unlikely(len < ETH_ZLEN || len > port->rmtu)) {
289 dev->stats.rx_length_errors++;
290 goto out_dropped;
291 }
292
293 skb = alloc_and_align_skb(dev, len);
294 err = -ENOMEM;
295 if (unlikely(!skb)) {
296 dev->stats.rx_missed_errors++;
297 goto out_dropped;
298 }
299
300 copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
301 skb_put(skb, copy_len);
302 err = ldc_copy(port->vio.lp, LDC_COPY_IN,
303 skb->data, copy_len, 0,
304 cookies, ncookies);
305 if (unlikely(err < 0)) {
306 dev->stats.rx_frame_errors++;
307 goto out_free_skb;
308 }
309
310 skb_pull(skb, VNET_PACKET_SKIP);
311 skb_trim(skb, len);
312 skb->protocol = eth_type_trans(skb, dev);
313
314 dev->stats.rx_packets++;
315 dev->stats.rx_bytes += len;
316 napi_gro_receive(&port->napi, skb);
317 return 0;
318
319 out_free_skb:
320 kfree_skb(skb);
321
322 out_dropped:
323 dev->stats.rx_dropped++;
324 return err;
325 }
326
327 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
328 u32 start, u32 end, u8 vio_dring_state)
329 {
330 struct vio_dring_data hdr = {
331 .tag = {
332 .type = VIO_TYPE_DATA,
333 .stype = VIO_SUBTYPE_ACK,
334 .stype_env = VIO_DRING_DATA,
335 .sid = vio_send_sid(&port->vio),
336 },
337 .dring_ident = dr->ident,
338 .start_idx = start,
339 .end_idx = end,
340 .state = vio_dring_state,
341 };
342 int err, delay;
343 int retries = 0;
344
345 hdr.seq = dr->snd_nxt;
346 delay = 1;
347 do {
348 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
349 if (err > 0) {
350 dr->snd_nxt++;
351 break;
352 }
353 udelay(delay);
354 if ((delay <<= 1) > 128)
355 delay = 128;
356 if (retries++ > VNET_MAX_RETRIES) {
357 pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
358 port->raddr[0], port->raddr[1],
359 port->raddr[2], port->raddr[3],
360 port->raddr[4], port->raddr[5]);
361 break;
362 }
363 } while (err == -EAGAIN);
364
365 if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
366 port->stop_rx_idx = end;
367 port->stop_rx = true;
368 } else {
369 port->stop_rx_idx = 0;
370 port->stop_rx = false;
371 }
372
373 return err;
374 }
375
376 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
377 {
378 if (++idx == dr->num_entries)
379 idx = 0;
380 return idx;
381 }
382
383 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
384 {
385 if (idx == 0)
386 idx = dr->num_entries - 1;
387 else
388 idx--;
389
390 return idx;
391 }
392
393 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
394 struct vio_dring_state *dr,
395 u32 index)
396 {
397 struct vio_net_desc *desc = port->vio.desc_buf;
398 int err;
399
400 err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
401 (index * dr->entry_size),
402 dr->cookies, dr->ncookies);
403 if (err < 0)
404 return ERR_PTR(err);
405
406 return desc;
407 }
408
409 static int put_rx_desc(struct vnet_port *port,
410 struct vio_dring_state *dr,
411 struct vio_net_desc *desc,
412 u32 index)
413 {
414 int err;
415
416 err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
417 (index * dr->entry_size),
418 dr->cookies, dr->ncookies);
419 if (err < 0)
420 return err;
421
422 return 0;
423 }
424
425 static int vnet_walk_rx_one(struct vnet_port *port,
426 struct vio_dring_state *dr,
427 u32 index, int *needs_ack)
428 {
429 struct vio_net_desc *desc = get_rx_desc(port, dr, index);
430 struct vio_driver_state *vio = &port->vio;
431 int err;
432
433 BUG_ON(desc == NULL);
434 if (IS_ERR(desc))
435 return PTR_ERR(desc);
436
437 if (desc->hdr.state != VIO_DESC_READY)
438 return 1;
439
440 rmb();
441
442 viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
443 desc->hdr.state, desc->hdr.ack,
444 desc->size, desc->ncookies,
445 desc->cookies[0].cookie_addr,
446 desc->cookies[0].cookie_size);
447
448 err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
449 if (err == -ECONNRESET)
450 return err;
451 desc->hdr.state = VIO_DESC_DONE;
452 err = put_rx_desc(port, dr, desc, index);
453 if (err < 0)
454 return err;
455 *needs_ack = desc->hdr.ack;
456 return 0;
457 }
458
459 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
460 u32 start, u32 end, int *npkts, int budget)
461 {
462 struct vio_driver_state *vio = &port->vio;
463 int ack_start = -1, ack_end = -1;
464 bool send_ack = true;
465
466 end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
467
468 viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
469
470 while (start != end) {
471 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
472 if (err == -ECONNRESET)
473 return err;
474 if (err != 0)
475 break;
476 (*npkts)++;
477 if (ack_start == -1)
478 ack_start = start;
479 ack_end = start;
480 start = next_idx(start, dr);
481 if (ack && start != end) {
482 err = vnet_send_ack(port, dr, ack_start, ack_end,
483 VIO_DRING_ACTIVE);
484 if (err == -ECONNRESET)
485 return err;
486 ack_start = -1;
487 }
488 if ((*npkts) >= budget) {
489 send_ack = false;
490 break;
491 }
492 }
493 if (unlikely(ack_start == -1))
494 ack_start = ack_end = prev_idx(start, dr);
495 if (send_ack) {
496 port->napi_resume = false;
497 return vnet_send_ack(port, dr, ack_start, ack_end,
498 VIO_DRING_STOPPED);
499 } else {
500 port->napi_resume = true;
501 port->napi_stop_idx = ack_end;
502 return 1;
503 }
504 }
505
506 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
507 int budget)
508 {
509 struct vio_dring_data *pkt = msgbuf;
510 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
511 struct vio_driver_state *vio = &port->vio;
512
513 viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
