1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2017 - Cambridge Greys Limited
4 * Copyright (C) 2011 - 2014 Cisco Systems Inc
5 * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
6 * Copyright (C) 2001 Lennert Buytenhek (buytenh@gnu.org) and
7 * James Leu (jleu@mindspring.net).
8 * Copyright (C) 2001 by various other people who didn't put their name here.
9 */
10
11 #include <linux/version.h>
12 #include <linux/memblock.h>
13 #include <linux/etherdevice.h>
14 #include <linux/ethtool.h>
15 #include <linux/inetdevice.h>
16 #include <linux/init.h>
17 #include <linux/list.h>
18 #include <linux/netdevice.h>
19 #include <linux/platform_device.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/skbuff.h>
22 #include <linux/slab.h>
23 #include <linux/interrupt.h>
24 #include <init.h>
25 #include <irq_kern.h>
26 #include <irq_user.h>
27 #include <net_kern.h>
28 #include <os.h>
29 #include "mconsole_kern.h"
30 #include "vector_user.h"
31 #include "vector_kern.h"
32
33 /*
34 * Adapted from network devices with the following major changes:
35 * All transports are static - simplifies the code significantly
36 * Multiple FDs/IRQs per device
37 * Vector IO optionally used for read/write, falling back to legacy
38 * based on configuration and/or availability
39 * Configuration is no longer positional - L2TPv3 and GRE require up to
40 * 10 parameters, passing this as positional is not fit for purpose.
41 * Only socket transports are supported
42 */
43
44
45 #define DRIVER_NAME "uml-vector"
46 #define DRIVER_VERSION "01"
47 struct vector_cmd_line_arg {
48 struct list_head list;
49 int unit;
50 char *arguments;
51 };
52
53 struct vector_device {
54 struct list_head list;
55 struct net_device *dev;
56 struct platform_device pdev;
57 int unit;
58 int opened;
59 };
60
61 static LIST_HEAD(vec_cmd_line);
62
63 static DEFINE_SPINLOCK(vector_devices_lock);
64 static LIST_HEAD(vector_devices);
65
66 static int driver_registered;
67
68 static void vector_eth_configure(int n, struct arglist *def);
69
70 /* Argument accessors to set variables (and/or set default values)
71 * mtu, buffer sizing, default headroom, etc
72 */
73
74 #define DEFAULT_HEADROOM 2
75 #define SAFETY_MARGIN 32
76 #define DEFAULT_VECTOR_SIZE 64
77 #define TX_SMALL_PACKET 128
78 #define MAX_IOV_SIZE (MAX_SKB_FRAGS + 1)
79 #define MAX_ITERATIONS 64
80
81 static const struct {
82 const char string[ETH_GSTRING_LEN];
83 } ethtool_stats_keys[] = {
84 { "rx_queue_max" },
85 { "rx_queue_running_average" },
86 { "tx_queue_max" },
87 { "tx_queue_running_average" },
88 { "rx_encaps_errors" },
89 { "tx_timeout_count" },
90 { "tx_restart_queue" },
91 { "tx_kicks" },
92 { "tx_flow_control_xon" },
93 { "tx_flow_control_xoff" },
94 { "rx_csum_offload_good" },
95 { "rx_csum_offload_errors"},
96 { "sg_ok"},
97 { "sg_linearized"},
98 };
99
100 #define VECTOR_NUM_STATS ARRAY_SIZE(ethtool_stats_keys)
101
vector_reset_stats(struct vector_private * vp)102 static void vector_reset_stats(struct vector_private *vp)
103 {
104 vp->estats.rx_queue_max = 0;
105 vp->estats.rx_queue_running_average = 0;
106 vp->estats.tx_queue_max = 0;
107 vp->estats.tx_queue_running_average = 0;
108 vp->estats.rx_encaps_errors = 0;
109 vp->estats.tx_timeout_count = 0;
110 vp->estats.tx_restart_queue = 0;
111 vp->estats.tx_kicks = 0;
112 vp->estats.tx_flow_control_xon = 0;
113 vp->estats.tx_flow_control_xoff = 0;
114 vp->estats.sg_ok = 0;
115 vp->estats.sg_linearized = 0;
116 }
117
get_mtu(struct arglist * def)118 static int get_mtu(struct arglist *def)
119 {
120 char *mtu = uml_vector_fetch_arg(def, "mtu");
121 long result;
122
123 if (mtu != NULL) {
124 if (kstrtoul(mtu, 10, &result) == 0)
125 if ((result < (1 << 16) - 1) && (result >= 576))
126 return result;
127 }
128 return ETH_MAX_PACKET;
129 }
130
get_depth(struct arglist * def)131 static int get_depth(struct arglist *def)
132 {
133 char *mtu = uml_vector_fetch_arg(def, "depth");
134 long result;
135
136 if (mtu != NULL) {
137 if (kstrtoul(mtu, 10, &result) == 0)
138 return result;
139 }
140 return DEFAULT_VECTOR_SIZE;
141 }
142
get_headroom(struct arglist * def)143 static int get_headroom(struct arglist *def)
144 {
145 char *mtu = uml_vector_fetch_arg(def, "headroom");
146 long result;
147
148 if (mtu != NULL) {
149 if (kstrtoul(mtu, 10, &result) == 0)
150 return result;
151 }
152 return DEFAULT_HEADROOM;
153 }
154
get_req_size(struct arglist * def)155 static int get_req_size(struct arglist *def)
156 {
157 char *gro = uml_vector_fetch_arg(def, "gro");
158 long result;
159
160 if (gro != NULL) {
161 if (kstrtoul(gro, 10, &result) == 0) {
162 if (result > 0)
163 return 65536;
164 }
165 }
166 return get_mtu(def) + ETH_HEADER_OTHER +
167 get_headroom(def) + SAFETY_MARGIN;
168 }
169
170
get_transport_options(struct arglist * def)171 static int get_transport_options(struct arglist *def)
172 {
173 char *transport = uml_vector_fetch_arg(def, "transport");
174 char *vector = uml_vector_fetch_arg(def, "vec");
175
176 int vec_rx = VECTOR_RX;
177 int vec_tx = VECTOR_TX;
178 long parsed;
179
180 if (vector != NULL) {
181 if (kstrtoul(vector, 10, &parsed) == 0) {
182 if (parsed == 0) {
183 vec_rx = 0;
184 vec_tx = 0;
185 }
186 }
187 }
188
189
190 if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
191 return 0;
192 if (strncmp(transport, TRANS_HYBRID, TRANS_HYBRID_LEN) == 0)
193 return (vec_rx | VECTOR_BPF);
194 if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
195 return (vec_rx | vec_tx | VECTOR_QDISC_BYPASS);
196 return (vec_rx | vec_tx);
197 }
198
199
200 /* A mini-buffer for packet drop read
201 * All of our supported transports are datagram oriented and we always
202 * read using recvmsg or recvmmsg. If we pass a buffer which is smaller
203 * than the packet size it still counts as full packet read and will
204 * clean the incoming stream to keep sigio/epoll happy
205 */
206
207 #define DROP_BUFFER_SIZE 32
208
209 static char *drop_buffer;
210
211 /* Array backed queues optimized for bulk enqueue/dequeue and
212 * 1:N (small values of N) or 1:1 enqueuer/dequeuer ratios.
