projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sunvnet: allow admin to set sunvnet MTU
[deliverable/linux.git] / drivers / net / ethernet / sun / sunvnet.c
CommitLineData
4c521e42
DM		 1/* sunvnet.c: Sun LDOM Virtual Network Driver.
2 *
3d452e55 3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4c521e42
DM		 4 */
5
4d5870ec
JP		 6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
4c521e42
DM		 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>
9184a046 17#include <linux/mutex.h>
e4defc77 18#include <linux/if_vlan.h>
4c521e42
DM		 19
20#include <asm/vio.h>
21#include <asm/ldc.h>
22
23#include "sunvnet.h"
24
25#define DRV_MODULE_NAME "sunvnet"
4c521e42
DM		 26#define DRV_MODULE_VERSION "1.0"
27#define DRV_MODULE_RELDATE "June 25, 2007"
28
f73d12bd 29static char version[] =
4c521e42
DM		 30 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
31MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
32MODULE_DESCRIPTION("Sun LDOM virtual network driver");
33MODULE_LICENSE("GPL");
34MODULE_VERSION(DRV_MODULE_VERSION);
35
adddc32d
SV		 36/* Heuristic for the number of times to exponentially backoff and
37 * retry sending an LDC trigger when EAGAIN is encountered
38 */
39#define VNET_MAX_RETRIES 10
40
d1015645
SV		 41static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
42
4c521e42
DM		 43/* Ordered from largest major to lowest */
44static struct vio_version vnet_versions[] = {
e4defc77 45 { .major = 1, .minor = 6 },
4c521e42
DM		 46 { .major = 1, .minor = 0 },
47};
48
49static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
50{
51 return vio_dring_avail(dr, VNET_TX_RING_SIZE);
52}
53
54static int vnet_handle_unknown(struct vnet_port *port, void *arg)
55{
56 struct vio_msg_tag *pkt = arg;
57
4d5870ec 58 pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
4c521e42 59 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
4d5870ec 60 pr_err("Resetting connection\n");
4c521e42
DM		 61
62 ldc_disconnect(port->vio.lp);
63
64 return -ECONNRESET;
65}
66
67static int vnet_send_attr(struct vio_driver_state *vio)
68{
69 struct vnet_port *port = to_vnet_port(vio);
70 struct net_device *dev = port->vp->dev;
71 struct vio_net_attr_info pkt;
e4defc77 72 int framelen = ETH_FRAME_LEN;
4c521e42
DM		 73 int i;
74
75 memset(&pkt, 0, sizeof(pkt));
76 pkt.tag.type = VIO_TYPE_CTRL;
77 pkt.tag.stype = VIO_SUBTYPE_INFO;
78 pkt.tag.stype_env = VIO_ATTR_INFO;
79 pkt.tag.sid = vio_send_sid(vio);
e4defc77
DS		 80 if (vio_version_before(vio, 1, 2))
81 pkt.xfer_mode = VIO_DRING_MODE;
82 else
83 pkt.xfer_mode = VIO_NEW_DRING_MODE;
4c521e42
DM		 84 pkt.addr_type = VNET_ADDR_ETHERMAC;
85 pkt.ack_freq = 0;
86 for (i = 0; i < 6; i++)
87 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
e4defc77
DS		 88 if (vio_version_after(vio, 1, 3)) {
89 if (port->rmtu) {
90 port->rmtu = min(VNET_MAXPACKET, port->rmtu);
91 pkt.mtu = port->rmtu;
92 } else {
93 port->rmtu = VNET_MAXPACKET;
94 pkt.mtu = port->rmtu;
95 }
96 if (vio_version_after_eq(vio, 1, 6))
97 pkt.options = VIO_TX_DRING;
98 } else if (vio_version_before(vio, 1, 3)) {
99 pkt.mtu = framelen;
100 } else { /* v1.3 */
101 pkt.mtu = framelen + VLAN_HLEN;
102 }
103
104 pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
105 pkt.cflags = 0;
4c521e42
DM		 106
107 viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
e4defc77
DS		 108 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
109 "cflags[0x%04x] lso_max[%u]\n",
4c521e42 110 pkt.xfer_mode, pkt.addr_type,
e4defc77
DS		 111 (unsigned long long)pkt.addr,
112 pkt.ack_freq, pkt.plnk_updt, pkt.options,
113 (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
114
4c521e42
DM		 115
116 return vio_ldc_send(vio, &pkt, sizeof(pkt));
117}
118
119static int handle_attr_info(struct vio_driver_state *vio,
120 struct vio_net_attr_info *pkt)
121{
e4defc77
DS		 122 struct vnet_port *port = to_vnet_port(vio);
123 u64 localmtu;
124 u8 xfer_mode;
125
126 viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
127 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
128 " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
4c521e42 129 pkt->xfer_mode, pkt->addr_type,
e4defc77
DS		 130 (unsigned long long)pkt->addr,
131 pkt->ack_freq, pkt->plnk_updt, pkt->options,
132 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
133 pkt->ipv4_lso_maxlen);
4c521e42
DM		 134
135 pkt->tag.sid = vio_send_sid(vio);
136
e4defc77
DS		 137 xfer_mode = pkt->xfer_mode;
138 /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
139 if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
140 xfer_mode = VIO_NEW_DRING_MODE;
141
142 /* MTU negotiation:
143 * < v1.3 - ETH_FRAME_LEN exactly
144 * > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
145 * pkt->mtu for ACK
146 * = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
147 */
148 if (vio_version_before(vio, 1, 3)) {
149 localmtu = ETH_FRAME_LEN;
150 } else if (vio_version_after(vio, 1, 3)) {
151 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
152 localmtu = min(pkt->mtu, localmtu);
153 pkt->mtu = localmtu;
154 } else { /* v1.3 */
155 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
156 }
157 port->rmtu = localmtu;
158
159 /* for version >= 1.6, ACK packet mode we support */
160 if (vio_version_after_eq(vio, 1, 6)) {
161 pkt->xfer_mode = VIO_NEW_DRING_MODE;
162 pkt->options = VIO_TX_DRING;
163 }
164
165 if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
4c521e42 166 pkt->addr_type != VNET_ADDR_ETHERMAC ||
e4defc77 167 pkt->mtu != localmtu) {
4c521e42
DM		 168 viodbg(HS, "SEND NET ATTR NACK\n");
169
170 pkt->tag.stype = VIO_SUBTYPE_NACK;
171
172 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
173
174 return -ECONNRESET;
175 } else {
e4defc77
DS		 176 viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
177 "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
178 "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
179 pkt->xfer_mode, pkt->addr_type,
180 (unsigned long long)pkt->addr,
181 pkt->ack_freq, pkt->plnk_updt, pkt->options,
182 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
183 pkt->ipv4_lso_maxlen);
4c521e42
DM		 184
185 pkt->tag.stype = VIO_SUBTYPE_ACK;
186
187 return vio_ldc_send(vio, pkt, sizeof(*pkt));
188 }
189
190}
191
192static int handle_attr_ack(struct vio_driver_state *vio,
193 struct vio_net_attr_info *pkt)
194{
195 viodbg(HS, "GOT NET ATTR ACK\n");
196
197 return 0;
198}
199
200static int handle_attr_nack(struct vio_driver_state *vio,
201 struct vio_net_attr_info *pkt)
202{
203 viodbg(HS, "GOT NET ATTR NACK\n");
204
205 return -ECONNRESET;
206}
207
208static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
209{
210 struct vio_net_attr_info *pkt = arg;
211
212 switch (pkt->tag.stype) {
213 case VIO_SUBTYPE_INFO:
214 return handle_attr_info(vio, pkt);
215
216 case VIO_SUBTYPE_ACK:
217 return handle_attr_ack(vio, pkt);
218
219 case VIO_SUBTYPE_NACK:
220 return handle_attr_nack(vio, pkt);
221
222 default:
223 return -ECONNRESET;
224 }
225}
226
227static void vnet_handshake_complete(struct vio_driver_state *vio)
228{
229 struct vio_dring_state *dr;
230
231 dr = &vio->drings[VIO_DRIVER_RX_RING];
232 dr->snd_nxt = dr->rcv_nxt = 1;
233
234 dr = &vio->drings[VIO_DRIVER_TX_RING];
235 dr->snd_nxt = dr->rcv_nxt = 1;
236}
237
238/* The hypervisor interface that implements copying to/from imported
239 * memory from another domain requires that copies are done to 8-byte
240 * aligned buffers, and that the lengths of such copies are also 8-byte
241 * multiples.
