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 | * | |
714c48f1 | 6 | * Copyright (C) 1999-2009 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 | 23 | #include <linux/module.h> |
a2d974da DN |
24 | #include <linux/netdevice.h> |
25 | #include <linux/etherdevice.h> | |
45d9ca49 | 26 | #include "xp.h" |
a2d974da | 27 | |
a2d974da DN |
28 | /* |
29 | * The message payload transferred by XPC. | |
30 | * | |
31 | * buf_pa is the physical address where the DMA should pull from. | |
32 | * | |
33 | * NOTE: for performance reasons, buf_pa should _ALWAYS_ begin on a | |
34 | * cacheline boundary. To accomplish this, we record the number of | |
35 | * bytes from the beginning of the first cacheline to the first useful | |
36 | * byte of the skb (leadin_ignore) and the number of bytes from the | |
37 | * last useful byte of the skb to the end of the last cacheline | |
38 | * (tailout_ignore). | |
39 | * | |
40 | * size is the number of bytes to transfer which includes the skb->len | |
41 | * (useful bytes of the senders skb) plus the leadin and tailout | |
42 | */ | |
43 | struct xpnet_message { | |
44 | u16 version; /* Version for this message */ | |
45 | u16 embedded_bytes; /* #of bytes embedded in XPC message */ | |
46 | u32 magic; /* Special number indicating this is xpnet */ | |
a812dcc3 | 47 | unsigned long buf_pa; /* phys address of buffer to retrieve */ |
a2d974da DN |
48 | u32 size; /* #of bytes in buffer */ |
49 | u8 leadin_ignore; /* #of bytes to ignore at the beginning */ | |
50 | u8 tailout_ignore; /* #of bytes to ignore at the end */ | |
51 | unsigned char data; /* body of small packets */ | |
52 | }; | |
53 | ||
54 | /* | |
55 | * Determine the size of our message, the cacheline aligned size, | |
56 | * and then the number of message will request from XPC. | |
57 | * | |
58 | * XPC expects each message to exist in an individual cacheline. | |
59 | */ | |
bd3e64c1 | 60 | #define XPNET_MSG_SIZE XPC_MSG_PAYLOAD_MAX_SIZE |
a2d974da | 61 | #define XPNET_MSG_DATA_MAX \ |
bd3e64c1 DN |
62 | (XPNET_MSG_SIZE - offsetof(struct xpnet_message, data)) |
63 | #define XPNET_MSG_NENTRIES (PAGE_SIZE / XPC_MSG_MAX_SIZE) | |
a2d974da | 64 | |
a2d974da DN |
65 | #define XPNET_MAX_KTHREADS (XPNET_MSG_NENTRIES + 1) |
66 | #define XPNET_MAX_IDLE_KTHREADS (XPNET_MSG_NENTRIES + 1) | |
67 | ||
68 | /* | |
69 | * Version number of XPNET implementation. XPNET can always talk to versions | |
70 | * with same major #, and never talk to versions with a different version. | |
71 | */ | |
72 | #define _XPNET_VERSION(_major, _minor) (((_major) << 4) | (_minor)) | |
73 | #define XPNET_VERSION_MAJOR(_v) ((_v) >> 4) | |
74 | #define XPNET_VERSION_MINOR(_v) ((_v) & 0xf) | |
75 | ||
2c2b94f9 DN |
76 | #define XPNET_VERSION _XPNET_VERSION(1, 0) /* version 1.0 */ |
77 | #define XPNET_VERSION_EMBED _XPNET_VERSION(1, 1) /* version 1.1 */ | |
35190506 | 78 | #define XPNET_MAGIC 0x88786984 /* "XNET" */ |
a2d974da DN |
79 | |
80 | #define XPNET_VALID_MSG(_m) \ | |
81 | ((XPNET_VERSION_MAJOR(_m->version) == XPNET_VERSION_MAJOR(XPNET_VERSION)) \ | |
