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) 2008 Silicon Graphics, Inc. All Rights Reserved.
10 * Cross Partition Communication (XPC) sn2-based functions.
12 * Architecture specific implementation of common functions.
16 #include <linux/kernel.h>
17 #include <linux/delay.h>
18 #include <asm/uncached.h>
19 #include <asm/sn/sn_sal.h>
22 static struct xpc_vars_sn2 *xpc_vars; /* >>> Add _sn2 suffix? */
23 static struct xpc_vars_part_sn2 *xpc_vars_part; /* >>> Add _sn2 suffix? */
26 * The following set of macros and functions are used for the sending and
27 * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
28 * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
29 * the other that is associated with channel activity (SGI_XPC_NOTIFY).
33 xpc_IPI_receive_sn2(AMO_t *amo)
35 return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_CLEAR);
39 xpc_IPI_send_sn2(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector)
42 unsigned long irq_flags;
44 local_irq_save(irq_flags);
46 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR, flag);
47 sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
50 * We must always use the nofault function regardless of whether we
51 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
52 * didn't, we'd never know that the other partition is down and would
53 * keep sending IPIs and AMOs to it until the heartbeat times out.
55 ret = xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
56 xp_nofault_PIOR_target));
58 local_irq_restore(irq_flags);
60 return ((ret == 0) ? xpSuccess : xpPioReadError);
64 xpc_IPI_init_sn2(int index)
66 AMO_t *amo = xpc_vars->amos_page + index;
68 (void)xpc_IPI_receive_sn2(amo); /* clear AMO variable */
73 * IPIs associated with SGI_XPC_ACTIVATE IRQ.
77 * Flag the appropriate AMO variable and send an IPI to the specified node.
80 xpc_activate_IRQ_send_sn2(u64 amos_page_pa, int from_nasid, int to_nasid,
83 int w_index = XPC_NASID_W_INDEX(from_nasid);
84 int b_index = XPC_NASID_B_INDEX(from_nasid);
85 AMO_t *amos = (AMO_t *)__va(amos_page_pa +
86 (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
88 (void)xpc_IPI_send_sn2(&amos[w_index], (1UL << b_index), to_nasid,
89 to_phys_cpuid, SGI_XPC_ACTIVATE);
93 xpc_activate_IRQ_send_local_sn2(int from_nasid)
95 int w_index = XPC_NASID_W_INDEX(from_nasid);
96 int b_index = XPC_NASID_B_INDEX(from_nasid);
97 AMO_t *amos = (AMO_t *)__va(xpc_vars->amos_page_pa +
98 (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
100 /* fake the sending and receipt of an activate IRQ from remote nasid */
101 FETCHOP_STORE_OP(TO_AMO((u64)&amos[w_index].variable), FETCHOP_OR,
103 atomic_inc(&xpc_act_IRQ_rcvd);
104 wake_up_interruptible(&xpc_act_IRQ_wq);
108 xpc_IPI_send_local_activate_sn2(int from_nasid)
110 xpc_activate_IRQ_send_local_sn2(from_nasid);
114 xpc_IPI_send_activated_sn2(struct xpc_partition *part)
116 xpc_activate_IRQ_send_sn2(part->remote_amos_page_pa,
117 cnodeid_to_nasid(0), part->remote_act_nasid,
118 part->remote_act_phys_cpuid);
122 xpc_IPI_send_local_reactivate_sn2(int from_nasid)
124 xpc_activate_IRQ_send_local_sn2(from_nasid);
128 xpc_IPI_send_disengage_sn2(struct xpc_partition *part)
130 xpc_activate_IRQ_send_sn2(part->remote_amos_page_pa,
131 cnodeid_to_nasid(0), part->remote_act_nasid,
132 part->remote_act_phys_cpuid);
136 * IPIs associated with SGI_XPC_NOTIFY IRQ.
