?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / arch / um / drivers / vector_kern.c
blob0ff86391f77d072cf7c3b884bc40a6ce252fa0b3
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2017 - 2019 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 <linux/firmware.h>
25 #include <linux/fs.h>
26 #include <uapi/linux/filter.h>
27 #include <init.h>
28 #include <irq_kern.h>
29 #include <irq_user.h>
30 #include <net_kern.h>
31 #include <os.h>
32 #include "mconsole_kern.h"
33 #include "vector_user.h"
34 #include "vector_kern.h"
35
36 /*
37 * Adapted from network devices with the following major changes:
38 * All transports are static - simplifies the code significantly
39 * Multiple FDs/IRQs per device
40 * Vector IO optionally used for read/write, falling back to legacy
41 * based on configuration and/or availability
42 * Configuration is no longer positional - L2TPv3 and GRE require up to
43 * 10 parameters, passing this as positional is not fit for purpose.
44 * Only socket transports are supported
45 */
46
47
48 #define DRIVER_NAME "uml-vector"
49 #define DRIVER_VERSION "01"
50 struct vector_cmd_line_arg {
51 struct list_head list;
52 int unit;
53 char *arguments;
54 };
55
56 struct vector_device {
57 struct list_head list;
58 struct net_device *dev;
59 struct platform_device pdev;
60 int unit;
61 int opened;
62 };
63
64 static LIST_HEAD(vec_cmd_line);
65
66 static DEFINE_SPINLOCK(vector_devices_lock);
67 static LIST_HEAD(vector_devices);
68
69 static int driver_registered;
70
71 static void vector_eth_configure(int n, struct arglist *def);
72
73 /* Argument accessors to set variables (and/or set default values)
74 * mtu, buffer sizing, default headroom, etc
75 */
76
77 #define DEFAULT_HEADROOM 2
78 #define SAFETY_MARGIN 32
79 #define DEFAULT_VECTOR_SIZE 64
80 #define TX_SMALL_PACKET 128
81 #define MAX_IOV_SIZE (MAX_SKB_FRAGS + 1)
82 #define MAX_ITERATIONS 64
83
84 static const struct {
85 const char string[ETH_GSTRING_LEN];
86 } ethtool_stats_keys[] = {
87 { "rx_queue_max" },
88 { "rx_queue_running_average" },
89 { "tx_queue_max" },
90 { "tx_queue_running_average" },
91 { "rx_encaps_errors" },
92 { "tx_timeout_count" },
93 { "tx_restart_queue" },
94 { "tx_kicks" },
95 { "tx_flow_control_xon" },
96 { "tx_flow_control_xoff" },
97 { "rx_csum_offload_good" },
98 { "rx_csum_offload_errors"},
99 { "sg_ok"},
100 { "sg_linearized"},
101 };
102
103 #define VECTOR_NUM_STATS ARRAY_SIZE(ethtool_stats_keys)
104
105 static void vector_reset_stats(struct vector_private *vp)
106 {
107 vp->estats.rx_queue_max = 0;
108 vp->estats.rx_queue_running_average = 0;
109 vp->estats.tx_queue_max = 0;
110 vp->estats.tx_queue_running_average = 0;
111 vp->estats.rx_encaps_errors = 0;
112 vp->estats.tx_timeout_count = 0;
113 vp->estats.tx_restart_queue = 0;
114 vp->estats.tx_kicks = 0;
115 vp->estats.tx_flow_control_xon = 0;
116 vp->estats.tx_flow_control_xoff = 0;
117 vp->estats.sg_ok = 0;
118 vp->estats.sg_linearized = 0;
119 }
120
121 static int get_mtu(struct arglist *def)
122 {
123 char *mtu = uml_vector_fetch_arg(def, "mtu");
124 long result;
125
126 if (mtu != NULL) {
127 if (kstrtoul(mtu, 10, &result) == 0)
128 if ((result < (1 << 16) - 1) && (result >= 576))
129 return result;
130 }
131 return ETH_MAX_PACKET;
132 }
133
134 static char *get_bpf_file(struct arglist *def)
135 {
136 return uml_vector_fetch_arg(def, "bpffile");
137 }
138
139 static bool get_bpf_flash(struct arglist *def)
140 {
141 char *allow = uml_vector_fetch_arg(def, "bpfflash");
142 long result;
143
144 if (allow != NULL) {
145 if (kstrtoul(allow, 10, &result) == 0)
146 return (allow > 0);
147 }
148 return false;
149 }
150
151 static int get_depth(struct arglist *def)
152 {
153 char *mtu = uml_vector_fetch_arg(def, "depth");
154 long result;
155
156 if (mtu != NULL) {
157 if (kstrtoul(mtu, 10, &result) == 0)
158 return result;
159 }
160 return DEFAULT_VECTOR_SIZE;
161 }
162
163 static int get_headroom(struct arglist *def)
164 {
165 char *mtu = uml_vector_fetch_arg(def, "headroom");
166 long result;
167
168 if (mtu != NULL) {
169 if (kstrtoul(mtu, 10, &result) == 0)
170 return result;
171 }
172 return DEFAULT_HEADROOM;
173 }
174
175 static int get_req_size(struct arglist *def)
176 {
177 char *gro = uml_vector_fetch_arg(def, "gro");
178 long result;
179
180 if (gro != NULL) {
181 if (kstrtoul(gro, 10, &result) == 0) {
182 if (result > 0)
183 return 65536;
184 }
185 }
186 return get_mtu(def) + ETH_HEADER_OTHER +
187 get_headroom(def) + SAFETY_MARGIN;
188 }
189
190
191 static int get_transport_options(struct arglist *def)
192 {
193 char *transport = uml_vector_fetch_arg(def, "transport");
194 char *vector = uml_vector_fetch_arg(def, "vec");
195
196 int vec_rx = VECTOR_RX;
197 int vec_tx = VECTOR_TX;
198 long parsed;
199 int result = 0;
200
201 if (vector != NULL) {
202 if (kstrtoul(vector, 10, &parsed) == 0) {
203 if (parsed == 0) {
204 vec_rx = 0;
205 vec_tx = 0;
206 }
207 }
208 }
209
210 if (get_bpf_flash(def))
211 result = VECTOR_BPF_FLASH;
212
213 if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
214 return result;
215 if (strncmp(transport, TRANS_HYBRID, TRANS_HYBRID_LEN) == 0)
216 return (result | vec_rx | VECTOR_BPF);
217 if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
218 return (result | vec_rx | vec_tx | VECTOR_QDISC_BYPASS);
219 return (result | vec_rx | vec_tx);
220 }
221
222
223 /* A mini-buffer for packet drop read
224 * All of our supported transports are datagram oriented and we always
225 * read using recvmsg or recvmmsg. If we pass a buffer which is smaller
226 * than the packet size it still counts as full packet read and will
227 * clean the incoming stream to keep sigio/epoll happy
228 */
229
230 #define DROP_BUFFER_SIZE 32
231
232 static char *drop_buffer;
233
234 /* Array backed queues optimized for bulk enqueue/dequeue and
235 * 1:N (small values of N) or 1:1 enqueuer/dequeuer ratios.
236 * For more details and full design rationale see
237 * http://foswiki.cambridgegreys.com/Main/EatYourTailAndEnjoyIt
238 */
239
240
241 /*
242 * Advance the mmsg queue head by n = advance. Resets the queue to
243 * maximum enqueue/dequeue-at-once capacity if possible. Called by
244 * dequeuers. Caller must hold the head_lock!