242 *
243 * So we align skb->data to an 8-byte multiple and pad-out the data
244 * area so we can round the copy length up to the next multiple of
245 * 8 for the copy.
246 *
247 * The transmitter puts the actual start of the packet 6 bytes into
248 * the buffer it sends over, so that the IP headers after the ethernet
249 * header are aligned properly. These 6 bytes are not in the descriptor
250 * length, they are simply implied. This offset is represented using
251 * the VNET_PACKET_SKIP macro.
252 */
253static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
254 unsigned int len)
255{
256 struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
257 unsigned long addr, off;
258
259 if (unlikely(!skb))
260 return NULL;
261
262 addr = (unsigned long) skb->data;
263 off = ((addr + 7UL) & ~7UL) - addr;
264 if (off)
265 skb_reserve(skb, off);
266
267 return skb;
268}
269
270static int vnet_rx_one(struct vnet_port *port, unsigned int len,
271 struct ldc_trans_cookie *cookies, int ncookies)
272{
273 struct net_device *dev = port->vp->dev;
274 unsigned int copy_len;
275 struct sk_buff *skb;
276 int err;
277
278 err = -EMSGSIZE;
e4defc77 279 if (unlikely(len < ETH_ZLEN || len > port->rmtu)) {
4c521e42
DM		 280 dev->stats.rx_length_errors++;
281 goto out_dropped;
282 }
283
284 skb = alloc_and_align_skb(dev, len);
285 err = -ENOMEM;
286 if (unlikely(!skb)) {
287 dev->stats.rx_missed_errors++;
288 goto out_dropped;
289 }
290
291 copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
292 skb_put(skb, copy_len);
293 err = ldc_copy(port->vio.lp, LDC_COPY_IN,
294 skb->data, copy_len, 0,
295 cookies, ncookies);
296 if (unlikely(err < 0)) {
297 dev->stats.rx_frame_errors++;
298 goto out_free_skb;
299 }
300
301 skb_pull(skb, VNET_PACKET_SKIP);
302 skb_trim(skb, len);
303 skb->protocol = eth_type_trans(skb, dev);
304
305 dev->stats.rx_packets++;
306 dev->stats.rx_bytes += len;
307
308 netif_rx(skb);
309
310 return 0;
311
312out_free_skb:
313 kfree_skb(skb);
314
315out_dropped:
316 dev->stats.rx_dropped++;
317 return err;
318}
319
320static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
321 u32 start, u32 end, u8 vio_dring_state)
322{
323 struct vio_dring_data hdr = {
324 .tag = {
325 .type = VIO_TYPE_DATA,
326 .stype = VIO_SUBTYPE_ACK,
327 .stype_env = VIO_DRING_DATA,
328 .sid = vio_send_sid(&port->vio),
329 },
330 .dring_ident = dr->ident,
331 .start_idx = start,
332 .end_idx = end,
333 .state = vio_dring_state,
334 };
335 int err, delay;
adddc32d 336 int retries = 0;
4c521e42
DM		 337
338 hdr.seq = dr->snd_nxt;
339 delay = 1;
340 do {
341 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
342 if (err > 0) {
343 dr->snd_nxt++;
344 break;
345 }
346 udelay(delay);
347 if ((delay <<= 1) > 128)
348 delay = 128;
adddc32d
SV		 349 if (retries++ > VNET_MAX_RETRIES) {
350 pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
351 port->raddr[0], port->raddr[1],
352 port->raddr[2], port->raddr[3],
353 port->raddr[4], port->raddr[5]);
d1015645 354 break;
adddc32d 355 }
4c521e42
DM		 356 } while (err == -EAGAIN);
357
d1015645
SV		 358 if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
359 port->stop_rx_idx = end;
360 port->stop_rx = true;
361 } else {
362 port->stop_rx_idx = 0;
363 port->stop_rx = false;
364 }
365
4c521e42
DM		 366 return err;
367}
368
369static u32 next_idx(u32 idx, struct vio_dring_state *dr)
370{
371 if (++idx == dr->num_entries)
372 idx = 0;
373 return idx;
374}
375
376static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
377{
378 if (idx == 0)
379 idx = dr->num_entries - 1;
380 else
381 idx--;
382
383 return idx;
384}
385
386static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
387 struct vio_dring_state *dr,
388 u32 index)
389{
390 struct vio_net_desc *desc = port->vio.desc_buf;
391 int err;
392
393 err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
394 (index * dr->entry_size),
395 dr->cookies, dr->ncookies);
396 if (err < 0)
397 return ERR_PTR(err);
398
399 return desc;
400}
401
402static int put_rx_desc(struct vnet_port *port,
403 struct vio_dring_state *dr,
404 struct vio_net_desc *desc,
405 u32 index)
406{
407 int err;
408
409 err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
410 (index * dr->entry_size),
411 dr->cookies, dr->ncookies);
412 if (err < 0)
413 return err;
414
415 return 0;
416}
417
418static int vnet_walk_rx_one(struct vnet_port *port,
419 struct vio_dring_state *dr,
420 u32 index, int *needs_ack)
421{
422 struct vio_net_desc *desc = get_rx_desc(port, dr, index);
423 struct vio_driver_state *vio = &port->vio;
424 int err;
425
426 if (IS_ERR(desc))
427 return PTR_ERR(desc);
428
78dcff7b
DS		 429 if (desc->hdr.state != VIO_DESC_READY)
430 return 1;
431
432 rmb();
433
3f4528d6 434 viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
4c521e42
DM		 435 desc->hdr.state, desc->hdr.ack,
436 desc->size, desc->ncookies,
437 desc->cookies[0].cookie_addr,
438 desc->cookies[0].cookie_size);
439
4c521e42
DM		 440 err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
441 if (err == -ECONNRESET)
442 return err;
