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