514 pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
515
516 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
517 return 0;
518 if (unlikely(pkt->seq != dr->rcv_nxt)) {
519 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
520 pkt->seq, dr->rcv_nxt);
521 return 0;
522 }
523
524 if (!port->napi_resume)
525 dr->rcv_nxt++;
526
527 /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
528
529 return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
530 npkts, budget);
531 }
532
533 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
534 {
535 u32 idx = dr->cons;
536 int found = 0;
537
538 while (idx != dr->prod) {
539 if (idx == end) {
540 found = 1;
541 break;
542 }
543 idx = next_idx(idx, dr);
544 }
545 return found;
546 }
547
548 static int vnet_ack(struct vnet_port *port, void *msgbuf)
549 {
550 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
551 struct vio_dring_data *pkt = msgbuf;
552 struct net_device *dev;
553 struct vnet *vp;
554 u32 end;
555 struct vio_net_desc *desc;
556 struct netdev_queue *txq;
557
558 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
559 return 0;
560
561 end = pkt->end_idx;
562 if (unlikely(!idx_is_pending(dr, end)))
563 return 0;
564
565 vp = port->vp;
566 dev = vp->dev;
567 /* sync for race conditions with vnet_start_xmit() and tell xmit it
568 * is time to send a trigger.
569 */
570 netif_tx_lock(dev);
571 dr->cons = next_idx(end, dr);
572 desc = vio_dring_entry(dr, dr->cons);
573 if (desc->hdr.state == VIO_DESC_READY && port->start_cons) {
574 /* vnet_start_xmit() just populated this dring but missed
575 * sending the "start" LDC message to the consumer.
576 * Send a "start" trigger on its behalf.
577 */
578 if (__vnet_tx_trigger(port, dr->cons) > 0)
579 port->start_cons = false;
580 else
581 port->start_cons = true;
582 } else {
583 port->start_cons = true;
584 }
585 netif_tx_unlock(dev);
586
587 txq = netdev_get_tx_queue(dev, port->q_index);
588 if (unlikely(netif_tx_queue_stopped(txq) &&
589 vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
590 return 1;
591
592 return 0;
593 }
594
595 static int vnet_nack(struct vnet_port *port, void *msgbuf)
596 {
597 /* XXX just reset or similar XXX */
598 return 0;
599 }
600
601 static int handle_mcast(struct vnet_port *port, void *msgbuf)
602 {
603 struct vio_net_mcast_info *pkt = msgbuf;
604
605 if (pkt->tag.stype != VIO_SUBTYPE_ACK)
606 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
607 port->vp->dev->name,
608 pkt->tag.type,
609 pkt->tag.stype,
610 pkt->tag.stype_env,
611 pkt->tag.sid);
612
613 return 0;
614 }
615
616 /* Got back a STOPPED LDC message on port. If the queue is stopped,
617 * wake it up so that we'll send out another START message at the
618 * next TX.
619 */
620 static void maybe_tx_wakeup(struct vnet_port *port)
621 {
622 struct netdev_queue *txq;
623
624 txq = netdev_get_tx_queue(port->vp->dev, port->q_index);
625 __netif_tx_lock(txq, smp_processor_id());
626 if (likely(netif_tx_queue_stopped(txq))) {
627 struct vio_dring_state *dr;
628
629 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
630 netif_tx_wake_queue(txq);
631 }
632 __netif_tx_unlock(txq);
633 }
634
635 static inline bool port_is_up(struct vnet_port *vnet)
636 {
637 struct vio_driver_state *vio = &vnet->vio;
638
639 return !!(vio->hs_state & VIO_HS_COMPLETE);
640 }
641
642 static int vnet_event_napi(struct vnet_port *port, int budget)
643 {
644 struct vio_driver_state *vio = &port->vio;
645 int tx_wakeup, err;
646 int npkts = 0;
647 int event = (port->rx_event & LDC_EVENT_RESET);
648
649 ldc_ctrl:
650 if (unlikely(event == LDC_EVENT_RESET ||
651 event == LDC_EVENT_UP)) {
652 vio_link_state_change(vio, event);
653
654 if (event == LDC_EVENT_RESET) {
655 port->rmtu = 0;
656 vio_port_up(vio);
657 }
658 port->rx_event = 0;
659 return 0;
660 }
661 /* We may have multiple LDC events in rx_event. Unroll send_events() */
662 event = (port->rx_event & LDC_EVENT_UP);
663 port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
664 if (event == LDC_EVENT_UP)
665 goto ldc_ctrl;
666 event = port->rx_event;
667 if (!(event & LDC_EVENT_DATA_READY))
668 return 0;
669
670 /* we dont expect any other bits than RESET, UP, DATA_READY */
671 BUG_ON(event != LDC_EVENT_DATA_READY);
672
673 tx_wakeup = err = 0;
674 while (1) {
675 union {
676 struct vio_msg_tag tag;
677 u64 raw[8];
678 } msgbuf;
679
680 if (port->napi_resume) {
681 struct vio_dring_data *pkt =
682 (struct vio_dring_data *)&msgbuf;
683 struct vio_dring_state *dr =
684 &port->vio.drings[VIO_DRIVER_RX_RING];
685
686 pkt->tag.type = VIO_TYPE_DATA;
687 pkt->tag.stype = VIO_SUBTYPE_INFO;
688 pkt->tag.stype_env = VIO_DRING_DATA;
689 pkt->seq = dr->rcv_nxt;
690 pkt->start_idx = next_idx(port->napi_stop_idx, dr);
691 pkt->end_idx = -1;
692 goto napi_resume;
693 }
694 ldc_read:
695 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
696 if (unlikely(err < 0)) {
697 if (err == -ECONNRESET)
698 vio_conn_reset(vio);
699 break;
700 }
701 if (err == 0)
702 break;
703 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
704 msgbuf.tag.type,
705 msgbuf.tag.stype,
706 msgbuf.tag.stype_env,
707 msgbuf.tag.sid);
708 err = vio_validate_sid(vio, &msgbuf.tag);
709 if (err < 0)
710 break;
711 napi_resume:
712 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
713 if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
714 if (!port_is_up(port)) {
715 /* failures like handshake_failure()
716 * may have cleaned up dring, but
717 * NAPI polling may bring us here.
