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
6 * Copyright (c) 2004-2006 Silicon Graphics, Inc. All Rights Reserved.
11 * Cross Partition Communication (XPC) channel support.
13 * This is the part of XPC that manages the channels and
14 * sends/receives messages across them to/from other partitions.
19 #include <linux/kernel.h>
20 #include <linux/init.h>
21 #include <linux/sched.h>
22 #include <linux/cache.h>
23 #include <linux/interrupt.h>
24 #include <linux/slab.h>
25 #include <linux/mutex.h>
26 #include <linux/completion.h>
27 #include <asm/sn/bte.h>
28 #include <asm/sn/sn_sal.h>
29 #include <asm/sn/xpc.h>
33 * Set up the initial values for the XPartition Communication channels.
36 xpc_initialize_channels(struct xpc_partition *part, partid_t partid)
39 struct xpc_channel *ch;
42 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
43 ch = &part->channels[ch_number];
46 ch->number = ch_number;
47 ch->flags = XPC_C_DISCONNECTED;
49 ch->local_GP = &part->local_GPs[ch_number];
50 ch->local_openclose_args =
51 &part->local_openclose_args[ch_number];
53 atomic_set(&ch->kthreads_assigned, 0);
54 atomic_set(&ch->kthreads_idle, 0);
55 atomic_set(&ch->kthreads_active, 0);
57 atomic_set(&ch->references, 0);
58 atomic_set(&ch->n_to_notify, 0);
60 spin_lock_init(&ch->lock);
61 mutex_init(&ch->msg_to_pull_mutex);
62 init_completion(&ch->wdisconnect_wait);
64 atomic_set(&ch->n_on_msg_allocate_wq, 0);
65 init_waitqueue_head(&ch->msg_allocate_wq);
66 init_waitqueue_head(&ch->idle_wq);
72 * Setup the infrastructure necessary to support XPartition Communication
73 * between the specified remote partition and the local one.
76 xpc_setup_infrastructure(struct xpc_partition *part)
79 struct timer_list *timer;
80 partid_t partid = XPC_PARTID(part);
84 * Zero out MOST of the entry for this partition. Only the fields
85 * starting with `nchannels' will be zeroed. The preceding fields must
86 * remain `viable' across partition ups and downs, since they may be
87 * referenced during this memset() operation.
89 memset(&part->nchannels, 0, sizeof(struct xpc_partition) -
90 offsetof(struct xpc_partition, nchannels));
93 * Allocate all of the channel structures as a contiguous chunk of
96 part->channels = kmalloc(sizeof(struct xpc_channel) * XPC_NCHANNELS,
98 if (part->channels == NULL) {
99 dev_err(xpc_chan, "can't get memory for channels\n");
102 memset(part->channels, 0, sizeof(struct xpc_channel) * XPC_NCHANNELS);
104 part->nchannels = XPC_NCHANNELS;
107 /* allocate all the required GET/PUT values */
109 part->local_GPs = xpc_kmalloc_cacheline_aligned(XPC_GP_SIZE,
110 GFP_KERNEL, &part->local_GPs_base);
111 if (part->local_GPs == NULL) {
112 kfree(part->channels);
113 part->channels = NULL;
114 dev_err(xpc_chan, "can't get memory for local get/put "
118 memset(part->local_GPs, 0, XPC_GP_SIZE);
120 part->remote_GPs = xpc_kmalloc_cacheline_aligned(XPC_GP_SIZE,
121 GFP_KERNEL, &part->remote_GPs_base);
122 if (part->remote_GPs == NULL) {
123 kfree(part->channels);
124 part->channels = NULL;
125 kfree(part->local_GPs_base);
126 part->local_GPs = NULL;
127 dev_err(xpc_chan, "can't get memory for remote get/put "
131 memset(part->remote_GPs, 0, XPC_GP_SIZE);
134 /* allocate all the required open and close args */
136 part->local_openclose_args = xpc_kmalloc_cacheline_aligned(
137 XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
138 &part->local_openclose_args_base);
139 if (part->local_openclose_args == NULL) {
140 kfree(part->channels);
141 part->channels = NULL;
142 kfree(part->local_GPs_base);
143 part->local_GPs = NULL;
144 kfree(part->remote_GPs_base);
145 part->remote_GPs = NULL;
146 dev_err(xpc_chan, "can't get memory for local connect args\n");
149 memset(part->local_openclose_args, 0, XPC_OPENCLOSE_ARGS_SIZE);
151 part->remote_openclose_args = xpc_kmalloc_cacheline_aligned(
152 XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
153 &part->remote_openclose_args_base);
154 if (part->remote_openclose_args == NULL) {
155 kfree(part->channels);
156 part->channels = NULL;
157 kfree(part->local_GPs_base);
158 part->local_GPs = NULL;
159 kfree(part->remote_GPs_base);
160 part->remote_GPs = NULL;
161 kfree(part->local_openclose_args_base);
162 part->local_openclose_args = NULL;
163 dev_err(xpc_chan, "can't get memory for remote connect args\n");
166 memset(part->remote_openclose_args, 0, XPC_OPENCLOSE_ARGS_SIZE);
169 xpc_initialize_channels(part, partid);
171 atomic_set(&part->nchannels_active, 0);
172 atomic_set(&part->nchannels_engaged, 0);
175 /* local_IPI_amo were set to 0 by an earlier memset() */
177 /* Initialize this partitions AMO_t structure */
178 part->local_IPI_amo_va = xpc_IPI_init(partid);
180 spin_lock_init(&part->IPI_lock);
182 atomic_set(&part->channel_mgr_requests, 1);
183 init_waitqueue_head(&part->channel_mgr_wq);
185 sprintf(part->IPI_owner, "xpc%02d", partid);
186 ret = request_irq(SGI_XPC_NOTIFY, xpc_notify_IRQ_handler, SA_SHIRQ,
187 part->IPI_owner, (void *) (u64) partid);
189 kfree(part->channels);
190 part->channels = NULL;
191 kfree(part->local_GPs_base);
192 part->local_GPs = NULL;
193 kfree(part->remote_GPs_base);
194 part->remote_GPs = NULL;
195 kfree(part->local_openclose_args_base);
196 part->local_openclose_args = NULL;
197 kfree(part->remote_openclose_args_base);
198 part->remote_openclose_args = NULL;
199 dev_err(xpc_chan, "can't register NOTIFY IRQ handler, "
201 return xpcLackOfResources;
204 /* Setup a timer to check for dropped IPIs */
205 timer = &part->dropped_IPI_timer;
207 timer->function = (void (*)(unsigned long)) xpc_dropped_IPI_check;
208 timer->data = (unsigned long) part;
209 timer->expires = jiffies + XPC_P_DROPPED_IPI_WAIT;
213 * With the setting of the partition setup_state to XPC_P_SETUP, we're
214 * declaring that this partition is ready to go.
216 part->setup_state = XPC_P_SETUP;
220 * Setup the per partition specific variables required by the
221 * remote partition to establish channel connections with us.
223 * The setting of the magic # indicates that these per partition
224 * specific variables are ready to be used.
226 xpc_vars_part[partid].GPs_pa = __pa(part->local_GPs);
227 xpc_vars_part[partid].openclose_args_pa =
228 __pa(part->local_openclose_args);
229 xpc_vars_part[partid].IPI_amo_pa = __pa(part->local_IPI_amo_va);
230 cpuid = raw_smp_processor_id(); /* any CPU in this partition will do */
231 xpc_vars_part[partid].IPI_nasid = cpuid_to_nasid(cpuid);
232 xpc_vars_part[partid].IPI_phys_cpuid = cpu_physical_id(cpuid);
233 xpc_vars_part[partid].nchannels = part->nchannels;
234 xpc_vars_part[partid].magic = XPC_VP_MAGIC1;
241 * Create a wrapper that hides the underlying mechanism for pulling a cacheline
242 * (or multiple cachelines) from a remote partition.
244 * src must be a cacheline aligned physical address on the remote partition.
245 * dst must be a cacheline aligned virtual address on this partition.
