]>
Commit | Line | Data |
---|---|---|
a2d974da DN |
1 | /* |
2 | * This file is subject to the terms and conditions of the GNU General Public | |
3 | * License. See the file "COPYING" in the main directory of this archive | |
4 | * for more details. | |
5 | * | |
45d9ca49 | 6 | * Copyright (C) 1999-2008 Silicon Graphics, Inc. All rights reserved. |
a2d974da DN |
7 | */ |
8 | ||
a2d974da DN |
9 | /* |
10 | * Cross Partition Network Interface (XPNET) support | |
11 | * | |
12 | * XPNET provides a virtual network layered on top of the Cross | |
13 | * Partition communication layer. | |
14 | * | |
15 | * XPNET provides direct point-to-point and broadcast-like support | |
16 | * for an ethernet-like device. The ethernet broadcast medium is | |
17 | * replaced with a point-to-point message structure which passes | |
18 | * pointers to a DMA-capable block that a remote partition should | |
19 | * retrieve and pass to the upper level networking layer. | |
20 | * | |
21 | */ | |
22 | ||
a2d974da DN |
23 | #include <linux/module.h> |
24 | #include <linux/kernel.h> | |
a2d974da DN |
25 | #include <linux/init.h> |
26 | #include <linux/ioport.h> | |
27 | #include <linux/netdevice.h> | |
28 | #include <linux/etherdevice.h> | |
29 | #include <linux/delay.h> | |
30 | #include <linux/ethtool.h> | |
31 | #include <linux/mii.h> | |
32 | #include <linux/smp.h> | |
33 | #include <linux/string.h> | |
a2d974da | 34 | #include <asm/atomic.h> |
45d9ca49 | 35 | #include "xp.h" |
a2d974da | 36 | |
a2d974da DN |
37 | /* |
38 | * The message payload transferred by XPC. | |
39 | * | |
40 | * buf_pa is the physical address where the DMA should pull from. | |
41 | * | |
42 | * NOTE: for performance reasons, buf_pa should _ALWAYS_ begin on a | |
43 | * cacheline boundary. To accomplish this, we record the number of | |
44 | * bytes from the beginning of the first cacheline to the first useful | |
45 | * byte of the skb (leadin_ignore) and the number of bytes from the | |
46 | * last useful byte of the skb to the end of the last cacheline | |
47 | * (tailout_ignore). | |
48 | * | |
49 | * size is the number of bytes to transfer which includes the skb->len | |
50 | * (useful bytes of the senders skb) plus the leadin and tailout | |
51 | */ | |
52 | struct xpnet_message { | |
53 | u16 version; /* Version for this message */ | |
54 | u16 embedded_bytes; /* #of bytes embedded in XPC message */ | |
55 | u32 magic; /* Special number indicating this is xpnet */ | |
56 | u64 buf_pa; /* phys address of buffer to retrieve */ | |
57 | u32 size; /* #of bytes in buffer */ | |
58 | u8 leadin_ignore; /* #of bytes to ignore at the beginning */ | |
59 | u8 tailout_ignore; /* #of bytes to ignore at the end */ | |
60 | unsigned char data; /* body of small packets */ | |
61 | }; | |
62 | ||
63 | /* | |
64 | * Determine the size of our message, the cacheline aligned size, | |
65 | * and then the number of message will request from XPC. | |
66 | * | |
67 | * XPC expects each message to exist in an individual cacheline. | |
68 | */ | |
69 | #define XPNET_MSG_SIZE (L1_CACHE_BYTES - XPC_MSG_PAYLOAD_OFFSET) | |
70 | #define XPNET_MSG_DATA_MAX \ | |
71 | (XPNET_MSG_SIZE - (u64)(&((struct xpnet_message *)0)->data)) | |
72 | #define XPNET_MSG_ALIGNED_SIZE (L1_CACHE_ALIGN(XPNET_MSG_SIZE)) | |