443 desc->hdr.state = VIO_DESC_DONE;
444 err = put_rx_desc(port, dr, desc, index);
445 if (err < 0)
446 return err;
447 *needs_ack = desc->hdr.ack;
448 return 0;
449}
450
451static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
452 u32 start, u32 end)
453{
454 struct vio_driver_state *vio = &port->vio;
455 int ack_start = -1, ack_end = -1;
456
457 end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
458
459 viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
460
461 while (start != end) {
462 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
463 if (err == -ECONNRESET)
464 return err;
465 if (err != 0)
466 break;
467 if (ack_start == -1)
468 ack_start = start;
469 ack_end = start;
470 start = next_idx(start, dr);
471 if (ack && start != end) {
472 err = vnet_send_ack(port, dr, ack_start, ack_end,
473 VIO_DRING_ACTIVE);
474 if (err == -ECONNRESET)
475 return err;
476 ack_start = -1;
477 }
478 }
479 if (unlikely(ack_start == -1))
480 ack_start = ack_end = prev_idx(start, dr);
481 return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
482}
483
484static int vnet_rx(struct vnet_port *port, void *msgbuf)
485{
486 struct vio_dring_data *pkt = msgbuf;
487 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
488 struct vio_driver_state *vio = &port->vio;
489
3f4528d6 490 viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
4c521e42
DM		 491 pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
492
493 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
494 return 0;
495 if (unlikely(pkt->seq != dr->rcv_nxt)) {
4d5870ec
JP		 496 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
497 pkt->seq, dr->rcv_nxt);
4c521e42
DM		 498 return 0;
499 }
500
501 dr->rcv_nxt++;
502
503 /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
504
505 return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
506}
507
508static int idx_is_pending(struct vio_dring_state *dr, u32 end)
509{
510 u32 idx = dr->cons;
511 int found = 0;
512
513 while (idx != dr->prod) {
514 if (idx == end) {
515 found = 1;
516 break;
517 }
518 idx = next_idx(idx, dr);
519 }
520 return found;
521}
522
523static int vnet_ack(struct vnet_port *port, void *msgbuf)
524{
525 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
526 struct vio_dring_data *pkt = msgbuf;
527 struct net_device *dev;
528 struct vnet *vp;
529 u32 end;
d1015645 530 struct vio_net_desc *desc;
4c521e42
DM		 531 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
532 return 0;
533
534 end = pkt->end_idx;
535 if (unlikely(!idx_is_pending(dr, end)))
536 return 0;
537
d1015645
SV		 538 /* sync for race conditions with vnet_start_xmit() and tell xmit it
539 * is time to send a trigger.
540 */
4c521e42 541 dr->cons = next_idx(end, dr);
d1015645
SV		 542 desc = vio_dring_entry(dr, dr->cons);
543 if (desc->hdr.state == VIO_DESC_READY && port->start_cons) {
544 /* vnet_start_xmit() just populated this dring but missed
545 * sending the "start" LDC message to the consumer.
546 * Send a "start" trigger on its behalf.
547 */
548 if (__vnet_tx_trigger(port, dr->cons) > 0)
549 port->start_cons = false;
550 else
551 port->start_cons = true;
552 } else {
553 port->start_cons = true;
554 }
555
4c521e42
DM		 556
557 vp = port->vp;
558 dev = vp->dev;
559 if (unlikely(netif_queue_stopped(dev) &&
560 vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
561 return 1;
562
563 return 0;
564}
565
566static int vnet_nack(struct vnet_port *port, void *msgbuf)
567{
568 /* XXX just reset or similar XXX */
569 return 0;
570}
571
028ebff2
DM		 572static int handle_mcast(struct vnet_port *port, void *msgbuf)
573{
574 struct vio_net_mcast_info *pkt = msgbuf;
575
576 if (pkt->tag.stype != VIO_SUBTYPE_ACK)
4d5870ec 577 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
028ebff2
DM		 578 port->vp->dev->name,
579 pkt->tag.type,
580 pkt->tag.stype,
581 pkt->tag.stype_env,
582 pkt->tag.sid);
583
584 return 0;
585}
586
1d311ad2 587static void maybe_tx_wakeup(unsigned long param)
4c521e42 588{
1d311ad2 589 struct vnet *vp = (struct vnet *)param;
4c521e42
DM		 590 struct net_device *dev = vp->dev;
591
592 netif_tx_lock(dev);
593 if (likely(netif_queue_stopped(dev))) {
594 struct vnet_port *port;
595 int wake = 1;
596
597 list_for_each_entry(port, &vp->port_list, list) {
598 struct vio_dring_state *dr;
599
600 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
601 if (vnet_tx_dring_avail(dr) <
602 VNET_TX_WAKEUP_THRESH(dr)) {
603 wake = 0;
604 break;
605 }
606 }
607 if (wake)
608 netif_wake_queue(dev);
609 }
610 netif_tx_unlock(dev);
611}
612
613static void vnet_event(void *arg, int event)
614{
615 struct vnet_port *port = arg;
616 struct vio_driver_state *vio = &port->vio;
617 unsigned long flags;
618 int tx_wakeup, err;
619
620 spin_lock_irqsave(&vio->lock, flags);
621
622 if (unlikely(event == LDC_EVENT_RESET ||
623 event == LDC_EVENT_UP)) {
624 vio_link_state_change(vio, event);
625 spin_unlock_irqrestore(&vio->lock, flags);
626
e4defc77
DS		 627 if (event == LDC_EVENT_RESET) {
628 port->rmtu = 0;
d762acdb 629 vio_port_up(vio);
e4defc77 630 }