718 */
719 err = -ECONNRESET;
720 break;
721 }
722 err = vnet_rx(port, &msgbuf, &npkts, budget);
723 if (npkts >= budget)
724 break;
725 if (npkts == 0 && err != -ECONNRESET)
726 goto ldc_read;
727 } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
728 err = vnet_ack(port, &msgbuf);
729 if (err > 0)
730 tx_wakeup |= err;
731 } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
732 err = vnet_nack(port, &msgbuf);
733 }
734 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
735 if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
736 err = handle_mcast(port, &msgbuf);
737 else
738 err = vio_control_pkt_engine(vio, &msgbuf);
739 if (err)
740 break;
741 } else {
742 err = vnet_handle_unknown(port, &msgbuf);
743 }
744 if (err == -ECONNRESET)
745 break;
746 }
747 if (unlikely(tx_wakeup && err != -ECONNRESET))
748 maybe_tx_wakeup(port);
749 return npkts;
750 }
751
752 static int vnet_poll(struct napi_struct *napi, int budget)
753 {
754 struct vnet_port *port = container_of(napi, struct vnet_port, napi);
755 struct vio_driver_state *vio = &port->vio;
756 int processed = vnet_event_napi(port, budget);
757
758 if (processed < budget) {
759 napi_complete(napi);
760 port->rx_event &= ~LDC_EVENT_DATA_READY;
761 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
762 }
763 return processed;
764 }
765
766 static void vnet_event(void *arg, int event)
767 {
768 struct vnet_port *port = arg;
769 struct vio_driver_state *vio = &port->vio;
770
771 port->rx_event |= event;
772 vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
773 napi_schedule(&port->napi);
774
775 }
776
777 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
778 {
779 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
780 struct vio_dring_data hdr = {
781 .tag = {
782 .type = VIO_TYPE_DATA,
783 .stype = VIO_SUBTYPE_INFO,
784 .stype_env = VIO_DRING_DATA,
785 .sid = vio_send_sid(&port->vio),
786 },
787 .dring_ident = dr->ident,
788 .start_idx = start,
789 .end_idx = (u32) -1,
790 };
791 int err, delay;
792 int retries = 0;
793
794 if (port->stop_rx) {
795 err = vnet_send_ack(port,
796 &port->vio.drings[VIO_DRIVER_RX_RING],
797 port->stop_rx_idx, -1,
798 VIO_DRING_STOPPED);
799 if (err <= 0)
800 return err;
801 }
802
803 hdr.seq = dr->snd_nxt;
804 delay = 1;
805 do {
806 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
807 if (err > 0) {
808 dr->snd_nxt++;
809 break;
810 }
811 udelay(delay);
812 if ((delay <<= 1) > 128)
813 delay = 128;
814 if (retries++ > VNET_MAX_RETRIES)
815 break;
816 } while (err == -EAGAIN);
817
818 return err;
819 }
820
821 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
822 {
823 unsigned int hash = vnet_hashfn(skb->data);
824 struct hlist_head *hp = &vp->port_hash[hash];
825 struct vnet_port *port;
826
827 hlist_for_each_entry_rcu(port, hp, hash) {
828 if (!port_is_up(port))
829 continue;
830 if (ether_addr_equal(port->raddr, skb->data))
831 return port;
832 }
833 list_for_each_entry_rcu(port, &vp->port_list, list) {
834 if (!port->switch_port)
835 continue;
836 if (!port_is_up(port))
837 continue;
838 return port;
839 }
840 return NULL;
841 }
842
843 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
844 unsigned *pending)
845 {
846 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
847 struct sk_buff *skb = NULL;
848 int i, txi;
849
850 *pending = 0;
851
852 txi = dr->prod-1;
853 if (txi < 0)
854 txi = VNET_TX_RING_SIZE-1;
855
856 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
857 struct vio_net_desc *d;
858
859 d = vio_dring_entry(dr, txi);
860
861 if (d->hdr.state == VIO_DESC_DONE) {
862 if (port->tx_bufs[txi].skb) {
863 BUG_ON(port->tx_bufs[txi].skb->next);
864
865 port->tx_bufs[txi].skb->next = skb;
866 skb = port->tx_bufs[txi].skb;
867 port->tx_bufs[txi].skb = NULL;
868
869 ldc_unmap(port->vio.lp,
870 port->tx_bufs[txi].cookies,
871 port->tx_bufs[txi].ncookies);
872 }
873 d->hdr.state = VIO_DESC_FREE;
874 } else if (d->hdr.state == VIO_DESC_READY) {
875 (*pending)++;
876 } else if (d->hdr.state == VIO_DESC_FREE) {
877 break;
878 }
879 --txi;
880 if (txi < 0)
881 txi = VNET_TX_RING_SIZE-1;
882 }
883 return skb;
884 }
885
886 static inline void vnet_free_skbs(struct sk_buff *skb)
887 {
888 struct sk_buff *next;
889
890 while (skb) {
891 next = skb->next;
892 skb->next = NULL;
893 dev_kfree_skb(skb);
894 skb = next;
895 }
896 }
897
898 static void vnet_clean_timer_expire(unsigned long port0)
899 {
900 struct vnet_port *port = (struct vnet_port *)port0;