140 * Send an IPI to the remote partition that is associated with the
144 xpc_notify_IRQ_send_sn2(struct xpc_channel *ch, u8 ipi_flag,
145 char *ipi_flag_string, unsigned long *irq_flags)
147 struct xpc_partition *part = &xpc_partitions[ch->partid];
150 if (likely(part->act_state != XPC_P_DEACTIVATING)) {
151 ret = xpc_IPI_send_sn2(part->remote_IPI_amo_va,
152 (u64)ipi_flag << (ch->number * 8),
153 part->remote_IPI_nasid,
154 part->remote_IPI_phys_cpuid,
156 dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
157 ipi_flag_string, ch->partid, ch->number, ret);
158 if (unlikely(ret != xpSuccess)) {
159 if (irq_flags != NULL)
160 spin_unlock_irqrestore(&ch->lock, *irq_flags);
161 XPC_DEACTIVATE_PARTITION(part, ret);
162 if (irq_flags != NULL)
163 spin_lock_irqsave(&ch->lock, *irq_flags);
168 #define XPC_NOTIFY_IRQ_SEND_SN2(_ch, _ipi_f, _irq_f) \
169 xpc_notify_IRQ_send_sn2(_ch, _ipi_f, #_ipi_f, _irq_f)
172 * Make it look like the remote partition, which is associated with the
173 * specified channel, sent us an IPI. This faked IPI will be handled
174 * by xpc_dropped_IPI_check().
177 xpc_notify_IRQ_send_local_sn2(struct xpc_channel *ch, u8 ipi_flag,
178 char *ipi_flag_string)
180 struct xpc_partition *part = &xpc_partitions[ch->partid];
182 FETCHOP_STORE_OP(TO_AMO((u64)&part->local_IPI_amo_va->variable),
183 FETCHOP_OR, ((u64)ipi_flag << (ch->number * 8)));
184 dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
185 ipi_flag_string, ch->partid, ch->number);
188 #define XPC_NOTIFY_IRQ_SEND_LOCAL_SN2(_ch, _ipi_f) \
189 xpc_notify_IRQ_send_local_sn2(_ch, _ipi_f, #_ipi_f)
192 xpc_IPI_send_closerequest_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
194 struct xpc_openclose_args *args = ch->local_openclose_args;
196 args->reason = ch->reason;
197 XPC_NOTIFY_IRQ_SEND_SN2(ch, XPC_IPI_CLOSEREQUEST, irq_flags);
201 xpc_IPI_send_closereply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
203 XPC_NOTIFY_IRQ_SEND_SN2(ch, XPC_IPI_CLOSEREPLY, irq_flags);
207 xpc_IPI_send_openrequest_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
209 struct xpc_openclose_args *args = ch->local_openclose_args;
211 args->msg_size = ch->msg_size;
212 args->local_nentries = ch->local_nentries;
213 XPC_NOTIFY_IRQ_SEND_SN2(ch, XPC_IPI_OPENREQUEST, irq_flags);
217 xpc_IPI_send_openreply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
219 struct xpc_openclose_args *args = ch->local_openclose_args;
221 args->remote_nentries = ch->remote_nentries;
222 args->local_nentries = ch->local_nentries;
223 args->local_msgqueue_pa = __pa(ch->local_msgqueue);
224 XPC_NOTIFY_IRQ_SEND_SN2(ch, XPC_IPI_OPENREPLY, irq_flags);
228 xpc_IPI_send_msgrequest_sn2(struct xpc_channel *ch)
230 XPC_NOTIFY_IRQ_SEND_SN2(ch, XPC_IPI_MSGREQUEST, NULL);
234 xpc_IPI_send_local_msgrequest_sn2(struct xpc_channel *ch)
236 XPC_NOTIFY_IRQ_SEND_LOCAL_SN2(ch, XPC_IPI_MSGREQUEST);
240 * This next set of functions are used to keep track of when a partition is
241 * potentially engaged in accessing memory belonging to another partition.
245 xpc_mark_partition_engaged_sn2(struct xpc_partition *part)
247 unsigned long irq_flags;
248 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
249 (XPC_ENGAGED_PARTITIONS_AMO *
252 local_irq_save(irq_flags);
254 /* set bit corresponding to our partid in remote partition's AMO */
255 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
256 (1UL << sn_partition_id));
258 * We must always use the nofault function regardless of whether we
259 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
260 * didn't, we'd never know that the other partition is down and would
261 * keep sending IPIs and AMOs to it until the heartbeat times out.
263 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
265 xp_nofault_PIOR_target));
267 local_irq_restore(irq_flags);
271 xpc_mark_partition_disengaged_sn2(struct xpc_partition *part)
273 unsigned long irq_flags;
274 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
275 (XPC_ENGAGED_PARTITIONS_AMO *
278 local_irq_save(irq_flags);
280 /* clear bit corresponding to our partid in remote partition's AMO */
281 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
282 ~(1UL << sn_partition_id));
284 * We must always use the nofault function regardless of whether we
285 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
286 * didn't, we'd never know that the other partition is down and would
287 * keep sending IPIs and AMOs to it until the heartbeat times out.