213 * For more details and full design rationale see
214 * http://foswiki.cambridgegreys.com/Main/EatYourTailAndEnjoyIt
215 */
216
217
218 /*
219 * Advance the mmsg queue head by n = advance. Resets the queue to
220 * maximum enqueue/dequeue-at-once capacity if possible. Called by
221 * dequeuers. Caller must hold the head_lock!
222 */
223
vector_advancehead(struct vector_queue * qi,int advance)224 static int vector_advancehead(struct vector_queue *qi, int advance)
225 {
226 int queue_depth;
227
228 qi->head =
229 (qi->head + advance)
230 % qi->max_depth;
231
232
233 spin_lock(&qi->tail_lock);
234 qi->queue_depth -= advance;
235
236 /* we are at 0, use this to
237 * reset head and tail so we can use max size vectors
238 */
239
240 if (qi->queue_depth == 0) {
241 qi->head = 0;
242 qi->tail = 0;
243 }
244 queue_depth = qi->queue_depth;
245 spin_unlock(&qi->tail_lock);
246 return queue_depth;
247 }
248
249 /* Advance the queue tail by n = advance.
250 * This is called by enqueuers which should hold the
251 * head lock already
252 */
253
vector_advancetail(struct vector_queue * qi,int advance)254 static int vector_advancetail(struct vector_queue *qi, int advance)
255 {
256 int queue_depth;
257
258 qi->tail =
259 (qi->tail + advance)
260 % qi->max_depth;
261 spin_lock(&qi->head_lock);
262 qi->queue_depth += advance;
263 queue_depth = qi->queue_depth;
264 spin_unlock(&qi->head_lock);
265 return queue_depth;
266 }
267
prep_msg(struct vector_private * vp,struct sk_buff * skb,struct iovec * iov)268 static int prep_msg(struct vector_private *vp,
269 struct sk_buff *skb,
270 struct iovec *iov)
271 {
272 int iov_index = 0;
273 int nr_frags, frag;
274 skb_frag_t *skb_frag;
275
276 nr_frags = skb_shinfo(skb)->nr_frags;
277 if (nr_frags > MAX_IOV_SIZE) {
278 if (skb_linearize(skb) != 0)
279 goto drop;
280 }
281 if (vp->header_size > 0) {
282 iov[iov_index].iov_len = vp->header_size;
283 vp->form_header(iov[iov_index].iov_base, skb, vp);
284 iov_index++;
285 }
286 iov[iov_index].iov_base = skb->data;
287 if (nr_frags > 0) {
288 iov[iov_index].iov_len = skb->len - skb->data_len;
289 vp->estats.sg_ok++;
290 } else
291 iov[iov_index].iov_len = skb->len;
292 iov_index++;
293 for (frag = 0; frag < nr_frags; frag++) {
294 skb_frag = &skb_shinfo(skb)->frags[frag];
295 iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
296 iov[iov_index].iov_len = skb_frag_size(skb_frag);
297 iov_index++;
298 }
299 return iov_index;
300 drop:
301 return -1;
302 }
303 /*
304 * Generic vector enqueue with support for forming headers using transport
305 * specific callback. Allows GRE, L2TPv3, RAW and other transports
306 * to use a common enqueue procedure in vector mode
307 */
308
vector_enqueue(struct vector_queue * qi,struct sk_buff * skb)309 static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
310 {
311 struct vector_private *vp = netdev_priv(qi->dev);
312 int queue_depth;
313 int packet_len;
314 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
315 int iov_count;
316
317 spin_lock(&qi->tail_lock);
318 spin_lock(&qi->head_lock);
319 queue_depth = qi->queue_depth;
320 spin_unlock(&qi->head_lock);
321
322 if (skb)
323 packet_len = skb->len;
324
325 if (queue_depth < qi->max_depth) {
326
327 *(qi->skbuff_vector + qi->tail) = skb;
328 mmsg_vector += qi->tail;
329 iov_count = prep_msg(
330 vp,
331 skb,
332 mmsg_vector->msg_hdr.msg_iov
333 );
334 if (iov_count < 1)
335 goto drop;
336 mmsg_vector->msg_hdr.msg_iovlen = iov_count;
337 mmsg_vector->msg_hdr.msg_name = vp->fds->remote_addr;
338 mmsg_vector->msg_hdr.msg_namelen = vp->fds->remote_addr_size;
339 queue_depth = vector_advancetail(qi, 1);
340 } else
341 goto drop;
342 spin_unlock(&qi->tail_lock);
343 return queue_depth;
344 drop:
345 qi->dev->stats.tx_dropped++;
346 if (skb != NULL) {
347 packet_len = skb->len;
348 dev_consume_skb_any(skb);
349 netdev_completed_queue(qi->dev, 1, packet_len);
350 }
351 spin_unlock(&qi->tail_lock);
352 return queue_depth;
353 }
354
consume_vector_skbs(struct vector_queue * qi,int count)355 static int consume_vector_skbs(struct vector_queue *qi, int count)
356 {
357 struct sk_buff *skb;
358 int skb_index;
359 int bytes_compl = 0;
360
361 for (skb_index = qi->head; skb_index < qi->head + count; skb_index++) {
362 skb = *(qi->skbuff_vector + skb_index);
363 /* mark as empty to ensure correct destruction if
364 * needed
365 */
366 bytes_compl += skb->len;
367 *(qi->skbuff_vector + skb_index) = NULL;
368 dev_consume_skb_any(skb);
369 }
370 qi->dev->stats.tx_bytes += bytes_compl;
371 qi->dev->stats.tx_packets += count;
372 netdev_completed_queue(qi->dev, count, bytes_compl);
373 return vector_advancehead(qi, count);
374 }
375
376 /*
377 * Generic vector deque via sendmmsg with support for forming headers
378 * using transport specific callback. Allows GRE, L2TPv3, RAW and
379 * other transports to use a common dequeue procedure in vector mode
380 */
381
382
vector_send(struct vector_queue * qi)383 static int vector_send(struct vector_queue *qi)
384 {
385 struct vector_private *vp = netdev_priv(qi->dev);
386 struct mmsghdr *send_from;
387 int result = 0, send_len, queue_depth = qi->max_depth;
388
389 if (spin_trylock(&qi->head_lock)) {
390 if (spin_trylock(&qi->tail_lock)) {
391 /* update queue_depth to current value */
392 queue_depth = qi->queue_depth;
393 spin_unlock(&qi->tail_lock);
394 while (queue_depth > 0) {
395 /* Calculate the start of the vector */
396 send_len = queue_depth;
397 send_from = qi->mmsg_vector;
398 send_from += qi->head;
399 /* Adjust vector size if wraparound */
400 if (send_len + qi->head > qi->max_depth)
401 send_len = qi->max_depth - qi->head;
402 /* Try to TX as many packets as possible */
403 if (send_len > 0) {
404 result = uml_vector_sendmmsg(
405 vp->fds->tx_fd,
406 send_from,
407 send_len,
408 0
409 );
410 vp->in_write_poll =
411 (result != send_len);
412 }
413 /* For some of the sendmmsg error scenarios
414 * we may end being unsure in the TX success
415 * for all packets. It is safer to declare
416 * them all TX-ed and blame the network.
417 */
418 if (result < 0) {
419 if (net_ratelimit())
420 netdev_err(vp->dev, "sendmmsg err=%i\n",
421 result);
422 vp->in_error = true;
423 result = send_len;
424 }
425 if (result > 0) {
426 queue_depth =
427 consume_vector_skbs(qi, result);
428 /* This is equivalent to an TX IRQ.