246 * cnt must be an cacheline sized
248 static enum xpc_retval
249 xpc_pull_remote_cachelines(struct xpc_partition *part, void *dst,
250 const void *src, size_t cnt)
252 bte_result_t bte_ret;
255 DBUG_ON((u64) src != L1_CACHE_ALIGN((u64) src));
256 DBUG_ON((u64) dst != L1_CACHE_ALIGN((u64) dst));
257 DBUG_ON(cnt != L1_CACHE_ALIGN(cnt));
259 if (part->act_state == XPC_P_DEACTIVATING) {
263 bte_ret = xp_bte_copy((u64) src, (u64) ia64_tpa((u64) dst),
264 (u64) cnt, (BTE_NORMAL | BTE_WACQUIRE), NULL);
265 if (bte_ret == BTE_SUCCESS) {
269 dev_dbg(xpc_chan, "xp_bte_copy() from partition %d failed, ret=%d\n",
270 XPC_PARTID(part), bte_ret);
272 return xpc_map_bte_errors(bte_ret);
277 * Pull the remote per partititon specific variables from the specified
281 xpc_pull_remote_vars_part(struct xpc_partition *part)
283 u8 buffer[L1_CACHE_BYTES * 2];
284 struct xpc_vars_part *pulled_entry_cacheline =
285 (struct xpc_vars_part *) L1_CACHE_ALIGN((u64) buffer);
286 struct xpc_vars_part *pulled_entry;
287 u64 remote_entry_cacheline_pa, remote_entry_pa;
288 partid_t partid = XPC_PARTID(part);
292 /* pull the cacheline that contains the variables we're interested in */
294 DBUG_ON(part->remote_vars_part_pa !=
295 L1_CACHE_ALIGN(part->remote_vars_part_pa));
296 DBUG_ON(sizeof(struct xpc_vars_part) != L1_CACHE_BYTES / 2);
298 remote_entry_pa = part->remote_vars_part_pa +
299 sn_partition_id * sizeof(struct xpc_vars_part);
301 remote_entry_cacheline_pa = (remote_entry_pa & ~(L1_CACHE_BYTES - 1));
303 pulled_entry = (struct xpc_vars_part *) ((u64) pulled_entry_cacheline +
304 (remote_entry_pa & (L1_CACHE_BYTES - 1)));
306 ret = xpc_pull_remote_cachelines(part, pulled_entry_cacheline,
307 (void *) remote_entry_cacheline_pa,
309 if (ret != xpcSuccess) {
310 dev_dbg(xpc_chan, "failed to pull XPC vars_part from "
311 "partition %d, ret=%d\n", partid, ret);
316 /* see if they've been set up yet */
318 if (pulled_entry->magic != XPC_VP_MAGIC1 &&
319 pulled_entry->magic != XPC_VP_MAGIC2) {
321 if (pulled_entry->magic != 0) {
322 dev_dbg(xpc_chan, "partition %d's XPC vars_part for "
323 "partition %d has bad magic value (=0x%lx)\n",
324 partid, sn_partition_id, pulled_entry->magic);
328 /* they've not been initialized yet */
332 if (xpc_vars_part[partid].magic == XPC_VP_MAGIC1) {
334 /* validate the variables */
336 if (pulled_entry->GPs_pa == 0 ||
337 pulled_entry->openclose_args_pa == 0 ||
338 pulled_entry->IPI_amo_pa == 0) {
340 dev_err(xpc_chan, "partition %d's XPC vars_part for "
341 "partition %d are not valid\n", partid,
343 return xpcInvalidAddress;
346 /* the variables we imported look to be valid */
348 part->remote_GPs_pa = pulled_entry->GPs_pa;
349 part->remote_openclose_args_pa =
350 pulled_entry->openclose_args_pa;
351 part->remote_IPI_amo_va =
352 (AMO_t *) __va(pulled_entry->IPI_amo_pa);
353 part->remote_IPI_nasid = pulled_entry->IPI_nasid;
354 part->remote_IPI_phys_cpuid = pulled_entry->IPI_phys_cpuid;
356 if (part->nchannels > pulled_entry->nchannels) {
357 part->nchannels = pulled_entry->nchannels;
360 /* let the other side know that we've pulled their variables */
362 xpc_vars_part[partid].magic = XPC_VP_MAGIC2;
365 if (pulled_entry->magic == XPC_VP_MAGIC1) {
374 * Get the IPI flags and pull the openclose args and/or remote GPs as needed.
377 xpc_get_IPI_flags(struct xpc_partition *part)
379 unsigned long irq_flags;
385 * See if there are any IPI flags to be handled.
388 spin_lock_irqsave(&part->IPI_lock, irq_flags);
389 if ((IPI_amo = part->local_IPI_amo) != 0) {
390 part->local_IPI_amo = 0;
392 spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
395 if (XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(IPI_amo)) {
396 ret = xpc_pull_remote_cachelines(part,
397 part->remote_openclose_args,
398 (void *) part->remote_openclose_args_pa,
399 XPC_OPENCLOSE_ARGS_SIZE);
400 if (ret != xpcSuccess) {
401 XPC_DEACTIVATE_PARTITION(part, ret);
403 dev_dbg(xpc_chan, "failed to pull openclose args from "
404 "partition %d, ret=%d\n", XPC_PARTID(part),
407 /* don't bother processing IPIs anymore */
412 if (XPC_ANY_MSG_IPI_FLAGS_SET(IPI_amo)) {
413 ret = xpc_pull_remote_cachelines(part, part->remote_GPs,
414 (void *) part->remote_GPs_pa,
416 if (ret != xpcSuccess) {
417 XPC_DEACTIVATE_PARTITION(part, ret);
419 dev_dbg(xpc_chan, "failed to pull GPs from partition "
420 "%d, ret=%d\n", XPC_PARTID(part), ret);
422 /* don't bother processing IPIs anymore */
432 * Allocate the local message queue and the notify queue.
434 static enum xpc_retval
435 xpc_allocate_local_msgqueue(struct xpc_channel *ch)
437 unsigned long irq_flags;
442 // >>> may want to check for ch->flags & XPC_C_DISCONNECTING between
443 // >>> iterations of the for-loop, bail if set?
445 // >>> should we impose a minumum #of entries? like 4 or 8?
446 for (nentries = ch->local_nentries; nentries > 0; nentries--) {
448 nbytes = nentries * ch->msg_size;
449 ch->local_msgqueue = xpc_kmalloc_cacheline_aligned(nbytes,
450 (GFP_KERNEL | GFP_DMA),
451 &ch->local_msgqueue_base);
452 if (ch->local_msgqueue == NULL) {
455 memset(ch->local_msgqueue, 0, nbytes);
457 nbytes = nentries * sizeof(struct xpc_notify);
458 ch->notify_queue = kmalloc(nbytes, (GFP_KERNEL | GFP_DMA));
459 if (ch->notify_queue == NULL) {
460 kfree(ch->local_msgqueue_base);
461 ch->local_msgqueue = NULL;
464 memset(ch->notify_queue, 0, nbytes);
466 spin_lock_irqsave(&ch->lock, irq_flags);
467 if (nentries < ch->local_nentries) {
468 dev_dbg(xpc_chan, "nentries=%d local_nentries=%d, "
469 "partid=%d, channel=%d\n", nentries,
470 ch->local_nentries, ch->partid, ch->number);
472 ch->local_nentries = nentries;
474 spin_unlock_irqrestore(&ch->lock, irq_flags);
478 dev_dbg(xpc_chan, "can't get memory for local message queue and notify "
479 "queue, partid=%d, channel=%d\n", ch->partid, ch->number);
485 * Allocate the cached remote message queue.
487 static enum xpc_retval
488 xpc_allocate_remote_msgqueue(struct xpc_channel *ch)
490 unsigned long irq_flags;
495 DBUG_ON(ch->remote_nentries <= 0);
497 // >>> may want to check for ch->flags & XPC_C_DISCONNECTING between
498 // >>> iterations of the for-loop, bail if set?
500 // >>> should we impose a minumum #of entries? like 4 or 8?
501 for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
503 nbytes = nentries * ch->msg_size;
504 ch->remote_msgqueue = xpc_kmalloc_cacheline_aligned(nbytes,
505 (GFP_KERNEL | GFP_DMA),
506 &ch->remote_msgqueue_base);
507 if (ch->remote_msgqueue == NULL) {
510 memset(ch->remote_msgqueue, 0, nbytes);
512 spin_lock_irqsave(&ch->lock, irq_flags);
513 if (nentries < ch->remote_nentries) {
514 dev_dbg(xpc_chan, "nentries=%d remote_nentries=%d, "
515 "partid=%d, channel=%d\n", nentries,
516 ch->remote_nentries, ch->partid, ch->number);
518 ch->remote_nentries = nentries;
520 spin_unlock_irqrestore(&ch->lock, irq_flags);
524 dev_dbg(xpc_chan, "can't get memory for cached remote message queue, "
525 "partid=%d, channel=%d\n", ch->partid, ch->number);
531 * Allocate message queues and other stuff associated with a channel.
533 * Note: Assumes all of the channel sizes are filled in.
535 static enum xpc_retval
536 xpc_allocate_msgqueues(struct xpc_channel *ch)
538 unsigned long irq_flags;
542 DBUG_ON(ch->flags & XPC_C_SETUP);
544 if ((ret = xpc_allocate_local_msgqueue(ch)) != xpcSuccess) {
548 if ((ret = xpc_allocate_remote_msgqueue(ch)) != xpcSuccess) {
549 kfree(ch->local_msgqueue_base);
550 ch->local_msgqueue = NULL;
551 kfree(ch->notify_queue);
552 ch->notify_queue = NULL;
556 spin_lock_irqsave(&ch->lock, irq_flags);
557 ch->flags |= XPC_C_SETUP;
558 spin_unlock_irqrestore(&ch->lock, irq_flags);
565 * Process a connect message from a remote partition.