82 | && (msg->magic == XPNET_MAGIC)) | |
83 | ||
84 | #define XPNET_DEVICE_NAME "xp0" | |
85 | ||
a2d974da DN |
86 | /* |
87 | * When messages are queued with xpc_send_notify, a kmalloc'd buffer | |
88 | * of the following type is passed as a notification cookie. When the | |
89 | * notification function is called, we use the cookie to decide | |
90 | * whether all outstanding message sends have completed. The skb can | |
91 | * then be released. | |
92 | */ | |
93 | struct xpnet_pending_msg { | |
a2d974da DN |
94 | struct sk_buff *skb; |
95 | atomic_t use_count; | |
96 | }; | |
97 | ||
a2d974da DN |
98 | struct net_device *xpnet_device; |
99 | ||
100 | /* | |
101 | * When we are notified of other partitions activating, we add them to | |
102 | * our bitmask of partitions to which we broadcast. | |
103 | */ | |
a7b4d509 | 104 | static unsigned long *xpnet_broadcast_partitions; |
a2d974da | 105 | /* protect above */ |
a9f6a0dd | 106 | static DEFINE_SPINLOCK(xpnet_broadcast_lock); |
a2d974da DN |
107 | |
108 | /* | |
109 | * Since the Block Transfer Engine (BTE) is being used for the transfer | |
110 | * and it relies upon cache-line size transfers, we need to reserve at | |
111 | * least one cache-line for head and tail alignment. The BTE is | |
112 | * limited to 8MB transfers. | |
113 | * | |
114 | * Testing has shown that changing MTU to greater than 64KB has no effect | |
115 | * on TCP as the two sides negotiate a Max Segment Size that is limited | |
116 | * to 64K. Other protocols May use packets greater than this, but for | |
117 | * now, the default is 64KB. | |
118 | */ | |
119 | #define XPNET_MAX_MTU (0x800000UL - L1_CACHE_BYTES) | |
120 | /* 32KB has been determined to be the ideal */ | |
121 | #define XPNET_DEF_MTU (0x8000UL) | |
122 | ||
a2d974da | 123 | /* |
a7b4d509 DN |
124 | * The partid is encapsulated in the MAC address beginning in the following |
125 | * octet and it consists of two octets. | |
a2d974da | 126 | */ |
a7b4d509 DN |
127 | #define XPNET_PARTID_OCTET 2 |
128 | ||
129 | /* Define the XPNET debug device structures to be used with dev_dbg() et al */ | |
a2d974da | 130 | |
a2d974da DN |
131 | struct device_driver xpnet_dbg_name = { |
132 | .name = "xpnet" | |
133 | }; | |
134 | ||
135 | struct device xpnet_dbg_subname = { | |
bb0dc43e | 136 | .init_name = "", /* set to "" */ |
a2d974da DN |
137 | .driver = &xpnet_dbg_name |
138 | }; | |
139 | ||
140 | struct device *xpnet = &xpnet_dbg_subname; | |
141 | ||
142 | /* | |
143 | * Packet was recevied by XPC and forwarded to us. | |
144 | */ | |
145 | static void | |
64d032ba | 146 | xpnet_receive(short partid, int channel, struct xpnet_message *msg) |
a2d974da DN |
147 | { |
148 | struct sk_buff *skb; | |
a812dcc3 | 149 | void *dst; |
908787db | 150 | enum xp_retval ret; |
a2d974da DN |
151 | |
152 | if (!XPNET_VALID_MSG(msg)) { | |
153 | /* | |
154 | * Packet with a different XPC version. Ignore. | |
155 | */ | |
35190506 | 156 | xpc_received(partid, channel, (void *)msg); |
a2d974da | 157 | |
e8ac9c55 | 158 | xpnet_device->stats.rx_errors++; |