4c521e42
DM		 631 return;
632 }
633
634 if (unlikely(event != LDC_EVENT_DATA_READY)) {
fe3881cf 635 pr_warn("Unexpected LDC event %d\n", event);
4c521e42
DM		 636 spin_unlock_irqrestore(&vio->lock, flags);
637 return;
638 }
639
640 tx_wakeup = err = 0;
641 while (1) {
642 union {
643 struct vio_msg_tag tag;
644 u64 raw[8];
645 } msgbuf;
646
647 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
648 if (unlikely(err < 0)) {
649 if (err == -ECONNRESET)
650 vio_conn_reset(vio);
651 break;
652 }
653 if (err == 0)
654 break;
655 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
656 msgbuf.tag.type,
657 msgbuf.tag.stype,
658 msgbuf.tag.stype_env,
659 msgbuf.tag.sid);
660 err = vio_validate_sid(vio, &msgbuf.tag);
661 if (err < 0)
662 break;
663
664 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
665 if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
666 err = vnet_rx(port, &msgbuf);
667 } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
668 err = vnet_ack(port, &msgbuf);
669 if (err > 0)
670 tx_wakeup |= err;
671 } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
672 err = vnet_nack(port, &msgbuf);
673 }
674 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
028ebff2
DM		 675 if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
676 err = handle_mcast(port, &msgbuf);
677 else
678 err = vio_control_pkt_engine(vio, &msgbuf);
4c521e42
DM		 679 if (err)
680 break;
681 } else {
682 err = vnet_handle_unknown(port, &msgbuf);
683 }
684 if (err == -ECONNRESET)
685 break;
686 }
687 spin_unlock(&vio->lock);
1d311ad2
SV		 688 /* Kick off a tasklet to wake the queue. We cannot call
689 * maybe_tx_wakeup directly here because we could deadlock on
690 * netif_tx_lock() with dev_watchdog()
691 */
4c521e42 692 if (unlikely(tx_wakeup && err != -ECONNRESET))
1d311ad2
SV		 693 tasklet_schedule(&port->vp->vnet_tx_wakeup);
694
4c521e42
DM		 695 local_irq_restore(flags);
696}
697
d1015645 698static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
4c521e42
DM		 699{
700 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
701 struct vio_dring_data hdr = {
702 .tag = {
703 .type = VIO_TYPE_DATA,
704 .stype = VIO_SUBTYPE_INFO,
705 .stype_env = VIO_DRING_DATA,
706 .sid = vio_send_sid(&port->vio),
707 },
708 .dring_ident = dr->ident,
d1015645 709 .start_idx = start,
4c521e42
DM		 710 .end_idx = (u32) -1,
711 };
712 int err, delay;
adddc32d 713 int retries = 0;
4c521e42 714
d1015645
SV		 715 if (port->stop_rx) {
716 err = vnet_send_ack(port,
717 &port->vio.drings[VIO_DRIVER_RX_RING],
718 port->stop_rx_idx, -1,
719 VIO_DRING_STOPPED);
720 if (err <= 0)
721 return err;
722 }
723
4c521e42
DM		 724 hdr.seq = dr->snd_nxt;
725 delay = 1;
726 do {
727 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
728 if (err > 0) {
729 dr->snd_nxt++;
730 break;
731 }
732 udelay(delay);
733 if ((delay <<= 1) > 128)
734 delay = 128;
adddc32d
SV		 735 if (retries++ > VNET_MAX_RETRIES)
736 break;
4c521e42
DM		 737 } while (err == -EAGAIN);
738
739 return err;
740}
741
8266f5fc
DS		 742static inline bool port_is_up(struct vnet_port *vnet)
743{
744 struct vio_driver_state *vio = &vnet->vio;
745
746 return !!(vio->hs_state & VIO_HS_COMPLETE);
747}
748
4c521e42
DM		 749struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
750{
751 unsigned int hash = vnet_hashfn(skb->data);
752 struct hlist_head *hp = &vp->port_hash[hash];
4c521e42
DM		 753 struct vnet_port *port;
754
b67bfe0d 755 hlist_for_each_entry(port, hp, hash) {
8266f5fc
DS		 756 if (!port_is_up(port))
757 continue;
2e42e474 758 if (ether_addr_equal(port->raddr, skb->data))
4c521e42
DM		 759 return port;
760 }
8266f5fc
DS		 761 list_for_each_entry(port, &vp->port_list, list) {
762 if (!port->switch_port)
763 continue;
764 if (!port_is_up(port))
765 continue;
766 return port;
767 }
768 return NULL;
4c521e42
DM		 769}
770
771struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
772{
773 struct vnet_port *ret;
774 unsigned long flags;
775
776 spin_lock_irqsave(&vp->lock, flags);
777 ret = __tx_port_find(vp, skb);
778 spin_unlock_irqrestore(&vp->lock, flags);
779
780 return ret;
781}
782
8e845f4c
DS		 783static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
784 unsigned *pending)
785{
786 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
787 struct sk_buff *skb = NULL;
788 int i, txi;
789
790 *pending = 0;
791
792 txi = dr->prod-1;
793 if (txi < 0)
794 txi = VNET_TX_RING_SIZE-1;
795
796 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
797 struct vio_net_desc *d;
798
799 d = vio_dring_entry(dr, txi);
800
801 if (d->hdr.state == VIO_DESC_DONE) {
802 if (port->tx_bufs[txi].skb) {
803 BUG_ON(port->tx_bufs[txi].skb->next);
804
805 port->tx_bufs[txi].skb->next = skb;
806 skb = port->tx_bufs[txi].skb;
807 port->tx_bufs[txi].skb = NULL;
808
809 ldc_unmap(port->vio.lp,
810 port->tx_bufs[txi].cookies,
811 port->tx_bufs[txi].ncookies);
812 }
813 d->hdr.state = VIO_DESC_FREE;
814 } else if (d->hdr.state == VIO_DESC_READY) {
815 (*pending)++;
816 } else if (d->hdr.state == VIO_DESC_FREE) {
817 break;
818 }
819 --txi;