567 * Note: xpc_process_connect() is expecting to be called with the
568 * spin_lock_irqsave held and will leave it locked upon return.
571 xpc_process_connect(struct xpc_channel *ch, unsigned long *irq_flags)
576 DBUG_ON(!spin_is_locked(&ch->lock));
578 if (!(ch->flags & XPC_C_OPENREQUEST) ||
579 !(ch->flags & XPC_C_ROPENREQUEST)) {
580 /* nothing more to do for now */
583 DBUG_ON(!(ch->flags & XPC_C_CONNECTING));
585 if (!(ch->flags & XPC_C_SETUP)) {
586 spin_unlock_irqrestore(&ch->lock, *irq_flags);
587 ret = xpc_allocate_msgqueues(ch);
588 spin_lock_irqsave(&ch->lock, *irq_flags);
590 if (ret != xpcSuccess) {
591 XPC_DISCONNECT_CHANNEL(ch, ret, irq_flags);
593 if (ch->flags & (XPC_C_CONNECTED | XPC_C_DISCONNECTING)) {
597 DBUG_ON(!(ch->flags & XPC_C_SETUP));
598 DBUG_ON(ch->local_msgqueue == NULL);
599 DBUG_ON(ch->remote_msgqueue == NULL);
602 if (!(ch->flags & XPC_C_OPENREPLY)) {
603 ch->flags |= XPC_C_OPENREPLY;
604 xpc_IPI_send_openreply(ch, irq_flags);
607 if (!(ch->flags & XPC_C_ROPENREPLY)) {
611 DBUG_ON(ch->remote_msgqueue_pa == 0);
613 ch->flags = (XPC_C_CONNECTED | XPC_C_SETUP); /* clear all else */
615 dev_info(xpc_chan, "channel %d to partition %d connected\n",
616 ch->number, ch->partid);
618 spin_unlock_irqrestore(&ch->lock, *irq_flags);
619 xpc_create_kthreads(ch, 1);
620 spin_lock_irqsave(&ch->lock, *irq_flags);
625 * Notify those who wanted to be notified upon delivery of their message.
628 xpc_notify_senders(struct xpc_channel *ch, enum xpc_retval reason, s64 put)
630 struct xpc_notify *notify;
632 s64 get = ch->w_remote_GP.get - 1;
635 while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
637 notify = &ch->notify_queue[get % ch->local_nentries];
640 * See if the notify entry indicates it was associated with
641 * a message who's sender wants to be notified. It is possible
642 * that it is, but someone else is doing or has done the
645 notify_type = notify->type;
646 if (notify_type == 0 ||
647 cmpxchg(¬ify->type, notify_type, 0) !=
652 DBUG_ON(notify_type != XPC_N_CALL);
654 atomic_dec(&ch->n_to_notify);
656 if (notify->func != NULL) {
657 dev_dbg(xpc_chan, "notify->func() called, notify=0x%p, "
658 "msg_number=%ld, partid=%d, channel=%d\n",
659 (void *) notify, get, ch->partid, ch->number);
661 notify->func(reason, ch->partid, ch->number,
664 dev_dbg(xpc_chan, "notify->func() returned, "
665 "notify=0x%p, msg_number=%ld, partid=%d, "
666 "channel=%d\n", (void *) notify, get,
667 ch->partid, ch->number);
674 * Free up message queues and other stuff that were allocated for the specified
677 * Note: ch->reason and ch->reason_line are left set for debugging purposes,
678 * they're cleared when XPC_C_DISCONNECTED is cleared.
681 xpc_free_msgqueues(struct xpc_channel *ch)
683 DBUG_ON(!spin_is_locked(&ch->lock));
684 DBUG_ON(atomic_read(&ch->n_to_notify) != 0);
686 ch->remote_msgqueue_pa = 0;
690 ch->local_nentries = 0;
691 ch->remote_nentries = 0;
692 ch->kthreads_assigned_limit = 0;
693 ch->kthreads_idle_limit = 0;
695 ch->local_GP->get = 0;
696 ch->local_GP->put = 0;
697 ch->remote_GP.get = 0;
698 ch->remote_GP.put = 0;
699 ch->w_local_GP.get = 0;
700 ch->w_local_GP.put = 0;
701 ch->w_remote_GP.get = 0;
702 ch->w_remote_GP.put = 0;
703 ch->next_msg_to_pull = 0;
705 if (ch->flags & XPC_C_SETUP) {
706 ch->flags &= ~XPC_C_SETUP;
708 dev_dbg(xpc_chan, "ch->flags=0x%x, partid=%d, channel=%d\n",
709 ch->flags, ch->partid, ch->number);
711 kfree(ch->local_msgqueue_base);
712 ch->local_msgqueue = NULL;
713 kfree(ch->remote_msgqueue_base);
714 ch->remote_msgqueue = NULL;
715 kfree(ch->notify_queue);
716 ch->notify_queue = NULL;
722 * spin_lock_irqsave() is expected to be held on entry.
725 xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags)
727 struct xpc_partition *part = &xpc_partitions[ch->partid];
728 u32 channel_was_connected = (ch->flags & XPC_C_WASCONNECTED);
731 DBUG_ON(!spin_is_locked(&ch->lock));
733 if (!(ch->flags & XPC_C_DISCONNECTING)) {
737 DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
739 /* make sure all activity has settled down first */
741 if (atomic_read(&ch->references) > 0) {
744 DBUG_ON(atomic_read(&ch->kthreads_assigned) != 0);
746 if (part->act_state == XPC_P_DEACTIVATING) {
747 /* can't proceed until the other side disengages from us */
748 if (xpc_partition_engaged(1UL << ch->partid)) {
754 /* as long as the other side is up do the full protocol */
756 if (!(ch->flags & XPC_C_RCLOSEREQUEST)) {
760 if (!(ch->flags & XPC_C_CLOSEREPLY)) {
761 ch->flags |= XPC_C_CLOSEREPLY;
762 xpc_IPI_send_closereply(ch, irq_flags);
765 if (!(ch->flags & XPC_C_RCLOSEREPLY)) {
770 /* wake those waiting for notify completion */
771 if (atomic_read(&ch->n_to_notify) > 0) {
772 /* >>> we do callout while holding ch->lock */
773 xpc_notify_senders(ch, ch->reason, ch->w_local_GP.put);
776 /* both sides are disconnected now */
778 if (ch->flags & XPC_C_CONNECTCALLOUT) {
779 spin_unlock_irqrestore(&ch->lock, *irq_flags);
780 xpc_disconnect_callout(ch, xpcDisconnected);
781 spin_lock_irqsave(&ch->lock, *irq_flags);
784 /* it's now safe to free the channel's message queues */
785 xpc_free_msgqueues(ch);
787 /* mark disconnected, clear all other flags except XPC_C_WDISCONNECT */
788 ch->flags = (XPC_C_DISCONNECTED | (ch->flags & XPC_C_WDISCONNECT));
790 atomic_dec(&part->nchannels_active);
792 if (channel_was_connected) {
793 dev_info(xpc_chan, "channel %d to partition %d disconnected, "
794 "reason=%d\n", ch->number, ch->partid, ch->reason);
797 if (ch->flags & XPC_C_WDISCONNECT) {
798 /* we won't lose the CPU since we're holding ch->lock */
799 complete(&ch->wdisconnect_wait);
800 } else if (ch->delayed_IPI_flags) {
801 if (part->act_state != XPC_P_DEACTIVATING) {
802 /* time to take action on any delayed IPI flags */
803 spin_lock(&part->IPI_lock);
804 XPC_SET_IPI_FLAGS(part->local_IPI_amo, ch->number,
805 ch->delayed_IPI_flags);
806 spin_unlock(&part->IPI_lock);
808 ch->delayed_IPI_flags = 0;
814 * Process a change in the channel's remote connection state.
817 xpc_process_openclose_IPI(struct xpc_partition *part, int ch_number,
820 unsigned long irq_flags;
821 struct xpc_openclose_args *args =
822 &part->remote_openclose_args[ch_number];
823 struct xpc_channel *ch = &part->channels[ch_number];
824 enum xpc_retval reason;
828 spin_lock_irqsave(&ch->lock, irq_flags);
832 if ((ch->flags & XPC_C_DISCONNECTED) &&
833 (ch->flags & XPC_C_WDISCONNECT)) {
835 * Delay processing IPI flags until thread waiting disconnect
836 * has had a chance to see that the channel is disconnected.
838 ch->delayed_IPI_flags |= IPI_flags;
839 spin_unlock_irqrestore(&ch->lock, irq_flags);
844 if (IPI_flags & XPC_IPI_CLOSEREQUEST) {
846 dev_dbg(xpc_chan, "XPC_IPI_CLOSEREQUEST (reason=%d) received "
847 "from partid=%d, channel=%d\n", args->reason,
848 ch->partid, ch->number);
851 * If RCLOSEREQUEST is set, we're probably waiting for
852 * RCLOSEREPLY. We should find it and a ROPENREQUEST packed
853 * with this RCLOSEREQUEST in the IPI_flags.