289 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
291 xp_nofault_PIOR_target));
293 local_irq_restore(irq_flags);
297 xpc_request_partition_disengage_sn2(struct xpc_partition *part)
299 unsigned long irq_flags;
300 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
301 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
303 local_irq_save(irq_flags);
305 /* set bit corresponding to our partid in remote partition's AMO */
306 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
307 (1UL << sn_partition_id));
309 * We must always use the nofault function regardless of whether we
310 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
311 * didn't, we'd never know that the other partition is down and would
312 * keep sending IPIs and AMOs to it until the heartbeat times out.
314 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
316 xp_nofault_PIOR_target));
318 local_irq_restore(irq_flags);
322 xpc_cancel_partition_disengage_request_sn2(struct xpc_partition *part)
324 unsigned long irq_flags;
325 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
326 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
328 local_irq_save(irq_flags);
330 /* clear bit corresponding to our partid in remote partition's AMO */
331 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
332 ~(1UL << sn_partition_id));
334 * We must always use the nofault function regardless of whether we
335 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
336 * didn't, we'd never know that the other partition is down and would
337 * keep sending IPIs and AMOs to it until the heartbeat times out.
339 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
341 xp_nofault_PIOR_target));
343 local_irq_restore(irq_flags);
347 xpc_partition_engaged_sn2(u64 partid_mask)
349 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
351 /* return our partition's AMO variable ANDed with partid_mask */
352 return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
357 xpc_partition_disengage_requested_sn2(u64 partid_mask)
359 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
361 /* return our partition's AMO variable ANDed with partid_mask */
362 return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
367 xpc_clear_partition_engaged_sn2(u64 partid_mask)
369 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
371 /* clear bit(s) based on partid_mask in our partition's AMO */
372 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
377 xpc_clear_partition_disengage_request_sn2(u64 partid_mask)
379 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
381 /* clear bit(s) based on partid_mask in our partition's AMO */
382 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
386 static enum xp_retval
387 xpc_rsvd_page_init_sn2(struct xpc_rsvd_page *rp)
394 xpc_vars = XPC_RP_VARS(rp);
396 rp->sn.vars_pa = __pa(xpc_vars);
398 /* vars_part array follows immediately after vars */
399 xpc_vars_part = (struct xpc_vars_part_sn2 *)((u8 *)XPC_RP_VARS(rp) +
404 * Before clearing xpc_vars, see if a page of AMOs had been previously
405 * allocated. If not we'll need to allocate one and set permissions
406 * so that cross-partition AMOs are allowed.
408 * The allocated AMO page needs MCA reporting to remain disabled after
409 * XPC has unloaded. To make this work, we keep a copy of the pointer
410 * to this page (i.e., amos_page) in the struct xpc_vars structure,
411 * which is pointed to by the reserved page, and re-use that saved copy
412 * on subsequent loads of XPC. This AMO page is never freed, and its
413 * memory protections are never restricted.
415 amos_page = xpc_vars->amos_page;
416 if (amos_page == NULL) {
417 amos_page = (AMO_t *)TO_AMO(uncached_alloc_page(0, 1));
418 if (amos_page == NULL) {
419 dev_err(xpc_part, "can't allocate page of AMOs\n");
424 * Open up AMO-R/W to cpu. This is done for Shub 1.1 systems
425 * when xpc_allow_IPI_ops() is called via xpc_hb_init().
