diff options
-rw-r--r-- | arch/ia64/sn/kernel/Makefile | 1 | ||||
-rw-r--r-- | arch/ia64/sn/kernel/xpnet.c | 715 |
2 files changed, 716 insertions, 0 deletions
diff --git a/arch/ia64/sn/kernel/Makefile b/arch/ia64/sn/kernel/Makefile index 6959736eadea..4351c4ff9845 100644 --- a/arch/ia64/sn/kernel/Makefile +++ b/arch/ia64/sn/kernel/Makefile | |||
@@ -15,3 +15,4 @@ obj-$(CONFIG_IA64_SGI_SN_XP) += xp.o | |||
15 | xp-y := xp_main.o xp_nofault.o | 15 | xp-y := xp_main.o xp_nofault.o |
16 | obj-$(CONFIG_IA64_SGI_SN_XP) += xpc.o | 16 | obj-$(CONFIG_IA64_SGI_SN_XP) += xpc.o |
17 | xpc-y := xpc_main.o xpc_channel.o xpc_partition.o | 17 | xpc-y := xpc_main.o xpc_channel.o xpc_partition.o |
18 | obj-$(CONFIG_IA64_SGI_SN_XP) += xpnet.o | ||
diff --git a/arch/ia64/sn/kernel/xpnet.c b/arch/ia64/sn/kernel/xpnet.c new file mode 100644 index 000000000000..78c13d676fa6 --- /dev/null +++ b/arch/ia64/sn/kernel/xpnet.c | |||
@@ -0,0 +1,715 @@ | |||
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 | * | ||
6 | * Copyright (C) 1999,2001-2005 Silicon Graphics, Inc. All rights reserved. | ||
7 | */ | ||
8 | |||
9 | |||
10 | /* | ||
11 | * Cross Partition Network Interface (XPNET) support | ||
12 | * | ||
13 | * XPNET provides a virtual network layered on top of the Cross | ||
14 | * Partition communication layer. | ||
15 | * | ||
16 | * XPNET provides direct point-to-point and broadcast-like support | ||
17 | * for an ethernet-like device. The ethernet broadcast medium is | ||
18 | * replaced with a point-to-point message structure which passes | ||
19 | * pointers to a DMA-capable block that a remote partition should | ||
20 | * retrieve and pass to the upper level networking layer. | ||
21 | * | ||
22 | */ | ||
23 | |||
24 | |||
25 | #include <linux/config.h> | ||
26 | #include <linux/module.h> | ||
27 | #include <linux/kernel.h> | ||
28 | #include <linux/pci.h> | ||
29 | #include <linux/init.h> | ||
30 | #include <linux/ioport.h> | ||
31 | #include <linux/netdevice.h> | ||
32 | #include <linux/etherdevice.h> | ||
33 | #include <linux/delay.h> | ||
34 | #include <linux/ethtool.h> | ||
35 | #include <linux/mii.h> | ||
36 | #include <linux/smp.h> | ||
37 | #include <linux/string.h> | ||
38 | #include <asm/sn/bte.h> | ||
39 | #include <asm/sn/io.h> | ||
40 | #include <asm/sn/sn_sal.h> | ||
41 | #include <asm/types.h> | ||
42 | #include <asm/atomic.h> | ||
43 | #include <asm/sn/xp.h> | ||
44 | |||
45 | |||
46 | /* | ||
47 | * The message payload transferred by XPC. | ||
48 | * | ||
49 | * buf_pa is the physical address where the DMA should pull from. | ||
50 | * | ||
51 | * NOTE: for performance reasons, buf_pa should _ALWAYS_ begin on a | ||
52 | * cacheline boundary. To accomplish this, we record the number of | ||
53 | * bytes from the beginning of the first cacheline to the first useful | ||
54 | * byte of the skb (leadin_ignore) and the number of bytes from the | ||
55 | * last useful byte of the skb to the end of the last cacheline | ||
56 | * (tailout_ignore). | ||
57 | * | ||
58 | * size is the number of bytes to transfer which includes the skb->len | ||
59 | * (useful bytes of the senders skb) plus the leadin and tailout | ||
60 | */ | ||
61 | struct xpnet_message { | ||
62 | u16 version; /* Version for this message */ | ||
63 | u16 embedded_bytes; /* #of bytes embedded in XPC message */ | ||