901 struct sk_buff *freeskbs;
902 unsigned pending;
903
904 netif_tx_lock(port->vp->dev);
905 freeskbs = vnet_clean_tx_ring(port, &pending);
906 netif_tx_unlock(port->vp->dev);
907
908 vnet_free_skbs(freeskbs);
909
910 if (pending)
911 (void)mod_timer(&port->clean_timer,
912 jiffies + VNET_CLEAN_TIMEOUT);
913 else
914 del_timer(&port->clean_timer);
915 }
916
917 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, void **pstart,
918 int *plen)
919 {
920 struct sk_buff *nskb;
921 int len, pad;
922
923 len = skb->len;
924 pad = 0;
925 if (len < ETH_ZLEN) {
926 pad += ETH_ZLEN - skb->len;
927 len += pad;
928 }
929 len += VNET_PACKET_SKIP;
930 pad += 8 - (len & 7);
931 len += 8 - (len & 7);
932
933 if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
934 skb_tailroom(skb) < pad ||
935 skb_headroom(skb) < VNET_PACKET_SKIP) {
936 nskb = alloc_and_align_skb(skb->dev, skb->len);
937 skb_reserve(nskb, VNET_PACKET_SKIP);
938 if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
939 dev_kfree_skb(nskb);
940 dev_kfree_skb(skb);
941 return NULL;
942 }
943 (void)skb_put(nskb, skb->len);
944 dev_kfree_skb(skb);
945 skb = nskb;
946 }
947
948 *pstart = skb->data - VNET_PACKET_SKIP;
949 *plen = len;
950 return skb;
951 }
952
953 static u16
954 vnet_select_queue(struct net_device *dev, struct sk_buff *skb,
955 void *accel_priv, select_queue_fallback_t fallback)
956 {
957 struct vnet *vp = netdev_priv(dev);
958 struct vnet_port *port = __tx_port_find(vp, skb);
959
960 return port->q_index;
961 }
962
963 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
964 {
965 struct vnet *vp = netdev_priv(dev);
966 struct vnet_port *port = NULL;
967 struct vio_dring_state *dr;
968 struct vio_net_desc *d;
969 unsigned int len;
970 struct sk_buff *freeskbs = NULL;
971 int i, err, txi;
972 void *start = NULL;
973 int nlen = 0;
974 unsigned pending = 0;
975 struct netdev_queue *txq;
976
977 skb = vnet_skb_shape(skb, &start, &nlen);
978 if (unlikely(!skb))
979 goto out_dropped;
980
981 rcu_read_lock();
982 port = __tx_port_find(vp, skb);
983 if (unlikely(!port))
984 goto out_dropped;
985
986 if (skb->len > port->rmtu) {
987 unsigned long localmtu = port->rmtu - ETH_HLEN;
988
989 if (vio_version_after_eq(&port->vio, 1, 3))
990 localmtu -= VLAN_HLEN;
991
992 if (skb->protocol == htons(ETH_P_IP)) {
993 struct flowi4 fl4;
994 struct rtable *rt = NULL;
995
996 memset(&fl4, 0, sizeof(fl4));
997 fl4.flowi4_oif = dev->ifindex;
998 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
999 fl4.daddr = ip_hdr(skb)->daddr;
1000 fl4.saddr = ip_hdr(skb)->saddr;
1001
1002 rt = ip_route_output_key(dev_net(dev), &fl4);
1003 rcu_read_unlock();
1004 if (!IS_ERR(rt)) {
1005 skb_dst_set(skb, &rt->dst);
1006 icmp_send(skb, ICMP_DEST_UNREACH,
1007 ICMP_FRAG_NEEDED,
1008 htonl(localmtu));
1009 }
1010 }
1011 #if IS_ENABLED(CONFIG_IPV6)
1012 else if (skb->protocol == htons(ETH_P_IPV6))
1013 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1014 #endif
1015 goto out_dropped;
1016 }
1017
1018 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1019 i = skb_get_queue_mapping(skb);
1020 txq = netdev_get_tx_queue(dev, i);
1021 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1022 if (!netif_tx_queue_stopped(txq)) {
1023 netif_tx_stop_queue(txq);
1024
1025 /* This is a hard error, log it. */
1026 netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1027 dev->stats.tx_errors++;
1028 }
1029 rcu_read_unlock();
1030 return NETDEV_TX_BUSY;
1031 }
1032
1033 d = vio_dring_cur(dr);
1034
1035 txi = dr->prod;
1036
1037 freeskbs = vnet_clean_tx_ring(port, &pending);
1038
1039 BUG_ON(port->tx_bufs[txi].skb);
1040
1041 len = skb->len;
1042 if (len < ETH_ZLEN)
1043 len = ETH_ZLEN;
1044
1045 port->tx_bufs[txi].skb = skb;
1046 skb = NULL;
1047
1048 err = ldc_map_single(port->vio.lp, start, nlen,
1049 port->tx_bufs[txi].cookies, VNET_MAXCOOKIES,
1050 (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1051 if (err < 0) {
1052 netdev_info(dev, "tx buffer map error %d\n", err);
1053 goto out_dropped;
1054 }
1055 port->tx_bufs[txi].ncookies = err;
1056
1057 /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1058 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1059 * the protocol itself does not require it as long as the peer
1060 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1061 *
1062 * An ACK for every packet in the ring is expensive as the
1063 * sending of LDC messages is slow and affects performance.
