projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sunvnet: add VIO v1.7 and v1.8 support
[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 19
a2b78e9b
DS		 20#if IS_ENABLED(CONFIG_IPV6)
21#include <linux/icmpv6.h>
22#endif
23
6d0ba919 24#include <net/ip.h>
a2b78e9b
DS		 25#include <net/icmp.h>
26#include <net/route.h>
27
4c521e42
DM		 28#include <asm/vio.h>
29#include <asm/ldc.h>
30
31#include "sunvnet.h"
32
33#define DRV_MODULE_NAME "sunvnet"
4c521e42
DM		 34#define DRV_MODULE_VERSION "1.0"
35#define DRV_MODULE_RELDATE "June 25, 2007"
36
f73d12bd 37static char version[] =
4c521e42
DM		 38 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
39MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
40MODULE_DESCRIPTION("Sun LDOM virtual network driver");
41MODULE_LICENSE("GPL");
42MODULE_VERSION(DRV_MODULE_VERSION);
43
d51bffd1
SV		 44#define VNET_MAX_TXQS 16
45
adddc32d
SV		 46/* Heuristic for the number of times to exponentially backoff and
47 * retry sending an LDC trigger when EAGAIN is encountered
48 */
49#define VNET_MAX_RETRIES 10
50
d1015645
SV		 51static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
52
4c521e42
DM		 53/* Ordered from largest major to lowest */
54static struct vio_version vnet_versions[] = {
6d0ba919
DS		 55 { .major = 1, .minor = 8 },
56 { .major = 1, .minor = 7 },
e4defc77 57 { .major = 1, .minor = 6 },
4c521e42
DM		 58 { .major = 1, .minor = 0 },
59};
60
61static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
62{
63 return vio_dring_avail(dr, VNET_TX_RING_SIZE);
64}
65
66static int vnet_handle_unknown(struct vnet_port *port, void *arg)
67{
68 struct vio_msg_tag *pkt = arg;
69
4d5870ec 70 pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
4c521e42 71 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
4d5870ec 72 pr_err("Resetting connection\n");
4c521e42
DM		 73
74 ldc_disconnect(port->vio.lp);
75
76 return -ECONNRESET;
77}
78
d6732489
DS		 79static int vnet_port_alloc_tx_ring(struct vnet_port *port);
80
4c521e42
DM		 81static int vnet_send_attr(struct vio_driver_state *vio)
82{
83 struct vnet_port *port = to_vnet_port(vio);
84 struct net_device *dev = port->vp->dev;
85 struct vio_net_attr_info pkt;
e4defc77 86 int framelen = ETH_FRAME_LEN;
d6732489
DS		 87 int i, err;
88
89 err = vnet_port_alloc_tx_ring(to_vnet_port(vio));
90 if (err)
91 return err;
4c521e42
DM		 92
93 memset(&pkt, 0, sizeof(pkt));
94 pkt.tag.type = VIO_TYPE_CTRL;
95 pkt.tag.stype = VIO_SUBTYPE_INFO;
96 pkt.tag.stype_env = VIO_ATTR_INFO;
97 pkt.tag.sid = vio_send_sid(vio);
e4defc77
DS		 98 if (vio_version_before(vio, 1, 2))
99 pkt.xfer_mode = VIO_DRING_MODE;
100 else
101 pkt.xfer_mode = VIO_NEW_DRING_MODE;
4c521e42
DM		 102 pkt.addr_type = VNET_ADDR_ETHERMAC;
103 pkt.ack_freq = 0;
104 for (i = 0; i < 6; i++)
105 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
e4defc77
DS		 106 if (vio_version_after(vio, 1, 3)) {
107 if (port->rmtu) {
108 port->rmtu = min(VNET_MAXPACKET, port->rmtu);
109 pkt.mtu = port->rmtu;
110 } else {
111 port->rmtu = VNET_MAXPACKET;
112 pkt.mtu = port->rmtu;
113 }
114 if (vio_version_after_eq(vio, 1, 6))
115 pkt.options = VIO_TX_DRING;
116 } else if (vio_version_before(vio, 1, 3)) {
117 pkt.mtu = framelen;
118 } else { /* v1.3 */
119 pkt.mtu = framelen + VLAN_HLEN;
120 }
121
122 pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
123 pkt.cflags = 0;
4c521e42
DM		 124
125 viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
e4defc77
DS		 126 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
127 "cflags[0x%04x] lso_max[%u]\n",
4c521e42 128 pkt.xfer_mode, pkt.addr_type,
e4defc77
DS		 129 (unsigned long long)pkt.addr,
130 pkt.ack_freq, pkt.plnk_updt, pkt.options,
131 (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
132
4c521e42
DM		 133
134 return vio_ldc_send(vio, &pkt, sizeof(pkt));
135}
136
137static int handle_attr_info(struct vio_driver_state *vio,
138 struct vio_net_attr_info *pkt)
139{
e4defc77
DS		 140 struct vnet_port *port = to_vnet_port(vio);
141 u64 localmtu;
142 u8 xfer_mode;
143
144 viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
145 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
146 " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
4c521e42 147 pkt->xfer_mode, pkt->addr_type,
e4defc77
DS		 148 (unsigned long long)pkt->addr,
149 pkt->ack_freq, pkt->plnk_updt, pkt->options,
150 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
151 pkt->ipv4_lso_maxlen);
4c521e42
DM		 152
153 pkt->tag.sid = vio_send_sid(vio);
154
e4defc77
DS		 155 xfer_mode = pkt->xfer_mode;
156 /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
157 if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
158 xfer_mode = VIO_NEW_DRING_MODE;
159
160 /* MTU negotiation:
161 * < v1.3 - ETH_FRAME_LEN exactly
162 * > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
163 * pkt->mtu for ACK
164 * = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
165 */
166 if (vio_version_before(vio, 1, 3)) {
167 localmtu = ETH_FRAME_LEN;
168 } else if (vio_version_after(vio, 1, 3)) {
169 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
170 localmtu = min(pkt->mtu, localmtu);
171 pkt->mtu = localmtu;
172 } else { /* v1.3 */
173 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
174 }
175 port->rmtu = localmtu;
176
177 /* for version >= 1.6, ACK packet mode we support */
178 if (vio_version_after_eq(vio, 1, 6)) {
179 pkt->xfer_mode = VIO_NEW_DRING_MODE;
180 pkt->options = VIO_TX_DRING;
181 }
182
183 if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
4c521e42 184 pkt->addr_type != VNET_ADDR_ETHERMAC ||
e4defc77 185 pkt->mtu != localmtu) {
4c521e42
DM		 186 viodbg(HS, "SEND NET ATTR NACK\n");
187
188 pkt->tag.stype = VIO_SUBTYPE_NACK;
189
190 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
191
192 return -ECONNRESET;
193 } else {
e4defc77
DS		 194 viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
195 "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
196 "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
197 pkt->xfer_mode, pkt->addr_type,
198 (unsigned long long)pkt->addr,
199 pkt->ack_freq, pkt->plnk_updt, pkt->options,
200 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
201 pkt->ipv4_lso_maxlen);
4c521e42
DM		 202
203 pkt->tag.stype = VIO_SUBTYPE_ACK;
204
205 return vio_ldc_send(vio, pkt, sizeof(*pkt));
206 }
207
208}
209
210static int handle_attr_ack(struct vio_driver_state *vio,
211 struct vio_net_attr_info *pkt)
212{
213 viodbg(HS, "GOT NET ATTR ACK\n");
214
215 return 0;
216}
217
218static int handle_attr_nack(struct vio_driver_state *vio,
219 struct vio_net_attr_info *pkt)
220{
221 viodbg(HS, "GOT NET ATTR NACK\n");
222
223 return -ECONNRESET;
224}
225
226static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
227{
228 struct vio_net_attr_info *pkt = arg;
229
230 switch (pkt->tag.stype) {
231 case VIO_SUBTYPE_INFO:
232 return handle_attr_info(vio, pkt);
233
234 case VIO_SUBTYPE_ACK:
235 return handle_attr_ack(vio, pkt);
236
237 case VIO_SUBTYPE_NACK:
238 return handle_attr_nack(vio, pkt);
239
240 default:
241 return -ECONNRESET;
242 }
243}
244
245static void vnet_handshake_complete(struct vio_driver_state *vio)
246{
247 struct vio_dring_state *dr;
248
249 dr = &vio->drings[VIO_DRIVER_RX_RING];
250 dr->snd_nxt = dr->rcv_nxt = 1;
251
252 dr = &vio->drings[VIO_DRIVER_TX_RING];
253 dr->snd_nxt = dr->rcv_nxt = 1;
254}
255
256/* The hypervisor interface that implements copying to/from imported
257 * memory from another domain requires that copies are done to 8-byte
258 * aligned buffers, and that the lengths of such copies are also 8-byte
259 * multiples.