73 | #define XPNET_MSG_NENTRIES (PAGE_SIZE / XPNET_MSG_ALIGNED_SIZE) | |
74 | ||
a2d974da DN |
75 | #define XPNET_MAX_KTHREADS (XPNET_MSG_NENTRIES + 1) |
76 | #define XPNET_MAX_IDLE_KTHREADS (XPNET_MSG_NENTRIES + 1) | |
77 | ||
78 | /* | |
79 | * Version number of XPNET implementation. XPNET can always talk to versions | |
80 | * with same major #, and never talk to versions with a different version. | |
81 | */ | |
82 | #define _XPNET_VERSION(_major, _minor) (((_major) << 4) | (_minor)) | |
83 | #define XPNET_VERSION_MAJOR(_v) ((_v) >> 4) | |
84 | #define XPNET_VERSION_MINOR(_v) ((_v) & 0xf) | |
85 | ||
2c2b94f9 DN |
86 | #define XPNET_VERSION _XPNET_VERSION(1, 0) /* version 1.0 */ |
87 | #define XPNET_VERSION_EMBED _XPNET_VERSION(1, 1) /* version 1.1 */ | |
35190506 | 88 | #define XPNET_MAGIC 0x88786984 /* "XNET" */ |
a2d974da DN |
89 | |
90 | #define XPNET_VALID_MSG(_m) \ | |
91 | ((XPNET_VERSION_MAJOR(_m->version) == XPNET_VERSION_MAJOR(XPNET_VERSION)) \ | |
92 | && (msg->magic == XPNET_MAGIC)) | |
93 | ||
94 | #define XPNET_DEVICE_NAME "xp0" | |
95 | ||
a2d974da DN |
96 | /* |
97 | * When messages are queued with xpc_send_notify, a kmalloc'd buffer | |
98 | * of the following type is passed as a notification cookie. When the | |
99 | * notification function is called, we use the cookie to decide | |
100 | * whether all outstanding message sends have completed. The skb can | |
101 | * then be released. | |
102 | */ | |
103 | struct xpnet_pending_msg { | |
a2d974da DN |
104 | struct sk_buff *skb; |
105 | atomic_t use_count; | |
106 | }; | |
107 | ||
108 | /* driver specific structure pointed to by the device structure */ | |
109 | struct xpnet_dev_private { | |
110 | struct net_device_stats stats; | |
111 | }; | |
112 | ||
113 | struct net_device *xpnet_device; | |
114 | ||
115 | /* | |
116 | * When we are notified of other partitions activating, we add them to | |
117 | * our bitmask of partitions to which we broadcast. | |
118 | */ | |
a7b4d509 | 119 | static unsigned long *xpnet_broadcast_partitions; |
a2d974da | 120 | /* protect above */ |
a9f6a0dd | 121 | static DEFINE_SPINLOCK(xpnet_broadcast_lock); |
a2d974da DN |
122 | |
123 | /* | |
124 | * Since the Block Transfer Engine (BTE) is being used for the transfer | |
125 | * and it relies upon cache-line size transfers, we need to reserve at | |
126 | * least one cache-line for head and tail alignment. The BTE is | |
127 | * limited to 8MB transfers. | |
128 | * | |
129 | * Testing has shown that changing MTU to greater than 64KB has no effect | |
130 | * on TCP as the two sides negotiate a Max Segment Size that is limited | |
131 | * to 64K. Other protocols May use packets greater than this, but for | |
132 | * now, the default is 64KB. | |
133 | */ | |
134 | #define XPNET_MAX_MTU (0x800000UL - L1_CACHE_BYTES) | |
135 | /* 32KB has been determined to be the ideal */ | |
136 | #define XPNET_DEF_MTU (0x8000UL) | |
137 | ||
a2d974da | 138 | /* |
a7b4d509 DN |
139 | * The partid is encapsulated in the MAC address beginning in the following |
140 | * octet and it consists of two octets. | |
a2d974da | 141 | */ |
a7b4d509 DN |
142 | #define XPNET_PARTID_OCTET 2 |
143 | ||
144 | /* Define the XPNET debug device structures to be used with dev_dbg() et al */ | |
a2d974da | 145 | |