245 */
246
247 static int vector_advancehead(struct vector_queue *qi, int advance)
248 {
249 int queue_depth;
250
251 qi->head =
252 (qi->head + advance)
253 % qi->max_depth;
254
255
256 spin_lock(&qi->tail_lock);
257 qi->queue_depth -= advance;
258
259 /* we are at 0, use this to
260 * reset head and tail so we can use max size vectors
261 */
262
263 if (qi->queue_depth == 0) {
264 qi->head = 0;
265 qi->tail = 0;
266 }
267 queue_depth = qi->queue_depth;
268 spin_unlock(&qi->tail_lock);
269 return queue_depth;
270 }
271
272 /* Advance the queue tail by n = advance.
273 * This is called by enqueuers which should hold the
274 * head lock already
275 */
276
277 static int vector_advancetail(struct vector_queue *qi, int advance)
278 {
279 int queue_depth;
280
281 qi->tail =
282 (qi->tail + advance)
283 % qi->max_depth;
284 spin_lock(&qi->head_lock);
285 qi->queue_depth += advance;
286 queue_depth = qi->queue_depth;
287 spin_unlock(&qi->head_lock);
288 return queue_depth;
289 }
290
291 static int prep_msg(struct vector_private *vp,
292 struct sk_buff *skb,
293 struct iovec *iov)
294 {
295 int iov_index = 0;
296 int nr_frags, frag;
297 skb_frag_t *skb_frag;
298
299 nr_frags = skb_shinfo(skb)->nr_frags;
300 if (nr_frags > MAX_IOV_SIZE) {
301 if (skb_linearize(skb) != 0)
302 goto drop;
303 }
304 if (vp->header_size > 0) {
305 iov[iov_index].iov_len = vp->header_size;
306 vp->form_header(iov[iov_index].iov_base, skb, vp);
307 iov_index++;
308 }
309 iov[iov_index].iov_base = skb->data;
310 if (nr_frags > 0) {
311 iov[iov_index].iov_len = skb->len - skb->data_len;
312 vp->estats.sg_ok++;
313 } else
314 iov[iov_index].iov_len = skb->len;
315 iov_index++;
316 for (frag = 0; frag < nr_frags; frag++) {
317 skb_frag = &skb_shinfo(skb)->frags[frag];
318 iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
319 iov[iov_index].iov_len = skb_frag_size(skb_frag);
320 iov_index++;
321 }
322 return iov_index;
323 drop:
324 return -1;
325 }
326 /*
327 * Generic vector enqueue with support for forming headers using transport
328 * specific callback. Allows GRE, L2TPv3, RAW and other transports
329 * to use a common enqueue procedure in vector mode
330 */
331
332 static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
333 {
334 struct vector_private *vp = netdev_priv(qi->dev);
335 int queue_depth;
336 int packet_len;
337 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
338 int iov_count;
339
340 spin_lock(&qi->tail_lock);
341 spin_lock(&qi->head_lock);
342 queue_depth = qi->queue_depth;
343 spin_unlock(&qi->head_lock);
344
345 if (skb)
346 packet_len = skb->len;
347
348 if (queue_depth < qi->max_depth) {
349
350 *(qi->skbuff_vector + qi->tail) = skb;
351 mmsg_vector += qi->tail;
352 iov_count = prep_msg(
353 vp,
354 skb,
355 mmsg_vector->msg_hdr.msg_iov
356 );
357 if (iov_count < 1)
358 goto drop;
359 mmsg_vector->msg_hdr.msg_iovlen = iov_count;
360 mmsg_vector->msg_hdr.msg_name = vp->fds->remote_addr;
361 mmsg_vector->msg_hdr.msg_namelen = vp->fds->remote_addr_size;
362 queue_depth = vector_advancetail(qi, 1);
363 } else
364 goto drop;
365 spin_unlock(&qi->tail_lock);
366 return queue_depth;
367 drop:
368 qi->dev->stats.tx_dropped++;
369 if (skb != NULL) {
370 packet_len = skb->len;
371 dev_consume_skb_any(skb);
372 netdev_completed_queue(qi->dev, 1, packet_len);
373 }
374 spin_unlock(&qi->tail_lock);
375 return queue_depth;
376 }
377
378 static int consume_vector_skbs(struct vector_queue *qi, int count)
379 {
380 struct sk_buff *skb;
381 int skb_index;
382 int bytes_compl = 0;
383
384 for (skb_index = qi->head; skb_index < qi->head + count; skb_index++) {
385 skb = *(qi->skbuff_vector + skb_index);
386 /* mark as empty to ensure correct destruction if
387 * needed
388 */
389 bytes_compl += skb->len;
390 *(qi->skbuff_vector + skb_index) = NULL;
391 dev_consume_skb_any(skb);
392 }
393 qi->dev->stats.tx_bytes += bytes_compl;
394 qi->dev->stats.tx_packets += count;
395 netdev_completed_queue(qi->dev, count, bytes_compl);
396 return vector_advancehead(qi, count);
397 }
398
399 /*
400 * Generic vector deque via sendmmsg with support for forming headers
401 * using transport specific callback. Allows GRE, L2TPv3, RAW and
402 * other transports to use a common dequeue procedure in vector mode
403 */
404
405
406 static int vector_send(struct vector_queue *qi)
407 {
408 struct vector_private *vp = netdev_priv(qi->dev);
409 struct mmsghdr *send_from;
410 int result = 0, send_len, queue_depth = qi->max_depth;
411
412 if (spin_trylock(&qi->head_lock)) {
413 if (spin_trylock(&qi->tail_lock)) {
414 /* update queue_depth to current value */
415 queue_depth = qi->queue_depth;
416 spin_unlock(&qi->tail_lock);
417 while (queue_depth > 0) {
418 /* Calculate the start of the vector */
419 send_len = queue_depth;
420 send_from = qi->mmsg_vector;
421 send_from += qi->head;
422 /* Adjust vector size if wraparound */
423 if (send_len + qi->head > qi->max_depth)
424 send_len = qi->max_depth - qi->head;
425 /* Try to TX as many packets as possible */
426 if (send_len > 0) {
427 result = uml_vector_sendmmsg(
428 vp->fds->tx_fd,
429 send_from,
430 send_len,
431 0
432 );
433 vp->in_write_poll =
434 (result != send_len);
435 }
436 /* For some of the sendmmsg error scenarios
437 * we may end being unsure in the TX success
438 * for all packets. It is safer to declare
439 * them all TX-ed and blame the network.
440 */
441 if (result < 0) {
442 if (net_ratelimit())
443 netdev_err(vp->dev, "sendmmsg err=%i\n",
444 result);
445 vp->in_error = true;
446 result = send_len;
447 }
448 if (result > 0) {
449 queue_depth =
450 consume_vector_skbs(qi, result);
451 /* This is equivalent to an TX IRQ.
452 * Restart the upper layers to feed us
453 * more packets.
454 */
455 if (result > vp->estats.tx_queue_max)
456 vp->estats.tx_queue_max = result;
457 vp->estats.tx_queue_running_average =
458 (vp->estats.tx_queue_running_average + result) >> 1;
459 }
460 netif_trans_update(qi->dev);
461 netif_wake_queue(qi->dev);
462 /* if TX is busy, break out of the send loop,
463 * poll write IRQ will reschedule xmit for us
464 */
465 if (result != send_len) {
466 vp->estats.tx_restart_queue++;
467 break;
468 }
469 }
470 }
471 spin_unlock(&qi->head_lock);
472 } else {
473 tasklet_schedule(&vp->tx_poll);
474 }
475 return queue_depth;
476 }
477
478 /* Queue destructor. Deliberately stateless so we can use
479 * it in queue cleanup if initialization fails.