260 *
261 * So we align skb->data to an 8-byte multiple and pad-out the data
262 * area so we can round the copy length up to the next multiple of
263 * 8 for the copy.
264 *
265 * The transmitter puts the actual start of the packet 6 bytes into
266 * the buffer it sends over, so that the IP headers after the ethernet
267 * header are aligned properly. These 6 bytes are not in the descriptor
268 * length, they are simply implied. This offset is represented using
269 * the VNET_PACKET_SKIP macro.
270 */
271static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
272 unsigned int len)
273{
274 struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
275 unsigned long addr, off;
276
277 if (unlikely(!skb))
278 return NULL;
279
280 addr = (unsigned long) skb->data;
281 off = ((addr + 7UL) & ~7UL) - addr;
282 if (off)
283 skb_reserve(skb, off);
284
285 return skb;
286}
287
6d0ba919
DS		 288static inline void vnet_fullcsum(struct sk_buff *skb)
289{
290 struct iphdr *iph = ip_hdr(skb);
291 int offset = skb_transport_offset(skb);
292
293 if (skb->protocol != htons(ETH_P_IP))
294 return;
295 if (iph->protocol != IPPROTO_TCP &&
296 iph->protocol != IPPROTO_UDP)
297 return;
298 skb->ip_summed = CHECKSUM_NONE;
299 skb->csum_level = 1;
300 skb->csum = 0;
301 if (iph->protocol == IPPROTO_TCP) {
302 struct tcphdr *ptcp = tcp_hdr(skb);
303
304 ptcp->check = 0;
305 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
306 ptcp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
307 skb->len - offset, IPPROTO_TCP,
308 skb->csum);
309 } else if (iph->protocol == IPPROTO_UDP) {
310 struct udphdr *pudp = udp_hdr(skb);
311
312 pudp->check = 0;
313 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
314 pudp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
315 skb->len - offset, IPPROTO_UDP,
316 skb->csum);
317 }
318}
319
320static int vnet_rx_one(struct vnet_port *port, struct vio_net_desc *desc)
4c521e42
DM		 321{
322 struct net_device *dev = port->vp->dev;
6d0ba919 323 unsigned int len = desc->size;
4c521e42
DM		 324 unsigned int copy_len;
325 struct sk_buff *skb;
326 int err;
327
328 err = -EMSGSIZE;
e4defc77 329 if (unlikely(len < ETH_ZLEN || len > port->rmtu)) {
4c521e42
DM		 330 dev->stats.rx_length_errors++;
331 goto out_dropped;
332 }
333
334 skb = alloc_and_align_skb(dev, len);
335 err = -ENOMEM;
336 if (unlikely(!skb)) {
337 dev->stats.rx_missed_errors++;
338 goto out_dropped;
339 }
340
341 copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
342 skb_put(skb, copy_len);
343 err = ldc_copy(port->vio.lp, LDC_COPY_IN,
344 skb->data, copy_len, 0,
6d0ba919 345 desc->cookies, desc->ncookies);
4c521e42
DM		 346 if (unlikely(err < 0)) {
347 dev->stats.rx_frame_errors++;
348 goto out_free_skb;
349 }
350
351 skb_pull(skb, VNET_PACKET_SKIP);
352 skb_trim(skb, len);
353 skb->protocol = eth_type_trans(skb, dev);
354
6d0ba919
DS		 355 if (vio_version_after_eq(&port->vio, 1, 8)) {
356 struct vio_net_dext *dext = vio_net_ext(desc);
357
358 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM) {
359 if (skb->protocol == ETH_P_IP) {
360 struct iphdr *iph = (struct iphdr *)skb->data;
361
362 iph->check = 0;
363 ip_send_check(iph);
364 }
365 }
366 if ((dext->flags & VNET_PKT_HCK_FULLCKSUM) &&
367 skb->ip_summed == CHECKSUM_NONE)
368 vnet_fullcsum(skb);
369 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM_OK) {
370 skb->ip_summed = CHECKSUM_PARTIAL;
371 skb->csum_level = 0;
372 if (dext->flags & VNET_PKT_HCK_FULLCKSUM_OK)
373 skb->csum_level = 1;
374 }
375 }
376
4c521e42
DM		 377 dev->stats.rx_packets++;
378 dev->stats.rx_bytes += len;
69088822 379 napi_gro_receive(&port->napi, skb);
4c521e42
DM		 380 return 0;
381
382out_free_skb:
383 kfree_skb(skb);
384
385out_dropped:
386 dev->stats.rx_dropped++;
387 return err;
388}
389
390static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
391 u32 start, u32 end, u8 vio_dring_state)
392{
393 struct vio_dring_data hdr = {
394 .tag = {
395 .type = VIO_TYPE_DATA,
396 .stype = VIO_SUBTYPE_ACK,
397 .stype_env = VIO_DRING_DATA,
398 .sid = vio_send_sid(&port->vio),
399 },
400 .dring_ident = dr->ident,
401 .start_idx = start,
402 .end_idx = end,
403 .state = vio_dring_state,
404 };
405 int err, delay;
adddc32d 406 int retries = 0;
4c521e42
DM		 407
408 hdr.seq = dr->snd_nxt;
409 delay = 1;
410 do {
411 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
412 if (err > 0) {
413 dr->snd_nxt++;
414 break;
415 }
416 udelay(delay);
417 if ((delay <<= 1) > 128)
418 delay = 128;
adddc32d
SV		 419 if (retries++ > VNET_MAX_RETRIES) {
420 pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
421 port->raddr[0], port->raddr[1],
422 port->raddr[2], port->raddr[3],
423 port->raddr[4], port->raddr[5]);
d1015645 424 break;
adddc32d 425 }
4c521e42
DM		 426 } while (err == -EAGAIN);
427
d1015645
SV		 428 if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
429 port->stop_rx_idx = end;
430 port->stop_rx = true;
431 } else {
432 port->stop_rx_idx = 0;
433 port->stop_rx = false;
434 }
435
4c521e42
DM		 436 return err;
437}
438
439static u32 next_idx(u32 idx, struct vio_dring_state *dr)
440{
441 if (++idx == dr->num_entries)
442 idx = 0;
443 return idx;
444}
445
446static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
447{
448 if (idx == 0)
449 idx = dr->num_entries - 1;
450 else
451 idx--;
452
453 return idx;
454}
455
456static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
457 struct vio_dring_state *dr,
458 u32 index)
459{
460 struct vio_net_desc *desc = port->vio.desc_buf;
461 int err;
462
463 err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
464 (index * dr->entry_size),
465 dr->cookies, dr->ncookies);
466 if (err < 0)
467 return ERR_PTR(err);
468
469 return desc;
470}
471
472static int put_rx_desc(struct vnet_port *port,
473 struct vio_dring_state *dr,
474 struct vio_net_desc *desc,
475 u32 index)
476{
477 int err;
478
479 err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
480 (index * dr->entry_size),
481 dr->cookies, dr->ncookies);
482 if (err < 0)
483 return err;
484
485 return 0;
486}
487
488static int vnet_walk_rx_one(struct vnet_port *port,
489 struct vio_dring_state *dr,
490 u32 index, int *needs_ack)
491{
492 struct vio_net_desc *desc = get_rx_desc(port, dr, index);
493 struct vio_driver_state *vio = &port->vio;
494 int err;
495
69088822 496 BUG_ON(desc == NULL);
4c521e42
DM		 497 if (IS_ERR(desc))
498 return PTR_ERR(desc);
499
78dcff7b
DS		 500 if (desc->hdr.state != VIO_DESC_READY)
501 return 1;
502
503 rmb();
504
3f4528d6 505 viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
4c521e42
DM		 506 desc->hdr.state, desc->hdr.ack,
507 desc->size, desc->ncookies,
508 desc->cookies[0].cookie_addr,
509 desc->cookies[0].cookie_size);
510
6d0ba919 511 err = vnet_rx_one(port, desc);
4c521e42
DM		 512 if (err == -ECONNRESET)
513 return err;
514 desc->hdr.state = VIO_DESC_DONE;
515 err = put_rx_desc(port, dr, desc, index);
516 if (err < 0)
517 return err;
518 *needs_ack = desc->hdr.ack;
519 return 0;
520}
521
522static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
69088822 523 u32 start, u32 end, int *npkts, int budget)
4c521e42
DM		 524{
525 struct vio_driver_state *vio = &port->vio;
526 int ack_start = -1, ack_end = -1;
69088822 527 bool send_ack = true;
4c521e42
DM		 528
529 end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
530
531 viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
532
