2 * Copyright (C) 2002 Cluster File Systems, Inc.
3 * Author: Eric Barton <eric@bartonsoftware.com>
5 * Copyright (C) 2002, Lawrence Livermore National Labs (LLNL)
6 * W. Marcus Miller - Based on ksocknal
8 * This file is part of Portals, http://www.sf.net/projects/lustre/
10 * Portals is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Portals is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Portals; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 ptl_handle_ni_t kqswnal_ni;
29 kqswnal_data_t kqswnal_data;
31 kpr_nal_interface_t kqswnal_router_interface = {
34 kprni_fwd: kqswnal_fwd_packet,
35 kprni_notify: NULL, /* we're connectionless */
39 #define QSWNAL_SYSCTL 201
41 #define QSWNAL_SYSCTL_OPTIMIZED_GETS 1
42 #define QSWNAL_SYSCTL_COPY_SMALL_FWD 2
44 static ctl_table kqswnal_ctl_table[] = {
45 {QSWNAL_SYSCTL_OPTIMIZED_GETS, "optimized_gets",
46 &kqswnal_data.kqn_optimized_gets, sizeof (int),
47 0644, NULL, &proc_dointvec},
48 {QSWNAL_SYSCTL_COPY_SMALL_FWD, "copy_small_fwd",
49 &kqswnal_data.kqn_copy_small_fwd, sizeof (int),
50 0644, NULL, &proc_dointvec},
54 static ctl_table kqswnal_top_ctl_table[] = {
55 {QSWNAL_SYSCTL, "qswnal", NULL, 0, 0555, kqswnal_ctl_table},
61 kqswnal_forward(nal_t *nal,
63 void *args, size_t args_len,
64 void *ret, size_t ret_len)
66 kqswnal_data_t *k = nal->nal_data;
67 nal_cb_t *nal_cb = k->kqn_cb;
69 LASSERT (nal == &kqswnal_api);
70 LASSERT (k == &kqswnal_data);
71 LASSERT (nal_cb == &kqswnal_lib);
73 lib_dispatch(nal_cb, k, id, args, ret); /* nal needs k */
78 kqswnal_lock (nal_t *nal, unsigned long *flags)
80 kqswnal_data_t *k = nal->nal_data;
81 nal_cb_t *nal_cb = k->kqn_cb;
83 LASSERT (nal == &kqswnal_api);
84 LASSERT (k == &kqswnal_data);
85 LASSERT (nal_cb == &kqswnal_lib);
87 nal_cb->cb_cli(nal_cb,flags);
91 kqswnal_unlock(nal_t *nal, unsigned long *flags)
93 kqswnal_data_t *k = nal->nal_data;
94 nal_cb_t *nal_cb = k->kqn_cb;
96 LASSERT (nal == &kqswnal_api);
97 LASSERT (k == &kqswnal_data);
98 LASSERT (nal_cb == &kqswnal_lib);
100 nal_cb->cb_sti(nal_cb,flags);
104 kqswnal_shutdown(nal_t *nal, int ni)
106 CDEBUG (D_NET, "shutdown\n");
108 LASSERT (nal == &kqswnal_api);
113 kqswnal_yield( nal_t *nal )
115 CDEBUG (D_NET, "yield\n");
117 if (current->need_resched)
123 kqswnal_init(int interface, ptl_pt_index_t ptl_size, ptl_ac_index_t ac_size,
124 ptl_pid_t requested_pid)
126 ptl_nid_t mynid = kqswnal_elanid2nid (kqswnal_data.kqn_elanid);
127 int nnids = kqswnal_data.kqn_nnodes;
129 CDEBUG(D_NET, "calling lib_init with nid "LPX64" of %d\n", mynid, nnids);
131 lib_init(&kqswnal_lib, mynid, 0, nnids, ptl_size, ac_size);
133 return (&kqswnal_api);
137 kqswnal_get_tx_desc (struct portals_cfg *pcfg)
140 struct list_head *tmp;
142 int index = pcfg->pcfg_count;
145 spin_lock_irqsave (&kqswnal_data.kqn_idletxd_lock, flags);
147 list_for_each (tmp, &kqswnal_data.kqn_activetxds) {
151 ktx = list_entry (tmp, kqswnal_tx_t, ktx_list);
153 pcfg->pcfg_pbuf1 = (char *)ktx;
154 pcfg->pcfg_count = NTOH__u32(ktx->ktx_wire_hdr->type);
155 pcfg->pcfg_size = NTOH__u32(ktx->ktx_wire_hdr->payload_length);
156 pcfg->pcfg_nid = NTOH__u64(ktx->ktx_wire_hdr->dest_nid);