a2d974da DN |
146 | struct device_driver xpnet_dbg_name = { |
147 | .name = "xpnet" | |
148 | }; | |
149 | ||
150 | struct device xpnet_dbg_subname = { | |
35190506 | 151 | .bus_id = {0}, /* set to "" */ |
a2d974da DN |
152 | .driver = &xpnet_dbg_name |
153 | }; | |
154 | ||
155 | struct device *xpnet = &xpnet_dbg_subname; | |
156 | ||
157 | /* | |
158 | * Packet was recevied by XPC and forwarded to us. | |
159 | */ | |
160 | static void | |
64d032ba | 161 | xpnet_receive(short partid, int channel, struct xpnet_message *msg) |
a2d974da DN |
162 | { |
163 | struct sk_buff *skb; | |
908787db | 164 | enum xp_retval ret; |
a2d974da | 165 | struct xpnet_dev_private *priv = |
35190506 | 166 | (struct xpnet_dev_private *)xpnet_device->priv; |
a2d974da DN |
167 | |
168 | if (!XPNET_VALID_MSG(msg)) { | |
169 | /* | |
170 | * Packet with a different XPC version. Ignore. | |
171 | */ | |
35190506 | 172 | xpc_received(partid, channel, (void *)msg); |
a2d974da DN |
173 | |
174 | priv->stats.rx_errors++; | |
175 | ||
176 | return; | |
177 | } | |
178 | dev_dbg(xpnet, "received 0x%lx, %d, %d, %d\n", msg->buf_pa, msg->size, | |
179 | msg->leadin_ignore, msg->tailout_ignore); | |
180 | ||
a2d974da DN |
181 | /* reserve an extra cache line */ |
182 | skb = dev_alloc_skb(msg->size + L1_CACHE_BYTES); | |
183 | if (!skb) { | |
184 | dev_err(xpnet, "failed on dev_alloc_skb(%d)\n", | |
185 | msg->size + L1_CACHE_BYTES); | |
186 | ||
35190506 | 187 | xpc_received(partid, channel, (void *)msg); |
a2d974da DN |
188 | |
189 | priv->stats.rx_errors++; | |
190 | ||
191 | return; | |
192 | } | |
193 | ||
194 | /* | |
195 | * The allocated skb has some reserved space. | |
908787db | 196 | * In order to use xp_remote_memcpy(), we need to get the |
a2d974da DN |
197 | * skb->data pointer moved forward. |
198 | */ | |
199 | skb_reserve(skb, (L1_CACHE_BYTES - ((u64)skb->data & | |
200 | (L1_CACHE_BYTES - 1)) + | |
201 | msg->leadin_ignore)); | |
202 | ||
203 | /* | |
204 | * Update the tail pointer to indicate data actually | |
205 | * transferred. | |
206 | */ | |
207 | skb_put(skb, (msg->size - msg->leadin_ignore - msg->tailout_ignore)); | |
208 | ||
209 | /* | |
4b3f686d | 210 | * Move the data over from the other side. |
a2d974da DN |
211 | */ |
212 | if ((XPNET_VERSION_MINOR(msg->version) == 1) && | |
35190506 | 213 | (msg->embedded_bytes != 0)) { |
a2d974da DN |
214 | dev_dbg(xpnet, "copying embedded message. memcpy(0x%p, 0x%p, " |
215 | "%lu)\n", skb->data, &msg->data, | |
35190506 | 216 | (size_t)msg->embedded_bytes); |
a2d974da | 217 | |
35190506 DN |
218 | skb_copy_to_linear_data(skb, &msg->data, |
219 | (size_t)msg->embedded_bytes); | |
a2d974da DN |
220 | } else { |
221 | dev_dbg(xpnet, "transferring buffer to the skb->data area;\n\t" | |
908787db DN |
222 | "xp_remote_memcpy(0x%p, 0x%p, %hu)\n", (void *) |
223 | ((u64)skb->data & ~(L1_CACHE_BYTES - 1)), | |
224 | (void *)msg->buf_pa, msg->size); | |
a2d974da | 225 | |
908787db | 226 | ret = xp_remote_memcpy((void *)((u64)skb->data & |
a7b4d509 | 227 | ~(L1_CACHE_BYTES - 1)), |
908787db | 228 | (void *)msg->buf_pa, msg->size); |
a2d974da | 229 | |
908787db | 230 | if (ret != xpSuccess) { |
2c2b94f9 | 231 | /* |
ea57f80c DN |
232 | * !!! Need better way of cleaning skb. Currently skb |