a2d974da DN |
159 | |
160 | return; | |
161 | } | |
a812dcc3 DN |
162 | dev_dbg(xpnet, "received 0x%lx, %d, %d, %d\n", msg->buf_pa, msg->size, |
163 | msg->leadin_ignore, msg->tailout_ignore); | |
a2d974da | 164 | |
a2d974da DN |
165 | /* reserve an extra cache line */ |
166 | skb = dev_alloc_skb(msg->size + L1_CACHE_BYTES); | |
167 | if (!skb) { | |
168 | dev_err(xpnet, "failed on dev_alloc_skb(%d)\n", | |
169 | msg->size + L1_CACHE_BYTES); | |
170 | ||
35190506 | 171 | xpc_received(partid, channel, (void *)msg); |
a2d974da | 172 | |
e8ac9c55 | 173 | xpnet_device->stats.rx_errors++; |
a2d974da DN |
174 | |
175 | return; | |
176 | } | |
177 | ||
178 | /* | |
179 | * The allocated skb has some reserved space. | |
908787db | 180 | * In order to use xp_remote_memcpy(), we need to get the |
a2d974da DN |
181 | * skb->data pointer moved forward. |
182 | */ | |
183 | skb_reserve(skb, (L1_CACHE_BYTES - ((u64)skb->data & | |
184 | (L1_CACHE_BYTES - 1)) + | |
185 | msg->leadin_ignore)); | |
186 | ||
187 | /* | |
188 | * Update the tail pointer to indicate data actually | |
189 | * transferred. | |
190 | */ | |
191 | skb_put(skb, (msg->size - msg->leadin_ignore - msg->tailout_ignore)); | |
192 | ||
193 | /* | |
4b3f686d | 194 | * Move the data over from the other side. |
a2d974da DN |
195 | */ |
196 | if ((XPNET_VERSION_MINOR(msg->version) == 1) && | |
35190506 | 197 | (msg->embedded_bytes != 0)) { |
a2d974da DN |
198 | dev_dbg(xpnet, "copying embedded message. memcpy(0x%p, 0x%p, " |
199 | "%lu)\n", skb->data, &msg->data, | |
35190506 | 200 | (size_t)msg->embedded_bytes); |
a2d974da | 201 | |
35190506 DN |
202 | skb_copy_to_linear_data(skb, &msg->data, |
203 | (size_t)msg->embedded_bytes); | |
a2d974da | 204 | } else { |
a812dcc3 | 205 | dst = (void *)((u64)skb->data & ~(L1_CACHE_BYTES - 1)); |
a2d974da | 206 | dev_dbg(xpnet, "transferring buffer to the skb->data area;\n\t" |
a812dcc3 | 207 | "xp_remote_memcpy(0x%p, 0x%p, %hu)\n", dst, |
908787db | 208 | (void *)msg->buf_pa, msg->size); |
a2d974da | 209 | |
a812dcc3 | 210 | ret = xp_remote_memcpy(xp_pa(dst), msg->buf_pa, msg->size); |
908787db | 211 | if (ret != xpSuccess) { |
2c2b94f9 | 212 | /* |
ea57f80c DN |
213 | * !!! Need better way of cleaning skb. Currently skb |
214 | * !!! appears in_use and we can't just call | |
215 | * !!! dev_kfree_skb. | |
2c2b94f9 | 216 | */ |
908787db | 217 | dev_err(xpnet, "xp_remote_memcpy(0x%p, 0x%p, 0x%hx) " |
a812dcc3 | 218 | "returned error=0x%x\n", dst, |
908787db | 219 | (void *)msg->buf_pa, msg->size, ret); |
a2d974da | 220 | |
35190506 | 221 | xpc_received(partid, channel, (void *)msg); |
a2d974da | 222 | |
e8ac9c55 | 223 | xpnet_device->stats.rx_errors++; |
a2d974da DN |
224 | |
225 | return; | |
226 | } | |
227 | } | |
228 | ||
229 | dev_dbg(xpnet, "<skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | |
35190506 | 230 | "skb->end=0x%p skb->len=%d\n", (void *)skb->head, |
4305b541 | 231 | (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb), |
a2d974da DN |
232 | skb->len); |
233 | ||
a2d974da DN |