427 if (!enable_shub_wars_1_1()) {
428 ret = sn_change_memprotect(ia64_tpa((u64)amos_page),
430 SN_MEMPROT_ACCESS_CLASS_1,
433 dev_err(xpc_part, "can't change memory "
435 uncached_free_page(__IA64_UNCACHED_OFFSET |
436 TO_PHYS((u64)amos_page), 1);
443 memset(xpc_vars, 0, sizeof(struct xpc_vars_sn2));
445 xpc_vars->version = XPC_V_VERSION;
446 xpc_vars->act_nasid = cpuid_to_nasid(0);
447 xpc_vars->act_phys_cpuid = cpu_physical_id(0);
448 xpc_vars->vars_part_pa = __pa(xpc_vars_part);
449 xpc_vars->amos_page_pa = ia64_tpa((u64)amos_page);
450 xpc_vars->amos_page = amos_page; /* save for next load of XPC */
452 /* clear xpc_vars_part */
453 memset((u64 *)xpc_vars_part, 0, sizeof(struct xpc_vars_part_sn2) *
456 /* initialize the activate IRQ related AMO variables */
457 for (i = 0; i < xp_nasid_mask_words; i++)
458 (void)xpc_IPI_init_sn2(XPC_ACTIVATE_IRQ_AMOS + i);
460 /* initialize the engaged remote partitions related AMO variables */
461 (void)xpc_IPI_init_sn2(XPC_ENGAGED_PARTITIONS_AMO);
462 (void)xpc_IPI_init_sn2(XPC_DISENGAGE_REQUEST_AMO);
468 xpc_increment_heartbeat_sn2(void)
470 xpc_vars->heartbeat++;
474 xpc_offline_heartbeat_sn2(void)
476 xpc_increment_heartbeat_sn2();
477 xpc_vars->heartbeat_offline = 1;
481 xpc_online_heartbeat_sn2(void)
483 xpc_increment_heartbeat_sn2();
484 xpc_vars->heartbeat_offline = 0;
488 xpc_heartbeat_init_sn2(void)
490 DBUG_ON(xpc_vars == NULL);
492 bitmap_zero(xpc_vars->heartbeating_to_mask, XP_MAX_NPARTITIONS_SN2);
493 xpc_heartbeating_to_mask = &xpc_vars->heartbeating_to_mask[0];
494 xpc_online_heartbeat_sn2();
498 xpc_heartbeat_exit_sn2(void)
500 xpc_offline_heartbeat_sn2();
504 * At periodic intervals, scan through all active partitions and ensure
505 * their heartbeat is still active. If not, the partition is deactivated.
508 xpc_check_remote_hb_sn2(void)
510 struct xpc_vars_sn2 *remote_vars;
511 struct xpc_partition *part;
515 remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer;
517 for (partid = 0; partid < xp_max_npartitions; partid++) {
522 if (partid == sn_partition_id)
525 part = &xpc_partitions[partid];
527 if (part->act_state == XPC_P_INACTIVE ||
528 part->act_state == XPC_P_DEACTIVATING) {
532 /* pull the remote_hb cache line */
533 ret = xp_remote_memcpy(remote_vars,
534 (void *)part->remote_vars_pa,
536 if (ret != xpSuccess) {
537 XPC_DEACTIVATE_PARTITION(part, ret);
541 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
542 " = %ld, heartbeat_offline = %ld, HB_mask[0] = 0x%lx\n",
543 partid, remote_vars->heartbeat, part->last_heartbeat,
544 remote_vars->heartbeat_offline,
545 remote_vars->heartbeating_to_mask[0]);
547 if (((remote_vars->heartbeat == part->last_heartbeat) &&
548 (remote_vars->heartbeat_offline == 0)) ||
549 !xpc_hb_allowed(sn_partition_id,
550 &remote_vars->heartbeating_to_mask)) {
552 XPC_DEACTIVATE_PARTITION(part, xpNoHeartbeat);
556 part->last_heartbeat = remote_vars->heartbeat;
561 * Get a copy of the remote partition's XPC variables from the reserved page.
563 * remote_vars points to a buffer that is cacheline aligned for BTE copies and
564 * assumed to be of size XPC_RP_VARS_SIZE.
566 static enum xp_retval
567 xpc_get_remote_vars_sn2(u64 remote_vars_pa, struct xpc_vars_sn2 *remote_vars)
571 if (remote_vars_pa == 0)
574 /* pull over the cross partition variables */
575 ret = xp_remote_memcpy(remote_vars, (void *)remote_vars_pa,
577 if (ret != xpSuccess)
580 if (XPC_VERSION_MAJOR(remote_vars->version) !=
581 XPC_VERSION_MAJOR(XPC_V_VERSION)) {
589 xpc_initiate_partition_activation_sn2(struct xpc_rsvd_page *remote_rp,
590 u64 remote_rp_pa, int nasid)
592 xpc_IPI_send_local_activate(nasid);
596 * Update the remote partition's info.