233 | * !!! appears in_use and we can't just call | |
234 | * !!! dev_kfree_skb. | |
2c2b94f9 | 235 | */ |
908787db DN |
236 | dev_err(xpnet, "xp_remote_memcpy(0x%p, 0x%p, 0x%hx) " |
237 | "returned error=0x%x\n", (void *) | |
238 | ((u64)skb->data & ~(L1_CACHE_BYTES - 1)), | |
239 | (void *)msg->buf_pa, msg->size, ret); | |
a2d974da | 240 | |
35190506 | 241 | xpc_received(partid, channel, (void *)msg); |
a2d974da DN |
242 | |
243 | priv->stats.rx_errors++; | |
244 | ||
245 | return; | |
246 | } | |
247 | } | |
248 | ||
249 | dev_dbg(xpnet, "<skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | |
35190506 | 250 | "skb->end=0x%p skb->len=%d\n", (void *)skb->head, |
4305b541 | 251 | (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb), |
a2d974da DN |
252 | skb->len); |
253 | ||
a2d974da DN |
254 | skb->protocol = eth_type_trans(skb, xpnet_device); |
255 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
256 | ||
898eb71c JP |
257 | dev_dbg(xpnet, "passing skb to network layer\n" |
258 | KERN_DEBUG "\tskb->head=0x%p skb->data=0x%p skb->tail=0x%p " | |
259 | "skb->end=0x%p skb->len=%d\n", | |
27a884dc | 260 | (void *)skb->head, (void *)skb->data, skb_tail_pointer(skb), |
4305b541 | 261 | skb_end_pointer(skb), skb->len); |
a2d974da | 262 | |
a2d974da DN |
263 | xpnet_device->last_rx = jiffies; |
264 | priv->stats.rx_packets++; | |
265 | priv->stats.rx_bytes += skb->len + ETH_HLEN; | |
266 | ||
267 | netif_rx_ni(skb); | |
35190506 | 268 | xpc_received(partid, channel, (void *)msg); |
a2d974da DN |
269 | } |
270 | ||
a2d974da DN |
271 | /* |
272 | * This is the handler which XPC calls during any sort of change in | |
273 | * state or message reception on a connection. | |
274 | */ | |
275 | static void | |
64d032ba | 276 | xpnet_connection_activity(enum xp_retval reason, short partid, int channel, |
a2d974da DN |
277 | void *data, void *key) |
278 | { | |
bc63d387 | 279 | DBUG_ON(partid < 0 || partid >= xp_max_npartitions); |
a2d974da DN |
280 | DBUG_ON(channel != XPC_NET_CHANNEL); |
281 | ||
35190506 | 282 | switch (reason) { |
65c17b80 | 283 | case xpMsgReceived: /* message received */ |
a2d974da DN |
284 | DBUG_ON(data == NULL); |
285 | ||
35190506 | 286 | xpnet_receive(partid, channel, (struct xpnet_message *)data); |
a2d974da DN |
287 | break; |
288 | ||
65c17b80 | 289 | case xpConnected: /* connection completed to a partition */ |
a2d974da | 290 | spin_lock_bh(&xpnet_broadcast_lock); |
a7b4d509 | 291 | __set_bit(partid, xpnet_broadcast_partitions); |
a2d974da DN |
292 | spin_unlock_bh(&xpnet_broadcast_lock); |
293 | ||
294 | netif_carrier_on(xpnet_device); | |
295 | ||
a7b4d509 DN |
296 | dev_dbg(xpnet, "%s connected to partition %d\n", |
297 | xpnet_device->name, partid); | |
a2d974da DN |
298 | break; |
299 | ||
300 | default: | |
301 | spin_lock_bh(&xpnet_broadcast_lock); | |
a7b4d509 | 302 | __clear_bit(partid, xpnet_broadcast_partitions); |
a2d974da DN |
303 | spin_unlock_bh(&xpnet_broadcast_lock); |
304 | ||
a7b4d509 DN |
305 | if (bitmap_empty((unsigned long *)xpnet_broadcast_partitions, |
306 | xp_max_npartitions)) { | |
a2d974da | 307 | netif_carrier_off(xpnet_device); |
a7b4d509 | 308 | } |
a2d974da | 309 | |
a7b4d509 DN |
310 | dev_dbg(xpnet, "%s disconnected from partition %d\n", |
311 | xpnet_device->name, partid); | |
a2d974da | 312 | break; |