480 */
481
482 static void destroy_queue(struct vector_queue *qi)
483 {
484 int i;
485 struct iovec *iov;
486 struct vector_private *vp = netdev_priv(qi->dev);
487 struct mmsghdr *mmsg_vector;
488
489 if (qi == NULL)
490 return;
491 /* deallocate any skbuffs - we rely on any unused to be
492 * set to NULL.
493 */
494 if (qi->skbuff_vector != NULL) {
495 for (i = 0; i < qi->max_depth; i++) {
496 if (*(qi->skbuff_vector + i) != NULL)
497 dev_kfree_skb_any(*(qi->skbuff_vector + i));
498 }
499 kfree(qi->skbuff_vector);
500 }
501 /* deallocate matching IOV structures including header buffs */
502 if (qi->mmsg_vector != NULL) {
503 mmsg_vector = qi->mmsg_vector;
504 for (i = 0; i < qi->max_depth; i++) {
505 iov = mmsg_vector->msg_hdr.msg_iov;
506 if (iov != NULL) {
507 if ((vp->header_size > 0) &&
508 (iov->iov_base != NULL))
509 kfree(iov->iov_base);
510 kfree(iov);
511 }
512 mmsg_vector++;
513 }
514 kfree(qi->mmsg_vector);
515 }
516 kfree(qi);
517 }
518
519 /*
520 * Queue constructor. Create a queue with a given side.
521 */
522 static struct vector_queue *create_queue(
523 struct vector_private *vp,
524 int max_size,
525 int header_size,
526 int num_extra_frags)
527 {
528 struct vector_queue *result;
529 int i;
530 struct iovec *iov;
531 struct mmsghdr *mmsg_vector;
532
533 result = kmalloc(sizeof(struct vector_queue), GFP_KERNEL);
534 if (result == NULL)
535 return NULL;
536 result->max_depth = max_size;
537 result->dev = vp->dev;
538 result->mmsg_vector = kmalloc(
539 (sizeof(struct mmsghdr) * max_size), GFP_KERNEL);
540 if (result->mmsg_vector == NULL)
541 goto out_mmsg_fail;
542 result->skbuff_vector = kmalloc(
543 (sizeof(void *) * max_size), GFP_KERNEL);
544 if (result->skbuff_vector == NULL)
545 goto out_skb_fail;
546
547 /* further failures can be handled safely by destroy_queue*/
548
549 mmsg_vector = result->mmsg_vector;
550 for (i = 0; i < max_size; i++) {
551 /* Clear all pointers - we use non-NULL as marking on
552 * what to free on destruction
553 */
554 *(result->skbuff_vector + i) = NULL;
555 mmsg_vector->msg_hdr.msg_iov = NULL;
556 mmsg_vector++;
557 }
558 mmsg_vector = result->mmsg_vector;
559 result->max_iov_frags = num_extra_frags;
560 for (i = 0; i < max_size; i++) {
561 if (vp->header_size > 0)
562 iov = kmalloc_array(3 + num_extra_frags,
563 sizeof(struct iovec),
564 GFP_KERNEL
565 );
566 else
567 iov = kmalloc_array(2 + num_extra_frags,
568 sizeof(struct iovec),
569 GFP_KERNEL
570 );
571 if (iov == NULL)
572 goto out_fail;
573 mmsg_vector->msg_hdr.msg_iov = iov;
574 mmsg_vector->msg_hdr.msg_iovlen = 1;
575 mmsg_vector->msg_hdr.msg_control = NULL;
576 mmsg_vector->msg_hdr.msg_controllen = 0;
577 mmsg_vector->msg_hdr.msg_flags = MSG_DONTWAIT;
578 mmsg_vector->msg_hdr.msg_name = NULL;
579 mmsg_vector->msg_hdr.msg_namelen = 0;
580 if (vp->header_size > 0) {
581 iov->iov_base = kmalloc(header_size, GFP_KERNEL);
582 if (iov->iov_base == NULL)
583 goto out_fail;
584 iov->iov_len = header_size;
585 mmsg_vector->msg_hdr.msg_iovlen = 2;
586 iov++;
587 }
588 iov->iov_base = NULL;
589 iov->iov_len = 0;
590 mmsg_vector++;
591 }
592 spin_lock_init(&result->head_lock);
593 spin_lock_init(&result->tail_lock);
594 result->queue_depth = 0;
595 result->head = 0;
596 result->tail = 0;
597 return result;
598 out_skb_fail:
599 kfree(result->mmsg_vector);
600 out_mmsg_fail:
601 kfree(result);
602 return NULL;
603 out_fail:
604 destroy_queue(result);
605 return NULL;
606 }
607
608 /*
609 * We do not use the RX queue as a proper wraparound queue for now
610 * This is not necessary because the consumption via netif_rx()
611 * happens in-line. While we can try using the return code of
612 * netif_rx() for flow control there are no drivers doing this today.
613 * For this RX specific use we ignore the tail/head locks and
614 * just read into a prepared queue filled with skbuffs.