599 xpc_update_partition_info_sn2(struct xpc_partition *part, u8 remote_rp_version,
600 unsigned long *remote_rp_stamp, u64 remote_rp_pa,
602 struct xpc_vars_sn2 *remote_vars)
604 part->remote_rp_version = remote_rp_version;
605 dev_dbg(xpc_part, " remote_rp_version = 0x%016x\n",
606 part->remote_rp_version);
608 part->remote_rp_stamp = *remote_rp_stamp;
609 dev_dbg(xpc_part, " remote_rp_stamp = 0x%016lx\n",
610 part->remote_rp_stamp);
612 part->remote_rp_pa = remote_rp_pa;
613 dev_dbg(xpc_part, " remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
615 part->remote_vars_pa = remote_vars_pa;
616 dev_dbg(xpc_part, " remote_vars_pa = 0x%016lx\n",
617 part->remote_vars_pa);
619 part->last_heartbeat = remote_vars->heartbeat;
620 dev_dbg(xpc_part, " last_heartbeat = 0x%016lx\n",
621 part->last_heartbeat);
623 part->remote_vars_part_pa = remote_vars->vars_part_pa;
624 dev_dbg(xpc_part, " remote_vars_part_pa = 0x%016lx\n",
625 part->remote_vars_part_pa);
627 part->remote_act_nasid = remote_vars->act_nasid;
628 dev_dbg(xpc_part, " remote_act_nasid = 0x%x\n",
629 part->remote_act_nasid);
631 part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
632 dev_dbg(xpc_part, " remote_act_phys_cpuid = 0x%x\n",
633 part->remote_act_phys_cpuid);
635 part->remote_amos_page_pa = remote_vars->amos_page_pa;
636 dev_dbg(xpc_part, " remote_amos_page_pa = 0x%lx\n",
637 part->remote_amos_page_pa);
639 part->remote_vars_version = remote_vars->version;
640 dev_dbg(xpc_part, " remote_vars_version = 0x%x\n",
641 part->remote_vars_version);
645 * Prior code has determined the nasid which generated an IPI. Inspect
646 * that nasid to determine if its partition needs to be activated or
649 * A partition is consider "awaiting activation" if our partition
650 * flags indicate it is not active and it has a heartbeat. A
651 * partition is considered "awaiting deactivation" if our partition
652 * flags indicate it is active but it has no heartbeat or it is not
653 * sending its heartbeat to us.
655 * To determine the heartbeat, the remote nasid must have a properly
656 * initialized reserved page.
659 xpc_identify_act_IRQ_req_sn2(int nasid)
661 struct xpc_rsvd_page *remote_rp;
662 struct xpc_vars_sn2 *remote_vars;
665 int remote_rp_version;
667 unsigned long remote_rp_stamp = 0;
669 struct xpc_partition *part;
672 /* pull over the reserved page structure */
674 remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer;
676 ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
677 if (ret != xpSuccess) {
678 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
679 "which sent interrupt, reason=%d\n", nasid, ret);
683 remote_vars_pa = remote_rp->sn.vars_pa;
684 remote_rp_version = remote_rp->version;
685 if (XPC_SUPPORTS_RP_STAMP(remote_rp_version))
686 remote_rp_stamp = remote_rp->stamp;
688 partid = remote_rp->SAL_partid;
689 part = &xpc_partitions[partid];
691 /* pull over the cross partition variables */
693 remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer;
695 ret = xpc_get_remote_vars_sn2(remote_vars_pa, remote_vars);
696 if (ret != xpSuccess) {
698 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
699 "which sent interrupt, reason=%d\n", nasid, ret);
701 XPC_DEACTIVATE_PARTITION(part, ret);
705 part->act_IRQ_rcvd++;
707 dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
708 "%ld:0x%lx\n", (int)nasid, (int)partid, part->act_IRQ_rcvd,
709 remote_vars->heartbeat, remote_vars->heartbeating_to_mask[0]);
711 if (xpc_partition_disengaged(part) &&
712 part->act_state == XPC_P_INACTIVE) {
714 xpc_update_partition_info_sn2(part, remote_rp_version,
715 &remote_rp_stamp, remote_rp_pa,
716 remote_vars_pa, remote_vars);
718 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
719 if (xpc_partition_disengage_requested(1UL << partid)) {
721 * Other side is waiting on us to disengage,
722 * even though we already have.