234 | skb->protocol = eth_type_trans(skb, xpnet_device); |
235 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
236 | ||
898eb71c | 237 | dev_dbg(xpnet, "passing skb to network layer\n" |
ad361c98 | 238 | "\tskb->head=0x%p skb->data=0x%p skb->tail=0x%p " |
898eb71c | 239 | "skb->end=0x%p skb->len=%d\n", |
27a884dc | 240 | (void *)skb->head, (void *)skb->data, skb_tail_pointer(skb), |
4305b541 | 241 | skb_end_pointer(skb), skb->len); |
a2d974da | 242 | |
e8ac9c55 SH |
243 | xpnet_device->stats.rx_packets++; |
244 | xpnet_device->stats.rx_bytes += skb->len + ETH_HLEN; | |
a2d974da DN |
245 | |
246 | netif_rx_ni(skb); | |
35190506 | 247 | xpc_received(partid, channel, (void *)msg); |
a2d974da DN |
248 | } |
249 | ||
a2d974da DN |
250 | /* |
251 | * This is the handler which XPC calls during any sort of change in | |
252 | * state or message reception on a connection. | |
253 | */ | |
254 | static void | |
64d032ba | 255 | xpnet_connection_activity(enum xp_retval reason, short partid, int channel, |
a2d974da DN |
256 | void *data, void *key) |
257 | { | |
bc63d387 | 258 | DBUG_ON(partid < 0 || partid >= xp_max_npartitions); |
a2d974da DN |
259 | DBUG_ON(channel != XPC_NET_CHANNEL); |
260 | ||
35190506 | 261 | switch (reason) { |
65c17b80 | 262 | case xpMsgReceived: /* message received */ |
a2d974da DN |
263 | DBUG_ON(data == NULL); |
264 | ||
35190506 | 265 | xpnet_receive(partid, channel, (struct xpnet_message *)data); |
a2d974da DN |
266 | break; |
267 | ||
65c17b80 | 268 | case xpConnected: /* connection completed to a partition */ |
a2d974da | 269 | spin_lock_bh(&xpnet_broadcast_lock); |
a7b4d509 | 270 | __set_bit(partid, xpnet_broadcast_partitions); |
a2d974da DN |
271 | spin_unlock_bh(&xpnet_broadcast_lock); |
272 | ||
273 | netif_carrier_on(xpnet_device); | |
274 | ||
a7b4d509 DN |
275 | dev_dbg(xpnet, "%s connected to partition %d\n", |
276 | xpnet_device->name, partid); | |
a2d974da DN |
277 | break; |
278 | ||
279 | default: | |
280 | spin_lock_bh(&xpnet_broadcast_lock); | |
a7b4d509 | 281 | __clear_bit(partid, xpnet_broadcast_partitions); |
a2d974da DN |
282 | spin_unlock_bh(&xpnet_broadcast_lock); |
283 | ||
a7b4d509 DN |
284 | if (bitmap_empty((unsigned long *)xpnet_broadcast_partitions, |
285 | xp_max_npartitions)) { | |
a2d974da | 286 | netif_carrier_off(xpnet_device); |
a7b4d509 | 287 | } |
a2d974da | 288 | |
a7b4d509 DN |
289 | dev_dbg(xpnet, "%s disconnected from partition %d\n", |
290 | xpnet_device->name, partid); | |
a2d974da | 291 | break; |
a2d974da DN |
292 | } |
293 | } | |
294 | ||
a2d974da DN |
295 | static int |
296 | xpnet_dev_open(struct net_device *dev) | |
297 | { | |
65c17b80 | 298 | enum xp_retval ret; |
a2d974da | 299 | |
b9ae3bd2 TL |
300 | dev_dbg(xpnet, "calling xpc_connect(%d, 0x%p, NULL, %ld, %ld, %ld, " |
301 | "%ld)\n", XPC_NET_CHANNEL, xpnet_connection_activity, | |
261f3b49 DN |
302 | (unsigned long)XPNET_MSG_SIZE, |
303 | (unsigned long)XPNET_MSG_NENTRIES, | |
304 | (unsigned long)XPNET_MAX_KTHREADS, | |
305 | (unsigned long)XPNET_MAX_IDLE_KTHREADS); | |
a2d974da DN |