a2d974da DN |
313 | } |
314 | } | |
315 | ||
a2d974da DN |
316 | static int |
317 | xpnet_dev_open(struct net_device *dev) | |
318 | { | |
65c17b80 | 319 | enum xp_retval ret; |
a2d974da | 320 | |
b9ae3bd2 TL |
321 | dev_dbg(xpnet, "calling xpc_connect(%d, 0x%p, NULL, %ld, %ld, %ld, " |
322 | "%ld)\n", XPC_NET_CHANNEL, xpnet_connection_activity, | |
a2d974da DN |
323 | XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, XPNET_MAX_KTHREADS, |
324 | XPNET_MAX_IDLE_KTHREADS); | |
325 | ||
326 | ret = xpc_connect(XPC_NET_CHANNEL, xpnet_connection_activity, NULL, | |
327 | XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, | |
328 | XPNET_MAX_KTHREADS, XPNET_MAX_IDLE_KTHREADS); | |
65c17b80 | 329 | if (ret != xpSuccess) { |
a2d974da DN |
330 | dev_err(xpnet, "ifconfig up of %s failed on XPC connect, " |
331 | "ret=%d\n", dev->name, ret); | |
332 | ||
333 | return -ENOMEM; | |
334 | } | |
335 | ||
336 | dev_dbg(xpnet, "ifconfig up of %s; XPC connected\n", dev->name); | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
a2d974da DN |
341 | static int |
342 | xpnet_dev_stop(struct net_device *dev) | |
343 | { | |
344 | xpc_disconnect(XPC_NET_CHANNEL); | |
345 | ||
346 | dev_dbg(xpnet, "ifconfig down of %s; XPC disconnected\n", dev->name); | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
a2d974da DN |
351 | static int |
352 | xpnet_dev_change_mtu(struct net_device *dev, int new_mtu) | |
353 | { | |
354 | /* 68 comes from min TCP+IP+MAC header */ | |
355 | if ((new_mtu < 68) || (new_mtu > XPNET_MAX_MTU)) { | |
356 | dev_err(xpnet, "ifconfig %s mtu %d failed; value must be " | |
357 | "between 68 and %ld\n", dev->name, new_mtu, | |
358 | XPNET_MAX_MTU); | |
359 | return -EINVAL; | |
360 | } | |
361 | ||
362 | dev->mtu = new_mtu; | |
363 | dev_dbg(xpnet, "ifconfig %s mtu set to %d\n", dev->name, new_mtu); | |
364 | return 0; | |
365 | } | |
366 | ||
a2d974da DN |
367 | /* |
368 | * Required for the net_device structure. | |
369 | */ | |
370 | static int | |
371 | xpnet_dev_set_config(struct net_device *dev, struct ifmap *new_map) | |
372 | { | |
373 | return 0; | |
374 | } | |
375 | ||
a2d974da DN |
376 | /* |
377 | * Return statistics to the caller. | |
378 | */ | |
379 | static struct net_device_stats * | |
380 | xpnet_dev_get_stats(struct net_device *dev) | |
381 | { | |
382 | struct xpnet_dev_private *priv; | |
383 | ||
35190506 | 384 | priv = (struct xpnet_dev_private *)dev->priv; |
a2d974da DN |
385 | |
386 | return &priv->stats; | |
387 | } | |
388 | ||
a2d974da DN |
389 | /* |
390 | * Notification that the other end has received the message and | |
391 | * DMA'd the skb information. At this point, they are done with | |
392 | * our side. When all recipients are done processing, we | |
393 | * release the skb and then release our pending message structure. | |
394 | */ | |
395 | static void | |
64d032ba | 396 | xpnet_send_completed(enum xp_retval reason, short partid, int channel, |
35190506 | 397 | void *__qm) |
a2d974da | 398 | { |
35190506 | 399 | struct xpnet_pending_msg *queued_msg = (struct xpnet_pending_msg *)__qm; |
a2d974da DN |
400 | |
401 | DBUG_ON(queued_msg == NULL); | |
402 | ||
403 | dev_dbg(xpnet, "message to %d notified with reason %d\n", | |
404 | partid, reason); | |
405 | ||