856 if (ch->flags & XPC_C_RCLOSEREQUEST) {
857 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
858 DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
859 DBUG_ON(!(ch->flags & XPC_C_CLOSEREPLY));
860 DBUG_ON(ch->flags & XPC_C_RCLOSEREPLY);
862 DBUG_ON(!(IPI_flags & XPC_IPI_CLOSEREPLY));
863 IPI_flags &= ~XPC_IPI_CLOSEREPLY;
864 ch->flags |= XPC_C_RCLOSEREPLY;
866 /* both sides have finished disconnecting */
867 xpc_process_disconnect(ch, &irq_flags);
868 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
872 if (ch->flags & XPC_C_DISCONNECTED) {
873 if (!(IPI_flags & XPC_IPI_OPENREQUEST)) {
874 if ((XPC_GET_IPI_FLAGS(part->local_IPI_amo,
875 ch_number) & XPC_IPI_OPENREQUEST)) {
877 DBUG_ON(ch->delayed_IPI_flags != 0);
878 spin_lock(&part->IPI_lock);
879 XPC_SET_IPI_FLAGS(part->local_IPI_amo,
881 XPC_IPI_CLOSEREQUEST);
882 spin_unlock(&part->IPI_lock);
884 spin_unlock_irqrestore(&ch->lock, irq_flags);
888 XPC_SET_REASON(ch, 0, 0);
889 ch->flags &= ~XPC_C_DISCONNECTED;
891 atomic_inc(&part->nchannels_active);
892 ch->flags |= (XPC_C_CONNECTING | XPC_C_ROPENREQUEST);
895 IPI_flags &= ~(XPC_IPI_OPENREQUEST | XPC_IPI_OPENREPLY);
898 * The meaningful CLOSEREQUEST connection state fields are:
899 * reason = reason connection is to be closed
902 ch->flags |= XPC_C_RCLOSEREQUEST;
904 if (!(ch->flags & XPC_C_DISCONNECTING)) {
905 reason = args->reason;
906 if (reason <= xpcSuccess || reason > xpcUnknownReason) {
907 reason = xpcUnknownReason;
908 } else if (reason == xpcUnregistering) {
909 reason = xpcOtherUnregistering;
912 XPC_DISCONNECT_CHANNEL(ch, reason, &irq_flags);
914 DBUG_ON(IPI_flags & XPC_IPI_CLOSEREPLY);
915 spin_unlock_irqrestore(&ch->lock, irq_flags);
919 xpc_process_disconnect(ch, &irq_flags);
923 if (IPI_flags & XPC_IPI_CLOSEREPLY) {
925 dev_dbg(xpc_chan, "XPC_IPI_CLOSEREPLY received from partid=%d,"
926 " channel=%d\n", ch->partid, ch->number);
928 if (ch->flags & XPC_C_DISCONNECTED) {
929 DBUG_ON(part->act_state != XPC_P_DEACTIVATING);
930 spin_unlock_irqrestore(&ch->lock, irq_flags);
934 DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
936 if (!(ch->flags & XPC_C_RCLOSEREQUEST)) {
937 if ((XPC_GET_IPI_FLAGS(part->local_IPI_amo, ch_number)
938 & XPC_IPI_CLOSEREQUEST)) {
940 DBUG_ON(ch->delayed_IPI_flags != 0);
941 spin_lock(&part->IPI_lock);
942 XPC_SET_IPI_FLAGS(part->local_IPI_amo,
943 ch_number, XPC_IPI_CLOSEREPLY);
944 spin_unlock(&part->IPI_lock);
946 spin_unlock_irqrestore(&ch->lock, irq_flags);
950 ch->flags |= XPC_C_RCLOSEREPLY;
952 if (ch->flags & XPC_C_CLOSEREPLY) {
953 /* both sides have finished disconnecting */
954 xpc_process_disconnect(ch, &irq_flags);
959 if (IPI_flags & XPC_IPI_OPENREQUEST) {
961 dev_dbg(xpc_chan, "XPC_IPI_OPENREQUEST (msg_size=%d, "
962 "local_nentries=%d) received from partid=%d, "
963 "channel=%d\n", args->msg_size, args->local_nentries,
964 ch->partid, ch->number);
966 if (part->act_state == XPC_P_DEACTIVATING ||
967 (ch->flags & XPC_C_ROPENREQUEST)) {
968 spin_unlock_irqrestore(&ch->lock, irq_flags);
972 if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_WDISCONNECT)) {
973 ch->delayed_IPI_flags |= XPC_IPI_OPENREQUEST;
974 spin_unlock_irqrestore(&ch->lock, irq_flags);
977 DBUG_ON(!(ch->flags & (XPC_C_DISCONNECTED |
978 XPC_C_OPENREQUEST)));
979 DBUG_ON(ch->flags & (XPC_C_ROPENREQUEST | XPC_C_ROPENREPLY |
980 XPC_C_OPENREPLY | XPC_C_CONNECTED));
983 * The meaningful OPENREQUEST connection state fields are:
984 * msg_size = size of channel's messages in bytes
985 * local_nentries = remote partition's local_nentries
987 if (args->msg_size == 0 || args->local_nentries == 0) {
988 /* assume OPENREQUEST was delayed by mistake */
989 spin_unlock_irqrestore(&ch->lock, irq_flags);
993 ch->flags |= (XPC_C_ROPENREQUEST | XPC_C_CONNECTING);
994 ch->remote_nentries = args->local_nentries;
997 if (ch->flags & XPC_C_OPENREQUEST) {
998 if (args->msg_size != ch->msg_size) {
999 XPC_DISCONNECT_CHANNEL(ch, xpcUnequalMsgSizes,
1001 spin_unlock_irqrestore(&ch->lock, irq_flags);
1005 ch->msg_size = args->msg_size;
1007 XPC_SET_REASON(ch, 0, 0);
1008 ch->flags &= ~XPC_C_DISCONNECTED;
1010 atomic_inc(&part->nchannels_active);
1013 xpc_process_connect(ch, &irq_flags);
1017 if (IPI_flags & XPC_IPI_OPENREPLY) {
1019 dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY (local_msgqueue_pa=0x%lx, "
1020 "local_nentries=%d, remote_nentries=%d) received from "
1021 "partid=%d, channel=%d\n", args->local_msgqueue_pa,
1022 args->local_nentries, args->remote_nentries,
1023 ch->partid, ch->number);
1025 if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_DISCONNECTED)) {
1026 spin_unlock_irqrestore(&ch->lock, irq_flags);
1029 if (!(ch->flags & XPC_C_OPENREQUEST)) {
1030 XPC_DISCONNECT_CHANNEL(ch, xpcOpenCloseError,
1032 spin_unlock_irqrestore(&ch->lock, irq_flags);
1036 DBUG_ON(!(ch->flags & XPC_C_ROPENREQUEST));
1037 DBUG_ON(ch->flags & XPC_C_CONNECTED);
1040 * The meaningful OPENREPLY connection state fields are:
1041 * local_msgqueue_pa = physical address of remote
1042 * partition's local_msgqueue
1043 * local_nentries = remote partition's local_nentries
1044 * remote_nentries = remote partition's remote_nentries
1046 DBUG_ON(args->local_msgqueue_pa == 0);
1047 DBUG_ON(args->local_nentries == 0);
1048 DBUG_ON(args->remote_nentries == 0);
1050 ch->flags |= XPC_C_ROPENREPLY;
1051 ch->remote_msgqueue_pa = args->local_msgqueue_pa;
1053 if (args->local_nentries < ch->remote_nentries) {
1054 dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY: new "
1055 "remote_nentries=%d, old remote_nentries=%d, "
1056 "partid=%d, channel=%d\n",
1057 args->local_nentries, ch->remote_nentries,
1058 ch->partid, ch->number);
1060 ch->remote_nentries = args->local_nentries;
1062 if (args->remote_nentries < ch->local_nentries) {
1063 dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY: new "
1064 "local_nentries=%d, old local_nentries=%d, "
1065 "partid=%d, channel=%d\n",
1066 args->remote_nentries, ch->local_nentries,
1067 ch->partid, ch->number);
1069 ch->local_nentries = args->remote_nentries;
1072 xpc_process_connect(ch, &irq_flags);
1075 spin_unlock_irqrestore(&ch->lock, irq_flags);
1080 * Attempt to establish a channel connection to a remote partition.