64 | u32 magic; /* Special number indicating this is xpnet */ | ||
65 | u64 buf_pa; /* phys address of buffer to retrieve */ | ||
66 | u32 size; /* #of bytes in buffer */ | ||
67 | u8 leadin_ignore; /* #of bytes to ignore at the beginning */ | ||
68 | u8 tailout_ignore; /* #of bytes to ignore at the end */ | ||
69 | unsigned char data; /* body of small packets */ | ||
70 | }; | ||
71 | |||
72 | /* | ||
73 | * Determine the size of our message, the cacheline aligned size, | ||
74 | * and then the number of message will request from XPC. | ||
75 | * | ||
76 | * XPC expects each message to exist in an individual cacheline. | ||
77 | */ | ||
78 | #define XPNET_MSG_SIZE (L1_CACHE_BYTES - XPC_MSG_PAYLOAD_OFFSET) | ||
79 | #define XPNET_MSG_DATA_MAX \ | ||
80 | (XPNET_MSG_SIZE - (u64)(&((struct xpnet_message *)0)->data)) | ||
81 | #define XPNET_MSG_ALIGNED_SIZE (L1_CACHE_ALIGN(XPNET_MSG_SIZE)) | ||
82 | #define XPNET_MSG_NENTRIES (PAGE_SIZE / XPNET_MSG_ALIGNED_SIZE) | ||
83 | |||
84 | |||
85 | #define XPNET_MAX_KTHREADS (XPNET_MSG_NENTRIES + 1) | ||
86 | #define XPNET_MAX_IDLE_KTHREADS (XPNET_MSG_NENTRIES + 1) | ||
87 | |||
88 | /* | ||
89 | * Version number of XPNET implementation. XPNET can always talk to versions | ||
90 | * with same major #, and never talk to versions with a different version. | ||
91 | */ | ||
92 | #define _XPNET_VERSION(_major, _minor) (((_major) << 4) | (_minor)) | ||
93 | #define XPNET_VERSION_MAJOR(_v) ((_v) >> 4) | ||
94 | #define XPNET_VERSION_MINOR(_v) ((_v) & 0xf) | ||
95 | |||
96 | #define XPNET_VERSION _XPNET_VERSION(1,0) /* version 1.0 */ | ||
97 | #define XPNET_VERSION_EMBED _XPNET_VERSION(1,1) /* version 1.1 */ | ||
98 | #define XPNET_MAGIC 0x88786984 /* "XNET" */ | ||
99 | |||
100 | #define XPNET_VALID_MSG(_m) \ | ||
101 | ((XPNET_VERSION_MAJOR(_m->version) == XPNET_VERSION_MAJOR(XPNET_VERSION)) \ | ||
102 | && (msg->magic == XPNET_MAGIC)) | ||
103 | |||
104 | #define XPNET_DEVICE_NAME "xp0" | ||
105 | |||
106 | |||
107 | /* | ||
108 | * When messages are queued with xpc_send_notify, a kmalloc'd buffer | ||
109 | * of the following type is passed as a notification cookie. When the | ||
110 | * notification function is called, we use the cookie to decide | ||
111 | * whether all outstanding message sends have completed. The skb can | ||
112 | * then be released. | ||
113 | */ | ||
114 | struct xpnet_pending_msg { | ||
115 | struct list_head free_list; | ||
116 | struct sk_buff *skb; | ||
117 | atomic_t use_count; | ||
118 | }; | ||
119 | |||
120 | /* driver specific structure pointed to by the device structure */ | ||
121 | struct xpnet_dev_private { | ||
122 | struct net_device_stats stats; | ||
123 | }; | ||
124 | |||
125 | struct net_device *xpnet_device; | ||
126 | |||
127 | /* | ||
128 | * When we are notified of other partitions activating, we add them to | ||
129 | * our bitmask of partitions to which we broadcast. | ||
130 | */ | ||
131 | static u64 xpnet_broadcast_partitions; | ||
132 | /* protect above */ | ||
133 | static spinlock_t xpnet_broadcast_lock = SPIN_LOCK_UNLOCKED; | ||
134 | |||
135 | /* | ||
136 | * Since the Block Transfer Engine (BTE) is being used for the transfer | ||
137 | * and it relies upon cache-line size transfers, we need to reserve at | ||
138 | * least one cache-line for head and tail alignment. The BTE is | ||
139 | * limited to 8MB transfers. | ||