306 | |
307 | ret = xpc_connect(XPC_NET_CHANNEL, xpnet_connection_activity, NULL, | |
308 | XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, | |
309 | XPNET_MAX_KTHREADS, XPNET_MAX_IDLE_KTHREADS); | |
65c17b80 | 310 | if (ret != xpSuccess) { |
a2d974da DN |
311 | dev_err(xpnet, "ifconfig up of %s failed on XPC connect, " |
312 | "ret=%d\n", dev->name, ret); | |
313 | ||
314 | return -ENOMEM; | |
315 | } | |
316 | ||
317 | dev_dbg(xpnet, "ifconfig up of %s; XPC connected\n", dev->name); | |
318 | ||
319 | return 0; | |
320 | } | |
321 | ||
a2d974da DN |
322 | static int |
323 | xpnet_dev_stop(struct net_device *dev) | |
324 | { | |
325 | xpc_disconnect(XPC_NET_CHANNEL); | |
326 | ||
327 | dev_dbg(xpnet, "ifconfig down of %s; XPC disconnected\n", dev->name); | |
328 | ||
329 | return 0; | |
330 | } | |
331 | ||
a2d974da DN |
332 | static int |
333 | xpnet_dev_change_mtu(struct net_device *dev, int new_mtu) | |
334 | { | |
335 | /* 68 comes from min TCP+IP+MAC header */ | |
336 | if ((new_mtu < 68) || (new_mtu > XPNET_MAX_MTU)) { | |
337 | dev_err(xpnet, "ifconfig %s mtu %d failed; value must be " | |
338 | "between 68 and %ld\n", dev->name, new_mtu, | |
339 | XPNET_MAX_MTU); | |
340 | return -EINVAL; | |
341 | } | |
342 | ||
343 | dev->mtu = new_mtu; | |
344 | dev_dbg(xpnet, "ifconfig %s mtu set to %d\n", dev->name, new_mtu); | |
345 | return 0; | |
346 | } | |
347 | ||
a2d974da DN |
348 | /* |
349 | * Notification that the other end has received the message and | |
350 | * DMA'd the skb information. At this point, they are done with | |
351 | * our side. When all recipients are done processing, we | |
352 | * release the skb and then release our pending message structure. | |
353 | */ | |
354 | static void | |
64d032ba | 355 | xpnet_send_completed(enum xp_retval reason, short partid, int channel, |
35190506 | 356 | void *__qm) |
a2d974da | 357 | { |
35190506 | 358 | struct xpnet_pending_msg *queued_msg = (struct xpnet_pending_msg *)__qm; |
a2d974da DN |
359 | |
360 | DBUG_ON(queued_msg == NULL); | |
361 | ||
362 | dev_dbg(xpnet, "message to %d notified with reason %d\n", | |
363 | partid, reason); | |
364 | ||
365 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | |
366 | dev_dbg(xpnet, "all acks for skb->head=-x%p\n", | |
35190506 | 367 | (void *)queued_msg->skb->head); |
a2d974da DN |
368 | |
369 | dev_kfree_skb_any(queued_msg->skb); | |
370 | kfree(queued_msg); | |
371 | } | |
372 | } | |
373 | ||
a7b4d509 DN |
374 | static void |
375 | xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg, | |
376 | u64 start_addr, u64 end_addr, u16 embedded_bytes, int dest_partid) | |
377 | { | |
378 | u8 msg_buffer[XPNET_MSG_SIZE]; | |
379 | struct xpnet_message *msg = (struct xpnet_message *)&msg_buffer; | |
bd3e64c1 | 380 | u16 msg_size = sizeof(struct xpnet_message); |
a7b4d509 DN |
381 | enum xp_retval ret; |
382 | ||
383 | msg->embedded_bytes = embedded_bytes; | |
384 | if (unlikely(embedded_bytes != 0)) { | |
385 | msg->version = XPNET_VERSION_EMBED; | |
386 | dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n", | |