533 while (start != end) {
534 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
535 if (err == -ECONNRESET)
536 return err;
537 if (err != 0)
538 break;
69088822 539 (*npkts)++;
4c521e42
DM		 540 if (ack_start == -1)
541 ack_start = start;
542 ack_end = start;
543 start = next_idx(start, dr);
544 if (ack && start != end) {
545 err = vnet_send_ack(port, dr, ack_start, ack_end,
546 VIO_DRING_ACTIVE);
547 if (err == -ECONNRESET)
548 return err;
549 ack_start = -1;
550 }
69088822
SV		 551 if ((*npkts) >= budget) {
552 send_ack = false;
553 break;
554 }
4c521e42
DM		 555 }
556 if (unlikely(ack_start == -1))
557 ack_start = ack_end = prev_idx(start, dr);
69088822
SV		 558 if (send_ack) {
559 port->napi_resume = false;
560 return vnet_send_ack(port, dr, ack_start, ack_end,
561 VIO_DRING_STOPPED);
562 } else {
563 port->napi_resume = true;
564 port->napi_stop_idx = ack_end;
565 return 1;
566 }
4c521e42
DM		 567}
568
69088822
SV		 569static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
570 int budget)
4c521e42
DM		 571{
572 struct vio_dring_data *pkt = msgbuf;
573 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
574 struct vio_driver_state *vio = &port->vio;
575
3f4528d6 576 viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
4c521e42
DM		 577 pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
578
579 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
580 return 0;
581 if (unlikely(pkt->seq != dr->rcv_nxt)) {
4d5870ec
JP		 582 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
583 pkt->seq, dr->rcv_nxt);
4c521e42
DM		 584 return 0;
585 }
586
69088822
SV		 587 if (!port->napi_resume)
588 dr->rcv_nxt++;
4c521e42
DM		 589
590 /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
591
69088822
SV		 592 return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
593 npkts, budget);
4c521e42
DM		 594}
595
596static int idx_is_pending(struct vio_dring_state *dr, u32 end)
597{
598 u32 idx = dr->cons;
599 int found = 0;
600
601 while (idx != dr->prod) {
602 if (idx == end) {
603 found = 1;
604 break;
605 }
606 idx = next_idx(idx, dr);
607 }
608 return found;
609}
610
611static int vnet_ack(struct vnet_port *port, void *msgbuf)
612{
613 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
614 struct vio_dring_data *pkt = msgbuf;
615 struct net_device *dev;
616 struct vnet *vp;
617 u32 end;
d1015645 618 struct vio_net_desc *desc;
d51bffd1
SV		 619 struct netdev_queue *txq;
620
4c521e42
DM		 621 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
622 return 0;
623
624 end = pkt->end_idx;
69088822
SV		 625 vp = port->vp;
626 dev = vp->dev;
b0cffed5
SV		 627 netif_tx_lock(dev);
628 if (unlikely(!idx_is_pending(dr, end))) {
629 netif_tx_unlock(dev);
630 return 0;
631 }
632
d1015645
SV		 633 /* sync for race conditions with vnet_start_xmit() and tell xmit it
634 * is time to send a trigger.
635 */
4c521e42 636 dr->cons = next_idx(end, dr);
d1015645 637 desc = vio_dring_entry(dr, dr->cons);
777362d7 638 if (desc->hdr.state == VIO_DESC_READY && !port->start_cons) {
d1015645
SV		 639 /* vnet_start_xmit() just populated this dring but missed
640 * sending the "start" LDC message to the consumer.
641 * Send a "start" trigger on its behalf.
642 */
643 if (__vnet_tx_trigger(port, dr->cons) > 0)
644 port->start_cons = false;
645 else
646 port->start_cons = true;
647 } else {
648 port->start_cons = true;
649 }
69088822 650 netif_tx_unlock(dev);
d1015645 651
d51bffd1
SV		 652 txq = netdev_get_tx_queue(dev, port->q_index);
653 if (unlikely(netif_tx_queue_stopped(txq) &&
4c521e42
DM		 654 vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
655 return 1;
656
657 return 0;
658}
659
660static int vnet_nack(struct vnet_port *port, void *msgbuf)
661{
662 /* XXX just reset or similar XXX */
663 return 0;
664}
665
028ebff2
DM		 666static int handle_mcast(struct vnet_port *port, void *msgbuf)
667{
668 struct vio_net_mcast_info *pkt = msgbuf;
669
670 if (pkt->tag.stype != VIO_SUBTYPE_ACK)
4d5870ec 671 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
028ebff2
DM		 672 port->vp->dev->name,
673 pkt->tag.type,
674 pkt->tag.stype,
675 pkt->tag.stype_env,
676 pkt->tag.sid);
677
678 return 0;
679}
680
d51bffd1
SV		 681/* Got back a STOPPED LDC message on port. If the queue is stopped,
682 * wake it up so that we'll send out another START message at the
683 * next TX.
684 */
685static void maybe_tx_wakeup(struct vnet_port *port)
4c521e42 686{
d51bffd1 687 struct netdev_queue *txq;
4c521e42 688
d51bffd1
SV		 689 txq = netdev_get_tx_queue(port->vp->dev, port->q_index);
690 __netif_tx_lock(txq, smp_processor_id());
691 if (likely(netif_tx_queue_stopped(txq))) {
692 struct vio_dring_state *dr;
693
694 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
6c3ce8a3 695 netif_tx_wake_queue(txq);
4c521e42 696 }
d51bffd1 697 __netif_tx_unlock(txq);
4c521e42
DM		 698}
699
69088822
SV		 700static inline bool port_is_up(struct vnet_port *vnet)
701{
702 struct vio_driver_state *vio = &vnet->vio;
703
704 return !!(vio->hs_state & VIO_HS_COMPLETE);
705}
706
707static int vnet_event_napi(struct vnet_port *port, int budget)
4c521e42 708{
4c521e42 709 struct vio_driver_state *vio = &port->vio;
4c521e42 710 int tx_wakeup, err;
69088822
SV		 711 int npkts = 0;
712 int event = (port->rx_event & LDC_EVENT_RESET);
4c521e42 713
69088822 714ldc_ctrl:
4c521e42
DM		 715 if (unlikely(event == LDC_EVENT_RESET ||
716 event == LDC_EVENT_UP)) {
717 vio_link_state_change(vio, event);
4c521e42 718
e4defc77
DS		 719 if (event == LDC_EVENT_RESET) {
720 port->rmtu = 0;
d762acdb 721 vio_port_up(vio);
e4defc77 722 }
69088822
SV		 723 port->rx_event = 0;
724 return 0;
4c521e42 725 }
69088822
SV		 726 /* We may have multiple LDC events in rx_event. Unroll send_events() */
727 event = (port->rx_event & LDC_EVENT_UP);
728 port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
729 if (event == LDC_EVENT_UP)
730 goto ldc_ctrl;
731 event = port->rx_event;
732 if (!(event & LDC_EVENT_DATA_READY))
733 return 0;
4c521e42 734
69088822
SV		 735 /* we dont expect any other bits than RESET, UP, DATA_READY */
736 BUG_ON(event != LDC_EVENT_DATA_READY);
4c521e42
DM		 737
738 tx_wakeup = err = 0;
739 while (1) {
740 union {
741 struct vio_msg_tag tag;
742 u64 raw[8];
743 } msgbuf;
744
69088822
SV		 745 if (port->napi_resume) {
746 struct vio_dring_data *pkt =
747 (struct vio_dring_data *)&msgbuf;
748 struct vio_dring_state *dr =
749 &port->vio.drings[VIO_DRIVER_RX_RING];
750
751 pkt->tag.type = VIO_TYPE_DATA;
752 pkt->tag.stype = VIO_SUBTYPE_INFO;
753 pkt->tag.stype_env = VIO_DRING_DATA;
754 pkt->seq = dr->rcv_nxt;
755 pkt->start_idx = next_idx(port->napi_stop_idx, dr);
756 pkt->end_idx = -1;
757 goto napi_resume;
758 }
4c521e42
DM		 759 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
760 if (unlikely(err < 0)) {
761 if (err == -ECONNRESET)
762 vio_conn_reset(vio);
763 break;
764 }
765 if (err == 0)
766 break;
767 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
768 msgbuf.tag.type,
769 msgbuf.tag.stype,
770 msgbuf.tag.stype_env,
771 msgbuf.tag.sid);
772 err = vio_validate_sid(vio, &msgbuf.tag);
773 if (err < 0)
774 break;
69088822 775napi_resume:
4c521e42
DM		 776 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
777 if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
69088822
SV		 778 if (!port_is_up(port)) {
779 /* failures like handshake_failure()
780 * may have cleaned up dring, but
781 * NAPI polling may bring us here.