429 * Restart the upper layers to feed us
430 * more packets.
431 */
432 if (result > vp->estats.tx_queue_max)
433 vp->estats.tx_queue_max = result;
434 vp->estats.tx_queue_running_average =
435 (vp->estats.tx_queue_running_average + result) >> 1;
436 }
437 netif_trans_update(qi->dev);
438 netif_wake_queue(qi->dev);
439 /* if TX is busy, break out of the send loop,
440 * poll write IRQ will reschedule xmit for us
441 */
442 if (result != send_len) {
443 vp->estats.tx_restart_queue++;
444 break;
445 }
446 }
447 }
448 spin_unlock(&qi->head_lock);
449 } else {
450 tasklet_schedule(&vp->tx_poll);
451 }
452 return queue_depth;
453 }
454
455 /* Queue destructor. Deliberately stateless so we can use
456 * it in queue cleanup if initialization fails.
457 */
458
destroy_queue(struct vector_queue * qi)459 static void destroy_queue(struct vector_queue *qi)
460 {
461 int i;
462 struct iovec *iov;
463 struct vector_private *vp = netdev_priv(qi->dev);
464 struct mmsghdr *mmsg_vector;
465
466 if (qi == NULL)
467 return;
468 /* deallocate any skbuffs - we rely on any unused to be
469 * set to NULL.
470 */
471 if (qi->skbuff_vector != NULL) {
472 for (i = 0; i < qi->max_depth; i++) {
473 if (*(qi->skbuff_vector + i) != NULL)
474 dev_kfree_skb_any(*(qi->skbuff_vector + i));
475 }
476 kfree(qi->skbuff_vector);
477 }
478 /* deallocate matching IOV structures including header buffs */
479 if (qi->mmsg_vector != NULL) {
480 mmsg_vector = qi->mmsg_vector;
481 for (i = 0; i < qi->max_depth; i++) {
482 iov = mmsg_vector->msg_hdr.msg_iov;
483 if (iov != NULL) {
484 if ((vp->header_size > 0) &&
485 (iov->iov_base != NULL))
486 kfree(iov->iov_base);
487 kfree(iov);
488 }
489 mmsg_vector++;
490 }
491 kfree(qi->mmsg_vector);
492 }
493 kfree(qi);
494 }
495
496 /*
497 * Queue constructor. Create a queue with a given side.
498 */
create_queue(struct vector_private * vp,int max_size,int header_size,int num_extra_frags)499 static struct vector_queue *create_queue(
500 struct vector_private *vp,
501 int max_size,
502 int header_size,
503 int num_extra_frags)
504 {
505 struct vector_queue *result;
506 int i;
507 struct iovec *iov;
508 struct mmsghdr *mmsg_vector;
509
510 result = kmalloc(sizeof(struct vector_queue), GFP_KERNEL);
511 if (result == NULL)
512 return NULL;
513 result->max_depth = max_size;
514 result->dev = vp->dev;
515 result->mmsg_vector = kmalloc(
516 (sizeof(struct mmsghdr) * max_size), GFP_KERNEL);
517 if (result->mmsg_vector == NULL)
518 goto out_mmsg_fail;
519 result->skbuff_vector = kmalloc(
520 (sizeof(void *) * max_size), GFP_KERNEL);
521 if (result->skbuff_vector == NULL)
522 goto out_skb_fail;
523
524 /* further failures can be handled safely by destroy_queue*/
525
526 mmsg_vector = result->mmsg_vector;
527 for (i = 0; i < max_size; i++) {
528 /* Clear all pointers - we use non-NULL as marking on
529 * what to free on destruction
530 */
531 *(result->skbuff_vector + i) = NULL;
532 mmsg_vector->msg_hdr.msg_iov = NULL;
533 mmsg_vector++;
534 }
535 mmsg_vector = result->mmsg_vector;
536 result->max_iov_frags = num_extra_frags;
537 for (i = 0; i < max_size; i++) {
538 if (vp->header_size > 0)
539 iov = kmalloc_array(3 + num_extra_frags,
540 sizeof(struct iovec),
541 GFP_KERNEL
542 );
543 else
544 iov = kmalloc_array(2 + num_extra_frags,
545 sizeof(struct iovec),
546 GFP_KERNEL
547 );
548 if (iov == NULL)
549 goto out_fail;
550 mmsg_vector->msg_hdr.msg_iov = iov;
551 mmsg_vector->msg_hdr.msg_iovlen = 1;
552 mmsg_vector->msg_hdr.msg_control = NULL;
553 mmsg_vector->msg_hdr.msg_controllen = 0;
554 mmsg_vector->msg_hdr.msg_flags = MSG_DONTWAIT;
555 mmsg_vector->msg_hdr.msg_name = NULL;
556 mmsg_vector->msg_hdr.msg_namelen = 0;
557 if (vp->header_size > 0) {
558 iov->iov_base = kmalloc(header_size, GFP_KERNEL);
559 if (iov->iov_base == NULL)
560 goto out_fail;
561 iov->iov_len = header_size;
562 mmsg_vector->msg_hdr.msg_iovlen = 2;
563 iov++;
564 }
565 iov->iov_base = NULL;
566 iov->iov_len = 0;
567 mmsg_vector++;
568 }
569 spin_lock_init(&result->head_lock);
570 spin_lock_init(&result->tail_lock);
571 result->queue_depth = 0;
572 result->head = 0;
573 result->tail = 0;
574 return result;
575 out_skb_fail:
576 kfree(result->mmsg_vector);
577 out_mmsg_fail:
578 kfree(result);
579 return NULL;
580 out_fail:
581 destroy_queue(result);
582 return NULL;
583 }
584
585 /*
586 * We do not use the RX queue as a proper wraparound queue for now
587 * This is not necessary because the consumption via netif_rx()
588 * happens in-line. While we can try using the return code of
589 * netif_rx() for flow control there are no drivers doing this today.
590 * For this RX specific use we ignore the tail/head locks and
591 * just read into a prepared queue filled with skbuffs.