157 pcfg->pcfg_nid2 = ktx->ktx_nid;
158 pcfg->pcfg_misc = ktx->ktx_launcher;
159 pcfg->pcfg_flags = (list_empty (&ktx->ktx_delayed_list) ? 0 : 1) |
160 (!ktx->ktx_isnblk ? 0 : 2) |
161 (ktx->ktx_state << 2);
166 spin_unlock_irqrestore (&kqswnal_data.kqn_idletxd_lock, flags);
171 kqswnal_cmd (struct portals_cfg *pcfg, void *private)
173 LASSERT (pcfg != NULL);
175 switch (pcfg->pcfg_command) {
176 case NAL_CMD_GET_TXDESC:
177 return (kqswnal_get_tx_desc (pcfg));
179 case NAL_CMD_REGISTER_MYNID:
180 CDEBUG (D_IOCTL, "setting NID offset to "LPX64" (was "LPX64")\n",
181 pcfg->pcfg_nid - kqswnal_data.kqn_elanid,
182 kqswnal_data.kqn_nid_offset);
183 kqswnal_data.kqn_nid_offset =
184 pcfg->pcfg_nid - kqswnal_data.kqn_elanid;
185 kqswnal_lib.ni.nid = pcfg->pcfg_nid;
194 kqswnal_finalise (void)
196 switch (kqswnal_data.kqn_init)
203 if (kqswnal_data.kqn_sysctl != NULL)
204 unregister_sysctl_table (kqswnal_data.kqn_sysctl);
206 PORTAL_SYMBOL_UNREGISTER (kqswnal_ni);
207 kportal_nal_unregister(QSWNAL);
211 PtlNIFini (kqswnal_ni);
212 lib_fini (&kqswnal_lib);
216 LASSERT(list_empty(&kqswnal_data.kqn_activetxds));
219 case KQN_INIT_NOTHING:
223 /**********************************************************************/
224 /* Make router stop her calling me and fail any more call-ins */
225 kpr_shutdown (&kqswnal_data.kqn_router);
227 /**********************************************************************/
228 /* flag threads we've started to terminate and wait for all to ack */
230 kqswnal_data.kqn_shuttingdown = 1;
231 wake_up_all (&kqswnal_data.kqn_sched_waitq);
233 while (atomic_read (&kqswnal_data.kqn_nthreads_running) != 0) {
234 CDEBUG(D_NET, "waiting for %d threads to start shutting down\n",
235 atomic_read (&kqswnal_data.kqn_nthreads_running));
236 set_current_state (TASK_UNINTERRUPTIBLE);
237 schedule_timeout (HZ);
240 /**********************************************************************/
241 /* close elan comms */
243 if (kqswnal_data.kqn_eprx_small != NULL)
244 ep_free_rcvr (kqswnal_data.kqn_eprx_small);
246 if (kqswnal_data.kqn_eprx_large != NULL)
247 ep_free_rcvr (kqswnal_data.kqn_eprx_large);
249 if (kqswnal_data.kqn_eptx != NULL)
250 ep_free_xmtr (kqswnal_data.kqn_eptx);
252 if (kqswnal_data.kqn_eprx_small != NULL)
253 ep_remove_large_rcvr (kqswnal_data.kqn_eprx_small);
255 if (kqswnal_data.kqn_eprx_large != NULL)
256 ep_remove_large_rcvr (kqswnal_data.kqn_eprx_large);
258 if (kqswnal_data.kqn_eptx != NULL)
259 ep_free_large_xmtr (kqswnal_data.kqn_eptx);
261 /**********************************************************************/
262 /* flag threads to terminate, wake them and wait for them to die */
264 kqswnal_data.kqn_shuttingdown = 2;
265 wake_up_all (&kqswnal_data.kqn_sched_waitq);
267 while (atomic_read (&kqswnal_data.kqn_nthreads) != 0) {
268 CDEBUG(D_NET, "waiting for %d threads to terminate\n",
269 atomic_read (&kqswnal_data.kqn_nthreads));
270 set_current_state (TASK_UNINTERRUPTIBLE);
271 schedule_timeout (HZ);
274 /**********************************************************************/
275 /* No more threads. No more portals, router or comms callbacks!