140 | * | ||
141 | * Testing has shown that changing MTU to greater than 64KB has no effect | ||
142 | * on TCP as the two sides negotiate a Max Segment Size that is limited | ||
143 | * to 64K. Other protocols May use packets greater than this, but for | ||
144 | * now, the default is 64KB. | ||
145 | */ | ||
146 | #define XPNET_MAX_MTU (0x800000UL - L1_CACHE_BYTES) | ||
147 | /* 32KB has been determined to be the ideal */ | ||
148 | #define XPNET_DEF_MTU (0x8000UL) | ||
149 | |||
150 | |||
151 | /* | ||
152 | * The partition id is encapsulated in the MAC address. The following | ||
153 | * define locates the octet the partid is in. | ||
154 | */ | ||
155 | #define XPNET_PARTID_OCTET 1 | ||
156 | #define XPNET_LICENSE_OCTET 2 | ||
157 | |||
158 | |||
159 | /* | ||
160 | * Define the XPNET debug device structure that is to be used with dev_dbg(), | ||
161 | * dev_err(), dev_warn(), and dev_info(). | ||
162 | */ | ||
163 | struct device_driver xpnet_dbg_name = { | ||
164 | .name = "xpnet" | ||
165 | }; | ||
166 | |||
167 | struct device xpnet_dbg_subname = { | ||
168 | .bus_id = {0}, /* set to "" */ | ||
169 | .driver = &xpnet_dbg_name | ||
170 | }; | ||
171 | |||
172 | struct device *xpnet = &xpnet_dbg_subname; | ||
173 | |||
174 | /* | ||
175 | * Packet was recevied by XPC and forwarded to us. | ||
176 | */ | ||
177 | static void | ||
178 | xpnet_receive(partid_t partid, int channel, struct xpnet_message *msg) | ||
179 | { | ||
180 | struct sk_buff *skb; | ||
181 | bte_result_t bret; | ||
182 | struct xpnet_dev_private *priv = | ||
183 | (struct xpnet_dev_private *) xpnet_device->priv; | ||
184 | |||
185 | |||
186 | if (!XPNET_VALID_MSG(msg)) { | ||
187 | /* | ||
188 | * Packet with a different XPC version. Ignore. | ||
189 | */ | ||
190 | xpc_received(partid, channel, (void *) msg); | ||
191 | |||
192 | priv->stats.rx_errors++; | ||
193 | |||
194 | return; | ||
195 | } | ||
196 | dev_dbg(xpnet, "received 0x%lx, %d, %d, %d\n", msg->buf_pa, msg->size, | ||
197 | msg->leadin_ignore, msg->tailout_ignore); | ||
198 | |||
199 | |||
200 | /* reserve an extra cache line */ | ||
201 | skb = dev_alloc_skb(msg->size + L1_CACHE_BYTES); | ||
202 | if (!skb) { | ||
203 | dev_err(xpnet, "failed on dev_alloc_skb(%d)\n", | ||
204 | msg->size + L1_CACHE_BYTES); | ||
205 | |||
206 | xpc_received(partid, channel, (void *) msg); | ||
207 | |||
208 | priv->stats.rx_errors++; | ||
209 | |||
210 | return; | ||
211 | } | ||
212 | |||
213 | /* | ||
214 | * The allocated skb has some reserved space. | ||
215 | * In order to use bte_copy, we need to get the | ||
216 | * skb->data pointer moved forward. | ||
217 | */ | ||
218 | skb_reserve(skb, (L1_CACHE_BYTES - ((u64)skb->data & | ||
219 | (L1_CACHE_BYTES - 1)) + | ||
220 | msg->leadin_ignore)); | ||
221 | |||
222 | /* | ||
223 | * Update the tail pointer to indicate data actually | ||
224 | * transferred. | ||
225 | */ | ||
226 | skb_put(skb, (msg->size - msg->leadin_ignore - msg->tailout_ignore)); | ||
227 | |||
228 | /* | ||
229 | * Move the data over from the the other side. | ||
230 | */ | ||
231 | if ((XPNET_VERSION_MINOR(msg->version) == 1) && | ||
232 | (msg->embedded_bytes != 0)) { | ||
233 | dev_dbg(xpnet, "copying embedded message. memcpy(0x%p, 0x%p, " | ||
234 | "%lu)\n", skb->data, &msg->data, | ||
235 | (size_t) msg->embedded_bytes); | ||