1082 static enum xpc_retval
1083 xpc_connect_channel(struct xpc_channel *ch)
1085 unsigned long irq_flags;
1086 struct xpc_registration *registration = &xpc_registrations[ch->number];
1089 if (mutex_trylock(®istration->mutex) == 0) {
1093 if (!XPC_CHANNEL_REGISTERED(ch->number)) {
1094 mutex_unlock(®istration->mutex);
1095 return xpcUnregistered;
1098 spin_lock_irqsave(&ch->lock, irq_flags);
1100 DBUG_ON(ch->flags & XPC_C_CONNECTED);
1101 DBUG_ON(ch->flags & XPC_C_OPENREQUEST);
1103 if (ch->flags & XPC_C_DISCONNECTING) {
1104 spin_unlock_irqrestore(&ch->lock, irq_flags);
1105 mutex_unlock(®istration->mutex);
1110 /* add info from the channel connect registration to the channel */
1112 ch->kthreads_assigned_limit = registration->assigned_limit;
1113 ch->kthreads_idle_limit = registration->idle_limit;
1114 DBUG_ON(atomic_read(&ch->kthreads_assigned) != 0);
1115 DBUG_ON(atomic_read(&ch->kthreads_idle) != 0);
1116 DBUG_ON(atomic_read(&ch->kthreads_active) != 0);
1118 ch->func = registration->func;
1119 DBUG_ON(registration->func == NULL);
1120 ch->key = registration->key;
1122 ch->local_nentries = registration->nentries;
1124 if (ch->flags & XPC_C_ROPENREQUEST) {
1125 if (registration->msg_size != ch->msg_size) {
1126 /* the local and remote sides aren't the same */
1129 * Because XPC_DISCONNECT_CHANNEL() can block we're
1130 * forced to up the registration sema before we unlock
1131 * the channel lock. But that's okay here because we're
1132 * done with the part that required the registration
1133 * sema. XPC_DISCONNECT_CHANNEL() requires that the
1134 * channel lock be locked and will unlock and relock
1135 * the channel lock as needed.
1137 mutex_unlock(®istration->mutex);
1138 XPC_DISCONNECT_CHANNEL(ch, xpcUnequalMsgSizes,
1140 spin_unlock_irqrestore(&ch->lock, irq_flags);
1141 return xpcUnequalMsgSizes;
1144 ch->msg_size = registration->msg_size;
1146 XPC_SET_REASON(ch, 0, 0);
1147 ch->flags &= ~XPC_C_DISCONNECTED;
1149 atomic_inc(&xpc_partitions[ch->partid].nchannels_active);
1152 mutex_unlock(®istration->mutex);
1155 /* initiate the connection */
1157 ch->flags |= (XPC_C_OPENREQUEST | XPC_C_CONNECTING);
1158 xpc_IPI_send_openrequest(ch, &irq_flags);
1160 xpc_process_connect(ch, &irq_flags);
1162 spin_unlock_irqrestore(&ch->lock, irq_flags);
1169 * Clear some of the msg flags in the local message queue.
1172 xpc_clear_local_msgqueue_flags(struct xpc_channel *ch)
1174 struct xpc_msg *msg;
1178 get = ch->w_remote_GP.get;
1180 msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
1181 (get % ch->local_nentries) * ch->msg_size);
1183 } while (++get < (volatile s64) ch->remote_GP.get);
1188 * Clear some of the msg flags in the remote message queue.
1191 xpc_clear_remote_msgqueue_flags(struct xpc_channel *ch)
1193 struct xpc_msg *msg;
1197 put = ch->w_remote_GP.put;
1199 msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
1200 (put % ch->remote_nentries) * ch->msg_size);
1202 } while (++put < (volatile s64) ch->remote_GP.put);
1207 xpc_process_msg_IPI(struct xpc_partition *part, int ch_number)
1209 struct xpc_channel *ch = &part->channels[ch_number];
1213 ch->remote_GP = part->remote_GPs[ch_number];
1216 /* See what, if anything, has changed for each connected channel */
1218 xpc_msgqueue_ref(ch);
1220 if (ch->w_remote_GP.get == ch->remote_GP.get &&
1221 ch->w_remote_GP.put == ch->remote_GP.put) {
1222 /* nothing changed since GPs were last pulled */
1223 xpc_msgqueue_deref(ch);
1227 if (!(ch->flags & XPC_C_CONNECTED)){
1228 xpc_msgqueue_deref(ch);
1234 * First check to see if messages recently sent by us have been
1235 * received by the other side. (The remote GET value will have
1236 * changed since we last looked at it.)
1239 if (ch->w_remote_GP.get != ch->remote_GP.get) {
1242 * We need to notify any senders that want to be notified
1243 * that their sent messages have been received by their
1244 * intended recipients. We need to do this before updating
1245 * w_remote_GP.get so that we don't allocate the same message
1246 * queue entries prematurely (see xpc_allocate_msg()).
1248 if (atomic_read(&ch->n_to_notify) > 0) {
1250 * Notify senders that messages sent have been
1251 * received and delivered by the other side.
1253 xpc_notify_senders(ch, xpcMsgDelivered,
1258 * Clear msg->flags in previously sent messages, so that
1259 * they're ready for xpc_allocate_msg().
1261 xpc_clear_local_msgqueue_flags(ch);
1263 ch->w_remote_GP.get = ch->remote_GP.get;
1265 dev_dbg(xpc_chan, "w_remote_GP.get changed to %ld, partid=%d, "
1266 "channel=%d\n", ch->w_remote_GP.get, ch->partid,
1270 * If anyone was waiting for message queue entries to become
1271 * available, wake them up.
1273 if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) {
1274 wake_up(&ch->msg_allocate_wq);
1280 * Now check for newly sent messages by the other side. (The remote
1281 * PUT value will have changed since we last looked at it.)
1284 if (ch->w_remote_GP.put != ch->remote_GP.put) {
1286 * Clear msg->flags in previously received messages, so that
1287 * they're ready for xpc_get_deliverable_msg().
1289 xpc_clear_remote_msgqueue_flags(ch);
1291 ch->w_remote_GP.put = ch->remote_GP.put;
1293 dev_dbg(xpc_chan, "w_remote_GP.put changed to %ld, partid=%d, "
1294 "channel=%d\n", ch->w_remote_GP.put, ch->partid,
1297 nmsgs_sent = ch->w_remote_GP.put - ch->w_local_GP.get;
1298 if (nmsgs_sent > 0) {
1299 dev_dbg(xpc_chan, "msgs waiting to be copied and "
1300 "delivered=%d, partid=%d, channel=%d\n",
1301 nmsgs_sent, ch->partid, ch->number);
1303 if (ch->flags & XPC_C_CONNECTCALLOUT) {
1304 xpc_activate_kthreads(ch, nmsgs_sent);
1309 xpc_msgqueue_deref(ch);
1314 xpc_process_channel_activity(struct xpc_partition *part)
1316 unsigned long irq_flags;
1317 u64 IPI_amo, IPI_flags;
1318 struct xpc_channel *ch;
1323 IPI_amo = xpc_get_IPI_flags(part);
1326 * Initiate channel connections for registered channels.
1328 * For each connected channel that has pending messages activate idle
1329 * kthreads and/or create new kthreads as needed.
1332 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1333 ch = &part->channels[ch_number];
1337 * Process any open or close related IPI flags, and then deal
1338 * with connecting or disconnecting the channel as required.
1341 IPI_flags = XPC_GET_IPI_FLAGS(IPI_amo, ch_number);
1343 if (XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(IPI_flags)) {
1344 xpc_process_openclose_IPI(part, ch_number, IPI_flags);
1347 ch_flags = ch->flags; /* need an atomic snapshot of flags */
1349 if (ch_flags & XPC_C_DISCONNECTING) {
1350 spin_lock_irqsave(&ch->lock, irq_flags);
1351 xpc_process_disconnect(ch, &irq_flags);
1352 spin_unlock_irqrestore(&ch->lock, irq_flags);
1356 if (part->act_state == XPC_P_DEACTIVATING) {
1360 if (!(ch_flags & XPC_C_CONNECTED)) {
1361 if (!(ch_flags & XPC_C_OPENREQUEST)) {
1362 DBUG_ON(ch_flags & XPC_C_SETUP);
1363 (void) xpc_connect_channel(ch);
1365 spin_lock_irqsave(&ch->lock, irq_flags);
1366 xpc_process_connect(ch, &irq_flags);
1367 spin_unlock_irqrestore(&ch->lock, irq_flags);
1374 * Process any message related IPI flags, this may involve the
1375 * activation of kthreads to deliver any pending messages sent
1376 * from the other partition.
1379 if (XPC_ANY_MSG_IPI_FLAGS_SET(IPI_flags)) {
1380 xpc_process_msg_IPI(part, ch_number);
1387 * XPC's heartbeat code calls this function to inform XPC that a partition is
1388 * going down. XPC responds by tearing down the XPartition Communication
1389 * infrastructure used for the just downed partition.