615 */
616
617 static struct sk_buff *prep_skb(
618 struct vector_private *vp,
619 struct user_msghdr *msg)
620 {
621 int linear = vp->max_packet + vp->headroom + SAFETY_MARGIN;
622 struct sk_buff *result;
623 int iov_index = 0, len;
624 struct iovec *iov = msg->msg_iov;
625 int err, nr_frags, frag;
626 skb_frag_t *skb_frag;
627
628 if (vp->req_size <= linear)
629 len = linear;
630 else
631 len = vp->req_size;
632 result = alloc_skb_with_frags(
633 linear,
634 len - vp->max_packet,
635 3,
636 &err,
637 GFP_ATOMIC
638 );
639 if (vp->header_size > 0)
640 iov_index++;
641 if (result == NULL) {
642 iov[iov_index].iov_base = NULL;
643 iov[iov_index].iov_len = 0;
644 goto done;
645 }
646 skb_reserve(result, vp->headroom);
647 result->dev = vp->dev;
648 skb_put(result, vp->max_packet);
649 result->data_len = len - vp->max_packet;
650 result->len += len - vp->max_packet;
651 skb_reset_mac_header(result);
652 result->ip_summed = CHECKSUM_NONE;
653 iov[iov_index].iov_base = result->data;
654 iov[iov_index].iov_len = vp->max_packet;
655 iov_index++;
656
657 nr_frags = skb_shinfo(result)->nr_frags;
658 for (frag = 0; frag < nr_frags; frag++) {
659 skb_frag = &skb_shinfo(result)->frags[frag];
660 iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
661 if (iov[iov_index].iov_base != NULL)
662 iov[iov_index].iov_len = skb_frag_size(skb_frag);
663 else
664 iov[iov_index].iov_len = 0;
665 iov_index++;
666 }
667 done:
668 msg->msg_iovlen = iov_index;
669 return result;
670 }
671
672
673 /* Prepare queue for recvmmsg one-shot rx - fill with fresh sk_buffs*/
674
675 static void prep_queue_for_rx(struct vector_queue *qi)
676 {
677 struct vector_private *vp = netdev_priv(qi->dev);
678 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
679 void **skbuff_vector = qi->skbuff_vector;
680 int i;
681
682 if (qi->queue_depth == 0)
683 return;
684 for (i = 0; i < qi->queue_depth; i++) {
685 /* it is OK if allocation fails - recvmmsg with NULL data in
686 * iov argument still performs an RX, just drops the packet
687 * This allows us stop faffing around with a "drop buffer"
688 */
689
690 *skbuff_vector = prep_skb(vp, &mmsg_vector->msg_hdr);
691 skbuff_vector++;
692 mmsg_vector++;
693 }
694 qi->queue_depth = 0;
695 }
696
697 static struct vector_device *find_device(int n)
698 {
699 struct vector_device *device;
700 struct list_head *ele;
701
702 spin_lock(&vector_devices_lock);
703 list_for_each(ele, &vector_devices) {
704 device = list_entry(ele, struct vector_device, list);
705 if (device->unit == n)
706 goto out;
707 }
708 device = NULL;
709 out:
710 spin_unlock(&vector_devices_lock);
711 return device;
712 }
713
714 static int vector_parse(char *str, int *index_out, char **str_out,
715 char **error_out)
716 {
717 int n, len, err;
718 char *start = str;
719
720 len = strlen(str);
721
722 while ((*str != ':') && (strlen(str) > 1))
723 str++;
724 if (*str != ':') {
725 *error_out = "Expected ':' after device number";
726 return -EINVAL;
727 }
728 *str = '\0';
729
730 err = kstrtouint(start, 0, &n);
731 if (err < 0) {
732 *error_out = "Bad device number";
733 return err;
734 }
735
736 str++;
737 if (find_device(n)) {
738 *error_out = "Device already configured";
739 return -EINVAL;
740 }
741
742 *index_out = n;
743 *str_out = str;
744 return 0;
745 }
746
747 static int vector_config(char *str, char **error_out)
748 {
749 int err, n;
750 char *params;
751 struct arglist *parsed;
752
753 err = vector_parse(str, &n, &params, error_out);
754 if (err != 0)
755 return err;
756
757 /* This string is broken up and the pieces used by the underlying
758 * driver. We should copy it to make sure things do not go wrong
759 * later.
760 */
761
762 params = kstrdup(params, GFP_KERNEL);
763 if (params == NULL) {
764 *error_out = "vector_config failed to strdup string";
765 return -ENOMEM;
766 }
767
768 parsed = uml_parse_vector_ifspec(params);
769
770 if (parsed == NULL) {
771 *error_out = "vector_config failed to parse parameters";
772 return -EINVAL;
773 }
774
775 vector_eth_configure(n, parsed);
776 return 0;
777 }
778
779 static int vector_id(char **str, int *start_out, int *end_out)
780 {
781 char *end;
782 int n;
783
784 n = simple_strtoul(*str, &end, 0);
785 if ((*end != '\0') || (end == *str))
786 return -1;
787
788 *start_out = n;
789 *end_out = n;
790 *str = end;
791 return n;
792 }
793
794 static int vector_remove(int n, char **error_out)
795 {
796 struct vector_device *vec_d;
797 struct net_device *dev;
798 struct vector_private *vp;
799
800 vec_d = find_device(n);
801 if (vec_d == NULL)
802 return -ENODEV;
803 dev = vec_d->dev;
804 vp = netdev_priv(dev);
805 if (vp->fds != NULL)
806 return -EBUSY;
807 unregister_netdev(dev);
808 platform_device_unregister(&vec_d->pdev);
809 return 0;
810 }
811
812 /*
813 * There is no shared per-transport initialization code, so
814 * we will just initialize each interface one by one and
815 * add them to a list
816 */
817
818 static struct platform_driver uml_net_driver = {
819 .driver = {
820 .name = DRIVER_NAME,
821 },
822 };
823
824
825 static void vector_device_release(struct device *dev)
826 {
827 struct vector_device *device = dev_get_drvdata(dev);
828 struct net_device *netdev = device->dev;
829
830 list_del(&device->list);
831 kfree(device);
832 free_netdev(netdev);
833 }
834
835 /* Bog standard recv using recvmsg - not used normally unless the user
836 * explicitly specifies not to use recvmmsg vector RX.
837 */
838
839 static int vector_legacy_rx(struct vector_private *vp)
840 {
841 int pkt_len;
842 struct user_msghdr hdr;
843 struct iovec iov[2 + MAX_IOV_SIZE]; /* header + data use case only */
844 int iovpos = 0;
845 struct sk_buff *skb;
846 int header_check;
847
848 hdr.msg_name = NULL;
849 hdr.msg_namelen = 0;
850 hdr.msg_iov = (struct iovec *) &iov;
851 hdr.msg_control = NULL;
852 hdr.msg_controllen = 0;
853 hdr.msg_flags = 0;
854
855 if (vp->header_size > 0) {
856 iov[0].iov_base = vp->header_rxbuffer;
857 iov[0].iov_len = vp->header_size;
858 }
859
860 skb = prep_skb(vp, &hdr);
861
862 if (skb == NULL) {
863 /* Read a packet into drop_buffer and don't do
864 * anything with it.
865 */
866 iov[iovpos].iov_base = drop_buffer;
867 iov[iovpos].iov_len = DROP_BUFFER_SIZE;
868 hdr.msg_iovlen = 1;
869 vp->dev->stats.rx_dropped++;
870 }
871
872 pkt_len = uml_vector_recvmsg(vp->fds->rx_fd, &hdr, 0);
873 if (pkt_len < 0) {
874 vp->in_error = true;
875 return pkt_len;
876 }
877
878 if (skb != NULL) {
879 if (pkt_len > vp->header_size) {
880 if (vp->header_size > 0) {
881 header_check = vp->verify_header(
882 vp->header_rxbuffer, skb, vp);
883 if (header_check < 0) {
884 dev_kfree_skb_irq(skb);
885 vp->dev->stats.rx_dropped++;
886 vp->estats.rx_encaps_errors++;
887 return 0;
888 }
889 if (header_check > 0) {
890 vp->estats.rx_csum_offload_good++;
891 skb->ip_summed = CHECKSUM_UNNECESSARY;
892 }
893 }
894 pskb_trim(skb, pkt_len - vp->rx_header_size);
895 skb->protocol = eth_type_trans(skb, skb->dev);
896 vp->dev->stats.rx_bytes += skb->len;
897 vp->dev->stats.rx_packets++;
898 netif_rx(skb);
899 } else {
900 dev_kfree_skb_irq(skb);
901 }
902 }
903 return pkt_len;
904 }
905
906 /*
907 * Packet at a time TX which falls back to vector TX if the
908 * underlying transport is busy.
909 */
910
911
912
913 static int writev_tx(struct vector_private *vp, struct sk_buff *skb)
914 {
915 struct iovec iov[3 + MAX_IOV_SIZE];
916 int iov_count, pkt_len = 0;
917
918 iov[0].iov_base = vp->header_txbuffer;
919 iov_count = prep_msg(vp, skb, (struct iovec *) &iov);
920
921 if (iov_count < 1)
922 goto drop;
923
924 pkt_len = uml_vector_writev(
925 vp->fds->tx_fd,
926 (struct iovec *) &iov,
927 iov_count
928 );
929
930 if (pkt_len < 0)
931 goto drop;
932
933 netif_trans_update(vp->dev);
934 netif_wake_queue(vp->dev);
935
936 if (pkt_len > 0) {
937 vp->dev->stats.tx_bytes += skb->len;
938 vp->dev->stats.tx_packets++;
939 } else {
940 vp->dev->stats.tx_dropped++;
941 }
942 consume_skb(skb);
943 return pkt_len;
944 drop:
945 vp->dev->stats.tx_dropped++;
946 consume_skb(skb);
947 if (pkt_len < 0)
948 vp->in_error = true;
949 return pkt_len;
950 }
951
952 /*
953 * Receive as many messages as we can in one call using the special
954 * mmsg vector matched to an skb vector which we prepared earlier.