236 | |||
237 | memcpy(skb->data, &msg->data, (size_t) msg->embedded_bytes); | ||
238 | } else { | ||
239 | dev_dbg(xpnet, "transferring buffer to the skb->data area;\n\t" | ||
240 | "bte_copy(0x%p, 0x%p, %hu)\n", (void *)msg->buf_pa, | ||
241 | (void *)__pa((u64)skb->data & ~(L1_CACHE_BYTES - 1)), | ||
242 | msg->size); | ||
243 | |||
244 | bret = bte_copy(msg->buf_pa, | ||
245 | __pa((u64)skb->data & ~(L1_CACHE_BYTES - 1)), | ||
246 | msg->size, (BTE_NOTIFY | BTE_WACQUIRE), NULL); | ||
247 | |||
248 | if (bret != BTE_SUCCESS) { | ||
249 | // >>> Need better way of cleaning skb. Currently skb | ||
250 | // >>> appears in_use and we can't just call | ||
251 | // >>> dev_kfree_skb. | ||
252 | dev_err(xpnet, "bte_copy(0x%p, 0x%p, 0x%hx) returned " | ||
253 | "error=0x%x\n", (void *)msg->buf_pa, | ||
254 | (void *)__pa((u64)skb->data & | ||
255 | ~(L1_CACHE_BYTES - 1)), | ||
256 | msg->size, bret); | ||
257 | |||
258 | xpc_received(partid, channel, (void *) msg); | ||
259 | |||
260 | priv->stats.rx_errors++; | ||
261 | |||
262 | return; | ||
263 | } | ||
264 | } | ||
265 | |||
266 | dev_dbg(xpnet, "<skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | ||
267 | "skb->end=0x%p skb->len=%d\n", (void *) skb->head, | ||
268 | (void *) skb->data, (void *) skb->tail, (void *) skb->end, | ||
269 | skb->len); | ||
270 | |||
271 | skb->dev = xpnet_device; | ||
272 | skb->protocol = eth_type_trans(skb, xpnet_device); | ||
273 | skb->ip_summed = CHECKSUM_UNNECESSARY; | ||
274 | |||
275 | dev_dbg(xpnet, "passing skb to network layer; \n\tskb->head=0x%p " | ||
276 | "skb->data=0x%p skb->tail=0x%p skb->end=0x%p skb->len=%d\n", | ||
277 | (void *) skb->head, (void *) skb->data, (void *) skb->tail, | ||
278 | (void *) skb->end, skb->len); | ||
279 | |||
280 | |||
281 | xpnet_device->last_rx = jiffies; | ||
282 | priv->stats.rx_packets++; | ||
283 | priv->stats.rx_bytes += skb->len + ETH_HLEN; | ||
284 | |||
285 | netif_rx_ni(skb); | ||
286 | xpc_received(partid, channel, (void *) msg); | ||
287 | } | ||
288 | |||
289 | |||
290 | /* | ||
291 | * This is the handler which XPC calls during any sort of change in | ||
292 | * state or message reception on a connection. | ||
293 | */ | ||
294 | static void | ||
295 | xpnet_connection_activity(enum xpc_retval reason, partid_t partid, int channel, | ||
296 | void *data, void *key) | ||
297 | { | ||
298 | long bp; | ||
299 | |||
300 | |||
301 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | ||
302 | DBUG_ON(channel != XPC_NET_CHANNEL); | ||
303 | |||
304 | switch(reason) { | ||
305 | case xpcMsgReceived: /* message received */ | ||
306 | DBUG_ON(data == NULL); | ||
307 | |||
308 | xpnet_receive(partid, channel, (struct xpnet_message *) data); | ||
309 | break; | ||
310 | |||
311 | case xpcConnected: /* connection completed to a partition */ | ||
312 | spin_lock_bh(&xpnet_broadcast_lock); | ||
313 | xpnet_broadcast_partitions |= 1UL << (partid -1 ); | ||
314 | bp = xpnet_broadcast_partitions; | ||
315 | spin_unlock_bh(&xpnet_broadcast_lock); | ||
316 | |||
317 | netif_carrier_on(xpnet_device); | ||
318 | |||
319 | dev_dbg(xpnet, "%s connection created to partition %d; " | ||
320 | "xpnet_broadcast_partitions=0x%lx\n", | ||
321 | xpnet_device->name, partid, bp); | ||
322 | break; | ||
323 | |||
324 | default: | ||
325 | spin_lock_bh(&xpnet_broadcast_lock); | ||