592 */
593
prep_skb(struct vector_private * vp,struct user_msghdr * msg)594 static struct sk_buff *prep_skb(
595 struct vector_private *vp,
596 struct user_msghdr *msg)
597 {
598 int linear = vp->max_packet + vp->headroom + SAFETY_MARGIN;
599 struct sk_buff *result;
600 int iov_index = 0, len;
601 struct iovec *iov = msg->msg_iov;
602 int err, nr_frags, frag;
603 skb_frag_t *skb_frag;
604
605 if (vp->req_size <= linear)
606 len = linear;
607 else
608 len = vp->req_size;
609 result = alloc_skb_with_frags(
610 linear,
611 len - vp->max_packet,
612 3,
613 &err,
614 GFP_ATOMIC
615 );
616 if (vp->header_size > 0)
617 iov_index++;
618 if (result == NULL) {
619 iov[iov_index].iov_base = NULL;
620 iov[iov_index].iov_len = 0;
621 goto done;
622 }
623 skb_reserve(result, vp->headroom);
624 result->dev = vp->dev;
625 skb_put(result, vp->max_packet);
626 result->data_len = len - vp->max_packet;
627 result->len += len - vp->max_packet;
628 skb_reset_mac_header(result);
629 result->ip_summed = CHECKSUM_NONE;
630 iov[iov_index].iov_base = result->data;
631 iov[iov_index].iov_len = vp->max_packet;
632 iov_index++;
633
634 nr_frags = skb_shinfo(result)->nr_frags;
635 for (frag = 0; frag < nr_frags; frag++) {
636 skb_frag = &skb_shinfo(result)->frags[frag];
637 iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
638 if (iov[iov_index].iov_base != NULL)
639 iov[iov_index].iov_len = skb_frag_size(skb_frag);
640 else
641 iov[iov_index].iov_len = 0;
642 iov_index++;
643 }
644 done:
645 msg->msg_iovlen = iov_index;
646 return result;
647 }
648
649
650 /* Prepare queue for recvmmsg one-shot rx - fill with fresh sk_buffs*/
651
prep_queue_for_rx(struct vector_queue * qi)652 static void prep_queue_for_rx(struct vector_queue *qi)
653 {
654 struct vector_private *vp = netdev_priv(qi->dev);
655 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
656 void **skbuff_vector = qi->skbuff_vector;
657 int i;
658
659 if (qi->queue_depth == 0)
660 return;
661 for (i = 0; i < qi->queue_depth; i++) {
662 /* it is OK if allocation fails - recvmmsg with NULL data in
663 * iov argument still performs an RX, just drops the packet
664 * This allows us stop faffing around with a "drop buffer"
665 */
666
667 *skbuff_vector = prep_skb(vp, &mmsg_vector->msg_hdr);
668 skbuff_vector++;
669 mmsg_vector++;
670 }
671 qi->queue_depth = 0;
672 }
673
find_device(int n)674 static struct vector_device *find_device(int n)
675 {
676 struct vector_device *device;
677 struct list_head *ele;
678
679 spin_lock(&vector_devices_lock);
680 list_for_each(ele, &vector_devices) {
681 device = list_entry(ele, struct vector_device, list);
682 if (device->unit == n)
683 goto out;
684 }
685 device = NULL;
686 out:
687 spin_unlock(&vector_devices_lock);
688 return device;
689 }
690
vector_parse(char * str,int * index_out,char ** str_out,char ** error_out)691 static int vector_parse(char *str, int *index_out, char **str_out,
692 char **error_out)
693 {
694 int n, len, err;
695 char *start = str;
696
697 len = strlen(str);
698
699 while ((*str != ':') && (strlen(str) > 1))
700 str++;
701 if (*str != ':') {
702 *error_out = "Expected ':' after device number";
703 return -EINVAL;
704 }
705 *str = '\0';
706
707 err = kstrtouint(start, 0, &n);
708 if (err < 0) {
709 *error_out = "Bad device number";
710 return err;
711 }
712
713 str++;
714 if (find_device(n)) {
715 *error_out = "Device already configured";
716 return -EINVAL;
717 }
718
719 *index_out = n;
720 *str_out = str;
721 return 0;
722 }
723
vector_config(char * str,char ** error_out)724 static int vector_config(char *str, char **error_out)
725 {
726 int err, n;
727 char *params;
728 struct arglist *parsed;
729
730 err = vector_parse(str, &n, ¶ms, error_out);
731 if (err != 0)
732 return err;
733
734 /* This string is broken up and the pieces used by the underlying
735 * driver. We should copy it to make sure things do not go wrong
736 * later.
737 */
738
739 params = kstrdup(params, GFP_KERNEL);
740 if (params == NULL) {
741 *error_out = "vector_config failed to strdup string";
742 return -ENOMEM;
743 }
744
745 parsed = uml_parse_vector_ifspec(params);
746
747 if (parsed == NULL) {
748 *error_out = "vector_config failed to parse parameters";
749 kfree(params);
750 return -EINVAL;
751 }
752
753 vector_eth_configure(n, parsed);
754 return 0;
755 }
756
vector_id(char ** str,int * start_out,int * end_out)757 static int vector_id(char **str, int *start_out, int *end_out)
758 {
759 char *end;
760 int n;
761
762 n = simple_strtoul(*str, &end, 0);
763 if ((*end != '\0') || (end == *str))
764 return -1;
765
766 *start_out = n;
767 *end_out = n;
768 *str = end;
769 return n;
770 }
771
vector_remove(int n,char ** error_out)772 static int vector_remove(int n, char **error_out)
773 {
774 struct vector_device *vec_d;
775 struct net_device *dev;
776 struct vector_private *vp;
777
778 vec_d = find_device(n);
779 if (vec_d == NULL)
780 return -ENODEV;
781 dev = vec_d->dev;
782 vp = netdev_priv(dev);
783 if (vp->fds != NULL)
784 return -EBUSY;
785 unregister_netdev(dev);
786 platform_device_unregister(&vec_d->pdev);
787 return 0;
788 }
789
790 /*
791 * There is no shared per-transport initialization code, so
792 * we will just initialize each interface one by one and
793 * add them to a list
794 */
795
796 static struct platform_driver uml_net_driver = {
797 .driver = {
798 .name = DRIVER_NAME,
799 },
800 };
801
802
vector_device_release(struct device * dev)803 static void vector_device_release(struct device *dev)
804 {
805 struct vector_device *device = dev_get_drvdata(dev);
806 struct net_device *netdev = device->dev;
807
808 list_del(&device->list);
809 kfree(device);
810 free_netdev(netdev);
811 }
812
813 /* Bog standard recv using recvmsg - not used normally unless the user
814 * explicitly specifies not to use recvmmsg vector RX.
815 */
816
vector_legacy_rx(struct vector_private * vp)817 static int vector_legacy_rx(struct vector_private *vp)
818 {
819 int pkt_len;
820 struct user_msghdr hdr;
821 struct iovec iov[2 + MAX_IOV_SIZE]; /* header + data use case only */
822 int iovpos = 0;
823 struct sk_buff *skb;
824 int header_check;
825
826 hdr.msg_name = NULL;
827 hdr.msg_namelen = 0;
828 hdr.msg_iov = (struct iovec *) &iov;
829 hdr.msg_control = NULL;
830 hdr.msg_controllen = 0;
831 hdr.msg_flags = 0;
832
833 if (vp->header_size > 0) {
834 iov[0].iov_base = vp->header_rxbuffer;
835 iov[0].iov_len = vp->header_size;
836 }
837
838 skb = prep_skb(vp, &hdr);
839
840 if (skb == NULL) {
841 /* Read a packet into drop_buffer and don't do
842 * anything with it.
843 */
844 iov[iovpos].iov_base = drop_buffer;
845 iov[iovpos].iov_len = DROP_BUFFER_SIZE;
846 hdr.msg_iovlen = 1;
847 vp->dev->stats.rx_dropped++;
848 }
849
850 pkt_len = uml_vector_recvmsg(vp->fds->rx_fd, &hdr, 0);
851 if (pkt_len < 0) {
852 vp->in_error = true;
853 return pkt_len;
854 }
855
856 if (skb != NULL) {
857 if (pkt_len > vp->header_size) {
858 if (vp->header_size > 0) {
859 header_check = vp->verify_header(
860 vp->header_rxbuffer, skb, vp);
861 if (header_check < 0) {
862 dev_kfree_skb_irq(skb);
863 vp->dev->stats.rx_dropped++;
864 vp->estats.rx_encaps_errors++;
865 return 0;
866 }
867 if (header_check > 0) {
868 vp->estats.rx_csum_offload_good++;
869 skb->ip_summed = CHECKSUM_UNNECESSARY;
870 }
871 }
872 pskb_trim(skb, pkt_len - vp->rx_header_size);
873 skb->protocol = eth_type_trans(skb, skb->dev);
874 vp->dev->stats.rx_bytes += skb->len;
875 vp->dev->stats.rx_packets++;
876 netif_rx(skb);
877 } else {
878 dev_kfree_skb_irq(skb);
879 }
880 }
881 return pkt_len;
882 }
883
884 /*
885 * Packet at a time TX which falls back to vector TX if the
886 * underlying transport is busy.