955 */
956
957 static int vector_mmsg_rx(struct vector_private *vp)
958 {
959 int packet_count, i;
960 struct vector_queue *qi = vp->rx_queue;
961 struct sk_buff *skb;
962 struct mmsghdr *mmsg_vector = qi->mmsg_vector;
963 void **skbuff_vector = qi->skbuff_vector;
964 int header_check;
965
966 /* Refresh the vector and make sure it is with new skbs and the
967 * iovs are updated to point to them.
968 */
969
970 prep_queue_for_rx(qi);
971
972 /* Fire the Lazy Gun - get as many packets as we can in one go. */
973
974 packet_count = uml_vector_recvmmsg(
975 vp->fds->rx_fd, qi->mmsg_vector, qi->max_depth, 0);
976
977 if (packet_count < 0)
978 vp->in_error = true;
979
980 if (packet_count <= 0)
981 return packet_count;
982
983 /* We treat packet processing as enqueue, buffer refresh as dequeue
984 * The queue_depth tells us how many buffers have been used and how
985 * many do we need to prep the next time prep_queue_for_rx() is called.
986 */
987
988 qi->queue_depth = packet_count;
989
990 for (i = 0; i < packet_count; i++) {
991 skb = (*skbuff_vector);
992 if (mmsg_vector->msg_len > vp->header_size) {
993 if (vp->header_size > 0) {
994 header_check = vp->verify_header(
995 mmsg_vector->msg_hdr.msg_iov->iov_base,
996 skb,
997 vp
998 );
999 if (header_check < 0) {
1000 /* Overlay header failed to verify - discard.
1001 * We can actually keep this skb and reuse it,
1002 * but that will make the prep logic too
1003 * complex.
1004 */
1005 dev_kfree_skb_irq(skb);
1006 vp->estats.rx_encaps_errors++;
1007 continue;
1008 }
1009 if (header_check > 0) {
1010 vp->estats.rx_csum_offload_good++;
1011 skb->ip_summed = CHECKSUM_UNNECESSARY;
1012 }
1013 }
1014 pskb_trim(skb,
1015 mmsg_vector->msg_len - vp->rx_header_size);
1016 skb->protocol = eth_type_trans(skb, skb->dev);
1017 /*
1018 * We do not need to lock on updating stats here
1019 * The interrupt loop is non-reentrant.
1020 */
1021 vp->dev->stats.rx_bytes += skb->len;
1022 vp->dev->stats.rx_packets++;
1023 netif_rx(skb);
1024 } else {
1025 /* Overlay header too short to do anything - discard.
1026 * We can actually keep this skb and reuse it,
1027 * but that will make the prep logic too complex.
1028 */
1029 if (skb != NULL)
1030 dev_kfree_skb_irq(skb);
1031 }
1032 (*skbuff_vector) = NULL;
1033 /* Move to the next buffer element */
1034 mmsg_vector++;
1035 skbuff_vector++;
1036 }
1037 if (packet_count > 0) {
1038 if (vp->estats.rx_queue_max < packet_count)
1039 vp->estats.rx_queue_max = packet_count;
1040 vp->estats.rx_queue_running_average =
1041 (vp->estats.rx_queue_running_average + packet_count) >> 1;
1042 }
1043 return packet_count;
1044 }
1045
1046 static void vector_rx(struct vector_private *vp)
1047 {
1048 int err;
1049 int iter = 0;
1050
1051 if ((vp->options & VECTOR_RX) > 0)
1052 while (((err = vector_mmsg_rx(vp)) > 0) && (iter < MAX_ITERATIONS))
1053 iter++;
1054 else
1055 while (((err = vector_legacy_rx(vp)) > 0) && (iter < MAX_ITERATIONS))
1056 iter++;
1057 if ((err != 0) && net_ratelimit())
1058 netdev_err(vp->dev, "vector_rx: error(%d)\n", err);
1059 if (iter == MAX_ITERATIONS)
1060 netdev_err(vp->dev, "vector_rx: device stuck, remote end may have closed the connection\n");
1061 }
1062
1063 static int vector_net_start_xmit(struct sk_buff *skb, struct net_device *dev)
1064 {
1065 struct vector_private *vp = netdev_priv(dev);
1066 int queue_depth = 0;
1067
1068 if (vp->in_error) {
1069 deactivate_fd(vp->fds->rx_fd, vp->rx_irq);
1070 if ((vp->fds->rx_fd != vp->fds->tx_fd) && (vp->tx_irq != 0))
1071 deactivate_fd(vp->fds->tx_fd, vp->tx_irq);
1072 return NETDEV_TX_BUSY;
1073 }
1074
1075 if ((vp->options & VECTOR_TX) == 0) {
1076 writev_tx(vp, skb);
1077 return NETDEV_TX_OK;
1078 }
1079
1080 /* We do BQL only in the vector path, no point doing it in
1081 * packet at a time mode as there is no device queue
1082 */
1083
1084 netdev_sent_queue(vp->dev, skb->len);
1085 queue_depth = vector_enqueue(vp->tx_queue, skb);
1086
1087 /* if the device queue is full, stop the upper layers and
1088 * flush it.