1064 */
1065 d->hdr.ack = VIO_ACK_DISABLE;
1066 d->size = len;
1067 d->ncookies = port->tx_bufs[txi].ncookies;
1068 for (i = 0; i < d->ncookies; i++)
1069 d->cookies[i] = port->tx_bufs[txi].cookies[i];
1070
1071 /* This has to be a non-SMP write barrier because we are writing
1072 * to memory which is shared with the peer LDOM.
1073 */
1074 wmb();
1075
1076 d->hdr.state = VIO_DESC_READY;
1077
1078 /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1079 * to notify the consumer that some descriptors are READY.
1080 * After that "start" trigger, no additional triggers are needed until
1081 * a DRING_STOPPED is received from the consumer. The dr->cons field
1082 * (set up by vnet_ack()) has the value of the next dring index
1083 * that has not yet been ack-ed. We send a "start" trigger here
1084 * if, and only if, start_cons is true (reset it afterward). Conversely,
1085 * vnet_ack() should check if the dring corresponding to cons
1086 * is marked READY, but start_cons was false.
1087 * If so, vnet_ack() should send out the missed "start" trigger.
1088 *
1089 * Note that the wmb() above makes sure the cookies et al. are
1090 * not globally visible before the VIO_DESC_READY, and that the
1091 * stores are ordered correctly by the compiler. The consumer will
1092 * not proceed until the VIO_DESC_READY is visible assuring that
1093 * the consumer does not observe anything related to descriptors
1094 * out of order. The HV trap from the LDC start trigger is the
1095 * producer to consumer announcement that work is available to the
1096 * consumer
1097 */
1098 if (!port->start_cons)
1099 goto ldc_start_done; /* previous trigger suffices */
1100
1101 err = __vnet_tx_trigger(port, dr->cons);
1102 if (unlikely(err < 0)) {
1103 netdev_info(dev, "TX trigger error %d\n", err);
1104 d->hdr.state = VIO_DESC_FREE;
1105 dev->stats.tx_carrier_errors++;
1106 goto out_dropped;
1107 }
1108
1109 ldc_start_done:
1110 port->start_cons = false;
1111
1112 dev->stats.tx_packets++;
1113 dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1114
1115 dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1116 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1117 netif_tx_stop_queue(txq);
1118 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1119 netif_tx_wake_queue(txq);
1120 }
1121
1122 (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1123 rcu_read_unlock();
1124
1125 vnet_free_skbs(freeskbs);
1126
1127 return NETDEV_TX_OK;
1128
1129 out_dropped:
1130 if (pending)
1131 (void)mod_timer(&port->clean_timer,
1132 jiffies + VNET_CLEAN_TIMEOUT);
1133 else if (port)
1134 del_timer(&port->clean_timer);
1135 if (port)
1136 rcu_read_unlock();
1137 if (skb)
1138 dev_kfree_skb(skb);
1139 vnet_free_skbs(freeskbs);
1140 dev->stats.tx_dropped++;
1141 return NETDEV_TX_OK;
1142 }
1143
1144 static void vnet_tx_timeout(struct net_device *dev)
1145 {
1146 /* XXX Implement me XXX */
1147 }
1148
1149 static int vnet_open(struct net_device *dev)
1150 {
1151 netif_carrier_on(dev);
1152 netif_tx_start_all_queues(dev);
1153
1154 return 0;
1155 }
1156
1157 static int vnet_close(struct net_device *dev)
1158 {
1159 netif_tx_stop_all_queues(dev);
1160 netif_carrier_off(dev);
1161
1162 return 0;
1163 }
1164
1165 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1166 {
1167 struct vnet_mcast_entry *m;
1168
1169 for (m = vp->mcast_list; m; m = m->next) {
1170 if (ether_addr_equal(m->addr, addr))
1171 return m;
1172 }
1173 return NULL;
1174 }
1175
1176 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1177 {
1178 struct netdev_hw_addr *ha;
1179
1180 netdev_for_each_mc_addr(ha, dev) {
1181 struct vnet_mcast_entry *m;
1182
1183 m = __vnet_mc_find(vp, ha->addr);
1184 if (m) {
1185 m->hit = 1;
1186 continue;
1187 }
1188
1189 if (!m) {
1190 m = kzalloc(sizeof(*m), GFP_ATOMIC);
1191 if (!m)
1192 continue;
1193 memcpy(m->addr, ha->addr, ETH_ALEN);
1194 m->hit = 1;
1195
1196 m->next = vp->mcast_list;
1197 vp->mcast_list = m;
1198 }
1199 }
1200 }
1201
1202 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1203 {
1204 struct vio_net_mcast_info info;
1205 struct vnet_mcast_entry *m, **pp;
1206 int n_addrs;
1207
1208 memset(&info, 0, sizeof(info));
1209
1210 info.tag.type = VIO_TYPE_CTRL;
1211 info.tag.stype = VIO_SUBTYPE_INFO;
1212 info.tag.stype_env = VNET_MCAST_INFO;
1213 info.tag.sid = vio_send_sid(&port->vio);
1214 info.set = 1;
1215
1216 n_addrs = 0;
1217 for (m = vp->mcast_list; m; m = m->next) {