782 */
783 err = -ECONNRESET;
784 break;
785 }
786 err = vnet_rx(port, &msgbuf, &npkts, budget);
787 if (npkts >= budget)
788 break;
8c4ee3e7
SV		 789 if (npkts == 0)
790 break;
4c521e42
DM		 791 } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
792 err = vnet_ack(port, &msgbuf);
793 if (err > 0)
794 tx_wakeup |= err;
795 } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
796 err = vnet_nack(port, &msgbuf);
797 }
798 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
028ebff2
DM		 799 if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
800 err = handle_mcast(port, &msgbuf);
801 else
802 err = vio_control_pkt_engine(vio, &msgbuf);
4c521e42
DM		 803 if (err)
804 break;
805 } else {
806 err = vnet_handle_unknown(port, &msgbuf);
807 }
808 if (err == -ECONNRESET)
809 break;
810 }
4c521e42 811 if (unlikely(tx_wakeup && err != -ECONNRESET))
d51bffd1 812 maybe_tx_wakeup(port);
69088822
SV		 813 return npkts;
814}
815
816static int vnet_poll(struct napi_struct *napi, int budget)
817{
818 struct vnet_port *port = container_of(napi, struct vnet_port, napi);
819 struct vio_driver_state *vio = &port->vio;
820 int processed = vnet_event_napi(port, budget);
821
822 if (processed < budget) {
823 napi_complete(napi);
7bd68bfd 824 port->rx_event &= ~LDC_EVENT_DATA_READY;
69088822
SV		 825 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
826 }
827 return processed;
828}
829
830static void vnet_event(void *arg, int event)
831{
832 struct vnet_port *port = arg;
833 struct vio_driver_state *vio = &port->vio;
834
835 port->rx_event |= event;
836 vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
837 napi_schedule(&port->napi);
1d311ad2 838
4c521e42
DM		 839}
840
d1015645 841static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
4c521e42
DM		 842{
843 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
844 struct vio_dring_data hdr = {
845 .tag = {
846 .type = VIO_TYPE_DATA,
847 .stype = VIO_SUBTYPE_INFO,
848 .stype_env = VIO_DRING_DATA,
849 .sid = vio_send_sid(&port->vio),
850 },
851 .dring_ident = dr->ident,
d1015645 852 .start_idx = start,
4c521e42
DM		 853 .end_idx = (u32) -1,
854 };
855 int err, delay;
adddc32d 856 int retries = 0;
4c521e42 857
d1015645
SV		 858 if (port->stop_rx) {
859 err = vnet_send_ack(port,
860 &port->vio.drings[VIO_DRIVER_RX_RING],
861 port->stop_rx_idx, -1,
862 VIO_DRING_STOPPED);
863 if (err <= 0)
864 return err;
865 }
866
4c521e42
DM		 867 hdr.seq = dr->snd_nxt;
868 delay = 1;
869 do {
870 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
871 if (err > 0) {
872 dr->snd_nxt++;
873 break;
874 }
875 udelay(delay);
876 if ((delay <<= 1) > 128)
877 delay = 128;
adddc32d
SV		 878 if (retries++ > VNET_MAX_RETRIES)
879 break;
4c521e42
DM		 880 } while (err == -EAGAIN);
881
882 return err;
883}
884
885struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
886{
887 unsigned int hash = vnet_hashfn(skb->data);
888 struct hlist_head *hp = &vp->port_hash[hash];
4c521e42
DM		 889 struct vnet_port *port;
890
2a968dd8 891 hlist_for_each_entry_rcu(port, hp, hash) {
8266f5fc
DS		 892 if (!port_is_up(port))
893 continue;
2e42e474 894 if (ether_addr_equal(port->raddr, skb->data))
4c521e42
DM		 895 return port;
896 }
2a968dd8 897 list_for_each_entry_rcu(port, &vp->port_list, list) {
8266f5fc
DS		 898 if (!port->switch_port)
899 continue;
900 if (!port_is_up(port))
901 continue;
902 return port;
903 }
904 return NULL;
4c521e42
DM		 905}
906
8e845f4c
DS		 907static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
908 unsigned *pending)
909{
910 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
911 struct sk_buff *skb = NULL;
912 int i, txi;
913
914 *pending = 0;
915
916 txi = dr->prod-1;
917 if (txi < 0)
918 txi = VNET_TX_RING_SIZE-1;
919
920 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
921 struct vio_net_desc *d;
922
923 d = vio_dring_entry(dr, txi);
924
925 if (d->hdr.state == VIO_DESC_DONE) {
926 if (port->tx_bufs[txi].skb) {
927 BUG_ON(port->tx_bufs[txi].skb->next);
928
929 port->tx_bufs[txi].skb->next = skb;
930 skb = port->tx_bufs[txi].skb;
931 port->tx_bufs[txi].skb = NULL;
932
933 ldc_unmap(port->vio.lp,
934 port->tx_bufs[txi].cookies,
935 port->tx_bufs[txi].ncookies);
936 }
937 d->hdr.state = VIO_DESC_FREE;
938 } else if (d->hdr.state == VIO_DESC_READY) {
939 (*pending)++;
940 } else if (d->hdr.state == VIO_DESC_FREE) {
941 break;
942 }
943 --txi;
944 if (txi < 0)
945 txi = VNET_TX_RING_SIZE-1;
946 }
947 return skb;
948}
949
950static inline void vnet_free_skbs(struct sk_buff *skb)
951{
952 struct sk_buff *next;
953
954 while (skb) {
955 next = skb->next;
956 skb->next = NULL;
957 dev_kfree_skb(skb);
958 skb = next;
959 }
960}
961
962static void vnet_clean_timer_expire(unsigned long port0)
963{
964 struct vnet_port *port = (struct vnet_port *)port0;
965 struct sk_buff *freeskbs;
966 unsigned pending;
8e845f4c 967
13b13dd9 968 netif_tx_lock(port->vp->dev);
8e845f4c 969 freeskbs = vnet_clean_tx_ring(port, &pending);
13b13dd9 970 netif_tx_unlock(port->vp->dev);
8e845f4c
DS		 971
972 vnet_free_skbs(freeskbs);
973
974 if (pending)
975 (void)mod_timer(&port->clean_timer,
976 jiffies + VNET_CLEAN_TIMEOUT);
977 else
978 del_timer(&port->clean_timer);
979}
980
981static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, void **pstart,
982 int *plen)
983{
984 struct sk_buff *nskb;
985 int len, pad;
986
987 len = skb->len;
988 pad = 0;
989 if (len < ETH_ZLEN) {
990 pad += ETH_ZLEN - skb->len;
991 len += pad;
992 }
993 len += VNET_PACKET_SKIP;
994 pad += 8 - (len & 7);
995 len += 8 - (len & 7);
996
997 if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
998 skb_tailroom(skb) < pad ||
999 skb_headroom(skb) < VNET_PACKET_SKIP) {
6d0ba919
DS		 1000 int offset;
1001
8e845f4c
DS		 1002 nskb = alloc_and_align_skb(skb->dev, skb->len);
1003 skb_reserve(nskb, VNET_PACKET_SKIP);
6d0ba919
DS		 1004
1005 nskb->protocol = skb->protocol;
1006 offset = skb_mac_header(skb) - skb->data;