820 if (txi < 0)
821 txi = VNET_TX_RING_SIZE-1;
822 }
823 return skb;
824}
825
826static inline void vnet_free_skbs(struct sk_buff *skb)
827{
828 struct sk_buff *next;
829
830 while (skb) {
831 next = skb->next;
832 skb->next = NULL;
833 dev_kfree_skb(skb);
834 skb = next;
835 }
836}
837
838static void vnet_clean_timer_expire(unsigned long port0)
839{
840 struct vnet_port *port = (struct vnet_port *)port0;
841 struct sk_buff *freeskbs;
842 unsigned pending;
843 unsigned long flags;
844
845 spin_lock_irqsave(&port->vio.lock, flags);
846 freeskbs = vnet_clean_tx_ring(port, &pending);
847 spin_unlock_irqrestore(&port->vio.lock, flags);
848
849 vnet_free_skbs(freeskbs);
850
851 if (pending)
852 (void)mod_timer(&port->clean_timer,
853 jiffies + VNET_CLEAN_TIMEOUT);
854 else
855 del_timer(&port->clean_timer);
856}
857
858static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, void **pstart,
859 int *plen)
860{
861 struct sk_buff *nskb;
862 int len, pad;
863
864 len = skb->len;
865 pad = 0;
866 if (len < ETH_ZLEN) {
867 pad += ETH_ZLEN - skb->len;
868 len += pad;
869 }
870 len += VNET_PACKET_SKIP;
871 pad += 8 - (len & 7);
872 len += 8 - (len & 7);
873
874 if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
875 skb_tailroom(skb) < pad ||
876 skb_headroom(skb) < VNET_PACKET_SKIP) {
877 nskb = alloc_and_align_skb(skb->dev, skb->len);
878 skb_reserve(nskb, VNET_PACKET_SKIP);
879 if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
880 dev_kfree_skb(nskb);
881 dev_kfree_skb(skb);
882 return NULL;
883 }
884 (void)skb_put(nskb, skb->len);
885 dev_kfree_skb(skb);
886 skb = nskb;
887 }
888
889 *pstart = skb->data - VNET_PACKET_SKIP;
890 *plen = len;
891 return skb;
892}
893
4c521e42
DM		 894static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
895{
896 struct vnet *vp = netdev_priv(dev);
897 struct vnet_port *port = tx_port_find(vp, skb);
898 struct vio_dring_state *dr;
899 struct vio_net_desc *d;
900 unsigned long flags;
901 unsigned int len;
8e845f4c
DS		 902 struct sk_buff *freeskbs = NULL;
903 int i, err, txi;
904 void *start = NULL;
905 int nlen = 0;
906 unsigned pending = 0;
4c521e42
DM		 907
908 if (unlikely(!port))
909 goto out_dropped;
910
8e845f4c
DS		 911 skb = vnet_skb_shape(skb, &start, &nlen);
912
913 if (unlikely(!skb))
914 goto out_dropped;
915
42db672d
DS		 916 if (skb->len > port->rmtu)
917 goto out_dropped;
918
4c521e42
DM		 919 spin_lock_irqsave(&port->vio.lock, flags);
920
921 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
922 if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
923 if (!netif_queue_stopped(dev)) {
924 netif_stop_queue(dev);
925
926 /* This is a hard error, log it. */
4d5870ec 927 netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
4c521e42
DM		 928 dev->stats.tx_errors++;
929 }
930 spin_unlock_irqrestore(&port->vio.lock, flags);
931 return NETDEV_TX_BUSY;
932 }
933
934 d = vio_dring_cur(dr);
935
8e845f4c
DS		 936 txi = dr->prod;
937
938 freeskbs = vnet_clean_tx_ring(port, &pending);
939
940 BUG_ON(port->tx_bufs[txi].skb);
4c521e42
DM		 941
942 len = skb->len;
8e845f4c 943 if (len < ETH_ZLEN)
4c521e42 944 len = ETH_ZLEN;
8e845f4c
DS		 945
946 port->tx_bufs[txi].skb = skb;
947 skb = NULL;
948
949 err = ldc_map_single(port->vio.lp, start, nlen,
42db672d 950 port->tx_bufs[txi].cookies, VNET_MAXCOOKIES,
8e845f4c
DS		 951 (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
952 if (err < 0) {
953 netdev_info(dev, "tx buffer map error %d\n", err);
954 goto out_dropped_unlock;
4c521e42 955 }
8e845f4c 956 port->tx_bufs[txi].ncookies = err;
4c521e42 957
1f6394e3
SV		 958 /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
959 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
960 * the protocol itself does not require it as long as the peer
961 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
962 *
963 * An ACK for every packet in the ring is expensive as the
964 * sending of LDC messages is slow and affects performance.
965 */
966 d->hdr.ack = VIO_ACK_DISABLE;
4c521e42 967 d->size = len;
8e845f4c 968 d->ncookies = port->tx_bufs[txi].ncookies;
4c521e42 969 for (i = 0; i < d->ncookies; i++)
8e845f4c 970 d->cookies[i] = port->tx_bufs[txi].cookies[i];
4c521e42
DM		 971
972 /* This has to be a non-SMP write barrier because we are writing
973 * to memory which is shared with the peer LDOM.
974 */
975 wmb();
976
977 d->hdr.state = VIO_DESC_READY;
978
d1015645
SV		 979 /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
980 * to notify the consumer that some descriptors are READY.
981 * After that "start" trigger, no additional triggers are needed until
982 * a DRING_STOPPED is received from the consumer. The dr->cons field
983 * (set up by vnet_ack()) has the value of the next dring index
984 * that has not yet been ack-ed. We send a "start" trigger here
985 * if, and only if, start_cons is true (reset it afterward). Conversely,
986 * vnet_ack() should check if the dring corresponding to cons
987 * is marked READY, but start_cons was false.
988 * If so, vnet_ack() should send out the missed "start" trigger.