276 * I control the horizontals and the verticals...
280 LASSERT (list_empty (&kqswnal_data.kqn_readyrxds));
283 /**********************************************************************/
284 /* Complete any blocked forwarding packets with error
287 while (!list_empty (&kqswnal_data.kqn_idletxd_fwdq))
289 kpr_fwd_desc_t *fwd = list_entry (kqswnal_data.kqn_idletxd_fwdq.next,
290 kpr_fwd_desc_t, kprfd_list);
291 list_del (&fwd->kprfd_list);
292 kpr_fwd_done (&kqswnal_data.kqn_router, fwd, -EHOSTUNREACH);
295 while (!list_empty (&kqswnal_data.kqn_delayedfwds))
297 kpr_fwd_desc_t *fwd = list_entry (kqswnal_data.kqn_delayedfwds.next,
298 kpr_fwd_desc_t, kprfd_list);
299 list_del (&fwd->kprfd_list);
300 kpr_fwd_done (&kqswnal_data.kqn_router, fwd, -EHOSTUNREACH);
303 /**********************************************************************/
304 /* Wait for router to complete any packets I sent her
307 kpr_deregister (&kqswnal_data.kqn_router);
310 /**********************************************************************/
311 /* Unmap message buffers and free all descriptors and buffers
315 /* FTTB, we need to unmap any remaining mapped memory. When
316 * ep_dvma_release() get fixed (and releases any mappings in the
317 * region), we can delete all the code from here --------> */
319 if (kqswnal_data.kqn_txds != NULL) {
322 for (i = 0; i < KQSW_NTXMSGS + KQSW_NNBLK_TXMSGS; i++) {
323 kqswnal_tx_t *ktx = &kqswnal_data.kqn_txds[i];
325 /* If ktx has a buffer, it got mapped; unmap now.
326 * NB only the pre-mapped stuff is still mapped
327 * since all tx descs must be idle */
329 if (ktx->ktx_buffer != NULL)
330 ep_dvma_unload(kqswnal_data.kqn_ep,
331 kqswnal_data.kqn_ep_tx_nmh,
336 if (kqswnal_data.kqn_rxds != NULL) {
339 for (i = 0; i < KQSW_NRXMSGS_SMALL + KQSW_NRXMSGS_LARGE; i++) {
340 kqswnal_rx_t *krx = &kqswnal_data.kqn_rxds[i];
342 /* If krx_pages[0] got allocated, it got mapped.
343 * NB subsequent pages get merged */
345 if (krx->krx_pages[0] != NULL)
346 ep_dvma_unload(kqswnal_data.kqn_ep,
347 kqswnal_data.kqn_ep_rx_nmh,
348 &krx->krx_elanbuffer);
351 /* <----------- to here */
353 if (kqswnal_data.kqn_ep_rx_nmh != NULL)
354 ep_dvma_release(kqswnal_data.kqn_ep, kqswnal_data.kqn_ep_rx_nmh);
356 if (kqswnal_data.kqn_ep_tx_nmh != NULL)
357 ep_dvma_release(kqswnal_data.kqn_ep, kqswnal_data.kqn_ep_tx_nmh);
359 if (kqswnal_data.kqn_eprxdmahandle != NULL)
361 elan3_dvma_unload(kqswnal_data.kqn_ep->DmaState,
362 kqswnal_data.kqn_eprxdmahandle, 0,
363 KQSW_NRXMSGPAGES_SMALL * KQSW_NRXMSGS_SMALL +
364 KQSW_NRXMSGPAGES_LARGE * KQSW_NRXMSGS_LARGE);
366 elan3_dma_release(kqswnal_data.kqn_ep->DmaState,
367 kqswnal_data.kqn_eprxdmahandle);
370 if (kqswnal_data.kqn_eptxdmahandle != NULL)
372 elan3_dvma_unload(kqswnal_data.kqn_ep->DmaState,
373 kqswnal_data.kqn_eptxdmahandle, 0,
374 KQSW_NTXMSGPAGES * (KQSW_NTXMSGS +
377 elan3_dma_release(kqswnal_data.kqn_ep->DmaState,