887 */
888
889
890
writev_tx(struct vector_private * vp,struct sk_buff * skb)891 static int writev_tx(struct vector_private *vp, struct sk_buff *skb)
892 {
893 struct iovec iov[3 + MAX_IOV_SIZE];
894 int iov_count, pkt_len = 0;
895
896 iov[0].iov_base = vp->header_txbuffer;
897 iov_count = prep_msg(vp, skb, (struct iovec *) &iov);
898
899 if (iov_count < 1)
900 goto drop;
901
902 pkt_len = uml_vector_writev(
903 vp->fds->tx_fd,
904 (struct iovec *) &iov,
905 iov_count
906 );
907
908 if (pkt_len < 0)
909 goto drop;
910
911 netif_trans_update(vp->dev);
912 netif_wake_queue(vp->dev);
913
914 if (pkt_len > 0) {
915 vp->dev->stats.tx_bytes += skb->len;
916 vp->dev->stats.tx_packets++;
917 } else {
918 vp->dev->stats.tx_dropped++;
919 }
920 consume_skb(skb);
921 return pkt_len;
922 drop:
923 vp->dev->stats.tx_dropped++;
924 consume_skb(skb);
925 if (pkt_len < 0)
926 vp->in_error = true;
927 return pkt_len;
928 }
929
930 /*
931 * Receive as many messages as we can in one call using the special
932 * mmsg vector matched to an skb vector which we prepared earlier.
933 */
934
vector_mmsg_rx(struct vector_private * vp)935 static int vector_mmsg_rx(struct vector_private *vp)
936 {
937 int packet_count, i;
938 struct vector_queue *qi = vp->rx_queue;
939 struct sk_buff *skb;
940 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
941 void **skbuff_vector = qi->skbuff_vector;
942 int header_check;
943
944 /* Refresh the vector and make sure it is with new skbs and the
945 * iovs are updated to point to them.
946 */
947
948 prep_queue_for_rx(qi);
949
950 /* Fire the Lazy Gun - get as many packets as we can in one go. */
951
952 packet_count = uml_vector_recvmmsg(
953 vp->fds->rx_fd, qi->mmsg_vector, qi->max_depth, 0);
954
955 if (packet_count < 0)
956 vp->in_error = true;
957
958 if (packet_count <= 0)
959 return packet_count;
960
961 /* We treat packet processing as enqueue, buffer refresh as dequeue
962 * The queue_depth tells us how many buffers have been used and how
963 * many do we need to prep the next time prep_queue_for_rx() is called.
964 */
965
966 qi->queue_depth = packet_count;
967
968 for (i = 0; i < packet_count; i++) {
969 skb = (*skbuff_vector);
970 if (mmsg_vector->msg_len > vp->header_size) {
971 if (vp->header_size > 0) {
972 header_check = vp->verify_header(
973 mmsg_vector->msg_hdr.msg_iov->iov_base,
974 skb,
975 vp
976 );
977 if (header_check < 0) {
978 /* Overlay header failed to verify - discard.
979 * We can actually keep this skb and reuse it,
980 * but that will make the prep logic too
981 * complex.
982 */
983 dev_kfree_skb_irq(skb);
984 vp->estats.rx_encaps_errors++;
985 continue;
986 }
987 if (header_check > 0) {
988 vp->estats.rx_csum_offload_good++;
989 skb->ip_summed = CHECKSUM_UNNECESSARY;
990 }
991 }
992 pskb_trim(skb,
993 mmsg_vector->msg_len - vp->rx_header_size);
994 skb->protocol = eth_type_trans(skb, skb->dev);
995 /*
996 * We do not need to lock on updating stats here
997 * The interrupt loop is non-reentrant.
998 */
999 vp->dev->stats.rx_bytes += skb->len;
1000 vp->dev->stats.rx_packets++;
1001 netif_rx(skb);
1002 } else {
1003 /* Overlay header too short to do anything - discard.
1004 * We can actually keep this skb and reuse it,
1005 * but that will make the prep logic too complex.
1006 */
1007 if (skb != NULL)
1008 dev_kfree_skb_irq(skb);
1009 }
1010 (*skbuff_vector) = NULL;
1011 /* Move to the next buffer element */
1012 mmsg_vector++;
1013 skbuff_vector++;
1014 }
1015 if (packet_count > 0) {
1016 if (vp->estats.rx_queue_max < packet_count)
1017 vp->estats.rx_queue_max = packet_count;
1018 vp->estats.rx_queue_running_average =
1019 (vp->estats.rx_queue_running_average + packet_count) >> 1;
1020 }
1021 return packet_count;
1022 }
1023
vector_rx(struct vector_private * vp)1024 static void vector_rx(struct vector_private *vp)
1025 {
1026 int err;
1027 int iter = 0;
1028
1029 if ((vp->options & VECTOR_RX) > 0)
1030 while (((err = vector_mmsg_rx(vp)) > 0) && (iter < MAX_ITERATIONS))
1031 iter++;
1032 else
1033 while (((err = vector_legacy_rx(vp)) > 0) && (iter < MAX_ITERATIONS))
1034 iter++;
1035 if ((err != 0) && net_ratelimit())
1036 netdev_err(vp->dev, "vector_rx: error(%d)\n", err);
1037 if (iter == MAX_ITERATIONS)
1038 netdev_err(vp->dev, "vector_rx: device stuck, remote end may have closed the connection\n");
1039 }
1040
vector_net_start_xmit(struct sk_buff * skb,struct net_device * dev)1041 static int vector_net_start_xmit(struct sk_buff *skb, struct net_device *dev)
1042 {
1043 struct vector_private *vp = netdev_priv(dev);
1044 int queue_depth = 0;
1045
1046 if (vp->in_error) {
1047 deactivate_fd(vp->fds->rx_fd, vp->rx_irq);
1048 if ((vp->fds->rx_fd != vp->fds->tx_fd) && (vp->tx_irq != 0))
1049 deactivate_fd(vp->fds->tx_fd, vp->tx_irq);
1050 return NETDEV_TX_BUSY;
1051 }
1052
1053 if ((vp->options & VECTOR_TX) == 0) {
1054 writev_tx(vp, skb);
1055 return NETDEV_TX_OK;
1056 }
1057
1058 /* We do BQL only in the vector path, no point doing it in
1059 * packet at a time mode as there is no device queue
1060 */
1061
1062 netdev_sent_queue(vp->dev, skb->len);
1063 queue_depth = vector_enqueue(vp->tx_queue, skb);
1064
1065 /* if the device queue is full, stop the upper layers and
1066 * flush it.