406 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | |
407 | dev_dbg(xpnet, "all acks for skb->head=-x%p\n", | |
35190506 | 408 | (void *)queued_msg->skb->head); |
a2d974da DN |
409 | |
410 | dev_kfree_skb_any(queued_msg->skb); | |
411 | kfree(queued_msg); | |
412 | } | |
413 | } | |
414 | ||
a7b4d509 DN |
415 | static void |
416 | xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg, | |
417 | u64 start_addr, u64 end_addr, u16 embedded_bytes, int dest_partid) | |
418 | { | |
419 | u8 msg_buffer[XPNET_MSG_SIZE]; | |
420 | struct xpnet_message *msg = (struct xpnet_message *)&msg_buffer; | |
421 | enum xp_retval ret; | |
422 | ||
423 | msg->embedded_bytes = embedded_bytes; | |
424 | if (unlikely(embedded_bytes != 0)) { | |
425 | msg->version = XPNET_VERSION_EMBED; | |
426 | dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n", | |
427 | &msg->data, skb->data, (size_t)embedded_bytes); | |
428 | skb_copy_from_linear_data(skb, &msg->data, | |
429 | (size_t)embedded_bytes); | |
430 | } else { | |
431 | msg->version = XPNET_VERSION; | |
432 | } | |
433 | msg->magic = XPNET_MAGIC; | |
434 | msg->size = end_addr - start_addr; | |
435 | msg->leadin_ignore = (u64)skb->data - start_addr; | |
436 | msg->tailout_ignore = end_addr - (u64)skb_tail_pointer(skb); | |
437 | msg->buf_pa = __pa(start_addr); | |
438 | ||
439 | dev_dbg(xpnet, "sending XPC message to %d:%d\n" | |
440 | KERN_DEBUG "msg->buf_pa=0x%lx, msg->size=%u, " | |
441 | "msg->leadin_ignore=%u, msg->tailout_ignore=%u\n", | |
442 | dest_partid, XPC_NET_CHANNEL, msg->buf_pa, msg->size, | |
443 | msg->leadin_ignore, msg->tailout_ignore); | |
444 | ||
445 | atomic_inc(&queued_msg->use_count); | |
446 | ||
447 | ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, XPC_NOWAIT, msg, | |
448 | XPNET_MSG_SIZE, xpnet_send_completed, queued_msg); | |
449 | if (unlikely(ret != xpSuccess)) | |
450 | atomic_dec(&queued_msg->use_count); | |
451 | } | |
452 | ||
a2d974da DN |
453 | /* |
454 | * Network layer has formatted a packet (skb) and is ready to place it | |
455 | * "on the wire". Prepare and send an xpnet_message to all partitions | |
456 | * which have connected with us and are targets of this packet. | |
457 | * | |
458 | * MAC-NOTE: For the XPNET driver, the MAC address contains the | |
a7b4d509 DN |
459 | * destination partid. If the destination partid octets are 0xffff, |
460 | * this packet is to be broadcast to all connected partitions. | |
a2d974da DN |
461 | */ |
462 | static int | |
463 | xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
464 | { | |
465 | struct xpnet_pending_msg *queued_msg; | |
a2d974da | 466 | u64 start_addr, end_addr; |
64d032ba | 467 | short dest_partid; |
a7b4d509 DN |
468 | struct xpnet_dev_private *priv = (struct xpnet_dev_private *)dev->priv; |
469 | u16 embedded_bytes = 0; | |
a2d974da DN |
470 | |
471 | dev_dbg(xpnet, ">skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | |
35190506 | 472 | "skb->end=0x%p skb->len=%d\n", (void *)skb->head, |
4305b541 | 473 | (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb), |
a2d974da DN |
474 | skb->len); |
475 | ||
a7b4d509 DN |
476 | if (skb->data[0] == 0x33) { |
477 | dev_kfree_skb(skb); | |
478 | return 0; /* nothing needed to be done */ | |
479 | } | |
480 | ||
a2d974da DN |
481 | /* |