1218 if (m->sent)
1219 continue;
1220 m->sent = 1;
1221 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1222 m->addr, ETH_ALEN);
1223 if (++n_addrs == VNET_NUM_MCAST) {
1224 info.count = n_addrs;
1225
1226 (void) vio_ldc_send(&port->vio, &info,
1227 sizeof(info));
1228 n_addrs = 0;
1229 }
1230 }
1231 if (n_addrs) {
1232 info.count = n_addrs;
1233 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1234 }
1235
1236 info.set = 0;
1237
1238 n_addrs = 0;
1239 pp = &vp->mcast_list;
1240 while ((m = *pp) != NULL) {
1241 if (m->hit) {
1242 m->hit = 0;
1243 pp = &m->next;
1244 continue;
1245 }
1246
1247 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1248 m->addr, ETH_ALEN);
1249 if (++n_addrs == VNET_NUM_MCAST) {
1250 info.count = n_addrs;
1251 (void) vio_ldc_send(&port->vio, &info,
1252 sizeof(info));
1253 n_addrs = 0;
1254 }
1255
1256 *pp = m->next;
1257 kfree(m);
1258 }
1259 if (n_addrs) {
1260 info.count = n_addrs;
1261 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1262 }
1263 }
1264
1265 static void vnet_set_rx_mode(struct net_device *dev)
1266 {
1267 struct vnet *vp = netdev_priv(dev);
1268 struct vnet_port *port;
1269
1270 rcu_read_lock();
1271 list_for_each_entry_rcu(port, &vp->port_list, list) {
1272
1273 if (port->switch_port) {
1274 __update_mc_list(vp, dev);
1275 __send_mc_list(vp, port);
1276 break;
1277 }
1278 }
1279 rcu_read_unlock();
1280 }
1281
1282 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1283 {
1284 if (new_mtu < 68 || new_mtu > 65535)
1285 return -EINVAL;
1286
1287 dev->mtu = new_mtu;
1288 return 0;
1289 }
1290
1291 static int vnet_set_mac_addr(struct net_device *dev, void *p)
1292 {
1293 return -EINVAL;
1294 }
1295
1296 static void vnet_get_drvinfo(struct net_device *dev,
1297 struct ethtool_drvinfo *info)
1298 {
1299 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1300 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
1301 }
1302
1303 static u32 vnet_get_msglevel(struct net_device *dev)
1304 {
1305 struct vnet *vp = netdev_priv(dev);
1306 return vp->msg_enable;
1307 }
1308
1309 static void vnet_set_msglevel(struct net_device *dev, u32 value)
1310 {
1311 struct vnet *vp = netdev_priv(dev);
1312 vp->msg_enable = value;
1313 }
1314
1315 static const struct ethtool_ops vnet_ethtool_ops = {
1316 .get_drvinfo = vnet_get_drvinfo,
1317 .get_msglevel = vnet_get_msglevel,
1318 .set_msglevel = vnet_set_msglevel,
1319 .get_link = ethtool_op_get_link,
1320 };
1321
1322 static void vnet_port_free_tx_bufs(struct vnet_port *port)
1323 {
1324 struct vio_dring_state *dr;
1325 int i;
1326
1327 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1328 if (dr->base) {
1329 ldc_free_exp_dring(port->vio.lp, dr->base,
1330 (dr->entry_size * dr->num_entries),
1331 dr->cookies, dr->ncookies);
1332 dr->base = NULL;
1333 dr->entry_size = 0;
1334 dr->num_entries = 0;
1335 dr->pending = 0;
1336 dr->ncookies = 0;
1337 }
1338
1339 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1340 struct vio_net_desc *d;
1341 void *skb = port->tx_bufs[i].skb;
1342
1343 if (!skb)
1344 continue;
1345
1346 d = vio_dring_entry(dr, i);
1347 if (d->hdr.state == VIO_DESC_READY)
1348 pr_warn("active transmit buffers freed\n");
1349
1350 ldc_unmap(port->vio.lp,
1351 port->tx_bufs[i].cookies,
1352 port->tx_bufs[i].ncookies);
1353 dev_kfree_skb(skb);
1354 port->tx_bufs[i].skb = NULL;
1355 d->hdr.state = VIO_DESC_FREE;
1356 }
1357 }
1358
1359 static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
1360 {
1361 struct vio_dring_state *dr;
1362 unsigned long len;
1363 int i, err, ncookies;
1364 void *dring;
1365
1366 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1367
1368 len = (VNET_TX_RING_SIZE *
1369 (sizeof(struct vio_net_desc) +
1370 (sizeof(struct ldc_trans_cookie) * 2)));
1371
1372 ncookies = VIO_MAX_RING_COOKIES;
1373 dring = ldc_alloc_exp_dring(port->vio.lp, len,
1374 dr->cookies, &ncookies,
1375 (LDC_MAP_SHADOW |
1376 LDC_MAP_DIRECT |
1377 LDC_MAP_RW));
1378 if (IS_ERR(dring)) {
1379 err = PTR_ERR(dring);
1380 goto err_out;
1381 }
1382
1383 dr->base = dring;
1384 dr->entry_size = (sizeof(struct vio_net_desc) +
1385 (sizeof(struct ldc_trans_cookie) * 2));
1386 dr->num_entries = VNET_TX_RING_SIZE;
1387 dr->prod = dr->cons = 0;
1388 port->start_cons = true; /* need an initial trigger */
1389 dr->pending = VNET_TX_RING_SIZE;
1390 dr->ncookies = ncookies;
1391
1392 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1393 struct vio_net_desc *d;