1067 */
1068
1069 if (queue_depth >= vp->tx_queue->max_depth - 1) {
1070 vp->estats.tx_kicks++;
1071 netif_stop_queue(dev);
1072 vector_send(vp->tx_queue);
1073 return NETDEV_TX_OK;
1074 }
1075 if (netdev_xmit_more()) {
1076 mod_timer(&vp->tl, vp->coalesce);
1077 return NETDEV_TX_OK;
1078 }
1079 if (skb->len < TX_SMALL_PACKET) {
1080 vp->estats.tx_kicks++;
1081 vector_send(vp->tx_queue);
1082 } else
1083 tasklet_schedule(&vp->tx_poll);
1084 return NETDEV_TX_OK;
1085 }
1086
vector_rx_interrupt(int irq,void * dev_id)1087 static irqreturn_t vector_rx_interrupt(int irq, void *dev_id)
1088 {
1089 struct net_device *dev = dev_id;
1090 struct vector_private *vp = netdev_priv(dev);
1091
1092 if (!netif_running(dev))
1093 return IRQ_NONE;
1094 vector_rx(vp);
1095 return IRQ_HANDLED;
1096
1097 }
1098
vector_tx_interrupt(int irq,void * dev_id)1099 static irqreturn_t vector_tx_interrupt(int irq, void *dev_id)
1100 {
1101 struct net_device *dev = dev_id;
1102 struct vector_private *vp = netdev_priv(dev);
1103
1104 if (!netif_running(dev))
1105 return IRQ_NONE;
1106 /* We need to pay attention to it only if we got
1107 * -EAGAIN or -ENOBUFFS from sendmmsg. Otherwise
1108 * we ignore it. In the future, it may be worth
1109 * it to improve the IRQ controller a bit to make
1110 * tweaking the IRQ mask less costly
1111 */
1112
1113 if (vp->in_write_poll)
1114 tasklet_schedule(&vp->tx_poll);
1115 return IRQ_HANDLED;
1116
1117 }
1118
1119 static int irq_rr;
1120
vector_net_close(struct net_device * dev)1121 static int vector_net_close(struct net_device *dev)
1122 {
1123 struct vector_private *vp = netdev_priv(dev);
1124 unsigned long flags;
1125
1126 netif_stop_queue(dev);
1127 del_timer(&vp->tl);
1128
1129 if (vp->fds == NULL)
1130 return 0;
1131
1132 /* Disable and free all IRQS */
1133 if (vp->rx_irq > 0) {
1134 um_free_irq(vp->rx_irq, dev);
1135 vp->rx_irq = 0;
1136 }
1137 if (vp->tx_irq > 0) {
1138 um_free_irq(vp->tx_irq, dev);
1139 vp->tx_irq = 0;
1140 }
1141 tasklet_kill(&vp->tx_poll);
1142 if (vp->fds->rx_fd > 0) {
1143 os_close_file(vp->fds->rx_fd);
1144 vp->fds->rx_fd = -1;
1145 }
1146 if (vp->fds->tx_fd > 0) {
1147 os_close_file(vp->fds->tx_fd);
1148 vp->fds->tx_fd = -1;
1149 }
1150 kfree(vp->bpf);
1151 kfree(vp->fds->remote_addr);
1152 kfree(vp->transport_data);
1153 kfree(vp->header_rxbuffer);
1154 kfree(vp->header_txbuffer);
1155 if (vp->rx_queue != NULL)
1156 destroy_queue(vp->rx_queue);
1157 if (vp->tx_queue != NULL)
1158 destroy_queue(vp->tx_queue);
1159 kfree(vp->fds);
1160 vp->fds = NULL;
1161 spin_lock_irqsave(&vp->lock, flags);
1162 vp->opened = false;
1163 vp->in_error = false;
1164 spin_unlock_irqrestore(&vp->lock, flags);
1165 return 0;
1166 }
1167
1168 /* TX tasklet */
1169
vector_tx_poll(unsigned long data)1170 static void vector_tx_poll(unsigned long data)
1171 {
1172 struct vector_private *vp = (struct vector_private *)data;
1173
1174 vp->estats.tx_kicks++;
1175 vector_send(vp->tx_queue);
1176 }
vector_reset_tx(struct work_struct * work)1177 static void vector_reset_tx(struct work_struct *work)
1178 {
1179 struct vector_private *vp =
1180 container_of(work, struct vector_private, reset_tx);
1181 netdev_reset_queue(vp->dev);
1182 netif_start_queue(vp->dev);
1183 netif_wake_queue(vp->dev);
1184 }
vector_net_open(struct net_device * dev)1185 static int vector_net_open(struct net_device *dev)
1186 {
1187 struct vector_private *vp = netdev_priv(dev);
1188 unsigned long flags;
1189 int err = -EINVAL;
1190 struct vector_device *vdevice;
1191
1192 spin_lock_irqsave(&vp->lock, flags);
1193 if (vp->opened) {
1194 spin_unlock_irqrestore(&vp->lock, flags);
1195 return -ENXIO;
1196 }
1197 vp->opened = true;
1198 spin_unlock_irqrestore(&vp->lock, flags);
1199
1200 vp->fds = uml_vector_user_open(vp->unit, vp->parsed);
1201
1202 if (vp->fds == NULL)
1203 goto out_close;
1204
1205 if (build_transport_data(vp) < 0)
1206 goto out_close;
1207
1208 if ((vp->options & VECTOR_RX) > 0) {
1209 vp->rx_queue = create_queue(
1210 vp,
1211 get_depth(vp->parsed),
1212 vp->rx_header_size,
1213 MAX_IOV_SIZE
1214 );
1215 vp->rx_queue->queue_depth = get_depth(vp->parsed);
1216 } else {
1217 vp->header_rxbuffer = kmalloc(
1218 vp->rx_header_size,
1219 GFP_KERNEL
1220 );
1221 if (vp->header_rxbuffer == NULL)
1222 goto out_close;
1223 }
1224 if ((vp->options & VECTOR_TX) > 0) {
1225 vp->tx_queue = create_queue(
1226 vp,
1227 get_depth(vp->parsed),
1228 vp->header_size,
1229 MAX_IOV_SIZE
1230 );
1231 } else {
1232 vp->header_txbuffer = kmalloc(vp->header_size, GFP_KERNEL);
1233 if (vp->header_txbuffer == NULL)
1234 goto out_close;
1235 }
1236
1237 /* READ IRQ */
1238 err = um_request_irq(
1239 irq_rr + VECTOR_BASE_IRQ, vp->fds->rx_fd,
1240 IRQ_READ, vector_rx_interrupt,
1241 IRQF_SHARED, dev->name, dev);
1242 if (err != 0) {
1243 netdev_err(dev, "vector_open: failed to get rx irq(%d)\n", err);
1244 err = -ENETUNREACH;
1245 goto out_close;
1246 }
1247 vp->rx_irq = irq_rr + VECTOR_BASE_IRQ;
1248 dev->irq = irq_rr + VECTOR_BASE_IRQ;
1249 irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
1250
1251 /* WRITE IRQ - we need it only if we have vector TX */
1252 if ((vp->options & VECTOR_TX) > 0) {
1253 err = um_request_irq(
1254 irq_rr + VECTOR_BASE_IRQ, vp->fds->tx_fd,
1255 IRQ_WRITE, vector_tx_interrupt,
1256 IRQF_SHARED, dev->name, dev);
1257 if (err != 0) {
1258 netdev_err(dev,
1259 "vector_open: failed to get tx irq(%d)\n", err);
1260 err = -ENETUNREACH;
1261 goto out_close;
1262 }
1263 vp->tx_irq = irq_rr + VECTOR_BASE_IRQ;
1264 irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
1265 }
1266
1267 if ((vp->options & VECTOR_QDISC_BYPASS) != 0) {
1268 if (!uml_raw_enable_qdisc_bypass(vp->fds->rx_fd))
1269 vp->options |= VECTOR_BPF;
1270 }
1271 if ((vp->options & VECTOR_BPF) != 0)
1272 vp->bpf = uml_vector_default_bpf(vp->fds->rx_fd, dev->dev_addr);
1273
1274 netif_start_queue(dev);
1275
1276 /* clear buffer - it can happen that the host side of the interface
1277 * is full when we get here. In this case, new data is never queued,
1278 * SIGIOs never arrive, and the net never works.