482 | * The xpnet_pending_msg tracks how many outstanding | |
483 | * xpc_send_notifies are relying on this skb. When none | |
484 | * remain, release the skb. | |
485 | */ | |
486 | queued_msg = kmalloc(sizeof(struct xpnet_pending_msg), GFP_ATOMIC); | |
487 | if (queued_msg == NULL) { | |
488 | dev_warn(xpnet, "failed to kmalloc %ld bytes; dropping " | |
35190506 | 489 | "packet\n", sizeof(struct xpnet_pending_msg)); |
a2d974da DN |
490 | |
491 | priv->stats.tx_errors++; | |
a2d974da DN |
492 | return -ENOMEM; |
493 | } | |
494 | ||
a2d974da | 495 | /* get the beginning of the first cacheline and end of last */ |
35190506 | 496 | start_addr = ((u64)skb->data & ~(L1_CACHE_BYTES - 1)); |
27a884dc | 497 | end_addr = L1_CACHE_ALIGN((u64)skb_tail_pointer(skb)); |
a2d974da DN |
498 | |
499 | /* calculate how many bytes to embed in the XPC message */ | |
a2d974da DN |
500 | if (unlikely(skb->len <= XPNET_MSG_DATA_MAX)) { |
501 | /* skb->data does fit so embed */ | |
502 | embedded_bytes = skb->len; | |
503 | } | |
504 | ||
a2d974da DN |
505 | /* |
506 | * Since the send occurs asynchronously, we set the count to one | |
507 | * and begin sending. Any sends that happen to complete before | |
508 | * we are done sending will not free the skb. We will be left | |
509 | * with that task during exit. This also handles the case of | |
510 | * a packet destined for a partition which is no longer up. | |
511 | */ | |
512 | atomic_set(&queued_msg->use_count, 1); | |
513 | queued_msg->skb = skb; | |
514 | ||
a7b4d509 | 515 | if (skb->data[0] == 0xff) { |
a2d974da | 516 | /* we are being asked to broadcast to all partitions */ |
a7b4d509 DN |
517 | for_each_bit(dest_partid, xpnet_broadcast_partitions, |
518 | xp_max_npartitions) { | |
a2d974da | 519 | |
a7b4d509 DN |
520 | xpnet_send(skb, queued_msg, start_addr, end_addr, |
521 | embedded_bytes, dest_partid); | |
a2d974da | 522 | } |
a7b4d509 DN |
523 | } else { |
524 | dest_partid = (short)skb->data[XPNET_PARTID_OCTET + 1]; | |
525 | dest_partid |= (short)skb->data[XPNET_PARTID_OCTET + 0] << 8; | |
a2d974da | 526 | |
a7b4d509 DN |
527 | if (dest_partid >= 0 && |
528 | dest_partid < xp_max_npartitions && | |
529 | test_bit(dest_partid, xpnet_broadcast_partitions) != 0) { | |
a2d974da | 530 | |
a7b4d509 DN |
531 | xpnet_send(skb, queued_msg, start_addr, end_addr, |
532 | embedded_bytes, dest_partid); | |
a2d974da | 533 | } |
a2d974da DN |
534 | } |
535 | ||
536 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | |
a2d974da DN |
537 | dev_kfree_skb(skb); |
538 | kfree(queued_msg); | |
539 | } | |
540 | ||
541 | priv->stats.tx_packets++; | |
542 | priv->stats.tx_bytes += skb->len; | |
543 | ||
544 | return 0; | |
545 | } | |
546 | ||
a2d974da DN |
547 | /* |
548 | * Deal with transmit timeouts coming from the network layer. | |
549 | */ | |
550 | static void | |
35190506 | 551 | xpnet_dev_tx_timeout(struct net_device *dev) |
a2d974da DN |
552 | { |
553 | struct xpnet_dev_private *priv; | |
554 | ||
35190506 | 555 | priv = (struct xpnet_dev_private *)dev->priv; |
a2d974da DN |
556 | |
557 | priv->stats.tx_errors++; | |
558 | return; | |
559 | } | |
560 | ||
a2d974da DN |
561 | static int __init |
562 | xpnet_init(void) | |
563 | { | |
a7b4d509 | 564 | int result; |
a2d974da | 565 | |
a7b4d509 | 566 | if (!is_shub() && !is_uv()) |