989 *
990 * Note that the wmb() above makes sure the cookies et al. are
991 * not globally visible before the VIO_DESC_READY, and that the
992 * stores are ordered correctly by the compiler. The consumer will
993 * not proceed until the VIO_DESC_READY is visible assuring that
994 * the consumer does not observe anything related to descriptors
995 * out of order. The HV trap from the LDC start trigger is the
996 * producer to consumer announcement that work is available to the
997 * consumer
998 */
999 if (!port->start_cons)
1000 goto ldc_start_done; /* previous trigger suffices */
1001
1002 err = __vnet_tx_trigger(port, dr->cons);
4c521e42 1003 if (unlikely(err < 0)) {
4d5870ec 1004 netdev_info(dev, "TX trigger error %d\n", err);
4c521e42
DM		 1005 d->hdr.state = VIO_DESC_FREE;
1006 dev->stats.tx_carrier_errors++;
1007 goto out_dropped_unlock;
1008 }
1009
d1015645
SV		 1010ldc_start_done:
1011 port->start_cons = false;
1012
4c521e42 1013 dev->stats.tx_packets++;
8e845f4c 1014 dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
4c521e42
DM		 1015
1016 dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1017 if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
1018 netif_stop_queue(dev);
1019 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1020 netif_wake_queue(dev);
1021 }
1022
1023 spin_unlock_irqrestore(&port->vio.lock, flags);
1024
8e845f4c
DS		 1025 vnet_free_skbs(freeskbs);
1026
1027 (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
4c521e42 1028
4c521e42
DM		 1029 return NETDEV_TX_OK;
1030
1031out_dropped_unlock:
1032 spin_unlock_irqrestore(&port->vio.lock, flags);
1033
1034out_dropped:
8e845f4c
DS		 1035 if (skb)
1036 dev_kfree_skb(skb);
1037 vnet_free_skbs(freeskbs);
1038 if (pending)
1039 (void)mod_timer(&port->clean_timer,
1040 jiffies + VNET_CLEAN_TIMEOUT);
1041 else
1042 del_timer(&port->clean_timer);
4c521e42
DM		 1043 dev->stats.tx_dropped++;
1044 return NETDEV_TX_OK;
1045}
1046
1047static void vnet_tx_timeout(struct net_device *dev)
1048{
1049 /* XXX Implement me XXX */
1050}
1051
1052static int vnet_open(struct net_device *dev)
1053{
1054 netif_carrier_on(dev);
1055 netif_start_queue(dev);
1056
1057 return 0;
1058}
1059
1060static int vnet_close(struct net_device *dev)
1061{
1062 netif_stop_queue(dev);
1063 netif_carrier_off(dev);
1064
1065 return 0;
1066}
1067
028ebff2
DM		 1068static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1069{
1070 struct vnet_mcast_entry *m;
1071
1072 for (m = vp->mcast_list; m; m = m->next) {
00fa4ce9 1073 if (ether_addr_equal(m->addr, addr))
028ebff2
DM		 1074 return m;
1075 }
1076 return NULL;
1077}
1078
1079static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1080{
22bedad3 1081 struct netdev_hw_addr *ha;
028ebff2 1082
22bedad3 1083 netdev_for_each_mc_addr(ha, dev) {
028ebff2
DM		 1084 struct vnet_mcast_entry *m;
1085
22bedad3 1086 m = __vnet_mc_find(vp, ha->addr);
028ebff2
DM		 1087 if (m) {
1088 m->hit = 1;
1089 continue;
1090 }
1091
1092 if (!m) {
1093 m = kzalloc(sizeof(*m), GFP_ATOMIC);
1094 if (!m)
1095 continue;
22bedad3 1096 memcpy(m->addr, ha->addr, ETH_ALEN);
028ebff2
DM		 1097 m->hit = 1;
1098
1099 m->next = vp->mcast_list;
1100 vp->mcast_list = m;
1101 }
1102 }
1103}
1104
1105static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1106{
1107 struct vio_net_mcast_info info;
1108 struct vnet_mcast_entry *m, **pp;
1109 int n_addrs;
1110
1111 memset(&info, 0, sizeof(info));
1112
1113 info.tag.type = VIO_TYPE_CTRL;
1114 info.tag.stype = VIO_SUBTYPE_INFO;
1115 info.tag.stype_env = VNET_MCAST_INFO;
1116 info.tag.sid = vio_send_sid(&port->vio);
1117 info.set = 1;
1118
1119 n_addrs = 0;
1120 for (m = vp->mcast_list; m; m = m->next) {
1121 if (m->sent)
1122 continue;
1123 m->sent = 1;
1124 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1125 m->addr, ETH_ALEN);
1126 if (++n_addrs == VNET_NUM_MCAST) {
1127 info.count = n_addrs;
1128
1129 (void) vio_ldc_send(&port->vio, &info,
1130 sizeof(info));
1131 n_addrs = 0;
1132 }
1133 }
1134 if (n_addrs) {
1135 info.count = n_addrs;
1136 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1137 }
1138
1139 info.set = 0;
1140
1141 n_addrs = 0;
1142 pp = &vp->mcast_list;
1143 while ((m = *pp) != NULL) {
1144 if (m->hit) {
1145 m->hit = 0;
1146 pp = &m->next;
1147 continue;
1148 }
1149
1150 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1151 m->addr, ETH_ALEN);
1152 if (++n_addrs == VNET_NUM_MCAST) {
1153 info.count = n_addrs;
1154 (void) vio_ldc_send(&port->vio, &info,
1155 sizeof(info));
1156 n_addrs = 0;
1157 }
1158
1159 *pp = m->next;
1160 kfree(m);
1161 }
1162 if (n_addrs) {
1163 info.count = n_addrs;
1164 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1165 }
1166}
1167
4c521e42
DM		 1168static void vnet_set_rx_mode(struct net_device *dev)
1169{
028ebff2
DM		 1170 struct vnet *vp = netdev_priv(dev);
1171 struct vnet_port *port;
1172 unsigned long flags;
1173
1174 spin_lock_irqsave(&vp->lock, flags);
1175 if (!list_empty(&vp->port_list)) {
1176 port = list_entry(vp->port_list.next, struct vnet_port, list);
1177
1178 if (port->switch_port) {
1179 __update_mc_list(vp, dev);
1180 __send_mc_list(vp, port);
1181 }
1182 }
1183 spin_unlock_irqrestore(&vp->lock, flags);
4c521e42
DM		 1184}
1185
1186static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1187{
42db672d 1188 if (new_mtu < 68 || new_mtu > 65535)
4c521e42
DM		 1189 return -EINVAL;
1190
1191 dev->mtu = new_mtu;
1192 return 0;
1193}
1194
1195static int vnet_set_mac_addr(struct net_device *dev, void *p)
1196{
1197 return -EINVAL;
1198}
1199
1200static void vnet_get_drvinfo(struct net_device *dev,
1201 struct ethtool_drvinfo *info)
1202{
7826d43f
JP		 1203 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1204 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4c521e42