1391 * XPC's heartbeat code will never call this function and xpc_partition_up()
1392 * at the same time. Nor will it ever make multiple calls to either function
1396 xpc_partition_going_down(struct xpc_partition *part, enum xpc_retval reason)
1398 unsigned long irq_flags;
1400 struct xpc_channel *ch;
1403 dev_dbg(xpc_chan, "deactivating partition %d, reason=%d\n",
1404 XPC_PARTID(part), reason);
1406 if (!xpc_part_ref(part)) {
1407 /* infrastructure for this partition isn't currently set up */
1412 /* disconnect channels associated with the partition going down */
1414 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1415 ch = &part->channels[ch_number];
1417 xpc_msgqueue_ref(ch);
1418 spin_lock_irqsave(&ch->lock, irq_flags);
1420 XPC_DISCONNECT_CHANNEL(ch, reason, &irq_flags);
1422 spin_unlock_irqrestore(&ch->lock, irq_flags);
1423 xpc_msgqueue_deref(ch);
1426 xpc_wakeup_channel_mgr(part);
1428 xpc_part_deref(part);
1433 * Teardown the infrastructure necessary to support XPartition Communication
1434 * between the specified remote partition and the local one.
1437 xpc_teardown_infrastructure(struct xpc_partition *part)
1439 partid_t partid = XPC_PARTID(part);
1443 * We start off by making this partition inaccessible to local
1444 * processes by marking it as no longer setup. Then we make it
1445 * inaccessible to remote processes by clearing the XPC per partition
1446 * specific variable's magic # (which indicates that these variables
1447 * are no longer valid) and by ignoring all XPC notify IPIs sent to
1451 DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
1452 DBUG_ON(atomic_read(&part->nchannels_active) != 0);
1453 DBUG_ON(part->setup_state != XPC_P_SETUP);
1454 part->setup_state = XPC_P_WTEARDOWN;
1456 xpc_vars_part[partid].magic = 0;
1459 free_irq(SGI_XPC_NOTIFY, (void *) (u64) partid);
1463 * Before proceding with the teardown we have to wait until all
1464 * existing references cease.
1466 wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
1469 /* now we can begin tearing down the infrastructure */
1471 part->setup_state = XPC_P_TORNDOWN;
1473 /* in case we've still got outstanding timers registered... */
1474 del_timer_sync(&part->dropped_IPI_timer);
1476 kfree(part->remote_openclose_args_base);
1477 part->remote_openclose_args = NULL;
1478 kfree(part->local_openclose_args_base);
1479 part->local_openclose_args = NULL;
1480 kfree(part->remote_GPs_base);
1481 part->remote_GPs = NULL;
1482 kfree(part->local_GPs_base);
1483 part->local_GPs = NULL;
1484 kfree(part->channels);
1485 part->channels = NULL;
1486 part->local_IPI_amo_va = NULL;
1491 * Called by XP at the time of channel connection registration to cause
1492 * XPC to establish connections to all currently active partitions.
1495 xpc_initiate_connect(int ch_number)
1498 struct xpc_partition *part;
1499 struct xpc_channel *ch;
1502 DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
1504 for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
1505 part = &xpc_partitions[partid];
1507 if (xpc_part_ref(part)) {
1508 ch = &part->channels[ch_number];
1511 * Initiate the establishment of a connection on the
1512 * newly registered channel to the remote partition.
1514 xpc_wakeup_channel_mgr(part);
1515 xpc_part_deref(part);
1522 xpc_connected_callout(struct xpc_channel *ch)
1524 /* let the registerer know that a connection has been established */
1526 if (ch->func != NULL) {
1527 dev_dbg(xpc_chan, "ch->func() called, reason=xpcConnected, "
1528 "partid=%d, channel=%d\n", ch->partid, ch->number);
1530 ch->func(xpcConnected, ch->partid, ch->number,
1531 (void *) (u64) ch->local_nentries, ch->key);
1533 dev_dbg(xpc_chan, "ch->func() returned, reason=xpcConnected, "
1534 "partid=%d, channel=%d\n", ch->partid, ch->number);
1540 * Called by XP at the time of channel connection unregistration to cause
1541 * XPC to teardown all current connections for the specified channel.
1543 * Before returning xpc_initiate_disconnect() will wait until all connections
1544 * on the specified channel have been closed/torndown. So the caller can be
1545 * assured that they will not be receiving any more callouts from XPC to the
1546 * function they registered via xpc_connect().
1550 * ch_number - channel # to unregister.
1553 xpc_initiate_disconnect(int ch_number)
1555 unsigned long irq_flags;
1557 struct xpc_partition *part;
1558 struct xpc_channel *ch;
1561 DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
1563 /* initiate the channel disconnect for every active partition */
1564 for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
1565 part = &xpc_partitions[partid];
1567 if (xpc_part_ref(part)) {
1568 ch = &part->channels[ch_number];
1569 xpc_msgqueue_ref(ch);
1571 spin_lock_irqsave(&ch->lock, irq_flags);
1573 if (!(ch->flags & XPC_C_DISCONNECTED)) {
1574 ch->flags |= XPC_C_WDISCONNECT;
1576 XPC_DISCONNECT_CHANNEL(ch, xpcUnregistering,
1580 spin_unlock_irqrestore(&ch->lock, irq_flags);
1582 xpc_msgqueue_deref(ch);
1583 xpc_part_deref(part);
1587 xpc_disconnect_wait(ch_number);
1592 * To disconnect a channel, and reflect it back to all who may be waiting.
1594 * An OPEN is not allowed until XPC_C_DISCONNECTING is cleared by
1595 * xpc_process_disconnect(), and if set, XPC_C_WDISCONNECT is cleared by
1596 * xpc_disconnect_wait().
1598 * THE CHANNEL IS TO BE LOCKED BY THE CALLER AND WILL REMAIN LOCKED UPON RETURN.
1601 xpc_disconnect_channel(const int line, struct xpc_channel *ch,
1602 enum xpc_retval reason, unsigned long *irq_flags)
1604 u32 channel_was_connected = (ch->flags & XPC_C_CONNECTED);
1607 DBUG_ON(!spin_is_locked(&ch->lock));
1609 if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_DISCONNECTED)) {
1612 DBUG_ON(!(ch->flags & (XPC_C_CONNECTING | XPC_C_CONNECTED)));
1614 dev_dbg(xpc_chan, "reason=%d, line=%d, partid=%d, channel=%d\n",
1615 reason, line, ch->partid, ch->number);
1617 XPC_SET_REASON(ch, reason, line);
1619 ch->flags |= (XPC_C_CLOSEREQUEST | XPC_C_DISCONNECTING);
1620 /* some of these may not have been set */
1621 ch->flags &= ~(XPC_C_OPENREQUEST | XPC_C_OPENREPLY |
1622 XPC_C_ROPENREQUEST | XPC_C_ROPENREPLY |
1623 XPC_C_CONNECTING | XPC_C_CONNECTED);
1625 xpc_IPI_send_closerequest(ch, irq_flags);
1627 if (channel_was_connected) {
1628 ch->flags |= XPC_C_WASCONNECTED;
1631 spin_unlock_irqrestore(&ch->lock, *irq_flags);
1633 /* wake all idle kthreads so they can exit */
1634 if (atomic_read(&ch->kthreads_idle) > 0) {
1635 wake_up_all(&ch->idle_wq);
1638 /* wake those waiting to allocate an entry from the local msg queue */
1639 if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) {
1640 wake_up(&ch->msg_allocate_wq);
1643 spin_lock_irqsave(&ch->lock, *irq_flags);
1648 xpc_disconnect_callout(struct xpc_channel *ch, enum xpc_retval reason)
1651 * Let the channel's registerer know that the channel is being
1652 * disconnected. We don't want to do this if the registerer was never
1653 * informed of a connection being made.
1656 if (ch->func != NULL) {
1657 dev_dbg(xpc_chan, "ch->func() called, reason=%d, partid=%d, "
1658 "channel=%d\n", reason, ch->partid, ch->number);
1660 ch->func(reason, ch->partid, ch->number, NULL, ch->key);
1662 dev_dbg(xpc_chan, "ch->func() returned, reason=%d, partid=%d, "
1663 "channel=%d\n", reason, ch->partid, ch->number);
1669 * Wait for a message entry to become available for the specified channel,
1670 * but don't wait any longer than 1 jiffy.
1672 static enum xpc_retval
1673 xpc_allocate_msg_wait(struct xpc_channel *ch)
1675 enum xpc_retval ret;
1678 if (ch->flags & XPC_C_DISCONNECTING) {
1679 DBUG_ON(ch->reason == xpcInterrupted); // >>> Is this true?
1683 atomic_inc(&ch->n_on_msg_allocate_wq);
1684 ret = interruptible_sleep_on_timeout(&ch->msg_allocate_wq, 1);
1685 atomic_dec(&ch->n_on_msg_allocate_wq);
1687 if (ch->flags & XPC_C_DISCONNECTING) {
1689 DBUG_ON(ch->reason == xpcInterrupted); // >>> Is this true?
1690 } else if (ret == 0) {
1693 ret = xpcInterrupted;
1701 * Allocate an entry for a message from the message queue associated with the
1702 * specified channel.