1007 skb_set_mac_header(nskb, offset);
1008 offset = skb_network_header(skb) - skb->data;
1009 skb_set_network_header(nskb, offset);
1010 offset = skb_transport_header(skb) - skb->data;
1011 skb_set_transport_header(nskb, offset);
1012
1013 nskb->csum_offset = skb->csum_offset;
1014 nskb->ip_summed = skb->ip_summed;
1015
8e845f4c
DS		 1016 if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1017 dev_kfree_skb(nskb);
1018 dev_kfree_skb(skb);
1019 return NULL;
1020 }
1021 (void)skb_put(nskb, skb->len);
1022 dev_kfree_skb(skb);
1023 skb = nskb;
1024 }
1025
1026 *pstart = skb->data - VNET_PACKET_SKIP;
1027 *plen = len;
1028 return skb;
1029}
1030
d51bffd1
SV		 1031static u16
1032vnet_select_queue(struct net_device *dev, struct sk_buff *skb,
1033 void *accel_priv, select_queue_fallback_t fallback)
1034{
1035 struct vnet *vp = netdev_priv(dev);
1036 struct vnet_port *port = __tx_port_find(vp, skb);
1037
c647cc3f
DS		 1038 if (port == NULL)
1039 return 0;
d51bffd1
SV		 1040 return port->q_index;
1041}
1042
4c521e42
DM		 1043static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
1044{
1045 struct vnet *vp = netdev_priv(dev);
2a968dd8 1046 struct vnet_port *port = NULL;
4c521e42
DM		 1047 struct vio_dring_state *dr;
1048 struct vio_net_desc *d;
4c521e42 1049 unsigned int len;
8e845f4c
DS		 1050 struct sk_buff *freeskbs = NULL;
1051 int i, err, txi;
1052 void *start = NULL;
1053 int nlen = 0;
1054 unsigned pending = 0;
d51bffd1 1055 struct netdev_queue *txq;
4c521e42 1056
8e845f4c 1057 skb = vnet_skb_shape(skb, &start, &nlen);
8e845f4c
DS		 1058 if (unlikely(!skb))
1059 goto out_dropped;
1060
2a968dd8 1061 rcu_read_lock();
13b13dd9 1062 port = __tx_port_find(vp, skb);
df20286a
SV		 1063 if (unlikely(!port)) {
1064 rcu_read_unlock();
2a968dd8 1065 goto out_dropped;
df20286a 1066 }
2a968dd8 1067
a2b78e9b
DS		 1068 if (skb->len > port->rmtu) {
1069 unsigned long localmtu = port->rmtu - ETH_HLEN;
1070
1071 if (vio_version_after_eq(&port->vio, 1, 3))
1072 localmtu -= VLAN_HLEN;
1073
1074 if (skb->protocol == htons(ETH_P_IP)) {
1075 struct flowi4 fl4;
1076 struct rtable *rt = NULL;
1077
1078 memset(&fl4, 0, sizeof(fl4));
1079 fl4.flowi4_oif = dev->ifindex;
1080 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
1081 fl4.daddr = ip_hdr(skb)->daddr;
1082 fl4.saddr = ip_hdr(skb)->saddr;
1083
1084 rt = ip_route_output_key(dev_net(dev), &fl4);
2a968dd8 1085 rcu_read_unlock();
a2b78e9b
DS		 1086 if (!IS_ERR(rt)) {
1087 skb_dst_set(skb, &rt->dst);
1088 icmp_send(skb, ICMP_DEST_UNREACH,
1089 ICMP_FRAG_NEEDED,
1090 htonl(localmtu));
1091 }
1092 }
1093#if IS_ENABLED(CONFIG_IPV6)
1094 else if (skb->protocol == htons(ETH_P_IPV6))
1095 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1096#endif
42db672d 1097 goto out_dropped;
a2b78e9b 1098 }
42db672d 1099
4c521e42 1100 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
d51bffd1
SV		 1101 i = skb_get_queue_mapping(skb);
1102 txq = netdev_get_tx_queue(dev, i);
d0aedcd4 1103 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
d51bffd1
SV		 1104 if (!netif_tx_queue_stopped(txq)) {
1105 netif_tx_stop_queue(txq);
4c521e42
DM		 1106
1107 /* This is a hard error, log it. */
4d5870ec 1108 netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
4c521e42
DM		 1109 dev->stats.tx_errors++;
1110 }
2a968dd8 1111 rcu_read_unlock();
4c521e42
DM		 1112 return NETDEV_TX_BUSY;
1113 }
1114
1115 d = vio_dring_cur(dr);
1116
8e845f4c
DS		 1117 txi = dr->prod;
1118
1119 freeskbs = vnet_clean_tx_ring(port, &pending);
1120
1121 BUG_ON(port->tx_bufs[txi].skb);
4c521e42
DM		 1122
1123 len = skb->len;
8e845f4c 1124 if (len < ETH_ZLEN)
4c521e42 1125 len = ETH_ZLEN;
8e845f4c
DS		 1126
1127 port->tx_bufs[txi].skb = skb;
1128 skb = NULL;
1129
1130 err = ldc_map_single(port->vio.lp, start, nlen,
42db672d 1131 port->tx_bufs[txi].cookies, VNET_MAXCOOKIES,
8e845f4c
DS		 1132 (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1133 if (err < 0) {
1134 netdev_info(dev, "tx buffer map error %d\n", err);
13b13dd9 1135 goto out_dropped;
4c521e42 1136 }
8e845f4c 1137 port->tx_bufs[txi].ncookies = err;
4c521e42 1138
1f6394e3
SV		 1139 /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1140 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1141 * the protocol itself does not require it as long as the peer
1142 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1143 *
1144 * An ACK for every packet in the ring is expensive as the
1145 * sending of LDC messages is slow and affects performance.
1146 */
1147 d->hdr.ack = VIO_ACK_DISABLE;
4c521e42 1148 d->size = len;
8e845f4c 1149 d->ncookies = port->tx_bufs[txi].ncookies;
4c521e42 1150 for (i = 0; i < d->ncookies; i++)
8e845f4c 1151 d->cookies[i] = port->tx_bufs[txi].cookies[i];
6d0ba919
DS		 1152 if (vio_version_after_eq(&port->vio, 1, 7)) {
1153 struct vio_net_dext *dext = vio_net_ext(d);
1154
1155 memset(dext, 0, sizeof(*dext));
1156 if (vio_version_after_eq(&port->vio, 1, 8) &&
1157 !port->switch_port) {
1158 dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1159 dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1160 }
1161 }
4c521e42
DM		 1162
1163 /* This has to be a non-SMP write barrier because we are writing
1164 * to memory which is shared with the peer LDOM.
1165 */
1166 wmb();
1167
1168 d->hdr.state = VIO_DESC_READY;
1169
d1015645
SV		 1170 /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1171 * to notify the consumer that some descriptors are READY.
1172 * After that "start" trigger, no additional triggers are needed until
1173 * a DRING_STOPPED is received from the consumer. The dr->cons field
1174 * (set up by vnet_ack()) has the value of the next dring index
1175 * that has not yet been ack-ed. We send a "start" trigger here
1176 * if, and only if, start_cons is true (reset it afterward). Conversely,
1177 * vnet_ack() should check if the dring corresponding to cons
1178 * is marked READY, but start_cons was false.