1279 */
1280
1281 vector_rx(vp);
1282
1283 vector_reset_stats(vp);
1284 vdevice = find_device(vp->unit);
1285 vdevice->opened = 1;
1286
1287 if ((vp->options & VECTOR_TX) != 0)
1288 add_timer(&vp->tl);
1289 return 0;
1290 out_close:
1291 vector_net_close(dev);
1292 return err;
1293 }
1294
1295
vector_net_set_multicast_list(struct net_device * dev)1296 static void vector_net_set_multicast_list(struct net_device *dev)
1297 {
1298 /* TODO: - we can do some BPF games here */
1299 return;
1300 }
1301
vector_net_tx_timeout(struct net_device * dev)1302 static void vector_net_tx_timeout(struct net_device *dev)
1303 {
1304 struct vector_private *vp = netdev_priv(dev);
1305
1306 vp->estats.tx_timeout_count++;
1307 netif_trans_update(dev);
1308 schedule_work(&vp->reset_tx);
1309 }
1310
vector_fix_features(struct net_device * dev,netdev_features_t features)1311 static netdev_features_t vector_fix_features(struct net_device *dev,
1312 netdev_features_t features)
1313 {
1314 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
1315 return features;
1316 }
1317
vector_set_features(struct net_device * dev,netdev_features_t features)1318 static int vector_set_features(struct net_device *dev,
1319 netdev_features_t features)
1320 {
1321 struct vector_private *vp = netdev_priv(dev);
1322 /* Adjust buffer sizes for GSO/GRO. Unfortunately, there is
1323 * no way to negotiate it on raw sockets, so we can change
1324 * only our side.
1325 */
1326 if (features & NETIF_F_GRO)
1327 /* All new frame buffers will be GRO-sized */
1328 vp->req_size = 65536;
1329 else
1330 /* All new frame buffers will be normal sized */
1331 vp->req_size = vp->max_packet + vp->headroom + SAFETY_MARGIN;
1332 return 0;
1333 }
1334
1335 #ifdef CONFIG_NET_POLL_CONTROLLER
vector_net_poll_controller(struct net_device * dev)1336 static void vector_net_poll_controller(struct net_device *dev)
1337 {
1338 disable_irq(dev->irq);
1339 vector_rx_interrupt(dev->irq, dev);
1340 enable_irq(dev->irq);
1341 }
1342 #endif
1343
vector_net_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * info)1344 static void vector_net_get_drvinfo(struct net_device *dev,
1345 struct ethtool_drvinfo *info)
1346 {
1347 strlcpy(info->driver, DRIVER_NAME, sizeof(info->driver));
1348 strlcpy(info->version, DRIVER_VERSION, sizeof(info->version));
1349 }
1350
vector_get_ringparam(struct net_device * netdev,struct ethtool_ringparam * ring)1351 static void vector_get_ringparam(struct net_device *netdev,
1352 struct ethtool_ringparam *ring)
1353 {
1354 struct vector_private *vp = netdev_priv(netdev);
1355
1356 ring->rx_max_pending = vp->rx_queue->max_depth;
1357 ring->tx_max_pending = vp->tx_queue->max_depth;
1358 ring->rx_pending = vp->rx_queue->max_depth;
1359 ring->tx_pending = vp->tx_queue->max_depth;
1360 }
1361
vector_get_strings(struct net_device * dev,u32 stringset,u8 * buf)1362 static void vector_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
1363 {
1364 switch (stringset) {
1365 case ETH_SS_TEST:
1366 *buf = '\0';
1367 break;
1368 case ETH_SS_STATS:
1369 memcpy(buf, ðtool_stats_keys, sizeof(ethtool_stats_keys));
1370 break;
1371 default:
1372 WARN_ON(1);
1373 break;
1374 }
1375 }
1376
vector_get_sset_count(struct net_device * dev,int sset)1377 static int vector_get_sset_count(struct net_device *dev, int sset)
1378 {
1379 switch (sset) {
1380 case ETH_SS_TEST:
1381 return 0;
1382 case ETH_SS_STATS:
1383 return VECTOR_NUM_STATS;
1384 default:
1385 return -EOPNOTSUPP;
1386 }
1387 }
1388
vector_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * estats,u64 * tmp_stats)1389 static void vector_get_ethtool_stats(struct net_device *dev,
1390 struct ethtool_stats *estats,
1391 u64 *tmp_stats)
1392 {
1393 struct vector_private *vp = netdev_priv(dev);
1394
1395 memcpy(tmp_stats, &vp->estats, sizeof(struct vector_estats));
1396 }
1397
vector_get_coalesce(struct net_device * netdev,struct ethtool_coalesce * ec)1398 static int vector_get_coalesce(struct net_device *netdev,
1399 struct ethtool_coalesce *ec)
1400 {
1401 struct vector_private *vp = netdev_priv(netdev);
1402
1403 ec->tx_coalesce_usecs = (vp->coalesce * 1000000) / HZ;
1404 return 0;
1405 }
1406
vector_set_coalesce(struct net_device * netdev,struct ethtool_coalesce * ec)1407 static int vector_set_coalesce(struct net_device *netdev,
1408 struct ethtool_coalesce *ec)
1409 {
1410 struct vector_private *vp = netdev_priv(netdev);
1411
1412 vp->coalesce = (ec->tx_coalesce_usecs * HZ) / 1000000;
1413 if (vp->coalesce == 0)
1414 vp->coalesce = 1;
1415 return 0;
1416 }
1417
1418 static const struct ethtool_ops vector_net_ethtool_ops = {
1419 .get_drvinfo = vector_net_get_drvinfo,
1420 .get_link = ethtool_op_get_link,
1421 .get_ts_info = ethtool_op_get_ts_info,
1422 .get_ringparam = vector_get_ringparam,
1423 .get_strings = vector_get_strings,
1424 .get_sset_count = vector_get_sset_count,
1425 .get_ethtool_stats = vector_get_ethtool_stats,
1426 .get_coalesce = vector_get_coalesce,
1427 .set_coalesce = vector_set_coalesce,
1428 };
1429
1430
1431 static const struct net_device_ops vector_netdev_ops = {
1432 .ndo_open = vector_net_open,
1433 .ndo_stop = vector_net_close,
1434 .ndo_start_xmit = vector_net_start_xmit,
1435 .ndo_set_rx_mode = vector_net_set_multicast_list,
1436 .ndo_tx_timeout = vector_net_tx_timeout,
1437 .ndo_set_mac_address = eth_mac_addr,
1438 .ndo_validate_addr = eth_validate_addr,
1439 .ndo_fix_features = vector_fix_features,
1440 .ndo_set_features = vector_set_features,
1441 #ifdef CONFIG_NET_POLL_CONTROLLER
1442 .ndo_poll_controller = vector_net_poll_controller,
1443 #endif
1444 };
1445
1446
vector_timer_expire(struct timer_list * t)1447 static void vector_timer_expire(struct timer_list *t)
1448 {
1449 struct vector_private *vp = from_timer(vp, t, tl);
1450
1451 vp->estats.tx_kicks++;
1452 vector_send(vp->tx_queue);
1453 }
1454
vector_eth_configure(int n,struct arglist * def)1455 static void vector_eth_configure(
1456 int n,
1457 struct arglist *def
1458 )
1459 {
1460 struct vector_device *device;
1461 struct net_device *dev;
1462 struct vector_private *vp;
1463 int err;
1464
1465 device = kzalloc(sizeof(*device), GFP_KERNEL);
1466 if (device == NULL) {
1467 printk(KERN_ERR "eth_configure failed to allocate struct "
1468 "vector_device\n");
1469 return;
1470 }
1471 dev = alloc_etherdev(sizeof(struct vector_private));
1472 if (dev == NULL) {
1473 printk(KERN_ERR "eth_configure: failed to allocate struct "
1474 "net_device for vec%d\n", n);
1475 goto out_free_device;
1476 }
1477
1478 dev->mtu = get_mtu(def);
1479
1480 INIT_LIST_HEAD(&device->list);
1481 device->unit = n;
1482
1483 /* If this name ends up conflicting with an existing registered
1484 * netdevice, that is OK, register_netdev{,ice}() will notice this
1485 * and fail.