378 kqswnal_data.kqn_eptxdmahandle);
382 if (kqswnal_data.kqn_txds != NULL)
386 for (i = 0; i < KQSW_NTXMSGS + KQSW_NNBLK_TXMSGS; i++)
388 kqswnal_tx_t *ktx = &kqswnal_data.kqn_txds[i];
390 if (ktx->ktx_buffer != NULL)
391 PORTAL_FREE(ktx->ktx_buffer,
392 KQSW_TX_BUFFER_SIZE);
395 PORTAL_FREE(kqswnal_data.kqn_txds,
396 sizeof (kqswnal_tx_t) * (KQSW_NTXMSGS +
400 if (kqswnal_data.kqn_rxds != NULL)
405 for (i = 0; i < KQSW_NRXMSGS_SMALL + KQSW_NRXMSGS_LARGE; i++)
407 kqswnal_rx_t *krx = &kqswnal_data.kqn_rxds[i];
409 for (j = 0; j < krx->krx_npages; j++)
410 if (krx->krx_pages[j] != NULL)
411 __free_page (krx->krx_pages[j]);
414 PORTAL_FREE(kqswnal_data.kqn_rxds,
415 sizeof(kqswnal_rx_t) * (KQSW_NRXMSGS_SMALL +
416 KQSW_NRXMSGS_LARGE));
419 /* resets flags, pointers to NULL etc */
420 memset(&kqswnal_data, 0, sizeof (kqswnal_data));
422 CDEBUG (D_MALLOC, "done kmem %d\n", atomic_read(&portal_kmemory));
424 printk (KERN_INFO "Lustre: Routing QSW NAL unloaded (final mem %d)\n",
425 atomic_read(&portal_kmemory));
429 kqswnal_initialise (void)
432 EP_RAILMASK all_rails = EP_RAILMASK_ALL;
434 ELAN3_DMA_REQUEST dmareq;
439 int pkmem = atomic_read(&portal_kmemory);
441 LASSERT (kqswnal_data.kqn_init == KQN_INIT_NOTHING);
443 CDEBUG (D_MALLOC, "start kmem %d\n", atomic_read(&portal_kmemory));
445 kqswnal_api.forward = kqswnal_forward;
446 kqswnal_api.shutdown = kqswnal_shutdown;
447 kqswnal_api.yield = kqswnal_yield;
448 kqswnal_api.validate = NULL; /* our api validate is a NOOP */
449 kqswnal_api.lock = kqswnal_lock;
450 kqswnal_api.unlock = kqswnal_unlock;
451 kqswnal_api.nal_data = &kqswnal_data;
453 kqswnal_lib.nal_data = &kqswnal_data;
455 memset(&kqswnal_rpc_success, 0, sizeof(kqswnal_rpc_success));
456 memset(&kqswnal_rpc_failed, 0, sizeof(kqswnal_rpc_failed));
458 kqswnal_rpc_failed.Data[0] = -ECONNREFUSED;
460 kqswnal_rpc_failed.Status = -ECONNREFUSED;
462 /* ensure all pointers NULL etc */
463 memset (&kqswnal_data, 0, sizeof (kqswnal_data));
465 kqswnal_data.kqn_optimized_gets = KQSW_OPTIMIZED_GETS;
466 kqswnal_data.kqn_copy_small_fwd = KQSW_COPY_SMALL_FWD;
468 kqswnal_data.kqn_cb = &kqswnal_lib;
470 INIT_LIST_HEAD (&kqswnal_data.kqn_idletxds);
471 INIT_LIST_HEAD (&kqswnal_data.kqn_nblk_idletxds);
472 INIT_LIST_HEAD (&kqswnal_data.kqn_activetxds);
473 spin_lock_init (&kqswnal_data.kqn_idletxd_lock);
474 init_waitqueue_head (&kqswnal_data.kqn_idletxd_waitq);
475 INIT_LIST_HEAD (&kqswnal_data.kqn_idletxd_fwdq);
477 INIT_LIST_HEAD (&kqswnal_data.kqn_delayedfwds);
478 INIT_LIST_HEAD (&kqswnal_data.kqn_delayedtxds);
479 INIT_LIST_HEAD (&kqswnal_data.kqn_readyrxds);
481 spin_lock_init (&kqswnal_data.kqn_sched_lock);
482 init_waitqueue_head (&kqswnal_data.kqn_sched_waitq);
484 spin_lock_init (&kqswnal_data.kqn_statelock);
486 /* pointers/lists/locks initialised */
487 kqswnal_data.kqn_init = KQN_INIT_DATA;