DM		 1205}
1206
1207static u32 vnet_get_msglevel(struct net_device *dev)
1208{
1209 struct vnet *vp = netdev_priv(dev);
1210 return vp->msg_enable;
1211}
1212
1213static void vnet_set_msglevel(struct net_device *dev, u32 value)
1214{
1215 struct vnet *vp = netdev_priv(dev);
1216 vp->msg_enable = value;
1217}
1218
1219static const struct ethtool_ops vnet_ethtool_ops = {
1220 .get_drvinfo = vnet_get_drvinfo,
1221 .get_msglevel = vnet_get_msglevel,
1222 .set_msglevel = vnet_set_msglevel,
1223 .get_link = ethtool_op_get_link,
4c521e42
DM		 1224};
1225
1226static void vnet_port_free_tx_bufs(struct vnet_port *port)
1227{
1228 struct vio_dring_state *dr;
1229 int i;
1230
1231 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1232 if (dr->base) {
1233 ldc_free_exp_dring(port->vio.lp, dr->base,
1234 (dr->entry_size * dr->num_entries),
1235 dr->cookies, dr->ncookies);
1236 dr->base = NULL;
1237 dr->entry_size = 0;
1238 dr->num_entries = 0;
1239 dr->pending = 0;
1240 dr->ncookies = 0;
1241 }
1242
1243 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
8e845f4c
DS		 1244 struct vio_net_desc *d;
1245 void *skb = port->tx_bufs[i].skb;
4c521e42 1246
8e845f4c 1247 if (!skb)
4c521e42
DM		 1248 continue;
1249
8e845f4c
DS		 1250 d = vio_dring_entry(dr, i);
1251 if (d->hdr.state == VIO_DESC_READY)
1252 pr_warn("active transmit buffers freed\n");
1253
4c521e42
DM		 1254 ldc_unmap(port->vio.lp,
1255 port->tx_bufs[i].cookies,
1256 port->tx_bufs[i].ncookies);
8e845f4c
DS		 1257 dev_kfree_skb(skb);
1258 port->tx_bufs[i].skb = NULL;
1259 d->hdr.state = VIO_DESC_FREE;
4c521e42
DM		 1260 }
1261}
1262
f73d12bd 1263static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
4c521e42
DM		 1264{
1265 struct vio_dring_state *dr;
1266 unsigned long len;
1267 int i, err, ncookies;
1268 void *dring;
1269
4c521e42
DM		 1270 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1271
1272 len = (VNET_TX_RING_SIZE *
1273 (sizeof(struct vio_net_desc) +
1274 (sizeof(struct ldc_trans_cookie) * 2)));
1275
1276 ncookies = VIO_MAX_RING_COOKIES;
1277 dring = ldc_alloc_exp_dring(port->vio.lp, len,
1278 dr->cookies, &ncookies,
1279 (LDC_MAP_SHADOW |
1280 LDC_MAP_DIRECT |
1281 LDC_MAP_RW));
1282 if (IS_ERR(dring)) {
1283 err = PTR_ERR(dring);
1284 goto err_out;
1285 }
1286
1287 dr->base = dring;
1288 dr->entry_size = (sizeof(struct vio_net_desc) +
1289 (sizeof(struct ldc_trans_cookie) * 2));
1290 dr->num_entries = VNET_TX_RING_SIZE;
1291 dr->prod = dr->cons = 0;
d1015645 1292 port->start_cons = true; /* need an initial trigger */
4c521e42
DM		 1293 dr->pending = VNET_TX_RING_SIZE;
1294 dr->ncookies = ncookies;
1295
8e845f4c
DS		 1296 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1297 struct vio_net_desc *d;
1298
1299 d = vio_dring_entry(dr, i);
1300 d->hdr.state = VIO_DESC_FREE;
1301 }
4c521e42
DM		 1302 return 0;
1303
1304err_out:
1305 vnet_port_free_tx_bufs(port);
1306
1307 return err;
1308}
1309
9184a046
DM		 1310static LIST_HEAD(vnet_list);
1311static DEFINE_MUTEX(vnet_list_mutex);
1312
8fd17f2e
DM		 1313static const struct net_device_ops vnet_ops = {
1314 .ndo_open = vnet_open,
1315 .ndo_stop = vnet_close,
afc4b13d 1316 .ndo_set_rx_mode = vnet_set_rx_mode,
8fd17f2e 1317 .ndo_set_mac_address = vnet_set_mac_addr,
240c102d 1318 .ndo_validate_addr = eth_validate_addr,
8fd17f2e
DM		 1319 .ndo_tx_timeout = vnet_tx_timeout,
1320 .ndo_change_mtu = vnet_change_mtu,
1321 .ndo_start_xmit = vnet_start_xmit,
1322};
1323
f73d12bd 1324static struct vnet *vnet_new(const u64 *local_mac)
9184a046
DM		 1325{
1326 struct net_device *dev;
1327 struct vnet *vp;
1328 int err, i;
1329
1330 dev = alloc_etherdev(sizeof(*vp));
41de8d4c 1331 if (!dev)
9184a046 1332 return ERR_PTR(-ENOMEM);
8e845f4c
DS		 1333 dev->needed_headroom = VNET_PACKET_SKIP + 8;
1334 dev->needed_tailroom = 8;
9184a046
DM		 1335
1336 for (i = 0; i < ETH_ALEN; i++)
1337 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1338
9184a046
DM		 1339 vp = netdev_priv(dev);
1340
1341 spin_lock_init(&vp->lock);
1d311ad2 1342 tasklet_init(&vp->vnet_tx_wakeup, maybe_tx_wakeup, (unsigned long)vp);
9184a046
DM		 1343 vp->dev = dev;
1344
1345 INIT_LIST_HEAD(&vp->port_list);
1346 for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1347 INIT_HLIST_HEAD(&vp->port_hash[i]);
1348 INIT_LIST_HEAD(&vp->list);
1349 vp->local_mac = *local_mac;
1350
8fd17f2e 1351 dev->netdev_ops = &vnet_ops;
9184a046
DM		 1352 dev->ethtool_ops = &vnet_ethtool_ops;
1353 dev->watchdog_timeo = VNET_TX_TIMEOUT;
9184a046
DM		 1354
1355 err = register_netdev(dev);
1356 if (err) {
4d5870ec 1357 pr_err("Cannot register net device, aborting\n");
9184a046
DM		 1358 goto err_out_free_dev;
1359 }
1360
4d5870ec 1361 netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
9184a046
DM		 1362
1363 list_add(&vp->list, &vnet_list);
1364
1365 return vp;
1366
1367err_out_free_dev:
1368 free_netdev(dev);
1369
1370 return ERR_PTR(err);
1371}
1372
f73d12bd 1373static struct vnet *vnet_find_or_create(const u64 *local_mac)
9184a046
DM		 1374{
1375 struct vnet *iter, *vp;
1376
1377 mutex_lock(&vnet_list_mutex);
1378 vp = NULL;
1379 list_for_each_entry(iter, &vnet_list, list) {
1380 if (iter->local_mac == *local_mac) {
1381 vp = iter;
1382 break;
1383 }
1384 }
1385 if (!vp)
1386 vp = vnet_new(local_mac);
1387 mutex_unlock(&vnet_list_mutex);
1388
1389 return vp;
1390}
1391
a4b70a07
SV		 1392static void vnet_cleanup(void)
1393{
1394 struct vnet *vp;
1395 struct net_device *dev;
1396
1397 mutex_lock(&vnet_list_mutex);
1398 while (!list_empty(&vnet_list)) {
1399 vp = list_first_entry(&vnet_list, struct vnet, list);
1400 list_del(&vp->list);
1401 dev = vp->dev;
1d311ad2 1402 tasklet_kill(&vp->vnet_tx_wakeup);