387 | &msg->data, skb->data, (size_t)embedded_bytes); | |
388 | skb_copy_from_linear_data(skb, &msg->data, | |
389 | (size_t)embedded_bytes); | |
bd3e64c1 | 390 | msg_size += embedded_bytes - 1; |
a7b4d509 DN |
391 | } else { |
392 | msg->version = XPNET_VERSION; | |
393 | } | |
394 | msg->magic = XPNET_MAGIC; | |
395 | msg->size = end_addr - start_addr; | |
396 | msg->leadin_ignore = (u64)skb->data - start_addr; | |
397 | msg->tailout_ignore = end_addr - (u64)skb_tail_pointer(skb); | |
a812dcc3 | 398 | msg->buf_pa = xp_pa((void *)start_addr); |
a7b4d509 DN |
399 | |
400 | dev_dbg(xpnet, "sending XPC message to %d:%d\n" | |
ad361c98 | 401 | "msg->buf_pa=0x%lx, msg->size=%u, " |
a7b4d509 | 402 | "msg->leadin_ignore=%u, msg->tailout_ignore=%u\n", |
a812dcc3 DN |
403 | dest_partid, XPC_NET_CHANNEL, msg->buf_pa, msg->size, |
404 | msg->leadin_ignore, msg->tailout_ignore); | |
a7b4d509 DN |
405 | |
406 | atomic_inc(&queued_msg->use_count); | |
407 | ||
408 | ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, XPC_NOWAIT, msg, | |
bd3e64c1 | 409 | msg_size, xpnet_send_completed, queued_msg); |
a7b4d509 DN |
410 | if (unlikely(ret != xpSuccess)) |
411 | atomic_dec(&queued_msg->use_count); | |
412 | } | |
413 | ||
a2d974da DN |
414 | /* |
415 | * Network layer has formatted a packet (skb) and is ready to place it | |
416 | * "on the wire". Prepare and send an xpnet_message to all partitions | |
417 | * which have connected with us and are targets of this packet. | |
418 | * | |
419 | * MAC-NOTE: For the XPNET driver, the MAC address contains the | |
a7b4d509 DN |
420 | * destination partid. If the destination partid octets are 0xffff, |
421 | * this packet is to be broadcast to all connected partitions. | |
a2d974da DN |
422 | */ |
423 | static int | |
424 | xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
425 | { | |
426 | struct xpnet_pending_msg *queued_msg; | |
a2d974da | 427 | u64 start_addr, end_addr; |
64d032ba | 428 | short dest_partid; |
a7b4d509 | 429 | u16 embedded_bytes = 0; |
a2d974da DN |
430 | |
431 | dev_dbg(xpnet, ">skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | |
35190506 | 432 | "skb->end=0x%p skb->len=%d\n", (void *)skb->head, |
4305b541 | 433 | (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb), |
a2d974da DN |
434 | skb->len); |
435 | ||
a7b4d509 DN |
436 | if (skb->data[0] == 0x33) { |
437 | dev_kfree_skb(skb); | |
6ed10654 | 438 | return NETDEV_TX_OK; /* nothing needed to be done */ |
a7b4d509 DN |
439 | } |
440 | ||
a2d974da DN |
441 | /* |
442 | * The xpnet_pending_msg tracks how many outstanding | |
443 | * xpc_send_notifies are relying on this skb. When none | |
444 | * remain, release the skb. | |
445 | */ | |
446 | queued_msg = kmalloc(sizeof(struct xpnet_pending_msg), GFP_ATOMIC); | |
447 | if (queued_msg == NULL) { | |
448 | dev_warn(xpnet, "failed to kmalloc %ld bytes; dropping " | |
35190506 | 449 | "packet\n", sizeof(struct xpnet_pending_msg)); |
a2d974da | 450 | |
e8ac9c55 | 451 | dev->stats.tx_errors++; |
3790c8cd PM |
452 | dev_kfree_skb(skb); |