1179 * If so, vnet_ack() should send out the missed "start" trigger.
1180 *
1181 * Note that the wmb() above makes sure the cookies et al. are
1182 * not globally visible before the VIO_DESC_READY, and that the
1183 * stores are ordered correctly by the compiler. The consumer will
1184 * not proceed until the VIO_DESC_READY is visible assuring that
1185 * the consumer does not observe anything related to descriptors
1186 * out of order. The HV trap from the LDC start trigger is the
1187 * producer to consumer announcement that work is available to the
1188 * consumer
1189 */
1190 if (!port->start_cons)
1191 goto ldc_start_done; /* previous trigger suffices */
1192
1193 err = __vnet_tx_trigger(port, dr->cons);
4c521e42 1194 if (unlikely(err < 0)) {
4d5870ec 1195 netdev_info(dev, "TX trigger error %d\n", err);
4c521e42
DM		 1196 d->hdr.state = VIO_DESC_FREE;
1197 dev->stats.tx_carrier_errors++;
13b13dd9 1198 goto out_dropped;
4c521e42
DM		 1199 }
1200
d1015645
SV		 1201ldc_start_done:
1202 port->start_cons = false;
1203
4c521e42 1204 dev->stats.tx_packets++;
8e845f4c 1205 dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
4c521e42
DM		 1206
1207 dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
d0aedcd4 1208 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
d51bffd1 1209 netif_tx_stop_queue(txq);
4c521e42 1210 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
d51bffd1 1211 netif_tx_wake_queue(txq);
4c521e42
DM		 1212 }
1213
2a968dd8
SV		 1214 (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1215 rcu_read_unlock();
4c521e42 1216
8e845f4c
DS		 1217 vnet_free_skbs(freeskbs);
1218
4c521e42
DM		 1219 return NETDEV_TX_OK;
1220
4c521e42 1221out_dropped:
8e845f4c
DS		 1222 if (pending)
1223 (void)mod_timer(&port->clean_timer,
1224 jiffies + VNET_CLEAN_TIMEOUT);
a29c9c43 1225 else if (port)
8e845f4c 1226 del_timer(&port->clean_timer);
2a968dd8
SV		 1227 if (port)
1228 rcu_read_unlock();
1229 if (skb)
1230 dev_kfree_skb(skb);
1231 vnet_free_skbs(freeskbs);
4c521e42
DM		 1232 dev->stats.tx_dropped++;
1233 return NETDEV_TX_OK;
1234}
1235
1236static void vnet_tx_timeout(struct net_device *dev)
1237{
1238 /* XXX Implement me XXX */
1239}
1240
1241static int vnet_open(struct net_device *dev)
1242{
1243 netif_carrier_on(dev);
d51bffd1 1244 netif_tx_start_all_queues(dev);
4c521e42
DM		 1245
1246 return 0;
1247}
1248
1249static int vnet_close(struct net_device *dev)
1250{
d51bffd1 1251 netif_tx_stop_all_queues(dev);
4c521e42
DM		 1252 netif_carrier_off(dev);
1253
1254 return 0;
1255}
1256
028ebff2
DM		 1257static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1258{
1259 struct vnet_mcast_entry *m;
1260
1261 for (m = vp->mcast_list; m; m = m->next) {
00fa4ce9 1262 if (ether_addr_equal(m->addr, addr))
028ebff2
DM		 1263 return m;
1264 }
1265 return NULL;
1266}
1267
1268static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1269{
22bedad3 1270 struct netdev_hw_addr *ha;
028ebff2 1271
22bedad3 1272 netdev_for_each_mc_addr(ha, dev) {
028ebff2
DM		 1273 struct vnet_mcast_entry *m;
1274
22bedad3 1275 m = __vnet_mc_find(vp, ha->addr);
028ebff2
DM		 1276 if (m) {
1277 m->hit = 1;
1278 continue;
1279 }
1280
1281 if (!m) {
1282 m = kzalloc(sizeof(*m), GFP_ATOMIC);
1283 if (!m)
1284 continue;
22bedad3 1285 memcpy(m->addr, ha->addr, ETH_ALEN);
028ebff2
DM		 1286 m->hit = 1;
1287
1288 m->next = vp->mcast_list;
1289 vp->mcast_list = m;
1290 }
1291 }
1292}
1293
1294static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1295{
1296 struct vio_net_mcast_info info;
1297 struct vnet_mcast_entry *m, **pp;
1298 int n_addrs;
1299
1300 memset(&info, 0, sizeof(info));
1301
1302 info.tag.type = VIO_TYPE_CTRL;
1303 info.tag.stype = VIO_SUBTYPE_INFO;
1304 info.tag.stype_env = VNET_MCAST_INFO;
1305 info.tag.sid = vio_send_sid(&port->vio);
1306 info.set = 1;
1307
1308 n_addrs = 0;
1309 for (m = vp->mcast_list; m; m = m->next) {
1310 if (m->sent)
1311 continue;
1312 m->sent = 1;
1313 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1314 m->addr, ETH_ALEN);
1315 if (++n_addrs == VNET_NUM_MCAST) {
1316 info.count = n_addrs;
1317
1318 (void) vio_ldc_send(&port->vio, &info,
1319 sizeof(info));
1320 n_addrs = 0;
1321 }
1322 }
1323 if (n_addrs) {
1324 info.count = n_addrs;
1325 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1326 }
1327
1328 info.set = 0;
1329
1330 n_addrs = 0;
1331 pp = &vp->mcast_list;
1332 while ((m = *pp) != NULL) {
1333 if (m->hit) {
1334 m->hit = 0;
1335 pp = &m->next;
1336 continue;
1337 }
1338
1339 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1340 m->addr, ETH_ALEN);
1341 if (++n_addrs == VNET_NUM_MCAST) {
1342 info.count = n_addrs;
1343 (void) vio_ldc_send(&port->vio, &info,
1344 sizeof(info));
1345 n_addrs = 0;
1346 }
1347
1348 *pp = m->next;
1349 kfree(m);
1350 }
1351 if (n_addrs) {
1352 info.count = n_addrs;
1353 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1354 }
1355}
1356
4c521e42
DM		 1357static void vnet_set_rx_mode(struct net_device *dev)
1358{
028ebff2
DM		 1359 struct vnet *vp = netdev_priv(dev);
1360 struct vnet_port *port;
028ebff2 1361
2a968dd8
SV		 1362 rcu_read_lock();
1363 list_for_each_entry_rcu(port, &vp->port_list, list) {
028ebff2
DM		 1364
1365 if (port->switch_port) {
1366 __update_mc_list(vp, dev);
1367 __send_mc_list(vp, port);
2a968dd8 1368 break;
028ebff2
DM		 1369 }
1370 }
2a968dd8 1371 rcu_read_unlock();
4c521e42
DM		 1372}
1373
1374static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1375{
42db672d 1376 if (new_mtu < 68 || new_mtu > 65535)
4c521e42
DM		 1377 return -EINVAL;
1378
1379 dev->mtu = new_mtu;
1380 return 0;
1381}
1382
1383static int vnet_set_mac_addr(struct net_device *dev, void *p)
1384{
1385 return -EINVAL;
1386}
1387
1388static void vnet_get_drvinfo(struct net_device *dev,
1389 struct ethtool_drvinfo *info)
1390{
7826d43f
JP		 1391 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1392 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4c521e42
DM		 1393}
1394
1395static u32 vnet_get_msglevel(struct net_device *dev)
1396{
1397 struct vnet *vp = netdev_priv(dev);
1398 return vp->msg_enable;
1399}
1400
1401static void vnet_set_msglevel(struct net_device *dev, u32 value)
1402{
1403 struct vnet *vp = netdev_priv(dev);