1704 static enum xpc_retval
1705 xpc_allocate_msg(struct xpc_channel *ch, u32 flags,
1706 struct xpc_msg **address_of_msg)
1708 struct xpc_msg *msg;
1709 enum xpc_retval ret;
1713 /* this reference will be dropped in xpc_send_msg() */
1714 xpc_msgqueue_ref(ch);
1716 if (ch->flags & XPC_C_DISCONNECTING) {
1717 xpc_msgqueue_deref(ch);
1720 if (!(ch->flags & XPC_C_CONNECTED)) {
1721 xpc_msgqueue_deref(ch);
1722 return xpcNotConnected;
1727 * Get the next available message entry from the local message queue.
1728 * If none are available, we'll make sure that we grab the latest
1735 put = (volatile s64) ch->w_local_GP.put;
1736 if (put - (volatile s64) ch->w_remote_GP.get <
1737 ch->local_nentries) {
1739 /* There are available message entries. We need to try
1740 * to secure one for ourselves. We'll do this by trying
1741 * to increment w_local_GP.put as long as someone else
1742 * doesn't beat us to it. If they do, we'll have to
1745 if (cmpxchg(&ch->w_local_GP.put, put, put + 1) ==
1747 /* we got the entry referenced by put */
1750 continue; /* try again */
1755 * There aren't any available msg entries at this time.
1757 * In waiting for a message entry to become available,
1758 * we set a timeout in case the other side is not
1759 * sending completion IPIs. This lets us fake an IPI
1760 * that will cause the IPI handler to fetch the latest
1761 * GP values as if an IPI was sent by the other side.
1763 if (ret == xpcTimeout) {
1764 xpc_IPI_send_local_msgrequest(ch);
1767 if (flags & XPC_NOWAIT) {
1768 xpc_msgqueue_deref(ch);
1772 ret = xpc_allocate_msg_wait(ch);
1773 if (ret != xpcInterrupted && ret != xpcTimeout) {
1774 xpc_msgqueue_deref(ch);
1780 /* get the message's address and initialize it */
1781 msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
1782 (put % ch->local_nentries) * ch->msg_size);
1785 DBUG_ON(msg->flags != 0);
1788 dev_dbg(xpc_chan, "w_local_GP.put changed to %ld; msg=0x%p, "
1789 "msg_number=%ld, partid=%d, channel=%d\n", put + 1,
1790 (void *) msg, msg->number, ch->partid, ch->number);
1792 *address_of_msg = msg;
1799 * Allocate an entry for a message from the message queue associated with the
1800 * specified channel. NOTE that this routine can sleep waiting for a message
1801 * entry to become available. To not sleep, pass in the XPC_NOWAIT flag.
1805 * partid - ID of partition to which the channel is connected.
1806 * ch_number - channel #.
1807 * flags - see xpc.h for valid flags.
1808 * payload - address of the allocated payload area pointer (filled in on
1809 * return) in which the user-defined message is constructed.
1812 xpc_initiate_allocate(partid_t partid, int ch_number, u32 flags, void **payload)
1814 struct xpc_partition *part = &xpc_partitions[partid];
1815 enum xpc_retval ret = xpcUnknownReason;
1816 struct xpc_msg *msg;
1819 DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
1820 DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
1824 if (xpc_part_ref(part)) {
1825 ret = xpc_allocate_msg(&part->channels[ch_number], flags, &msg);
1826 xpc_part_deref(part);
1829 *payload = &msg->payload;
1838 * Now we actually send the messages that are ready to be sent by advancing
1839 * the local message queue's Put value and then send an IPI to the recipient
1843 xpc_send_msgs(struct xpc_channel *ch, s64 initial_put)
1845 struct xpc_msg *msg;
1846 s64 put = initial_put + 1;
1853 if (put == (volatile s64) ch->w_local_GP.put) {
1857 msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
1858 (put % ch->local_nentries) * ch->msg_size);
1860 if (!(msg->flags & XPC_M_READY)) {
1867 if (put == initial_put) {
1868 /* nothing's changed */
1872 if (cmpxchg_rel(&ch->local_GP->put, initial_put, put) !=
1874 /* someone else beat us to it */
1875 DBUG_ON((volatile s64) ch->local_GP->put < initial_put);
1879 /* we just set the new value of local_GP->put */
1881 dev_dbg(xpc_chan, "local_GP->put changed to %ld, partid=%d, "
1882 "channel=%d\n", put, ch->partid, ch->number);
1887 * We need to ensure that the message referenced by
1888 * local_GP->put is not XPC_M_READY or that local_GP->put
1889 * equals w_local_GP.put, so we'll go have a look.
1895 xpc_IPI_send_msgrequest(ch);
1901 * Common code that does the actual sending of the message by advancing the
1902 * local message queue's Put value and sends an IPI to the partition the
1903 * message is being sent to.
1905 static enum xpc_retval
1906 xpc_send_msg(struct xpc_channel *ch, struct xpc_msg *msg, u8 notify_type,
1907 xpc_notify_func func, void *key)
1909 enum xpc_retval ret = xpcSuccess;
1910 struct xpc_notify *notify = notify;
1911 s64 put, msg_number = msg->number;
1914 DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
1915 DBUG_ON((((u64) msg - (u64) ch->local_msgqueue) / ch->msg_size) !=
1916 msg_number % ch->local_nentries);
1917 DBUG_ON(msg->flags & XPC_M_READY);
1919 if (ch->flags & XPC_C_DISCONNECTING) {
1920 /* drop the reference grabbed in xpc_allocate_msg() */
1921 xpc_msgqueue_deref(ch);
1925 if (notify_type != 0) {
1927 * Tell the remote side to send an ACK interrupt when the
1928 * message has been delivered.
1930 msg->flags |= XPC_M_INTERRUPT;
1932 atomic_inc(&ch->n_to_notify);
1934 notify = &ch->notify_queue[msg_number % ch->local_nentries];
1935 notify->func = func;
1937 notify->type = notify_type;
1939 // >>> is a mb() needed here?
1941 if (ch->flags & XPC_C_DISCONNECTING) {
1943 * An error occurred between our last error check and
1944 * this one. We will try to clear the type field from
1945 * the notify entry. If we succeed then
1946 * xpc_disconnect_channel() didn't already process
1949 if (cmpxchg(¬ify->type, notify_type, 0) ==
1951 atomic_dec(&ch->n_to_notify);
1955 /* drop the reference grabbed in xpc_allocate_msg() */
1956 xpc_msgqueue_deref(ch);
1961 msg->flags |= XPC_M_READY;
1964 * The preceding store of msg->flags must occur before the following
1965 * load of ch->local_GP->put.
1969 /* see if the message is next in line to be sent, if so send it */
1971 put = ch->local_GP->put;
1972 if (put == msg_number) {
1973 xpc_send_msgs(ch, put);
1976 /* drop the reference grabbed in xpc_allocate_msg() */
1977 xpc_msgqueue_deref(ch);
1983 * Send a message previously allocated using xpc_initiate_allocate() on the
1984 * specified channel connected to the specified partition.
1986 * This routine will not wait for the message to be received, nor will
1987 * notification be given when it does happen. Once this routine has returned
1988 * the message entry allocated via xpc_initiate_allocate() is no longer
1989 * accessable to the caller.
1991 * This routine, although called by users, does not call xpc_part_ref() to
1992 * ensure that the partition infrastructure is in place. It relies on the
1993 * fact that we called xpc_msgqueue_ref() in xpc_allocate_msg().
1997 * partid - ID of partition to which the channel is connected.
1998 * ch_number - channel # to send message on.
1999 * payload - pointer to the payload area allocated via
2000 * xpc_initiate_allocate().
2003 xpc_initiate_send(partid_t partid, int ch_number, void *payload)
2005 struct xpc_partition *part = &xpc_partitions[partid];
2006 struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
2007 enum xpc_retval ret;
2010 dev_dbg(xpc_chan, "msg=0x%p, partid=%d, channel=%d\n", (void *) msg,
2013 DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
2014 DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
2015 DBUG_ON(msg == NULL);
2017 ret = xpc_send_msg(&part->channels[ch_number], msg, 0, NULL, NULL);
2024 * Send a message previously allocated using xpc_initiate_allocate on the
2025 * specified channel connected to the specified partition.
2027 * This routine will not wait for the message to be sent. Once this routine
2028 * has returned the message entry allocated via xpc_initiate_allocate() is no
2029 * longer accessable to the caller.
2031 * Once the remote end of the channel has received the message, the function
2032 * passed as an argument to xpc_initiate_send_notify() will be called. This
2033 * allows the sender to free up or re-use any buffers referenced by the
2034 * message, but does NOT mean the message has been processed at the remote
2035 * end by a receiver.
2037 * If this routine returns an error, the caller's function will NOT be called.