326 | xpnet_broadcast_partitions &= ~(1UL << (partid -1 )); | ||
327 | bp = xpnet_broadcast_partitions; | ||
328 | spin_unlock_bh(&xpnet_broadcast_lock); | ||
329 | |||
330 | if (bp == 0) { | ||
331 | netif_carrier_off(xpnet_device); | ||
332 | } | ||
333 | |||
334 | dev_dbg(xpnet, "%s disconnected from partition %d; " | ||
335 | "xpnet_broadcast_partitions=0x%lx\n", | ||
336 | xpnet_device->name, partid, bp); | ||
337 | break; | ||
338 | |||
339 | } | ||
340 | } | ||
341 | |||
342 | |||
343 | static int | ||
344 | xpnet_dev_open(struct net_device *dev) | ||
345 | { | ||
346 | enum xpc_retval ret; | ||
347 | |||
348 | |||
349 | dev_dbg(xpnet, "calling xpc_connect(%d, 0x%p, NULL, %ld, %ld, %d, " | ||
350 | "%d)\n", XPC_NET_CHANNEL, xpnet_connection_activity, | ||
351 | XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, XPNET_MAX_KTHREADS, | ||
352 | XPNET_MAX_IDLE_KTHREADS); | ||
353 | |||
354 | ret = xpc_connect(XPC_NET_CHANNEL, xpnet_connection_activity, NULL, | ||
355 | XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, | ||
356 | XPNET_MAX_KTHREADS, XPNET_MAX_IDLE_KTHREADS); | ||
357 | if (ret != xpcSuccess) { | ||
358 | dev_err(xpnet, "ifconfig up of %s failed on XPC connect, " | ||
359 | "ret=%d\n", dev->name, ret); | ||
360 | |||
361 | return -ENOMEM; | ||
362 | } | ||
363 | |||
364 | dev_dbg(xpnet, "ifconfig up of %s; XPC connected\n", dev->name); | ||
365 | |||
366 | return 0; | ||
367 | } | ||
368 | |||
369 | |||
370 | static int | ||
371 | xpnet_dev_stop(struct net_device *dev) | ||
372 | { | ||
373 | xpc_disconnect(XPC_NET_CHANNEL); | ||
374 | |||
375 | dev_dbg(xpnet, "ifconfig down of %s; XPC disconnected\n", dev->name); | ||
376 | |||
377 | return 0; | ||
378 | } | ||
379 | |||
380 | |||
381 | static int | ||
382 | xpnet_dev_change_mtu(struct net_device *dev, int new_mtu) | ||
383 | { | ||
384 | /* 68 comes from min TCP+IP+MAC header */ | ||
385 | if ((new_mtu < 68) || (new_mtu > XPNET_MAX_MTU)) { | ||
386 | dev_err(xpnet, "ifconfig %s mtu %d failed; value must be " | ||
387 | "between 68 and %ld\n", dev->name, new_mtu, | ||
388 | XPNET_MAX_MTU); | ||
389 | return -EINVAL; | ||
390 | } | ||
391 | |||
392 | dev->mtu = new_mtu; | ||
393 | dev_dbg(xpnet, "ifconfig %s mtu set to %d\n", dev->name, new_mtu); | ||
394 | return 0; | ||
395 | } | ||
396 | |||
397 | |||
398 | /* | ||
399 | * Required for the net_device structure. | ||
400 | */ | ||
401 | static int | ||
402 | xpnet_dev_set_config(struct net_device *dev, struct ifmap *new_map) | ||
403 | { | ||
404 | return 0; | ||
405 | } | ||
406 | |||
407 | |||
408 | /* | ||
409 | * Return statistics to the caller. | ||
410 | */ | ||
411 | static struct net_device_stats * | ||
412 | xpnet_dev_get_stats(struct net_device *dev) | ||
413 | { | ||
414 | struct xpnet_dev_private *priv; | ||
415 | |||
416 | |||
417 | priv = (struct xpnet_dev_private *) dev->priv; | ||
418 | |||
419 | return &priv->stats; | ||
420 | } | ||
421 | |||
422 | |||
423 | /* | ||
424 | * Notification that the other end has received the message and | ||
425 | * DMA'd the skb information. At this point, they are done with | ||
426 | * our side. When all recipients are done processing, we | ||
427 | * release the skb and then release our pending message structure. | ||
428 | */ | ||
429 | static void | ||
430 | xpnet_send_completed(enum xpc_retval reason, partid_t partid, int channel, | ||