1404 vp->msg_enable = value;
1405}
1406
1407static const struct ethtool_ops vnet_ethtool_ops = {
1408 .get_drvinfo = vnet_get_drvinfo,
1409 .get_msglevel = vnet_get_msglevel,
1410 .set_msglevel = vnet_set_msglevel,
1411 .get_link = ethtool_op_get_link,
4c521e42
DM		 1412};
1413
1414static void vnet_port_free_tx_bufs(struct vnet_port *port)
1415{
1416 struct vio_dring_state *dr;
1417 int i;
1418
1419 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1420 if (dr->base) {
1421 ldc_free_exp_dring(port->vio.lp, dr->base,
1422 (dr->entry_size * dr->num_entries),
1423 dr->cookies, dr->ncookies);
1424 dr->base = NULL;
1425 dr->entry_size = 0;
1426 dr->num_entries = 0;
1427 dr->pending = 0;
1428 dr->ncookies = 0;
1429 }
1430
1431 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
8e845f4c
DS		 1432 struct vio_net_desc *d;
1433 void *skb = port->tx_bufs[i].skb;
4c521e42 1434
8e845f4c 1435 if (!skb)
4c521e42
DM		 1436 continue;
1437
8e845f4c
DS		 1438 d = vio_dring_entry(dr, i);
1439 if (d->hdr.state == VIO_DESC_READY)
1440 pr_warn("active transmit buffers freed\n");
1441
4c521e42
DM		 1442 ldc_unmap(port->vio.lp,
1443 port->tx_bufs[i].cookies,
1444 port->tx_bufs[i].ncookies);
8e845f4c
DS		 1445 dev_kfree_skb(skb);
1446 port->tx_bufs[i].skb = NULL;
1447 d->hdr.state = VIO_DESC_FREE;
4c521e42
DM		 1448 }
1449}
1450
d6732489 1451static int vnet_port_alloc_tx_ring(struct vnet_port *port)
4c521e42
DM		 1452{
1453 struct vio_dring_state *dr;
6d0ba919 1454 unsigned long len, elen;
4c521e42
DM		 1455 int i, err, ncookies;
1456 void *dring;
1457
4c521e42
DM		 1458 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1459
6d0ba919
DS		 1460 elen = sizeof(struct vio_net_desc) +
1461 sizeof(struct ldc_trans_cookie) * 2;
1462 if (vio_version_after_eq(&port->vio, 1, 7))
1463 elen += sizeof(struct vio_net_dext);
1464 len = VNET_TX_RING_SIZE * elen;
4c521e42
DM		 1465
1466 ncookies = VIO_MAX_RING_COOKIES;
1467 dring = ldc_alloc_exp_dring(port->vio.lp, len,
1468 dr->cookies, &ncookies,
1469 (LDC_MAP_SHADOW |
1470 LDC_MAP_DIRECT |
1471 LDC_MAP_RW));
1472 if (IS_ERR(dring)) {
1473 err = PTR_ERR(dring);
1474 goto err_out;
1475 }
1476
1477 dr->base = dring;
6d0ba919 1478 dr->entry_size = elen;
4c521e42
DM		 1479 dr->num_entries = VNET_TX_RING_SIZE;
1480 dr->prod = dr->cons = 0;
d1015645 1481 port->start_cons = true; /* need an initial trigger */
4c521e42
DM		 1482 dr->pending = VNET_TX_RING_SIZE;
1483 dr->ncookies = ncookies;
1484
8e845f4c
DS		 1485 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1486 struct vio_net_desc *d;
1487
1488 d = vio_dring_entry(dr, i);
1489 d->hdr.state = VIO_DESC_FREE;
1490 }
4c521e42
DM		 1491 return 0;
1492
1493err_out:
1494 vnet_port_free_tx_bufs(port);
1495
1496 return err;
1497}
1498
69088822
SV		 1499#ifdef CONFIG_NET_POLL_CONTROLLER
1500static void vnet_poll_controller(struct net_device *dev)
1501{
1502 struct vnet *vp = netdev_priv(dev);
1503 struct vnet_port *port;
1504 unsigned long flags;
1505
1506 spin_lock_irqsave(&vp->lock, flags);
1507 if (!list_empty(&vp->port_list)) {
1508 port = list_entry(vp->port_list.next, struct vnet_port, list);
1509 napi_schedule(&port->napi);
1510 }
1511 spin_unlock_irqrestore(&vp->lock, flags);
1512}
1513#endif
9184a046
DM		 1514static LIST_HEAD(vnet_list);
1515static DEFINE_MUTEX(vnet_list_mutex);
1516
8fd17f2e
DM		 1517static const struct net_device_ops vnet_ops = {
1518 .ndo_open = vnet_open,
1519 .ndo_stop = vnet_close,
afc4b13d 1520 .ndo_set_rx_mode = vnet_set_rx_mode,
8fd17f2e 1521 .ndo_set_mac_address = vnet_set_mac_addr,
240c102d 1522 .ndo_validate_addr = eth_validate_addr,
8fd17f2e
DM		 1523 .ndo_tx_timeout = vnet_tx_timeout,
1524 .ndo_change_mtu = vnet_change_mtu,
1525 .ndo_start_xmit = vnet_start_xmit,
d51bffd1 1526 .ndo_select_queue = vnet_select_queue,
69088822
SV		 1527#ifdef CONFIG_NET_POLL_CONTROLLER
1528 .ndo_poll_controller = vnet_poll_controller,
1529#endif
8fd17f2e
DM		 1530};
1531
f73d12bd 1532static struct vnet *vnet_new(const u64 *local_mac)
9184a046
DM		 1533{
1534 struct net_device *dev;
1535 struct vnet *vp;
1536 int err, i;
1537
d51bffd1 1538 dev = alloc_etherdev_mqs(sizeof(*vp), VNET_MAX_TXQS, 1);
41de8d4c 1539 if (!dev)
9184a046 1540 return ERR_PTR(-ENOMEM);
8e845f4c
DS		 1541 dev->needed_headroom = VNET_PACKET_SKIP + 8;
1542 dev->needed_tailroom = 8;
9184a046
DM		 1543
1544 for (i = 0; i < ETH_ALEN; i++)
1545 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1546
9184a046
DM		 1547 vp = netdev_priv(dev);
1548
1549 spin_lock_init(&vp->lock);
1550 vp->dev = dev;
1551
1552 INIT_LIST_HEAD(&vp->port_list);
1553 for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1554 INIT_HLIST_HEAD(&vp->port_hash[i]);
1555 INIT_LIST_HEAD(&vp->list);
1556 vp->local_mac = *local_mac;
1557
8fd17f2e 1558 dev->netdev_ops = &vnet_ops;
9184a046
DM		 1559 dev->ethtool_ops = &vnet_ethtool_ops;
1560 dev->watchdog_timeo = VNET_TX_TIMEOUT;
9184a046
DM		 1561
1562 err = register_netdev(dev);
1563 if (err) {
4d5870ec 1564 pr_err("Cannot register net device, aborting\n");
9184a046
DM		 1565 goto err_out_free_dev;
1566 }
1567
4d5870ec 1568 netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
9184a046
DM		 1569
1570 list_add(&vp->list, &vnet_list);
1571
1572 return vp;
1573
1574err_out_free_dev:
1575 free_netdev(dev);
1576
1577 return ERR_PTR(err);
1578}
1579
f73d12bd 1580static struct vnet *vnet_find_or_create(const u64 *local_mac)
9184a046
DM		 1581{
1582 struct vnet *iter, *vp;
1583
1584 mutex_lock(&vnet_list_mutex);
1585 vp = NULL;
1586 list_for_each_entry(iter, &vnet_list, list) {
1587 if (iter->local_mac == *local_mac) {
1588 vp = iter;
1589 break;
1590 }
1591 }
1592 if (!vp)
1593 vp = vnet_new(local_mac);
1594 mutex_unlock(&vnet_list_mutex);
1595
1596 return vp;
1597}
1598
a4b70a07
SV		 1599static void vnet_cleanup(void)
1600{
1601 struct vnet *vp;
1602 struct net_device *dev;
1603
1604 mutex_lock(&vnet_list_mutex);
1605 while (!list_empty(&vnet_list)) {
1606 vp = list_first_entry(&vnet_list, struct vnet, list);
1607 list_del(&vp->list);
1608 dev = vp->dev;
1609 /* vio_unregister_driver() should have cleaned up port_list */
1610 BUG_ON(!list_empty(&vp->port_list));
1611 unregister_netdev(dev);