453 | return NETDEV_TX_OK; | |
a2d974da DN |
454 | } |
455 | ||
a2d974da | 456 | /* get the beginning of the first cacheline and end of last */ |
35190506 | 457 | start_addr = ((u64)skb->data & ~(L1_CACHE_BYTES - 1)); |
27a884dc | 458 | end_addr = L1_CACHE_ALIGN((u64)skb_tail_pointer(skb)); |
a2d974da DN |
459 | |
460 | /* calculate how many bytes to embed in the XPC message */ | |
a2d974da DN |
461 | if (unlikely(skb->len <= XPNET_MSG_DATA_MAX)) { |
462 | /* skb->data does fit so embed */ | |
463 | embedded_bytes = skb->len; | |
464 | } | |
465 | ||
a2d974da DN |
466 | /* |
467 | * Since the send occurs asynchronously, we set the count to one | |
468 | * and begin sending. Any sends that happen to complete before | |
469 | * we are done sending will not free the skb. We will be left | |
470 | * with that task during exit. This also handles the case of | |
471 | * a packet destined for a partition which is no longer up. | |
472 | */ | |
473 | atomic_set(&queued_msg->use_count, 1); | |
474 | queued_msg->skb = skb; | |
475 | ||
a7b4d509 | 476 | if (skb->data[0] == 0xff) { |
a2d974da | 477 | /* we are being asked to broadcast to all partitions */ |
a7b4d509 DN |
478 | for_each_bit(dest_partid, xpnet_broadcast_partitions, |
479 | xp_max_npartitions) { | |
a2d974da | 480 | |
a7b4d509 DN |
481 | xpnet_send(skb, queued_msg, start_addr, end_addr, |
482 | embedded_bytes, dest_partid); | |
a2d974da | 483 | } |
a7b4d509 DN |
484 | } else { |
485 | dest_partid = (short)skb->data[XPNET_PARTID_OCTET + 1]; | |
486 | dest_partid |= (short)skb->data[XPNET_PARTID_OCTET + 0] << 8; | |
a2d974da | 487 | |
a7b4d509 DN |
488 | if (dest_partid >= 0 && |
489 | dest_partid < xp_max_npartitions && | |
490 | test_bit(dest_partid, xpnet_broadcast_partitions) != 0) { | |
a2d974da | 491 | |
a7b4d509 DN |
492 | xpnet_send(skb, queued_msg, start_addr, end_addr, |
493 | embedded_bytes, dest_partid); | |
a2d974da | 494 | } |
a2d974da DN |
495 | } |
496 | ||
497 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | |
a2d974da DN |
498 | dev_kfree_skb(skb); |
499 | kfree(queued_msg); | |
500 | } | |
501 | ||
e8ac9c55 SH |
502 | dev->stats.tx_packets++; |
503 | dev->stats.tx_bytes += skb->len; | |
a2d974da | 504 | |
6ed10654 | 505 | return NETDEV_TX_OK; |
a2d974da DN |
506 | } |
507 | ||
a2d974da DN |
508 | /* |
509 | * Deal with transmit timeouts coming from the network layer. | |
510 | */ | |
511 | static void | |
35190506 | 512 | xpnet_dev_tx_timeout(struct net_device *dev) |
a2d974da | 513 | { |
e8ac9c55 | 514 | dev->stats.tx_errors++; |
a2d974da DN |
515 | } |
516 | ||
e8ac9c55 SH |
517 | static const struct net_device_ops xpnet_netdev_ops = { |
518 | .ndo_open = xpnet_dev_open, | |
519 | .ndo_stop = xpnet_dev_stop, | |
520 | .ndo_start_xmit = xpnet_dev_hard_start_xmit, | |
521 | .ndo_change_mtu = xpnet_dev_change_mtu, | |
522 | .ndo_tx_timeout = xpnet_dev_tx_timeout, | |
523 | .ndo_set_mac_address = eth_mac_addr, | |
524 | .ndo_validate_addr = eth_validate_addr, | |
525 | }; | |
526 | ||
a2d974da DN |
527 | static int __init |
528 | xpnet_init(void) | |