1089 */
1090
1091 if (queue_depth >= vp->tx_queue->max_depth - 1) {
1092 vp->estats.tx_kicks++;
1093 netif_stop_queue(dev);
1094 vector_send(vp->tx_queue);
1095 return NETDEV_TX_OK;
1096 }
1097 if (netdev_xmit_more()) {
1098 mod_timer(&vp->tl, vp->coalesce);
1099 return NETDEV_TX_OK;
1100 }
1101 if (skb->len < TX_SMALL_PACKET) {
1102 vp->estats.tx_kicks++;
1103 vector_send(vp->tx_queue);
1104 } else
1105 tasklet_schedule(&vp->tx_poll);
1106 return NETDEV_TX_OK;
1107 }
1108
1109 static irqreturn_t vector_rx_interrupt(int irq, void *dev_id)
1110 {
1111 struct net_device *dev = dev_id;
1112 struct vector_private *vp = netdev_priv(dev);
1113
1114 if (!netif_running(dev))
1115 return IRQ_NONE;
1116 vector_rx(vp);
1117 return IRQ_HANDLED;
1118
1119 }
1120
1121 static irqreturn_t vector_tx_interrupt(int irq, void *dev_id)
1122 {
1123 struct net_device *dev = dev_id;
1124 struct vector_private *vp = netdev_priv(dev);
1125
1126 if (!netif_running(dev))
1127 return IRQ_NONE;
1128 /* We need to pay attention to it only if we got
1129 * -EAGAIN or -ENOBUFFS from sendmmsg. Otherwise
1130 * we ignore it. In the future, it may be worth
1131 * it to improve the IRQ controller a bit to make
1132 * tweaking the IRQ mask less costly
1133 */
1134
1135 if (vp->in_write_poll)
1136 tasklet_schedule(&vp->tx_poll);
1137 return IRQ_HANDLED;
1138
1139 }
1140
1141 static int irq_rr;
1142
1143 static int vector_net_close(struct net_device *dev)
1144 {
1145 struct vector_private *vp = netdev_priv(dev);
1146 unsigned long flags;
1147
1148 netif_stop_queue(dev);
1149 del_timer(&vp->tl);
1150
1151 if (vp->fds == NULL)
1152 return 0;
1153
1154 /* Disable and free all IRQS */
1155 if (vp->rx_irq > 0) {
1156 um_free_irq(vp->rx_irq, dev);
1157 vp->rx_irq = 0;
1158 }
1159 if (vp->tx_irq > 0) {
1160 um_free_irq(vp->tx_irq, dev);
1161 vp->tx_irq = 0;
1162 }
1163 tasklet_kill(&vp->tx_poll);
1164 if (vp->fds->rx_fd > 0) {
1165 if (vp->bpf)
1166 uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
1167 os_close_file(vp->fds->rx_fd);
1168 vp->fds->rx_fd = -1;
1169 }
1170 if (vp->fds->tx_fd > 0) {
1171 os_close_file(vp->fds->tx_fd);
1172 vp->fds->tx_fd = -1;
1173 }
1174 if (vp->bpf != NULL)
1175 kfree(vp->bpf->filter);
1176 kfree(vp->bpf);
1177 vp->bpf = NULL;
1178 kfree(vp->fds->remote_addr);
1179 kfree(vp->transport_data);
1180 kfree(vp->header_rxbuffer);
1181 kfree(vp->header_txbuffer);
1182 if (vp->rx_queue != NULL)
1183 destroy_queue(vp->rx_queue);
1184 if (vp->tx_queue != NULL)
1185 destroy_queue(vp->tx_queue);
1186 kfree(vp->fds);
1187 vp->fds = NULL;
1188 spin_lock_irqsave(&vp->lock, flags);
1189 vp->opened = false;
1190 vp->in_error = false;
1191 spin_unlock_irqrestore(&vp->lock, flags);
1192 return 0;
1193 }
1194
1195 /* TX tasklet */
1196
1197 static void vector_tx_poll(unsigned long data)
1198 {
1199 struct vector_private *vp = (struct vector_private *)data;
1200
1201 vp->estats.tx_kicks++;
1202 vector_send(vp->tx_queue);
1203 }
1204 static void vector_reset_tx(struct work_struct *work)
1205 {
1206 struct vector_private *vp =
1207 container_of(work, struct vector_private, reset_tx);
1208 netdev_reset_queue(vp->dev);
1209 netif_start_queue(vp->dev);
1210 netif_wake_queue(vp->dev);
1211 }
1212
1213 static int vector_net_open(struct net_device *dev)
1214 {
1215 struct vector_private *vp = netdev_priv(dev);
1216 unsigned long flags;
1217 int err = -EINVAL;
1218 struct vector_device *vdevice;
1219
1220 spin_lock_irqsave(&vp->lock, flags);
1221 if (vp->opened) {
1222 spin_unlock_irqrestore(&vp->lock, flags);
1223 return -ENXIO;
1224 }
1225 vp->opened = true;
1226 spin_unlock_irqrestore(&vp->lock, flags);
1227
1228 vp->bpf = uml_vector_user_bpf(get_bpf_file(vp->parsed));
1229
1230 vp->fds = uml_vector_user_open(vp->unit, vp->parsed);
1231
1232 if (vp->fds == NULL)
1233 goto out_close;
1234
1235 if (build_transport_data(vp) < 0)
1236 goto out_close;
1237
1238 if ((vp->options & VECTOR_RX) > 0) {
1239 vp->rx_queue = create_queue(
1240 vp,
1241 get_depth(vp->parsed),
1242 vp->rx_header_size,
1243 MAX_IOV_SIZE
1244 );
1245 vp->rx_queue->queue_depth = get_depth(vp->parsed);
1246 } else {
1247 vp->header_rxbuffer = kmalloc(
1248 vp->rx_header_size,
1249 GFP_KERNEL
1250 );
1251 if (vp->header_rxbuffer == NULL)
1252 goto out_close;
1253 }
1254 if ((vp->options & VECTOR_TX) > 0) {
1255 vp->tx_queue = create_queue(
1256 vp,
1257 get_depth(vp->parsed),
1258 vp->header_size,
1259 MAX_IOV_SIZE
1260 );
1261 } else {
1262 vp->header_txbuffer = kmalloc(vp->header_size, GFP_KERNEL);
1263 if (vp->header_txbuffer == NULL)
1264 goto out_close;
1265 }
1266
1267 /* READ IRQ */
1268 err = um_request_irq(
1269 irq_rr + VECTOR_BASE_IRQ, vp->fds->rx_fd,
1270 IRQ_READ, vector_rx_interrupt,
1271 IRQF_SHARED, dev->name, dev);
1272 if (err != 0) {
1273 netdev_err(dev, "vector_open: failed to get rx irq(%d)\n", err);
1274 err = -ENETUNREACH;
1275 goto out_close;
1276 }
1277 vp->rx_irq = irq_rr + VECTOR_BASE_IRQ;
1278 dev->irq = irq_rr + VECTOR_BASE_IRQ;
1279 irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
1280
1281 /* WRITE IRQ - we need it only if we have vector TX */
1282 if ((vp->options & VECTOR_TX) > 0) {
1283 err = um_request_irq(
1284 irq_rr + VECTOR_BASE_IRQ, vp->fds->tx_fd,
1285 IRQ_WRITE, vector_tx_interrupt,
1286 IRQF_SHARED, dev->name, dev);
1287 if (err != 0) {
1288 netdev_err(dev,
1289 "vector_open: failed to get tx irq(%d)\n", err);
1290 err = -ENETUNREACH;
1291 goto out_close;
1292 }
1293 vp->tx_irq = irq_rr + VECTOR_BASE_IRQ;
1294 irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
1295 }
1296
1297 if ((vp->options & VECTOR_QDISC_BYPASS) != 0) {
1298 if (!uml_raw_enable_qdisc_bypass(vp->fds->rx_fd))
1299 vp->options |= VECTOR_BPF;
1300 }
1301 if (((vp->options & VECTOR_BPF) != 0) && (vp->bpf == NULL))
1302 vp->bpf = uml_vector_default_bpf(dev->dev_addr);
1303
1304 if (vp->bpf != NULL)
1305 uml_vector_attach_bpf(vp->fds->rx_fd, vp->bpf);
1306
1307 netif_start_queue(dev);
1308
1309 /* clear buffer - it can happen that the host side of the interface
1310 * is full when we get here. In this case, new data is never queued,
1311 * SIGIOs never arrive, and the net never works.