431 | void *__qm) | ||
432 | { | ||
433 | struct xpnet_pending_msg *queued_msg = | ||
434 | (struct xpnet_pending_msg *) __qm; | ||
435 | |||
436 | |||
437 | DBUG_ON(queued_msg == NULL); | ||
438 | |||
439 | dev_dbg(xpnet, "message to %d notified with reason %d\n", | ||
440 | partid, reason); | ||
441 | |||
442 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | ||
443 | dev_dbg(xpnet, "all acks for skb->head=-x%p\n", | ||
444 | (void *) queued_msg->skb->head); | ||
445 | |||
446 | dev_kfree_skb_any(queued_msg->skb); | ||
447 | kfree(queued_msg); | ||
448 | } | ||
449 | } | ||
450 | |||
451 | |||
452 | /* | ||
453 | * Network layer has formatted a packet (skb) and is ready to place it | ||
454 | * "on the wire". Prepare and send an xpnet_message to all partitions | ||
455 | * which have connected with us and are targets of this packet. | ||
456 | * | ||
457 | * MAC-NOTE: For the XPNET driver, the MAC address contains the | ||
458 | * destination partition_id. If the destination partition id word | ||
459 | * is 0xff, this packet is to broadcast to all partitions. | ||
460 | */ | ||
461 | static int | ||
462 | xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | ||
463 | { | ||
464 | struct xpnet_pending_msg *queued_msg; | ||
465 | enum xpc_retval ret; | ||
466 | struct xpnet_message *msg; | ||
467 | u64 start_addr, end_addr; | ||
468 | long dp; | ||
469 | u8 second_mac_octet; | ||
470 | partid_t dest_partid; | ||
471 | struct xpnet_dev_private *priv; | ||
472 | u16 embedded_bytes; | ||
473 | |||
474 | |||
475 | priv = (struct xpnet_dev_private *) dev->priv; | ||
476 | |||
477 | |||
478 | dev_dbg(xpnet, ">skb->head=0x%p skb->data=0x%p skb->tail=0x%p " | ||
479 | "skb->end=0x%p skb->len=%d\n", (void *) skb->head, | ||
480 | (void *) skb->data, (void *) skb->tail, (void *) skb->end, | ||
481 | skb->len); | ||
482 | |||
483 | |||
484 | /* | ||
485 | * The xpnet_pending_msg tracks how many outstanding | ||
486 | * xpc_send_notifies are relying on this skb. When none | ||
487 | * remain, release the skb. | ||
488 | */ | ||
489 | queued_msg = kmalloc(sizeof(struct xpnet_pending_msg), GFP_ATOMIC); | ||
490 | if (queued_msg == NULL) { | ||
491 | dev_warn(xpnet, "failed to kmalloc %ld bytes; dropping " | ||
492 | "packet\n", sizeof(struct xpnet_pending_msg)); | ||
493 | |||
494 | priv->stats.tx_errors++; | ||
495 | |||
496 | return -ENOMEM; | ||
497 | } | ||
498 | |||
499 | |||
500 | /* get the beginning of the first cacheline and end of last */ | ||
501 | start_addr = ((u64) skb->data & ~(L1_CACHE_BYTES - 1)); | ||
502 | end_addr = L1_CACHE_ALIGN((u64) skb->tail); | ||
503 | |||
504 | /* calculate how many bytes to embed in the XPC message */ | ||
505 | embedded_bytes = 0; | ||
506 | if (unlikely(skb->len <= XPNET_MSG_DATA_MAX)) { | ||
507 | /* skb->data does fit so embed */ | ||
508 | embedded_bytes = skb->len; | ||
509 | } | ||
510 | |||
511 | |||
512 | /* | ||
513 | * Since the send occurs asynchronously, we set the count to one | ||
514 | * and begin sending. Any sends that happen to complete before | ||
515 | * we are done sending will not free the skb. We will be left | ||
516 | * with that task during exit. This also handles the case of | ||
517 | * a packet destined for a partition which is no longer up. | ||
518 | */ | ||