529 | { | |
a7b4d509 | 530 | int result; |
a2d974da | 531 | |
a7b4d509 | 532 | if (!is_shub() && !is_uv()) |
408865ce | 533 | return -ENODEV; |
408865ce | 534 | |
a2d974da DN |
535 | dev_info(xpnet, "registering network device %s\n", XPNET_DEVICE_NAME); |
536 | ||
a7b4d509 DN |
537 | xpnet_broadcast_partitions = kzalloc(BITS_TO_LONGS(xp_max_npartitions) * |
538 | sizeof(long), GFP_KERNEL); | |
539 | if (xpnet_broadcast_partitions == NULL) | |
540 | return -ENOMEM; | |
541 | ||
a2d974da DN |
542 | /* |
543 | * use ether_setup() to init the majority of our device | |
544 | * structure and then override the necessary pieces. | |
545 | */ | |
e8ac9c55 | 546 | xpnet_device = alloc_netdev(0, XPNET_DEVICE_NAME, ether_setup); |
a7b4d509 DN |
547 | if (xpnet_device == NULL) { |
548 | kfree(xpnet_broadcast_partitions); | |
a2d974da | 549 | return -ENOMEM; |
a7b4d509 | 550 | } |
a2d974da DN |
551 | |
552 | netif_carrier_off(xpnet_device); | |
553 | ||
714c48f1 | 554 | xpnet_device->netdev_ops = &xpnet_netdev_ops; |
a2d974da | 555 | xpnet_device->mtu = XPNET_DEF_MTU; |
a2d974da DN |
556 | |
557 | /* | |
558 | * Multicast assumes the LSB of the first octet is set for multicast | |
559 | * MAC addresses. We chose the first octet of the MAC to be unlikely | |
560 | * to collide with any vendor's officially issued MAC. | |
561 | */ | |
a7b4d509 DN |
562 | xpnet_device->dev_addr[0] = 0x02; /* locally administered, no OUI */ |
563 | ||
261f3b49 DN |
564 | xpnet_device->dev_addr[XPNET_PARTID_OCTET + 1] = xp_partition_id; |
565 | xpnet_device->dev_addr[XPNET_PARTID_OCTET + 0] = (xp_partition_id >> 8); | |
a2d974da DN |
566 | |
567 | /* | |
568 | * ether_setup() sets this to a multicast device. We are | |
569 | * really not supporting multicast at this time. | |
570 | */ | |
571 | xpnet_device->flags &= ~IFF_MULTICAST; | |
572 | ||
573 | /* | |
574 | * No need to checksum as it is a DMA transfer. The BTE will | |
575 | * report an error if the data is not retrievable and the | |
576 | * packet will be dropped. | |
577 | */ | |
578 | xpnet_device->features = NETIF_F_NO_CSUM; | |
579 | ||
580 | result = register_netdev(xpnet_device); | |
a7b4d509 | 581 | if (result != 0) { |
a2d974da | 582 | free_netdev(xpnet_device); |
a7b4d509 DN |
583 | kfree(xpnet_broadcast_partitions); |
584 | } | |
a2d974da DN |
585 | |
586 | return result; | |
587 | } | |
a2d974da | 588 | |
35190506 | 589 | module_init(xpnet_init); |
a2d974da DN |
590 | |
591 | static void __exit | |
592 | xpnet_exit(void) | |
593 | { | |
594 | dev_info(xpnet, "unregistering network device %s\n", | |
35190506 | 595 | xpnet_device[0].name); |
a2d974da DN |
596 | |
597 | unregister_netdev(xpnet_device); | |
a2d974da | 598 | free_netdev(xpnet_device); |
a7b4d509 | 599 | kfree(xpnet_broadcast_partitions); |
a2d974da | 600 | } |
a2d974da | 601 | |
35190506 | 602 | module_exit(xpnet_exit); |
a2d974da DN |
603 | |
604 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
605 | MODULE_DESCRIPTION("Cross Partition Network adapter (XPNET)"); | |
606 | MODULE_LICENSE("GPL"); |