1312 */
1313
1314 vector_rx(vp);
1315
1316 vector_reset_stats(vp);
1317 vdevice = find_device(vp->unit);
1318 vdevice->opened = 1;
1319
1320 if ((vp->options & VECTOR_TX) != 0)
1321 add_timer(&vp->tl);
1322 return 0;
1323 out_close:
1324 vector_net_close(dev);
1325 return err;
1326 }
1327
1328
1329 static void vector_net_set_multicast_list(struct net_device *dev)
1330 {
1331 /* TODO: - we can do some BPF games here */
1332 return;
1333 }
1334
1335 static void vector_net_tx_timeout(struct net_device *dev, unsigned int txqueue)
1336 {
1337 struct vector_private *vp = netdev_priv(dev);
1338
1339 vp->estats.tx_timeout_count++;
1340 netif_trans_update(dev);
1341 schedule_work(&vp->reset_tx);
1342 }
1343
1344 static netdev_features_t vector_fix_features(struct net_device *dev,
1345 netdev_features_t features)
1346 {
1347 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
1348 return features;
1349 }
1350
1351 static int vector_set_features(struct net_device *dev,
1352 netdev_features_t features)
1353 {
1354 struct vector_private *vp = netdev_priv(dev);
1355 /* Adjust buffer sizes for GSO/GRO. Unfortunately, there is
1356 * no way to negotiate it on raw sockets, so we can change
1357 * only our side.
1358 */
1359 if (features & NETIF_F_GRO)
1360 /* All new frame buffers will be GRO-sized */
1361 vp->req_size = 65536;
1362 else
1363 /* All new frame buffers will be normal sized */
1364 vp->req_size = vp->max_packet + vp->headroom + SAFETY_MARGIN;
1365 return 0;
1366 }
1367
1368 #ifdef CONFIG_NET_POLL_CONTROLLER
1369 static void vector_net_poll_controller(struct net_device *dev)
1370 {
1371 disable_irq(dev->irq);
1372 vector_rx_interrupt(dev->irq, dev);
1373 enable_irq(dev->irq);
1374 }
1375 #endif
1376
1377 static void vector_net_get_drvinfo(struct net_device *dev,
1378 struct ethtool_drvinfo *info)
1379 {
1380 strlcpy(info->driver, DRIVER_NAME, sizeof(info->driver));
1381 strlcpy(info->version, DRIVER_VERSION, sizeof(info->version));
1382 }
1383
1384 static int vector_net_load_bpf_flash(struct net_device *dev,
1385 struct ethtool_flash *efl)
1386 {
1387 struct vector_private *vp = netdev_priv(dev);
1388 struct vector_device *vdevice;
1389 const struct firmware *fw;
1390 int result = 0;
1391
1392 if (!(vp->options & VECTOR_BPF_FLASH)) {
1393 netdev_err(dev, "loading firmware not permitted: %s\n", efl->data);
1394 return -1;
1395 }
1396
1397 spin_lock(&vp->lock);
1398
1399 if (vp->bpf != NULL) {
1400 if (vp->opened)
1401 uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
1402 kfree(vp->bpf->filter);
1403 vp->bpf->filter = NULL;
1404 } else {
1405 vp->bpf = kmalloc(sizeof(struct sock_fprog), GFP_KERNEL);
1406 if (vp->bpf == NULL) {
1407 netdev_err(dev, "failed to allocate memory for firmware\n");
1408 goto flash_fail;
1409 }
1410 }
1411
1412 vdevice = find_device(vp->unit);
1413
1414 if (request_firmware(&fw, efl->data, &vdevice->pdev.dev))
1415 goto flash_fail;
1416
1417 vp->bpf->filter = kmemdup(fw->data, fw->size, GFP_KERNEL);
1418 if (!vp->bpf->filter)
1419 goto free_buffer;
1420
1421 vp->bpf->len = fw->size / sizeof(struct sock_filter);
1422 release_firmware(fw);
1423
1424 if (vp->opened)
1425 result = uml_vector_attach_bpf(vp->fds->rx_fd, vp->bpf);
1426
1427 spin_unlock(&vp->lock);
1428
1429 return result;
1430
1431 free_buffer:
1432 release_firmware(fw);
1433
1434 flash_fail:
1435 spin_unlock(&vp->lock);
1436 if (vp->bpf != NULL)
1437 kfree(vp->bpf->filter);
1438 kfree(vp->bpf);
1439 vp->bpf = NULL;
1440 return -1;
1441 }
1442
1443 static void vector_get_ringparam(struct net_device *netdev,
1444 struct ethtool_ringparam *ring)
1445 {
1446 struct vector_private *vp = netdev_priv(netdev);
1447
1448 ring->rx_max_pending = vp->rx_queue->max_depth;
1449 ring->tx_max_pending = vp->tx_queue->max_depth;
1450 ring->rx_pending = vp->rx_queue->max_depth;
1451 ring->tx_pending = vp->tx_queue->max_depth;
1452 }
1453
1454 static void vector_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
1455 {
1456 switch (stringset) {
1457 case ETH_SS_TEST:
1458 *buf = '\0';
1459 break;
1460 case ETH_SS_STATS:
1461 memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
1462 break;
1463 default:
1464 WARN_ON(1);
1465 break;
1466 }
1467 }
1468
1469 static int vector_get_sset_count(struct net_device *dev, int sset)
1470 {
1471 switch (sset) {
1472 case ETH_SS_TEST:
1473 return 0;
1474 case ETH_SS_STATS:
1475 return VECTOR_NUM_STATS;
1476 default:
1477 return -EOPNOTSUPP;
1478 }
1479 }
1480
1481 static void vector_get_ethtool_stats(struct net_device *dev,
1482 struct ethtool_stats *estats,
1483 u64 *tmp_stats)
1484 {
1485 struct vector_private *vp = netdev_priv(dev);
1486
1487 memcpy(tmp_stats, &vp->estats, sizeof(struct vector_estats));
1488 }
1489
1490 static int vector_get_coalesce(struct net_device *netdev,
1491 struct ethtool_coalesce *ec)
1492 {
1493 struct vector_private *vp = netdev_priv(netdev);
1494
1495 ec->tx_coalesce_usecs = (vp->coalesce * 1000000) / HZ;
1496 return 0;
1497 }
1498
1499 static int vector_set_coalesce(struct net_device *netdev,
1500 struct ethtool_coalesce *ec)
1501 {
1502 struct vector_private *vp = netdev_priv(netdev);
1503
1504 vp->coalesce = (ec->tx_coalesce_usecs * HZ) / 1000000;
1505 if (vp->coalesce == 0)
1506 vp->coalesce = 1;
1507 return 0;
1508 }
1509
1510 static const struct ethtool_ops vector_net_ethtool_ops = {
1511 .get_drvinfo = vector_net_get_drvinfo,
1512 .get_link = ethtool_op_get_link,
1513 .get_ts_info = ethtool_op_get_ts_info,
1514 .get_ringparam = vector_get_ringparam,
1515 .get_strings = vector_get_strings,
1516 .get_sset_count = vector_get_sset_count,
1517 .get_ethtool_stats = vector_get_ethtool_stats,
1518 .get_coalesce = vector_get_coalesce,
1519 .set_coalesce = vector_set_coalesce,
1520 .flash_device = vector_net_load_bpf_flash,
1521 };
1522
1523
1524 static const struct net_device_ops vector_netdev_ops = {
1525 .ndo_open = vector_net_open,
1526 .ndo_stop = vector_net_close,
1527 .ndo_start_xmit = vector_net_start_xmit,
1528 .ndo_set_rx_mode = vector_net_set_multicast_list,
1529 .ndo_tx_timeout = vector_net_tx_timeout,
1530 .ndo_set_mac_address = eth_mac_addr,
1531 .ndo_validate_addr = eth_validate_addr,
1532 .ndo_fix_features = vector_fix_features,
1533 .ndo_set_features = vector_set_features,
1534 #ifdef CONFIG_NET_POLL_CONTROLLER
1535 .ndo_poll_controller = vector_net_poll_controller,
1536 #endif
1537 };
1538
1539
1540 static void vector_timer_expire(struct timer_list *t)
1541 {
1542 struct vector_private *vp = from_timer(vp, t, tl);
1543
1544 vp->estats.tx_kicks++;
1545 vector_send(vp->tx_queue);
1546 }
1547
1548 static void vector_eth_configure(
1549 int n,
1550 struct arglist *def
1551 )
1552 {
1553 struct vector_device *device;
1554 struct net_device *dev;
1555 struct vector_private *vp;
1556 int err;
1557
1558 device = kzalloc(sizeof(*device), GFP_KERNEL);
1559 if (device == NULL) {
1560 printk(KERN_ERR "eth_configure failed to allocate struct "
1561 "vector_device\n");
1562 return;
1563 }
1564 dev = alloc_etherdev(sizeof(struct vector_private));
1565 if (dev == NULL) {
1566 printk(KERN_ERR "eth_configure: failed to allocate struct "
1567 "net_device for vec%d\n", n);
1568 goto out_free_device;
1569 }
1570
1571 dev->mtu = get_mtu(def);
1572
1573 INIT_LIST_HEAD(&device->list);
1574 device->unit = n;
1575
1576 /* If this name ends up conflicting with an existing registered
1577 * netdevice, that is OK, register_netdev{,ice}() will notice this
1578 * and fail.