519 | atomic_set(&queued_msg->use_count, 1); | ||
520 | queued_msg->skb = skb; | ||
521 | |||
522 | |||
523 | second_mac_octet = skb->data[XPNET_PARTID_OCTET]; | ||
524 | if (second_mac_octet == 0xff) { | ||
525 | /* we are being asked to broadcast to all partitions */ | ||
526 | dp = xpnet_broadcast_partitions; | ||
527 | } else if (second_mac_octet != 0) { | ||
528 | dp = xpnet_broadcast_partitions & | ||
529 | (1UL << (second_mac_octet - 1)); | ||
530 | } else { | ||
531 | /* 0 is an invalid partid. Ignore */ | ||
532 | dp = 0; | ||
533 | } | ||
534 | dev_dbg(xpnet, "destination Partitions mask (dp) = 0x%lx\n", dp); | ||
535 | |||
536 | /* | ||
537 | * If we wanted to allow promiscous mode to work like an | ||
538 | * unswitched network, this would be a good point to OR in a | ||
539 | * mask of partitions which should be receiving all packets. | ||
540 | */ | ||
541 | |||
542 | /* | ||
543 | * Main send loop. | ||
544 | */ | ||
545 | for (dest_partid = 1; dp && dest_partid < XP_MAX_PARTITIONS; | ||
546 | dest_partid++) { | ||
547 | |||
548 | |||
549 | if (!(dp & (1UL << (dest_partid - 1)))) { | ||
550 | /* not destined for this partition */ | ||
551 | continue; | ||
552 | } | ||
553 | |||
554 | /* remove this partition from the destinations mask */ | ||
555 | dp &= ~(1UL << (dest_partid - 1)); | ||
556 | |||
557 | |||
558 | /* found a partition to send to */ | ||
559 | |||
560 | ret = xpc_allocate(dest_partid, XPC_NET_CHANNEL, | ||
561 | XPC_NOWAIT, (void **)&msg); | ||
562 | if (unlikely(ret != xpcSuccess)) { | ||
563 | continue; | ||
564 | } | ||
565 | |||
566 | msg->embedded_bytes = embedded_bytes; | ||
567 | if (unlikely(embedded_bytes != 0)) { | ||
568 | msg->version = XPNET_VERSION_EMBED; | ||
569 | dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n", | ||
570 | &msg->data, skb->data, (size_t) embedded_bytes); | ||
571 | memcpy(&msg->data, skb->data, (size_t) embedded_bytes); | ||
572 | } else { | ||
573 | msg->version = XPNET_VERSION; | ||
574 | } | ||
575 | msg->magic = XPNET_MAGIC; | ||
576 | msg->size = end_addr - start_addr; | ||
577 | msg->leadin_ignore = (u64) skb->data - start_addr; | ||
578 | msg->tailout_ignore = end_addr - (u64) skb->tail; | ||
579 | msg->buf_pa = __pa(start_addr); | ||
580 | |||
581 | dev_dbg(xpnet, "sending XPC message to %d:%d\nmsg->buf_pa=" | ||
582 | "0x%lx, msg->size=%u, msg->leadin_ignore=%u, " | ||
583 | "msg->tailout_ignore=%u\n", dest_partid, | ||
584 | XPC_NET_CHANNEL, msg->buf_pa, msg->size, | ||
585 | msg->leadin_ignore, msg->tailout_ignore); | ||
586 | |||
587 | |||
588 | atomic_inc(&queued_msg->use_count); | ||
589 | |||
590 | ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, msg, | ||
591 | xpnet_send_completed, queued_msg); | ||
592 | if (unlikely(ret != xpcSuccess)) { | ||
593 | atomic_dec(&queued_msg->use_count); | ||
594 | continue; | ||
595 | } | ||
596 | |||
597 | } | ||
598 | |||
599 | if (atomic_dec_return(&queued_msg->use_count) == 0) { | ||
600 | dev_dbg(xpnet, "no partitions to receive packet destined for " | ||
601 | "%d\n", dest_partid); | ||
602 | |||
603 | |||
604 | dev_kfree_skb(skb); | ||
605 | kfree(queued_msg); | ||
606 | } | ||
607 | |||
608 | priv->stats.tx_packets++; | ||
609 | priv->stats.tx_bytes += skb->len; | ||
610 | |||
611 | return 0; | ||
612 | } | ||
613 | |||
614 | |||
615 | /* | ||
616 | * Deal with transmit timeouts coming from the network layer. | ||