728 /* other side doesn't support disengage requests */
729 xpc_clear_partition_disengage_request(1UL << partid);
732 xpc_activate_partition(part);
736 DBUG_ON(part->remote_rp_version == 0);
737 DBUG_ON(part->remote_vars_version == 0);
739 if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
740 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
741 remote_vars_version));
743 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
744 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
746 /* see if the other side rebooted */
747 if (part->remote_amos_page_pa ==
748 remote_vars->amos_page_pa &&
749 xpc_hb_allowed(sn_partition_id,
750 &remote_vars->heartbeating_to_mask)) {
751 /* doesn't look that way, so ignore the IPI */
757 * Other side rebooted and previous XPC didn't support the
758 * disengage request, so we don't need to do anything special.
761 xpc_update_partition_info_sn2(part, remote_rp_version,
762 &remote_rp_stamp, remote_rp_pa,
763 remote_vars_pa, remote_vars);
764 part->reactivate_nasid = nasid;
765 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
769 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
771 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
772 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
775 * Other side rebooted and previous XPC did support the
776 * disengage request, but the new one doesn't.
779 xpc_clear_partition_engaged(1UL << partid);
780 xpc_clear_partition_disengage_request(1UL << partid);
782 xpc_update_partition_info_sn2(part, remote_rp_version,
783 &remote_rp_stamp, remote_rp_pa,
784 remote_vars_pa, remote_vars);
788 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
790 if (remote_rp_stamp != part->remote_rp_stamp) {
793 * Other side rebooted and the previous XPC did support
794 * the disengage request, as does the new one.
797 DBUG_ON(xpc_partition_engaged(1UL << partid));
798 DBUG_ON(xpc_partition_disengage_requested(1UL <<
801 xpc_update_partition_info_sn2(part, remote_rp_version,
810 if (part->disengage_request_timeout > 0 &&
811 !xpc_partition_disengaged(part)) {
812 /* still waiting on other side to disengage from us */
817 part->reactivate_nasid = nasid;
818 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
820 } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
821 xpc_partition_disengage_requested(1UL << partid)) {
822 XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
827 * Loop through the activation AMO variables and process any bits
828 * which are set. Each bit indicates a nasid sending a partition
829 * activation or deactivation request.
831 * Return #of IRQs detected.
834 xpc_identify_act_IRQ_sender_sn2(void)
838 u64 nasid; /* remote nasid */
839 int n_IRQs_detected = 0;
842 act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
844 /* scan through act AMO variable looking for non-zero entries */
845 for (word = 0; word < xp_nasid_mask_words; word++) {
850 nasid_mask = xpc_IPI_receive_sn2(&act_amos[word]);
851 if (nasid_mask == 0) {
852 /* no IRQs from nasids in this variable */
856 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
860 * If this nasid has been added to the machine since
861 * our partition was reset, this will retain the
862 * remote nasid in our reserved pages machine mask.
863 * This is used in the event of module reload.
865 xpc_mach_nasids[word] |= nasid_mask;
867 /* locate the nasid(s) which sent interrupts */
869 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
870 if (nasid_mask & (1UL << bit)) {
872 nasid = XPC_NASID_FROM_W_B(word, bit);
873 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
875 xpc_identify_act_IRQ_req_sn2(nasid);
879 return n_IRQs_detected;
883 xpc_process_act_IRQ_rcvd_sn2(int n_IRQs_expected)
887 n_IRQs_detected = xpc_identify_act_IRQ_sender_sn2();
888 if (n_IRQs_detected < n_IRQs_expected) {
889 /* retry once to help avoid missing AMO */
890 (void)xpc_identify_act_IRQ_sender_sn2();
895 * Setup the infrastructure necessary to support XPartition Communication
896 * between the specified remote partition and the local one.