490 kqswnal_data.kqn_ep = ep_system();
491 if (kqswnal_data.kqn_ep == NULL) {
492 CERROR("Can't initialise EKC\n");
496 if (ep_waitfor_nodeid(kqswnal_data.kqn_ep) == ELAN_INVALID_NODE) {
497 CERROR("Can't get elan ID\n");
502 /**********************************************************************/
503 /* Find the first Elan device */
505 kqswnal_data.kqn_ep = ep_device (0);
506 if (kqswnal_data.kqn_ep == NULL)
508 CERROR ("Can't get elan device 0\n");
513 kqswnal_data.kqn_nid_offset = 0;
514 kqswnal_data.kqn_nnodes = ep_numnodes (kqswnal_data.kqn_ep);
515 kqswnal_data.kqn_elanid = ep_nodeid (kqswnal_data.kqn_ep);
517 /**********************************************************************/
518 /* Get the transmitter */
520 kqswnal_data.kqn_eptx = ep_alloc_xmtr (kqswnal_data.kqn_ep);
521 if (kqswnal_data.kqn_eptx == NULL)
523 CERROR ("Can't allocate transmitter\n");
528 /**********************************************************************/
529 /* Get the receivers */
531 kqswnal_data.kqn_eprx_small = ep_alloc_rcvr (kqswnal_data.kqn_ep,
532 EP_MSG_SVC_PORTALS_SMALL,
533 KQSW_EP_ENVELOPES_SMALL);
534 if (kqswnal_data.kqn_eprx_small == NULL)
536 CERROR ("Can't install small msg receiver\n");
541 kqswnal_data.kqn_eprx_large = ep_alloc_rcvr (kqswnal_data.kqn_ep,
542 EP_MSG_SVC_PORTALS_LARGE,
543 KQSW_EP_ENVELOPES_LARGE);
544 if (kqswnal_data.kqn_eprx_large == NULL)
546 CERROR ("Can't install large msg receiver\n");
551 /**********************************************************************/
552 /* Reserve Elan address space for transmit descriptors NB we may
553 * either send the contents of associated buffers immediately, or
554 * map them for the peer to suck/blow... */
556 kqswnal_data.kqn_ep_tx_nmh =
557 ep_dvma_reserve(kqswnal_data.kqn_ep,
558 KQSW_NTXMSGPAGES*(KQSW_NTXMSGS+KQSW_NNBLK_TXMSGS),
560 if (kqswnal_data.kqn_ep_tx_nmh == NULL) {
561 CERROR("Can't reserve tx dma space\n");
566 dmareq.Waitfn = DDI_DMA_SLEEP;
567 dmareq.ElanAddr = (E3_Addr) 0;
568 dmareq.Attr = PTE_LOAD_LITTLE_ENDIAN;
569 dmareq.Perm = ELAN_PERM_REMOTEWRITE;
571 rc = elan3_dma_reserve(kqswnal_data.kqn_ep->DmaState,
572 KQSW_NTXMSGPAGES*(KQSW_NTXMSGS+KQSW_NNBLK_TXMSGS),
573 &dmareq, &kqswnal_data.kqn_eptxdmahandle);
574 if (rc != DDI_SUCCESS)
576 CERROR ("Can't reserve rx dma space\n");
581 /**********************************************************************/
582 /* Reserve Elan address space for receive buffers */
584 kqswnal_data.kqn_ep_rx_nmh =
585 ep_dvma_reserve(kqswnal_data.kqn_ep,
586 KQSW_NRXMSGPAGES_SMALL * KQSW_NRXMSGS_SMALL +
587 KQSW_NRXMSGPAGES_LARGE * KQSW_NRXMSGS_LARGE,
589 if (kqswnal_data.kqn_ep_tx_nmh == NULL) {
590 CERROR("Can't reserve rx dma space\n");
595 dmareq.Waitfn = DDI_DMA_SLEEP;
596 dmareq.ElanAddr = (E3_Addr) 0;
597 dmareq.Attr = PTE_LOAD_LITTLE_ENDIAN;
598 dmareq.Perm = ELAN_PERM_REMOTEWRITE;
600 rc = elan3_dma_reserve (kqswnal_data.kqn_ep->DmaState,