1394
1395 d = vio_dring_entry(dr, i);
1396 d->hdr.state = VIO_DESC_FREE;
1397 }
1398 return 0;
1399
1400 err_out:
1401 vnet_port_free_tx_bufs(port);
1402
1403 return err;
1404 }
1405
1406 #ifdef CONFIG_NET_POLL_CONTROLLER
1407 static void vnet_poll_controller(struct net_device *dev)
1408 {
1409 struct vnet *vp = netdev_priv(dev);
1410 struct vnet_port *port;
1411 unsigned long flags;
1412
1413 spin_lock_irqsave(&vp->lock, flags);
1414 if (!list_empty(&vp->port_list)) {
1415 port = list_entry(vp->port_list.next, struct vnet_port, list);
1416 napi_schedule(&port->napi);
1417 }
1418 spin_unlock_irqrestore(&vp->lock, flags);
1419 }
1420 #endif
1421 static LIST_HEAD(vnet_list);
1422 static DEFINE_MUTEX(vnet_list_mutex);
1423
1424 static const struct net_device_ops vnet_ops = {
1425 .ndo_open = vnet_open,
1426 .ndo_stop = vnet_close,
1427 .ndo_set_rx_mode = vnet_set_rx_mode,
1428 .ndo_set_mac_address = vnet_set_mac_addr,
1429 .ndo_validate_addr = eth_validate_addr,
1430 .ndo_tx_timeout = vnet_tx_timeout,
1431 .ndo_change_mtu = vnet_change_mtu,
1432 .ndo_start_xmit = vnet_start_xmit,
1433 .ndo_select_queue = vnet_select_queue,
1434 #ifdef CONFIG_NET_POLL_CONTROLLER
1435 .ndo_poll_controller = vnet_poll_controller,
1436 #endif
1437 };
1438
1439 static struct vnet *vnet_new(const u64 *local_mac)
1440 {
1441 struct net_device *dev;
1442 struct vnet *vp;
1443 int err, i;
1444
1445 dev = alloc_etherdev_mqs(sizeof(*vp), VNET_MAX_TXQS, 1);
1446 if (!dev)
1447 return ERR_PTR(-ENOMEM);
1448 dev->needed_headroom = VNET_PACKET_SKIP + 8;
1449 dev->needed_tailroom = 8;
1450
1451 for (i = 0; i < ETH_ALEN; i++)
1452 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1453
1454 vp = netdev_priv(dev);
1455
1456 spin_lock_init(&vp->lock);
1457 vp->dev = dev;
1458
1459 INIT_LIST_HEAD(&vp->port_list);
1460 for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1461 INIT_HLIST_HEAD(&vp->port_hash[i]);
1462 INIT_LIST_HEAD(&vp->list);
1463 vp->local_mac = *local_mac;
1464
1465 dev->netdev_ops = &vnet_ops;
1466 dev->ethtool_ops = &vnet_ethtool_ops;
1467 dev->watchdog_timeo = VNET_TX_TIMEOUT;
1468
1469 err = register_netdev(dev);
1470 if (err) {
1471 pr_err("Cannot register net device, aborting\n");
1472 goto err_out_free_dev;
1473 }
1474
1475 netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1476
1477 list_add(&vp->list, &vnet_list);
1478
1479 return vp;
1480
1481 err_out_free_dev:
1482 free_netdev(dev);
1483
1484 return ERR_PTR(err);
1485 }
1486
1487 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1488 {
1489 struct vnet *iter, *vp;
1490
1491 mutex_lock(&vnet_list_mutex);
1492 vp = NULL;
1493 list_for_each_entry(iter, &vnet_list, list) {
1494 if (iter->local_mac == *local_mac) {
1495 vp = iter;
1496 break;
1497 }
1498 }
1499 if (!vp)
1500 vp = vnet_new(local_mac);
1501 mutex_unlock(&vnet_list_mutex);
1502
1503 return vp;
1504 }
1505
1506 static void vnet_cleanup(void)
1507 {
1508 struct vnet *vp;
1509 struct net_device *dev;
1510
1511 mutex_lock(&vnet_list_mutex);
1512 while (!list_empty(&vnet_list)) {
1513 vp = list_first_entry(&vnet_list, struct vnet, list);
1514 list_del(&vp->list);
1515 dev = vp->dev;
1516 /* vio_unregister_driver() should have cleaned up port_list */
1517 BUG_ON(!list_empty(&vp->port_list));
1518 unregister_netdev(dev);
1519 free_netdev(dev);
1520 }
1521 mutex_unlock(&vnet_list_mutex);
1522 }
1523
1524 static const char *local_mac_prop = "local-mac-address";
1525
1526 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1527 u64 port_node)
1528 {
1529 const u64 *local_mac = NULL;
1530 u64 a;
1531
1532 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1533 u64 target = mdesc_arc_target(hp, a);
1534 const char *name;
1535
1536 name = mdesc_get_property(hp, target, "name", NULL);
1537 if (!name || strcmp(name, "network"))
1538 continue;
1539
1540 local_mac = mdesc_get_property(hp, target,
1541 local_mac_prop, NULL);
1542 if (local_mac)
1543 break;
1544 }
1545 if (!local_mac)
1546 return ERR_PTR(-ENODEV);
1547
1548 return vnet_find_or_create(local_mac);
1549 }
1550
1551 static struct ldc_channel_config vnet_ldc_cfg = {
1552 .event = vnet_event,
1553 .mtu = 64,
1554 .mode = LDC_MODE_UNRELIABLE,
1555 };
1556
1557 static struct vio_driver_ops vnet_vio_ops = {
1558 .send_attr = vnet_send_attr,
1559 .handle_attr = vnet_handle_attr,
1560 .handshake_complete = vnet_handshake_complete,
1561 };
1562
1563 static void print_version(void)
1564 {