1612 free_netdev(dev);
1613 }
1614 mutex_unlock(&vnet_list_mutex);
1615}
1616
9184a046
DM		 1617static const char *local_mac_prop = "local-mac-address";
1618
f73d12bd 1619static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
9184a046
DM		 1620 u64 port_node)
1621{
1622 const u64 *local_mac = NULL;
1623 u64 a;
1624
1625 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1626 u64 target = mdesc_arc_target(hp, a);
1627 const char *name;
1628
1629 name = mdesc_get_property(hp, target, "name", NULL);
1630 if (!name || strcmp(name, "network"))
1631 continue;
1632
1633 local_mac = mdesc_get_property(hp, target,
1634 local_mac_prop, NULL);
1635 if (local_mac)
1636 break;
1637 }
1638 if (!local_mac)
1639 return ERR_PTR(-ENODEV);
1640
1641 return vnet_find_or_create(local_mac);
1642}
1643
4c521e42
DM		 1644static struct ldc_channel_config vnet_ldc_cfg = {
1645 .event = vnet_event,
1646 .mtu = 64,
1647 .mode = LDC_MODE_UNRELIABLE,
1648};
1649
1650static struct vio_driver_ops vnet_vio_ops = {
1651 .send_attr = vnet_send_attr,
1652 .handle_attr = vnet_handle_attr,
1653 .handshake_complete = vnet_handshake_complete,
1654};
1655
f73d12bd 1656static void print_version(void)
9184a046 1657{
4d5870ec 1658 printk_once(KERN_INFO "%s", version);
9184a046
DM		 1659}
1660
4c521e42
DM		 1661const char *remote_macaddr_prop = "remote-mac-address";
1662
d51bffd1
SV		 1663static void
1664vnet_port_add_txq(struct vnet_port *port)
1665{
1666 struct vnet *vp = port->vp;
1667 int n;
1668
1669 n = vp->nports++;
1670 n = n & (VNET_MAX_TXQS - 1);
1671 port->q_index = n;
1672 netif_tx_wake_queue(netdev_get_tx_queue(vp->dev, port->q_index));
1673}
1674
1675static void
1676vnet_port_rm_txq(struct vnet_port *port)
1677{
1678 port->vp->nports--;
1679 netif_tx_stop_queue(netdev_get_tx_queue(port->vp->dev, port->q_index));
1680}
1681
1dd06ae8 1682static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
4c521e42 1683{
43fdf274 1684 struct mdesc_handle *hp;
4c521e42
DM		 1685 struct vnet_port *port;
1686 unsigned long flags;
1687 struct vnet *vp;
1688 const u64 *rmac;
1689 int len, i, err, switch_port;
1690
9184a046 1691 print_version();
4c521e42 1692
43fdf274
DM		 1693 hp = mdesc_grab();
1694
9184a046
DM		 1695 vp = vnet_find_parent(hp, vdev->mp);
1696 if (IS_ERR(vp)) {
4d5870ec 1697 pr_err("Cannot find port parent vnet\n");
9184a046
DM		 1698 err = PTR_ERR(vp);
1699 goto err_out_put_mdesc;
1700 }
1701
43fdf274
DM		 1702 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1703 err = -ENODEV;
4c521e42 1704 if (!rmac) {
4d5870ec 1705 pr_err("Port lacks %s property\n", remote_macaddr_prop);
43fdf274 1706 goto err_out_put_mdesc;
4c521e42
DM		 1707 }
1708
1709 port = kzalloc(sizeof(*port), GFP_KERNEL);
43fdf274 1710 err = -ENOMEM;
e404decb 1711 if (!port)
43fdf274 1712 goto err_out_put_mdesc;
4c521e42
DM		 1713
1714 for (i = 0; i < ETH_ALEN; i++)
1715 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1716
1717 port->vp = vp;
1718
43fdf274 1719 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
4c521e42
DM		 1720 vnet_versions, ARRAY_SIZE(vnet_versions),
1721 &vnet_vio_ops, vp->dev->name);
1722 if (err)
1723 goto err_out_free_port;
1724
1725 err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1726 if (err)
1727 goto err_out_free_port;
1728
69088822
SV		 1729 netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1730
4c521e42
DM		 1731 INIT_HLIST_NODE(&port->hash);
1732 INIT_LIST_HEAD(&port->list);
1733
1734 switch_port = 0;
43fdf274 1735 if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
4c521e42 1736 switch_port = 1;
028ebff2 1737 port->switch_port = switch_port;
4c521e42
DM		 1738
1739 spin_lock_irqsave(&vp->lock, flags);
1740 if (switch_port)
2a968dd8 1741 list_add_rcu(&port->list, &vp->port_list);
4c521e42 1742 else
2a968dd8
SV		 1743 list_add_tail_rcu(&port->list, &vp->port_list);
1744 hlist_add_head_rcu(&port->hash,
1745 &vp->port_hash[vnet_hashfn(port->raddr)]);
d51bffd1 1746 vnet_port_add_txq(port);
4c521e42
DM		 1747 spin_unlock_irqrestore(&vp->lock, flags);
1748
1749 dev_set_drvdata(&vdev->dev, port);
1750
4d5870ec
JP		 1751 pr_info("%s: PORT ( remote-mac %pM%s )\n",
1752 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
4c521e42 1753
8e845f4c
DS		 1754 setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1755 (unsigned long)port);
1756
69088822 1757 napi_enable(&port->napi);
4c521e42
DM		 1758 vio_port_up(&port->vio);
1759
43fdf274
DM		 1760 mdesc_release(hp);
1761
4c521e42
DM		 1762 return 0;
1763
4c521e42
DM		 1764err_out_free_port:
1765 kfree(port);
1766
43fdf274
DM		 1767err_out_put_mdesc:
1768 mdesc_release(hp);
4c521e42
DM		 1769 return err;
1770}
1771
1772static int vnet_port_remove(struct vio_dev *vdev)
1773{
1774 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1775
1776 if (port) {
4c521e42
DM		 1777
1778 del_timer_sync(&port->vio.timer);
1779
69088822 1780 napi_disable(&port->napi);
4c521e42 1781
2a968dd8
SV		 1782 list_del_rcu(&port->list);
1783 hlist_del_rcu(&port->hash);
1784
1785 synchronize_rcu();
1786 del_timer_sync(&port->clean_timer);
d51bffd1 1787 vnet_port_rm_txq(port);
69088822 1788 netif_napi_del(&port->napi);
4c521e42
DM		 1789 vnet_port_free_tx_bufs(port);
1790 vio_ldc_free(&port->vio);
1791
1792 dev_set_drvdata(&vdev->dev, NULL);
1793
1794 kfree(port);
aabb9875 1795
4c521e42
DM		 1796 }
1797 return 0;
1798}
1799
3d452e55 1800static const struct vio_device_id vnet_port_match[] = {
4c521e42
DM		 1801 {
1802 .type = "vnet-port",
1803 },
1804 {},
1805};
da68e081 1806MODULE_DEVICE_TABLE(vio, vnet_port_match);
4c521e42
DM		 1807
1808static struct vio_driver vnet_port_driver = {
1809 .id_table = vnet_port_match,
1810 .probe = vnet_port_probe,
1811 .remove = vnet_port_remove,
cb52d897 1812 .name = "vnet_port",
4c521e42
DM		 1813};
1814
4c521e42
DM		 1815static int __init vnet_init(void)
1816{
9184a046 1817 return vio_register_driver(&vnet_port_driver);
4c521e42
DM		 1818}
1819
1820static void __exit vnet_exit(void)
1821{
1822 vio_unregister_driver(&vnet_port_driver);
a4b70a07 1823 vnet_cleanup();
4c521e42
DM		 1824}
1825
1826module_init(vnet_init);
1827module_exit(vnet_exit);
Linux kernel - Deliverables
This page took 0.858691 seconds and 5 git commands to generate.