601 KQSW_NRXMSGPAGES_SMALL * KQSW_NRXMSGS_SMALL +
602 KQSW_NRXMSGPAGES_LARGE * KQSW_NRXMSGS_LARGE,
603 &dmareq, &kqswnal_data.kqn_eprxdmahandle);
604 if (rc != DDI_SUCCESS)
606 CERROR ("Can't reserve rx dma space\n");
611 /**********************************************************************/
612 /* Allocate/Initialise transmit descriptors */
614 PORTAL_ALLOC(kqswnal_data.kqn_txds,
615 sizeof(kqswnal_tx_t) * (KQSW_NTXMSGS + KQSW_NNBLK_TXMSGS));
616 if (kqswnal_data.kqn_txds == NULL)
622 /* clear flags, null pointers etc */
623 memset(kqswnal_data.kqn_txds, 0,
624 sizeof(kqswnal_tx_t) * (KQSW_NTXMSGS + KQSW_NNBLK_TXMSGS));
625 for (i = 0; i < (KQSW_NTXMSGS + KQSW_NNBLK_TXMSGS); i++)
628 kqswnal_tx_t *ktx = &kqswnal_data.kqn_txds[i];
629 int basepage = i * KQSW_NTXMSGPAGES;
631 PORTAL_ALLOC (ktx->ktx_buffer, KQSW_TX_BUFFER_SIZE);
632 if (ktx->ktx_buffer == NULL)
638 /* Map pre-allocated buffer NOW, to save latency on transmit */
639 premapped_pages = kqswnal_pages_spanned(ktx->ktx_buffer,
640 KQSW_TX_BUFFER_SIZE);
642 ep_dvma_load(kqswnal_data.kqn_ep, NULL,
643 ktx->ktx_buffer, KQSW_TX_BUFFER_SIZE,
644 kqswnal_data.kqn_ep_tx_nmh, basepage,
645 &all_rails, &ktx->ktx_ebuffer);
647 elan3_dvma_kaddr_load (kqswnal_data.kqn_ep->DmaState,
648 kqswnal_data.kqn_eptxdmahandle,
649 ktx->ktx_buffer, KQSW_TX_BUFFER_SIZE,
650 basepage, &ktx->ktx_ebuffer);
652 ktx->ktx_basepage = basepage + premapped_pages; /* message mapping starts here */
653 ktx->ktx_npages = KQSW_NTXMSGPAGES - premapped_pages; /* for this many pages */
655 INIT_LIST_HEAD (&ktx->ktx_delayed_list);
657 ktx->ktx_state = KTX_IDLE;
658 ktx->ktx_isnblk = (i >= KQSW_NTXMSGS);
659 list_add_tail (&ktx->ktx_list,
660 ktx->ktx_isnblk ? &kqswnal_data.kqn_nblk_idletxds :
661 &kqswnal_data.kqn_idletxds);
664 /**********************************************************************/
665 /* Allocate/Initialise receive descriptors */
667 PORTAL_ALLOC (kqswnal_data.kqn_rxds,
668 sizeof (kqswnal_rx_t) * (KQSW_NRXMSGS_SMALL + KQSW_NRXMSGS_LARGE));
669 if (kqswnal_data.kqn_rxds == NULL)
675 memset(kqswnal_data.kqn_rxds, 0, /* clear flags, null pointers etc */
676 sizeof(kqswnal_rx_t) * (KQSW_NRXMSGS_SMALL+KQSW_NRXMSGS_LARGE));
679 for (i = 0; i < KQSW_NRXMSGS_SMALL + KQSW_NRXMSGS_LARGE; i++)
687 kqswnal_rx_t *krx = &kqswnal_data.kqn_rxds[i];
689 if (i < KQSW_NRXMSGS_SMALL)
691 krx->krx_npages = KQSW_NRXMSGPAGES_SMALL;
692 krx->krx_eprx = kqswnal_data.kqn_eprx_small;
696 krx->krx_npages = KQSW_NRXMSGPAGES_LARGE;
697 krx->krx_eprx = kqswnal_data.kqn_eprx_large;
700 LASSERT (krx->krx_npages > 0);
701 for (j = 0; j < krx->krx_npages; j++)
703 krx->krx_pages[j] = alloc_page(GFP_KERNEL);
704 if (krx->krx_pages[j] == NULL)
710 LASSERT(page_address(krx->krx_pages[j]) != NULL);
713 ep_dvma_load(kqswnal_data.kqn_ep, NULL,
714 page_address(krx->krx_pages[j]),
715 PAGE_SIZE, kqswnal_data.kqn_ep_rx_nmh,