1565 printk_once(KERN_INFO "%s", version);
1566 }
1567
1568 const char *remote_macaddr_prop = "remote-mac-address";
1569
1570 static void
1571 vnet_port_add_txq(struct vnet_port *port)
1572 {
1573 struct vnet *vp = port->vp;
1574 int n;
1575
1576 n = vp->nports++;
1577 n = n & (VNET_MAX_TXQS - 1);
1578 port->q_index = n;
1579 netif_tx_wake_queue(netdev_get_tx_queue(vp->dev, port->q_index));
1580 }
1581
1582 static void
1583 vnet_port_rm_txq(struct vnet_port *port)
1584 {
1585 port->vp->nports--;
1586 netif_tx_stop_queue(netdev_get_tx_queue(port->vp->dev, port->q_index));
1587 }
1588
1589 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1590 {
1591 struct mdesc_handle *hp;
1592 struct vnet_port *port;
1593 unsigned long flags;
1594 struct vnet *vp;
1595 const u64 *rmac;
1596 int len, i, err, switch_port;
1597
1598 print_version();
1599
1600 hp = mdesc_grab();
1601
1602 vp = vnet_find_parent(hp, vdev->mp);
1603 if (IS_ERR(vp)) {
1604 pr_err("Cannot find port parent vnet\n");
1605 err = PTR_ERR(vp);
1606 goto err_out_put_mdesc;
1607 }
1608
1609 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1610 err = -ENODEV;
1611 if (!rmac) {
1612 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1613 goto err_out_put_mdesc;
1614 }
1615
1616 port = kzalloc(sizeof(*port), GFP_KERNEL);
1617 err = -ENOMEM;
1618 if (!port)
1619 goto err_out_put_mdesc;
1620
1621 for (i = 0; i < ETH_ALEN; i++)
1622 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1623
1624 port->vp = vp;
1625
1626 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1627 vnet_versions, ARRAY_SIZE(vnet_versions),
1628 &vnet_vio_ops, vp->dev->name);
1629 if (err)
1630 goto err_out_free_port;
1631
1632 err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1633 if (err)
1634 goto err_out_free_port;
1635
1636 netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1637
1638 err = vnet_port_alloc_tx_bufs(port);
1639 if (err)
1640 goto err_out_free_ldc;
1641
1642 INIT_HLIST_NODE(&port->hash);
1643 INIT_LIST_HEAD(&port->list);
1644
1645 switch_port = 0;
1646 if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1647 switch_port = 1;
1648 port->switch_port = switch_port;
1649
1650 spin_lock_irqsave(&vp->lock, flags);
1651 if (switch_port)
1652 list_add_rcu(&port->list, &vp->port_list);
1653 else
1654 list_add_tail_rcu(&port->list, &vp->port_list);
1655 hlist_add_head_rcu(&port->hash,
1656 &vp->port_hash[vnet_hashfn(port->raddr)]);
1657 vnet_port_add_txq(port);
1658 spin_unlock_irqrestore(&vp->lock, flags);
1659
1660 dev_set_drvdata(&vdev->dev, port);
1661
1662 pr_info("%s: PORT ( remote-mac %pM%s )\n",
1663 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1664
1665 setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1666 (unsigned long)port);
1667
1668 napi_enable(&port->napi);
1669 vio_port_up(&port->vio);
1670
1671 mdesc_release(hp);
1672
1673 return 0;
1674
1675 err_out_free_ldc:
1676 netif_napi_del(&port->napi);
1677 vio_ldc_free(&port->vio);
1678
1679 err_out_free_port:
1680 kfree(port);
1681
1682 err_out_put_mdesc:
1683 mdesc_release(hp);
1684 return err;
1685 }
1686
1687 static int vnet_port_remove(struct vio_dev *vdev)
1688 {
1689 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1690
1691 if (port) {
1692
1693 del_timer_sync(&port->vio.timer);
1694
1695 napi_disable(&port->napi);
1696
1697 list_del_rcu(&port->list);
1698 hlist_del_rcu(&port->hash);
1699
1700 synchronize_rcu();
1701 del_timer_sync(&port->clean_timer);
1702 vnet_port_rm_txq(port);
1703 netif_napi_del(&port->napi);
1704 vnet_port_free_tx_bufs(port);
1705 vio_ldc_free(&port->vio);
1706
1707 dev_set_drvdata(&vdev->dev, NULL);
1708
1709 kfree(port);
1710
1711 }
1712 return 0;
1713 }
1714
1715 static const struct vio_device_id vnet_port_match[] = {
1716 {
1717 .type = "vnet-port",
1718 },
1719 {},
1720 };
1721 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1722
1723 static struct vio_driver vnet_port_driver = {
1724 .id_table = vnet_port_match,
1725 .probe = vnet_port_probe,
1726 .remove = vnet_port_remove,
1727 .name = "vnet_port",
1728 };
1729
1730 static int __init vnet_init(void)
1731 {
1732 return vio_register_driver(&vnet_port_driver);
1733 }
1734
1735 static void __exit vnet_exit(void)
1736 {
1737 vio_unregister_driver(&vnet_port_driver);
1738 vnet_cleanup();
1739 }
1740
1741 module_init(vnet_init);
1742 module_exit(vnet_exit);
Linux kernel - Deliverables
This page took 0.240156 seconds and 5 git commands to generate.