a4b70a07
SV		 1403 /* vio_unregister_driver() should have cleaned up port_list */
1404 BUG_ON(!list_empty(&vp->port_list));
1405 unregister_netdev(dev);
1406 free_netdev(dev);
1407 }
1408 mutex_unlock(&vnet_list_mutex);
1409}
1410
9184a046
DM		 1411static const char *local_mac_prop = "local-mac-address";
1412
f73d12bd 1413static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
9184a046
DM		 1414 u64 port_node)
1415{
1416 const u64 *local_mac = NULL;
1417 u64 a;
1418
1419 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1420 u64 target = mdesc_arc_target(hp, a);
1421 const char *name;
1422
1423 name = mdesc_get_property(hp, target, "name", NULL);
1424 if (!name || strcmp(name, "network"))
1425 continue;
1426
1427 local_mac = mdesc_get_property(hp, target,
1428 local_mac_prop, NULL);
1429 if (local_mac)
1430 break;
1431 }
1432 if (!local_mac)
1433 return ERR_PTR(-ENODEV);
1434
1435 return vnet_find_or_create(local_mac);
1436}
1437
4c521e42
DM		 1438static struct ldc_channel_config vnet_ldc_cfg = {
1439 .event = vnet_event,
1440 .mtu = 64,
1441 .mode = LDC_MODE_UNRELIABLE,
1442};
1443
1444static struct vio_driver_ops vnet_vio_ops = {
1445 .send_attr = vnet_send_attr,
1446 .handle_attr = vnet_handle_attr,
1447 .handshake_complete = vnet_handshake_complete,
1448};
1449
f73d12bd 1450static void print_version(void)
9184a046 1451{
4d5870ec 1452 printk_once(KERN_INFO "%s", version);
9184a046
DM		 1453}
1454
4c521e42
DM		 1455const char *remote_macaddr_prop = "remote-mac-address";
1456
1dd06ae8 1457static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
4c521e42 1458{
43fdf274 1459 struct mdesc_handle *hp;
4c521e42
DM		 1460 struct vnet_port *port;
1461 unsigned long flags;
1462 struct vnet *vp;
1463 const u64 *rmac;
1464 int len, i, err, switch_port;
1465
9184a046 1466 print_version();
4c521e42 1467
43fdf274
DM		 1468 hp = mdesc_grab();
1469
9184a046
DM		 1470 vp = vnet_find_parent(hp, vdev->mp);
1471 if (IS_ERR(vp)) {
4d5870ec 1472 pr_err("Cannot find port parent vnet\n");
9184a046
DM		 1473 err = PTR_ERR(vp);
1474 goto err_out_put_mdesc;
1475 }
1476
43fdf274
DM		 1477 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1478 err = -ENODEV;
4c521e42 1479 if (!rmac) {
4d5870ec 1480 pr_err("Port lacks %s property\n", remote_macaddr_prop);
43fdf274 1481 goto err_out_put_mdesc;
4c521e42
DM		 1482 }
1483
1484 port = kzalloc(sizeof(*port), GFP_KERNEL);
43fdf274 1485 err = -ENOMEM;
e404decb 1486 if (!port)
43fdf274 1487 goto err_out_put_mdesc;
4c521e42
DM		 1488
1489 for (i = 0; i < ETH_ALEN; i++)
1490 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1491
1492 port->vp = vp;
1493
43fdf274 1494 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
4c521e42
DM		 1495 vnet_versions, ARRAY_SIZE(vnet_versions),
1496 &vnet_vio_ops, vp->dev->name);
1497 if (err)
1498 goto err_out_free_port;
1499
1500 err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1501 if (err)
1502 goto err_out_free_port;
1503
1504 err = vnet_port_alloc_tx_bufs(port);
1505 if (err)
1506 goto err_out_free_ldc;
1507
1508 INIT_HLIST_NODE(&port->hash);
1509 INIT_LIST_HEAD(&port->list);
1510
1511 switch_port = 0;
43fdf274 1512 if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
4c521e42 1513 switch_port = 1;
028ebff2 1514 port->switch_port = switch_port;
4c521e42
DM		 1515
1516 spin_lock_irqsave(&vp->lock, flags);
1517 if (switch_port)
1518 list_add(&port->list, &vp->port_list);
1519 else
1520 list_add_tail(&port->list, &vp->port_list);
1521 hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1522 spin_unlock_irqrestore(&vp->lock, flags);
1523
1524 dev_set_drvdata(&vdev->dev, port);
1525
4d5870ec
JP		 1526 pr_info("%s: PORT ( remote-mac %pM%s )\n",
1527 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
4c521e42 1528
8e845f4c
DS		 1529 setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1530 (unsigned long)port);
1531
4c521e42
DM		 1532 vio_port_up(&port->vio);
1533
43fdf274
DM		 1534 mdesc_release(hp);
1535
4c521e42
DM		 1536 return 0;
1537
1538err_out_free_ldc:
1539 vio_ldc_free(&port->vio);
1540
1541err_out_free_port:
1542 kfree(port);
1543
43fdf274
DM		 1544err_out_put_mdesc:
1545 mdesc_release(hp);
4c521e42
DM		 1546 return err;
1547}
1548
1549static int vnet_port_remove(struct vio_dev *vdev)
1550{
1551 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1552
1553 if (port) {
1554 struct vnet *vp = port->vp;
1555 unsigned long flags;
1556
1557 del_timer_sync(&port->vio.timer);
8e845f4c 1558 del_timer_sync(&port->clean_timer);
4c521e42
DM		 1559
1560 spin_lock_irqsave(&vp->lock, flags);
1561 list_del(&port->list);
1562 hlist_del(&port->hash);
1563 spin_unlock_irqrestore(&vp->lock, flags);
1564
1565 vnet_port_free_tx_bufs(port);
1566 vio_ldc_free(&port->vio);
1567
1568 dev_set_drvdata(&vdev->dev, NULL);
1569
1570 kfree(port);
aabb9875 1571
4c521e42
DM		 1572 }
1573 return 0;
1574}
1575
3d452e55 1576static const struct vio_device_id vnet_port_match[] = {
4c521e42
DM		 1577 {
1578 .type = "vnet-port",
1579 },
1580 {},
1581};
da68e081 1582MODULE_DEVICE_TABLE(vio, vnet_port_match);
4c521e42
DM		 1583
1584static struct vio_driver vnet_port_driver = {
1585 .id_table = vnet_port_match,
1586 .probe = vnet_port_probe,
1587 .remove = vnet_port_remove,
cb52d897 1588 .name = "vnet_port",
4c521e42
DM		 1589};
1590
4c521e42
DM		 1591static int __init vnet_init(void)
1592{
9184a046 1593 return vio_register_driver(&vnet_port_driver);
4c521e42
DM		 1594}
1595
1596static void __exit vnet_exit(void)
1597{
1598 vio_unregister_driver(&vnet_port_driver);
a4b70a07 1599 vnet_cleanup();
4c521e42
DM		 1600}
1601
1602module_init(vnet_init);
1603module_exit(vnet_exit);
Linux kernel - Deliverables
This page took 0.862016 seconds and 5 git commands to generate.