408865ce | 567 | return -ENODEV; |
408865ce | 568 | |
a2d974da DN |
569 | dev_info(xpnet, "registering network device %s\n", XPNET_DEVICE_NAME); |
570 | ||
a7b4d509 DN |
571 | xpnet_broadcast_partitions = kzalloc(BITS_TO_LONGS(xp_max_npartitions) * |
572 | sizeof(long), GFP_KERNEL); | |
573 | if (xpnet_broadcast_partitions == NULL) | |
574 | return -ENOMEM; | |
575 | ||
a2d974da DN |
576 | /* |
577 | * use ether_setup() to init the majority of our device | |
578 | * structure and then override the necessary pieces. | |
579 | */ | |
580 | xpnet_device = alloc_netdev(sizeof(struct xpnet_dev_private), | |
581 | XPNET_DEVICE_NAME, ether_setup); | |
a7b4d509 DN |
582 | if (xpnet_device == NULL) { |
583 | kfree(xpnet_broadcast_partitions); | |
a2d974da | 584 | return -ENOMEM; |
a7b4d509 | 585 | } |
a2d974da DN |
586 | |
587 | netif_carrier_off(xpnet_device); | |
588 | ||
589 | xpnet_device->mtu = XPNET_DEF_MTU; | |
590 | xpnet_device->change_mtu = xpnet_dev_change_mtu; | |
591 | xpnet_device->open = xpnet_dev_open; | |
592 | xpnet_device->get_stats = xpnet_dev_get_stats; | |
593 | xpnet_device->stop = xpnet_dev_stop; | |
594 | xpnet_device->hard_start_xmit = xpnet_dev_hard_start_xmit; | |
595 | xpnet_device->tx_timeout = xpnet_dev_tx_timeout; | |
596 | xpnet_device->set_config = xpnet_dev_set_config; | |
597 | ||
598 | /* | |
599 | * Multicast assumes the LSB of the first octet is set for multicast | |
600 | * MAC addresses. We chose the first octet of the MAC to be unlikely | |
601 | * to collide with any vendor's officially issued MAC. | |
602 | */ | |
a7b4d509 DN |
603 | xpnet_device->dev_addr[0] = 0x02; /* locally administered, no OUI */ |
604 | ||
605 | xpnet_device->dev_addr[XPNET_PARTID_OCTET + 1] = sn_partition_id; | |
606 | xpnet_device->dev_addr[XPNET_PARTID_OCTET + 0] = (sn_partition_id >> 8); | |
a2d974da DN |
607 | |
608 | /* | |
609 | * ether_setup() sets this to a multicast device. We are | |
610 | * really not supporting multicast at this time. | |
611 | */ | |
612 | xpnet_device->flags &= ~IFF_MULTICAST; | |
613 | ||
614 | /* | |
615 | * No need to checksum as it is a DMA transfer. The BTE will | |
616 | * report an error if the data is not retrievable and the | |
617 | * packet will be dropped. | |
618 | */ | |
619 | xpnet_device->features = NETIF_F_NO_CSUM; | |
620 | ||
621 | result = register_netdev(xpnet_device); | |
a7b4d509 | 622 | if (result != 0) { |
a2d974da | 623 | free_netdev(xpnet_device); |
a7b4d509 DN |
624 | kfree(xpnet_broadcast_partitions); |
625 | } | |
a2d974da DN |
626 | |
627 | return result; | |
628 | } | |
a2d974da | 629 | |
35190506 | 630 | module_init(xpnet_init); |
a2d974da DN |
631 | |
632 | static void __exit | |
633 | xpnet_exit(void) | |
634 | { | |
635 | dev_info(xpnet, "unregistering network device %s\n", | |
35190506 | 636 | xpnet_device[0].name); |
a2d974da DN |
637 | |
638 | unregister_netdev(xpnet_device); | |
a2d974da | 639 | free_netdev(xpnet_device); |
a7b4d509 | 640 | kfree(xpnet_broadcast_partitions); |
a2d974da | 641 | } |
a2d974da | 642 | |
35190506 | 643 | module_exit(xpnet_exit); |
a2d974da DN |
644 | |
645 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
646 | MODULE_DESCRIPTION("Cross Partition Network adapter (XPNET)"); | |
647 | MODULE_LICENSE("GPL"); |