1486 */
1487 snprintf(dev->name, sizeof(dev->name), "vec%d", n);
1488 uml_net_setup_etheraddr(dev, uml_vector_fetch_arg(def, "mac"));
1489 vp = netdev_priv(dev);
1490
1491 /* sysfs register */
1492 if (!driver_registered) {
1493 platform_driver_register(¨_net_driver);
1494 driver_registered = 1;
1495 }
1496 device->pdev.id = n;
1497 device->pdev.name = DRIVER_NAME;
1498 device->pdev.dev.release = vector_device_release;
1499 dev_set_drvdata(&device->pdev.dev, device);
1500 if (platform_device_register(&device->pdev))
1501 goto out_free_netdev;
1502 SET_NETDEV_DEV(dev, &device->pdev.dev);
1503
1504 device->dev = dev;
1505
1506 *vp = ((struct vector_private)
1507 {
1508 .list = LIST_HEAD_INIT(vp->list),
1509 .dev = dev,
1510 .unit = n,
1511 .options = get_transport_options(def),
1512 .rx_irq = 0,
1513 .tx_irq = 0,
1514 .parsed = def,
1515 .max_packet = get_mtu(def) + ETH_HEADER_OTHER,
1516 /* TODO - we need to calculate headroom so that ip header
1517 * is 16 byte aligned all the time
1518 */
1519 .headroom = get_headroom(def),
1520 .form_header = NULL,
1521 .verify_header = NULL,
1522 .header_rxbuffer = NULL,
1523 .header_txbuffer = NULL,
1524 .header_size = 0,
1525 .rx_header_size = 0,
1526 .rexmit_scheduled = false,
1527 .opened = false,
1528 .transport_data = NULL,
1529 .in_write_poll = false,
1530 .coalesce = 2,
1531 .req_size = get_req_size(def),
1532 .in_error = false
1533 });
1534
1535 dev->features = dev->hw_features = (NETIF_F_SG | NETIF_F_FRAGLIST);
1536 tasklet_init(&vp->tx_poll, vector_tx_poll, (unsigned long)vp);
1537 INIT_WORK(&vp->reset_tx, vector_reset_tx);
1538
1539 timer_setup(&vp->tl, vector_timer_expire, 0);
1540 spin_lock_init(&vp->lock);
1541
1542 /* FIXME */
1543 dev->netdev_ops = &vector_netdev_ops;
1544 dev->ethtool_ops = &vector_net_ethtool_ops;
1545 dev->watchdog_timeo = (HZ >> 1);
1546 /* primary IRQ - fixme */
1547 dev->irq = 0; /* we will adjust this once opened */
1548
1549 rtnl_lock();
1550 err = register_netdevice(dev);
1551 rtnl_unlock();
1552 if (err)
1553 goto out_undo_user_init;
1554
1555 spin_lock(&vector_devices_lock);
1556 list_add(&device->list, &vector_devices);
1557 spin_unlock(&vector_devices_lock);
1558
1559 return;
1560
1561 out_undo_user_init:
1562 return;
1563 out_free_netdev:
1564 free_netdev(dev);
1565 out_free_device:
1566 kfree(device);
1567 }
1568
1569
1570
1571
1572 /*
1573 * Invoked late in the init
1574 */
1575
vector_init(void)1576 static int __init vector_init(void)
1577 {
1578 struct list_head *ele;
1579 struct vector_cmd_line_arg *def;
1580 struct arglist *parsed;
1581
1582 list_for_each(ele, &vec_cmd_line) {
1583 def = list_entry(ele, struct vector_cmd_line_arg, list);
1584 parsed = uml_parse_vector_ifspec(def->arguments);
1585 if (parsed != NULL)
1586 vector_eth_configure(def->unit, parsed);
1587 }
1588 return 0;
1589 }
1590
1591
1592 /* Invoked at initial argument parsing, only stores
1593 * arguments until a proper vector_init is called
1594 * later
1595 */
1596
vector_setup(char * str)1597 static int __init vector_setup(char *str)
1598 {
1599 char *error;
1600 int n, err;
1601 struct vector_cmd_line_arg *new;
1602
1603 err = vector_parse(str, &n, &str, &error);
1604 if (err) {
1605 printk(KERN_ERR "vector_setup - Couldn't parse '%s' : %s\n",
1606 str, error);
1607 return 1;
1608 }
1609 new = memblock_alloc(sizeof(*new), SMP_CACHE_BYTES);
1610 if (!new)
1611 panic("%s: Failed to allocate %zu bytes\n", __func__,
1612 sizeof(*new));
1613 INIT_LIST_HEAD(&new->list);
1614 new->unit = n;
1615 new->arguments = str;
1616 list_add_tail(&new->list, &vec_cmd_line);
1617 return 1;
1618 }
1619
1620 __setup("vec", vector_setup);
1621 __uml_help(vector_setup,
1622 "vec[0-9]+:<option>=<value>,<option>=<value>\n"
1623 " Configure a vector io network device.\n\n"
1624 );
1625
1626 late_initcall(vector_init);
1627
1628 static struct mc_device vector_mc = {
1629 .list = LIST_HEAD_INIT(vector_mc.list),
1630 .name = "vec",
1631 .config = vector_config,
1632 .get_config = NULL,
1633 .id = vector_id,
1634 .remove = vector_remove,
1635 };
1636
1637 #ifdef CONFIG_INET
vector_inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)1638 static int vector_inetaddr_event(
1639 struct notifier_block *this,
1640 unsigned long event,
1641 void *ptr)
1642 {
1643 return NOTIFY_DONE;
1644 }
1645
1646 static struct notifier_block vector_inetaddr_notifier = {
1647 .notifier_call = vector_inetaddr_event,
1648 };
1649
inet_register(void)1650 static void inet_register(void)
1651 {
1652 register_inetaddr_notifier(&vector_inetaddr_notifier);
1653 }
1654 #else
inet_register(void)1655 static inline void inet_register(void)
1656 {
1657 }
1658 #endif
1659
vector_net_init(void)1660 static int vector_net_init(void)
1661 {
1662 mconsole_register_dev(&vector_mc);
1663 inet_register();
1664 return 0;
1665 }
1666
1667 __initcall(vector_net_init);
1668
1669
1670
1671