898 static enum xp_retval
899 xpc_setup_infrastructure_sn2(struct xpc_partition *part)
901 enum xp_retval retval;
905 struct xpc_channel *ch;
906 struct timer_list *timer;
907 short partid = XPC_PARTID(part);
910 * Allocate all of the channel structures as a contiguous chunk of
913 DBUG_ON(part->channels != NULL);
914 part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_MAX_NCHANNELS,
916 if (part->channels == NULL) {
917 dev_err(xpc_chan, "can't get memory for channels\n");
921 /* allocate all the required GET/PUT values */
923 part->local_GPs = xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE,
925 &part->local_GPs_base);
926 if (part->local_GPs == NULL) {
927 dev_err(xpc_chan, "can't get memory for local get/put "
933 part->remote_GPs = xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE,
937 if (part->remote_GPs == NULL) {
938 dev_err(xpc_chan, "can't get memory for remote get/put "
944 part->remote_GPs_pa = 0;
946 /* allocate all the required open and close args */
948 part->local_openclose_args =
949 xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
950 &part->local_openclose_args_base);
951 if (part->local_openclose_args == NULL) {
952 dev_err(xpc_chan, "can't get memory for local connect args\n");
957 part->remote_openclose_args =
958 xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
959 &part->remote_openclose_args_base);
960 if (part->remote_openclose_args == NULL) {
961 dev_err(xpc_chan, "can't get memory for remote connect args\n");
966 part->remote_openclose_args_pa = 0;
968 part->local_IPI_amo_va = xpc_IPI_init_sn2(partid);
969 part->local_IPI_amo = 0;
970 spin_lock_init(&part->IPI_lock);
972 part->remote_IPI_nasid = 0;
973 part->remote_IPI_phys_cpuid = 0;
974 part->remote_IPI_amo_va = NULL;
976 atomic_set(&part->channel_mgr_requests, 1);
977 init_waitqueue_head(&part->channel_mgr_wq);
979 sprintf(part->IPI_owner, "xpc%02d", partid);
980 ret = request_irq(SGI_XPC_NOTIFY, xpc_notify_IRQ_handler, IRQF_SHARED,
981 part->IPI_owner, (void *)(u64)partid);
983 dev_err(xpc_chan, "can't register NOTIFY IRQ handler, "
985 retval = xpLackOfResources;
989 /* Setup a timer to check for dropped IPIs */
990 timer = &part->dropped_IPI_timer;
992 timer->function = (void (*)(unsigned long))xpc_dropped_IPI_check;
993 timer->data = (unsigned long)part;
994 timer->expires = jiffies + XPC_P_DROPPED_IPI_WAIT_INTERVAL;
997 part->nchannels = XPC_MAX_NCHANNELS;
999 atomic_set(&part->nchannels_active, 0);
1000 atomic_set(&part->nchannels_engaged, 0);
1002 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1003 ch = &part->channels[ch_number];
1005 ch->partid = partid;
1006 ch->number = ch_number;
1007 ch->flags = XPC_C_DISCONNECTED;
1009 ch->local_GP = &part->local_GPs[ch_number];
1010 ch->local_openclose_args =
1011 &part->local_openclose_args[ch_number];
1013 atomic_set(&ch->kthreads_assigned, 0);
1014 atomic_set(&ch->kthreads_idle, 0);
1015 atomic_set(&ch->kthreads_active, 0);
1017 atomic_set(&ch->references, 0);
1018 atomic_set(&ch->n_to_notify, 0);
1020 spin_lock_init(&ch->lock);
1021 mutex_init(&ch->msg_to_pull_mutex);
1022 init_completion(&ch->wdisconnect_wait);
1024 atomic_set(&ch->n_on_msg_allocate_wq, 0);
1025 init_waitqueue_head(&ch->msg_allocate_wq);
1026 init_waitqueue_head(&ch->idle_wq);
1030 * With the setting of the partition setup_state to XPC_P_SETUP, we're
1031 * declaring that this partition is ready to go.
1033 part->setup_state = XPC_P_SETUP;
1036 * Setup the per partition specific variables required by the
1037 * remote partition to establish channel connections with us.
1039 * The setting of the magic # indicates that these per partition
1040 * specific variables are ready to be used.
1042 xpc_vars_part[partid].GPs_pa = __pa(part->local_GPs);
1043 xpc_vars_part[partid].openclose_args_pa =
1044 __pa(part->local_openclose_args);
1045 xpc_vars_part[partid].IPI_amo_pa = __pa(part->local_IPI_amo_va);
1046 cpuid = raw_smp_processor_id(); /* any CPU in this partition will do */
1047 xpc_vars_part[partid].IPI_nasid = cpuid_to_nasid(cpuid);
1048 xpc_vars_part[partid].IPI_phys_cpuid = cpu_physical_id(cpuid);
1049 xpc_vars_part[partid].nchannels = part->nchannels;
1050 xpc_vars_part[partid].magic = XPC_VP_MAGIC1;
1054 /* setup of infrastructure failed */
1056 kfree(part->remote_openclose_args_base);
1057 part->remote_openclose_args = NULL;
1059 kfree(part->local_openclose_args_base);
1060 part->local_openclose_args = NULL;
1062 kfree(part->remote_GPs_base);
1063 part->remote_GPs = NULL;
1065 kfree(part->local_GPs_base);
1066 part->local_GPs = NULL;
1068 kfree(part->channels);
1069 part->channels = NULL;
1074 * Teardown the infrastructure necessary to support XPartition Communication
1075 * between the specified remote partition and the local one.