1579 */
1580 snprintf(dev->name, sizeof(dev->name), "vec%d", n);
1581 uml_net_setup_etheraddr(dev, uml_vector_fetch_arg(def, "mac"));
1582 vp = netdev_priv(dev);
1583
1584 /* sysfs register */
1585 if (!driver_registered) {
1586 platform_driver_register(&uml_net_driver);
1587 driver_registered = 1;
1588 }
1589 device->pdev.id = n;
1590 device->pdev.name = DRIVER_NAME;
1591 device->pdev.dev.release = vector_device_release;
1592 dev_set_drvdata(&device->pdev.dev, device);
1593 if (platform_device_register(&device->pdev))
1594 goto out_free_netdev;
1595 SET_NETDEV_DEV(dev, &device->pdev.dev);
1596
1597 device->dev = dev;
1598
1599 *vp = ((struct vector_private)
1600 {
1601 .list = LIST_HEAD_INIT(vp->list),
1602 .dev = dev,
1603 .unit = n,
1604 .options = get_transport_options(def),
1605 .rx_irq = 0,
1606 .tx_irq = 0,
1607 .parsed = def,
1608 .max_packet = get_mtu(def) + ETH_HEADER_OTHER,
1609 /* TODO - we need to calculate headroom so that ip header
1610 * is 16 byte aligned all the time
1611 */
1612 .headroom = get_headroom(def),
1613 .form_header = NULL,
1614 .verify_header = NULL,
1615 .header_rxbuffer = NULL,
1616 .header_txbuffer = NULL,
1617 .header_size = 0,
1618 .rx_header_size = 0,
1619 .rexmit_scheduled = false,
1620 .opened = false,
1621 .transport_data = NULL,
1622 .in_write_poll = false,
1623 .coalesce = 2,
1624 .req_size = get_req_size(def),
1625 .in_error = false,
1626 .bpf = NULL
1627 });
1628
1629 dev->features = dev->hw_features = (NETIF_F_SG | NETIF_F_FRAGLIST);
1630 tasklet_init(&vp->tx_poll, vector_tx_poll, (unsigned long)vp);
1631 INIT_WORK(&vp->reset_tx, vector_reset_tx);
1632
1633 timer_setup(&vp->tl, vector_timer_expire, 0);
1634 spin_lock_init(&vp->lock);
1635
1636 /* FIXME */
1637 dev->netdev_ops = &vector_netdev_ops;
1638 dev->ethtool_ops = &vector_net_ethtool_ops;
1639 dev->watchdog_timeo = (HZ >> 1);
1640 /* primary IRQ - fixme */
1641 dev->irq = 0; /* we will adjust this once opened */
1642
1643 rtnl_lock();
1644 err = register_netdevice(dev);
1645 rtnl_unlock();
1646 if (err)
1647 goto out_undo_user_init;
1648
1649 spin_lock(&vector_devices_lock);
1650 list_add(&device->list, &vector_devices);
1651 spin_unlock(&vector_devices_lock);
1652
1653 return;
1654
1655 out_undo_user_init:
1656 return;
1657 out_free_netdev:
1658 free_netdev(dev);
1659 out_free_device:
1660 kfree(device);
1661 }
1662
1663
1664
1665
1666 /*
1667 * Invoked late in the init
1668 */
1669
1670 static int __init vector_init(void)
1671 {
1672 struct list_head *ele;
1673 struct vector_cmd_line_arg *def;
1674 struct arglist *parsed;
1675
1676 list_for_each(ele, &vec_cmd_line) {
1677 def = list_entry(ele, struct vector_cmd_line_arg, list);
1678 parsed = uml_parse_vector_ifspec(def->arguments);
1679 if (parsed != NULL)
1680 vector_eth_configure(def->unit, parsed);
1681 }
1682 return 0;
1683 }
1684
1685
1686 /* Invoked at initial argument parsing, only stores
1687 * arguments until a proper vector_init is called
1688 * later
1689 */
1690
1691 static int __init vector_setup(char *str)
1692 {
1693 char *error;
1694 int n, err;
1695 struct vector_cmd_line_arg *new;
1696
1697 err = vector_parse(str, &n, &str, &error);
1698 if (err) {
1699 printk(KERN_ERR "vector_setup - Couldn't parse '%s' : %s\n",
1700 str, error);
1701 return 1;
1702 }
1703 new = memblock_alloc(sizeof(*new), SMP_CACHE_BYTES);
1704 if (!new)
1705 panic("%s: Failed to allocate %zu bytes\n", __func__,
1706 sizeof(*new));
1707 INIT_LIST_HEAD(&new->list);
1708 new->unit = n;
1709 new->arguments = str;
1710 list_add_tail(&new->list, &vec_cmd_line);
1711 return 1;
1712 }
1713
1714 __setup("vec", vector_setup);
1715 __uml_help(vector_setup,
1716 "vec[0-9]+:<option>=<value>,<option>=<value>\n"
1717 " Configure a vector io network device.\n\n"
1718 );
1719
1720 late_initcall(vector_init);
1721
1722 static struct mc_device vector_mc = {
1723 .list = LIST_HEAD_INIT(vector_mc.list),
1724 .name = "vec",
1725 .config = vector_config,
1726 .get_config = NULL,
1727 .id = vector_id,
1728 .remove = vector_remove,
1729 };
1730
1731 #ifdef CONFIG_INET
1732 static int vector_inetaddr_event(
1733 struct notifier_block *this,
1734 unsigned long event,
1735 void *ptr)
1736 {
1737 return NOTIFY_DONE;
1738 }
1739
1740 static struct notifier_block vector_inetaddr_notifier = {
1741 .notifier_call = vector_inetaddr_event,
1742 };
1743
1744 static void inet_register(void)
1745 {
1746 register_inetaddr_notifier(&vector_inetaddr_notifier);
1747 }
1748 #else
1749 static inline void inet_register(void)
1750 {
1751 }
1752 #endif
1753
1754 static int vector_net_init(void)
1755 {
1756 mconsole_register_dev(&vector_mc);
1757 inet_register();
1758 return 0;
1759 }
1760
1761 __initcall(vector_net_init);
1762
1763
1764
Linux with added FPC-III support