716 elan_page_idx, &all_rails, &elanbuffer);
719 krx->krx_elanbuffer = elanbuffer;
721 rc = ep_nmd_merge(&krx->krx_elanbuffer,
722 &krx->krx_elanbuffer,
724 /* NB contiguous mapping */
728 elan3_dvma_kaddr_load(kqswnal_data.kqn_ep->DmaState,
729 kqswnal_data.kqn_eprxdmahandle,
730 page_address(krx->krx_pages[j]),
731 PAGE_SIZE, elan_page_idx,
734 krx->krx_elanbuffer = elanbuffer;
736 /* NB contiguous mapping */
737 LASSERT (elanbuffer == krx->krx_elanbuffer + j * PAGE_SIZE);
743 LASSERT (elan_page_idx ==
744 (KQSW_NRXMSGS_SMALL * KQSW_NRXMSGPAGES_SMALL) +
745 (KQSW_NRXMSGS_LARGE * KQSW_NRXMSGPAGES_LARGE));
747 /**********************************************************************/
748 /* Network interface ready to initialise */
750 rc = PtlNIInit(kqswnal_init, 32, 4, 0, &kqswnal_ni);
753 CERROR ("PtlNIInit failed %d\n", rc);
758 kqswnal_data.kqn_init = KQN_INIT_PTL;
760 /**********************************************************************/
761 /* Queue receives, now that it's OK to run their completion callbacks */
763 for (i = 0; i < KQSW_NRXMSGS_SMALL + KQSW_NRXMSGS_LARGE; i++)
765 kqswnal_rx_t *krx = &kqswnal_data.kqn_rxds[i];
767 /* NB this enqueue can allocate/sleep (attr == 0) */
769 rc = ep_queue_receive(krx->krx_eprx, kqswnal_rxhandler, krx,
770 &krx->krx_elanbuffer, 0);
772 rc = ep_queue_receive(krx->krx_eprx, kqswnal_rxhandler, krx,
774 krx->krx_npages * PAGE_SIZE, 0);
776 if (rc != EP_SUCCESS)
778 CERROR ("failed ep_queue_receive %d\n", rc);
784 /**********************************************************************/
785 /* Spawn scheduling threads */
786 for (i = 0; i < smp_num_cpus; i++)
788 rc = kqswnal_thread_start (kqswnal_scheduler, NULL);
791 CERROR ("failed to spawn scheduling thread: %d\n", rc);
797 /**********************************************************************/
798 /* Connect to the router */
799 rc = kpr_register (&kqswnal_data.kqn_router, &kqswnal_router_interface);
800 CDEBUG(D_NET, "Can't initialise routing interface (rc = %d): not routing\n",rc);
802 rc = kportal_nal_register (QSWNAL, &kqswnal_cmd, NULL);
804 CERROR ("Can't initialise command interface (rc = %d)\n", rc);
810 /* Press on regardless even if registering sysctl doesn't work */
811 kqswnal_data.kqn_sysctl = register_sysctl_table (kqswnal_top_ctl_table, 0);
814 PORTAL_SYMBOL_REGISTER(kqswnal_ni);
815 kqswnal_data.kqn_init = KQN_INIT_ALL;
817 printk(KERN_INFO "Lustre: Routing QSW NAL loaded on node %d of %d "
818 "(Routing %s, initial mem %d)\n",
819 kqswnal_data.kqn_elanid, kqswnal_data.kqn_nnodes,
820 kpr_routing (&kqswnal_data.kqn_router) ? "enabled" : "disabled",
827 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
828 MODULE_DESCRIPTION("Kernel Quadrics/Elan NAL v1.01");
829 MODULE_LICENSE("GPL");
831 module_init (kqswnal_initialise);
832 module_exit (kqswnal_finalise);
834 EXPORT_SYMBOL (kqswnal_ni);