2039 * This routine, although called by users, does not call xpc_part_ref() to
2040 * ensure that the partition infrastructure is in place. It relies on the
2041 * fact that we called xpc_msgqueue_ref() in xpc_allocate_msg().
2045 * partid - ID of partition to which the channel is connected.
2046 * ch_number - channel # to send message on.
2047 * payload - pointer to the payload area allocated via
2048 * xpc_initiate_allocate().
2049 * func - function to call with asynchronous notification of message
2050 * receipt. THIS FUNCTION MUST BE NON-BLOCKING.
2051 * key - user-defined key to be passed to the function when it's called.
2054 xpc_initiate_send_notify(partid_t partid, int ch_number, void *payload,
2055 xpc_notify_func func, void *key)
2057 struct xpc_partition *part = &xpc_partitions[partid];
2058 struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
2059 enum xpc_retval ret;
2062 dev_dbg(xpc_chan, "msg=0x%p, partid=%d, channel=%d\n", (void *) msg,
2065 DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
2066 DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
2067 DBUG_ON(msg == NULL);
2068 DBUG_ON(func == NULL);
2070 ret = xpc_send_msg(&part->channels[ch_number], msg, XPC_N_CALL,
2076 static struct xpc_msg *
2077 xpc_pull_remote_msg(struct xpc_channel *ch, s64 get)
2079 struct xpc_partition *part = &xpc_partitions[ch->partid];
2080 struct xpc_msg *remote_msg, *msg;
2081 u32 msg_index, nmsgs;
2083 enum xpc_retval ret;
2086 if (mutex_lock_interruptible(&ch->msg_to_pull_mutex) != 0) {
2087 /* we were interrupted by a signal */
2091 while (get >= ch->next_msg_to_pull) {
2093 /* pull as many messages as are ready and able to be pulled */
2095 msg_index = ch->next_msg_to_pull % ch->remote_nentries;
2097 DBUG_ON(ch->next_msg_to_pull >=
2098 (volatile s64) ch->w_remote_GP.put);
2099 nmsgs = (volatile s64) ch->w_remote_GP.put -
2100 ch->next_msg_to_pull;
2101 if (msg_index + nmsgs > ch->remote_nentries) {
2102 /* ignore the ones that wrap the msg queue for now */
2103 nmsgs = ch->remote_nentries - msg_index;
2106 msg_offset = msg_index * ch->msg_size;
2107 msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
2109 remote_msg = (struct xpc_msg *) (ch->remote_msgqueue_pa +
2112 if ((ret = xpc_pull_remote_cachelines(part, msg, remote_msg,
2113 nmsgs * ch->msg_size)) != xpcSuccess) {
2115 dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
2116 " msg %ld from partition %d, channel=%d, "
2117 "ret=%d\n", nmsgs, ch->next_msg_to_pull,
2118 ch->partid, ch->number, ret);
2120 XPC_DEACTIVATE_PARTITION(part, ret);
2122 mutex_unlock(&ch->msg_to_pull_mutex);
2126 mb(); /* >>> this may not be needed, we're not sure */
2128 ch->next_msg_to_pull += nmsgs;
2131 mutex_unlock(&ch->msg_to_pull_mutex);
2133 /* return the message we were looking for */
2134 msg_offset = (get % ch->remote_nentries) * ch->msg_size;
2135 msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue + msg_offset);
2142 * Get a message to be delivered.
2144 static struct xpc_msg *
2145 xpc_get_deliverable_msg(struct xpc_channel *ch)
2147 struct xpc_msg *msg = NULL;
2152 if ((volatile u32) ch->flags & XPC_C_DISCONNECTING) {
2156 get = (volatile s64) ch->w_local_GP.get;
2157 if (get == (volatile s64) ch->w_remote_GP.put) {
2161 /* There are messages waiting to be pulled and delivered.
2162 * We need to try to secure one for ourselves. We'll do this
2163 * by trying to increment w_local_GP.get and hope that no one
2164 * else beats us to it. If they do, we'll we'll simply have
2165 * to try again for the next one.
2168 if (cmpxchg(&ch->w_local_GP.get, get, get + 1) == get) {
2169 /* we got the entry referenced by get */
2171 dev_dbg(xpc_chan, "w_local_GP.get changed to %ld, "
2172 "partid=%d, channel=%d\n", get + 1,
2173 ch->partid, ch->number);
2175 /* pull the message from the remote partition */
2177 msg = xpc_pull_remote_msg(ch, get);
2179 DBUG_ON(msg != NULL && msg->number != get);
2180 DBUG_ON(msg != NULL && (msg->flags & XPC_M_DONE));
2181 DBUG_ON(msg != NULL && !(msg->flags & XPC_M_READY));
2193 * Deliver a message to its intended recipient.
2196 xpc_deliver_msg(struct xpc_channel *ch)
2198 struct xpc_msg *msg;
2201 if ((msg = xpc_get_deliverable_msg(ch)) != NULL) {
2204 * This ref is taken to protect the payload itself from being
2205 * freed before the user is finished with it, which the user
2206 * indicates by calling xpc_initiate_received().
2208 xpc_msgqueue_ref(ch);
2210 atomic_inc(&ch->kthreads_active);
2212 if (ch->func != NULL) {
2213 dev_dbg(xpc_chan, "ch->func() called, msg=0x%p, "
2214 "msg_number=%ld, partid=%d, channel=%d\n",
2215 (void *) msg, msg->number, ch->partid,
2218 /* deliver the message to its intended recipient */
2219 ch->func(xpcMsgReceived, ch->partid, ch->number,
2220 &msg->payload, ch->key);
2222 dev_dbg(xpc_chan, "ch->func() returned, msg=0x%p, "
2223 "msg_number=%ld, partid=%d, channel=%d\n",
2224 (void *) msg, msg->number, ch->partid,
2228 atomic_dec(&ch->kthreads_active);
2234 * Now we actually acknowledge the messages that have been delivered and ack'd
2235 * by advancing the cached remote message queue's Get value and if requested
2236 * send an IPI to the message sender's partition.
2239 xpc_acknowledge_msgs(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
2241 struct xpc_msg *msg;
2242 s64 get = initial_get + 1;
2249 if (get == (volatile s64) ch->w_local_GP.get) {
2253 msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
2254 (get % ch->remote_nentries) * ch->msg_size);
2256 if (!(msg->flags & XPC_M_DONE)) {
2260 msg_flags |= msg->flags;
2264 if (get == initial_get) {
2265 /* nothing's changed */
2269 if (cmpxchg_rel(&ch->local_GP->get, initial_get, get) !=
2271 /* someone else beat us to it */
2272 DBUG_ON((volatile s64) ch->local_GP->get <=
2277 /* we just set the new value of local_GP->get */
2279 dev_dbg(xpc_chan, "local_GP->get changed to %ld, partid=%d, "
2280 "channel=%d\n", get, ch->partid, ch->number);
2282 send_IPI = (msg_flags & XPC_M_INTERRUPT);
2285 * We need to ensure that the message referenced by
2286 * local_GP->get is not XPC_M_DONE or that local_GP->get
2287 * equals w_local_GP.get, so we'll go have a look.
2293 xpc_IPI_send_msgrequest(ch);
2299 * Acknowledge receipt of a delivered message.
2301 * If a message has XPC_M_INTERRUPT set, send an interrupt to the partition
2302 * that sent the message.
2304 * This function, although called by users, does not call xpc_part_ref() to
2305 * ensure that the partition infrastructure is in place. It relies on the
2306 * fact that we called xpc_msgqueue_ref() in xpc_deliver_msg().
2310 * partid - ID of partition to which the channel is connected.
2311 * ch_number - channel # message received on.
2312 * payload - pointer to the payload area allocated via
2313 * xpc_initiate_allocate().
2316 xpc_initiate_received(partid_t partid, int ch_number, void *payload)
2318 struct xpc_partition *part = &xpc_partitions[partid];
2319 struct xpc_channel *ch;
2320 struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
2321 s64 get, msg_number = msg->number;
2324 DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
2325 DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
2327 ch = &part->channels[ch_number];
2329 dev_dbg(xpc_chan, "msg=0x%p, msg_number=%ld, partid=%d, channel=%d\n",
2330 (void *) msg, msg_number, ch->partid, ch->number);
2332 DBUG_ON((((u64) msg - (u64) ch->remote_msgqueue) / ch->msg_size) !=
2333 msg_number % ch->remote_nentries);
2334 DBUG_ON(msg->flags & XPC_M_DONE);
2336 msg->flags |= XPC_M_DONE;
2339 * The preceding store of msg->flags must occur before the following
2340 * load of ch->local_GP->get.
2345 * See if this message is next in line to be acknowledged as having
2348 get = ch->local_GP->get;
2349 if (get == msg_number) {
2350 xpc_acknowledge_msgs(ch, get, msg->flags);
2353 /* the call to xpc_msgqueue_ref() was done by xpc_deliver_msg() */
2354 xpc_msgqueue_deref(ch);