617 | */ | ||
618 | static void | ||
619 | xpnet_dev_tx_timeout (struct net_device *dev) | ||
620 | { | ||
621 | struct xpnet_dev_private *priv; | ||
622 | |||
623 | |||
624 | priv = (struct xpnet_dev_private *) dev->priv; | ||
625 | |||
626 | priv->stats.tx_errors++; | ||
627 | return; | ||
628 | } | ||
629 | |||
630 | |||
631 | static int __init | ||
632 | xpnet_init(void) | ||
633 | { | ||
634 | int i; | ||
635 | u32 license_num; | ||
636 | int result = -ENOMEM; | ||
637 | |||
638 | |||
639 | dev_info(xpnet, "registering network device %s\n", XPNET_DEVICE_NAME); | ||
640 | |||
641 | /* | ||
642 | * use ether_setup() to init the majority of our device | ||
643 | * structure and then override the necessary pieces. | ||
644 | */ | ||
645 | xpnet_device = alloc_netdev(sizeof(struct xpnet_dev_private), | ||
646 | XPNET_DEVICE_NAME, ether_setup); | ||
647 | if (xpnet_device == NULL) { | ||
648 | return -ENOMEM; | ||
649 | } | ||
650 | |||
651 | netif_carrier_off(xpnet_device); | ||
652 | |||
653 | xpnet_device->mtu = XPNET_DEF_MTU; | ||
654 | xpnet_device->change_mtu = xpnet_dev_change_mtu; | ||
655 | xpnet_device->open = xpnet_dev_open; | ||
656 | xpnet_device->get_stats = xpnet_dev_get_stats; | ||
657 | xpnet_device->stop = xpnet_dev_stop; | ||
658 | xpnet_device->hard_start_xmit = xpnet_dev_hard_start_xmit; | ||
659 | xpnet_device->tx_timeout = xpnet_dev_tx_timeout; | ||
660 | xpnet_device->set_config = xpnet_dev_set_config; | ||
661 | |||
662 | /* | ||
663 | * Multicast assumes the LSB of the first octet is set for multicast | ||
664 | * MAC addresses. We chose the first octet of the MAC to be unlikely | ||
665 | * to collide with any vendor's officially issued MAC. | ||
666 | */ | ||
667 | xpnet_device->dev_addr[0] = 0xfe; | ||
668 | xpnet_device->dev_addr[XPNET_PARTID_OCTET] = sn_partition_id; | ||
669 | license_num = sn_partition_serial_number_val(); | ||
670 | for (i = 3; i >= 0; i--) { | ||
671 | xpnet_device->dev_addr[XPNET_LICENSE_OCTET + i] = | ||
672 | license_num & 0xff; | ||
673 | license_num = license_num >> 8; | ||
674 | } | ||
675 | |||
676 | /* | ||
677 | * ether_setup() sets this to a multicast device. We are | ||
678 | * really not supporting multicast at this time. | ||
679 | */ | ||
680 | xpnet_device->flags &= ~IFF_MULTICAST; | ||
681 | |||
682 | /* | ||
683 | * No need to checksum as it is a DMA transfer. The BTE will | ||
684 | * report an error if the data is not retrievable and the | ||
685 | * packet will be dropped. | ||
686 | */ | ||
687 | xpnet_device->features = NETIF_F_NO_CSUM; | ||
688 | |||
689 | result = register_netdev(xpnet_device); | ||
690 | if (result != 0) { | ||
691 | free_netdev(xpnet_device); | ||
692 | } | ||
693 | |||
694 | return result; | ||
695 | } | ||
696 | module_init(xpnet_init); | ||
697 | |||
698 | |||
699 | static void __exit | ||
700 | xpnet_exit(void) | ||
701 | { | ||
702 | dev_info(xpnet, "unregistering network device %s\n", | ||
703 | xpnet_device[0].name); | ||
704 | |||
705 | unregister_netdev(xpnet_device); | ||
706 | |||
707 | free_netdev(xpnet_device); | ||
708 | } | ||
709 | module_exit(xpnet_exit); | ||
710 | |||
711 | |||
712 | MODULE_AUTHOR("Silicon Graphics, Inc."); | ||
713 | MODULE_DESCRIPTION("Cross Partition Network adapter (XPNET)"); | ||
714 | MODULE_LICENSE("GPL"); | ||
715 | |||