1078 xpc_teardown_infrastructure_sn2(struct xpc_partition *part)
1080 short partid = XPC_PARTID(part);
1083 * We start off by making this partition inaccessible to local
1084 * processes by marking it as no longer setup. Then we make it
1085 * inaccessible to remote processes by clearing the XPC per partition
1086 * specific variable's magic # (which indicates that these variables
1087 * are no longer valid) and by ignoring all XPC notify IPIs sent to
1091 DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
1092 DBUG_ON(atomic_read(&part->nchannels_active) != 0);
1093 DBUG_ON(part->setup_state != XPC_P_SETUP);
1094 part->setup_state = XPC_P_WTEARDOWN;
1096 xpc_vars_part[partid].magic = 0;
1098 free_irq(SGI_XPC_NOTIFY, (void *)(u64)partid);
1101 * Before proceeding with the teardown we have to wait until all
1102 * existing references cease.
1104 wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
1106 /* now we can begin tearing down the infrastructure */
1108 part->setup_state = XPC_P_TORNDOWN;
1110 /* in case we've still got outstanding timers registered... */
1111 del_timer_sync(&part->dropped_IPI_timer);
1113 kfree(part->remote_openclose_args_base);
1114 part->remote_openclose_args = NULL;
1115 kfree(part->local_openclose_args_base);
1116 part->local_openclose_args = NULL;
1117 kfree(part->remote_GPs_base);
1118 part->remote_GPs = NULL;
1119 kfree(part->local_GPs_base);
1120 part->local_GPs = NULL;
1121 kfree(part->channels);
1122 part->channels = NULL;
1123 part->local_IPI_amo_va = NULL;
1127 * Create a wrapper that hides the underlying mechanism for pulling a cacheline
1128 * (or multiple cachelines) from a remote partition.
1130 * src must be a cacheline aligned physical address on the remote partition.
1131 * dst must be a cacheline aligned virtual address on this partition.
1132 * cnt must be cacheline sized
1134 /* >>> Replace this function by call to xp_remote_memcpy() or bte_copy()? */
1135 static enum xp_retval
1136 xpc_pull_remote_cachelines_sn2(struct xpc_partition *part, void *dst,
1137 const void *src, size_t cnt)
1141 DBUG_ON((u64)src != L1_CACHE_ALIGN((u64)src));
1142 DBUG_ON((u64)dst != L1_CACHE_ALIGN((u64)dst));
1143 DBUG_ON(cnt != L1_CACHE_ALIGN(cnt));
1145 if (part->act_state == XPC_P_DEACTIVATING)
1146 return part->reason;
1148 ret = xp_remote_memcpy(dst, src, cnt);
1149 if (ret != xpSuccess) {
1150 dev_dbg(xpc_chan, "xp_remote_memcpy() from partition %d failed,"
1151 " ret=%d\n", XPC_PARTID(part), ret);
1157 * Pull the remote per partition specific variables from the specified
1160 static enum xp_retval
1161 xpc_pull_remote_vars_part_sn2(struct xpc_partition *part)
1163 u8 buffer[L1_CACHE_BYTES * 2];
1164 struct xpc_vars_part_sn2 *pulled_entry_cacheline =
1165 (struct xpc_vars_part_sn2 *)L1_CACHE_ALIGN((u64)buffer);
1166 struct xpc_vars_part_sn2 *pulled_entry;
1167 u64 remote_entry_cacheline_pa, remote_entry_pa;
1168 short partid = XPC_PARTID(part);
1171 /* pull the cacheline that contains the variables we're interested in */
1173 DBUG_ON(part->remote_vars_part_pa !=
1174 L1_CACHE_ALIGN(part->remote_vars_part_pa));
1175 DBUG_ON(sizeof(struct xpc_vars_part_sn2) != L1_CACHE_BYTES / 2);
1177 remote_entry_pa = part->remote_vars_part_pa +
1178 sn_partition_id * sizeof(struct xpc_vars_part_sn2);
1180 remote_entry_cacheline_pa = (remote_entry_pa & ~(L1_CACHE_BYTES - 1));
1182 pulled_entry = (struct xpc_vars_part_sn2 *)((u64)pulled_entry_cacheline
1183 + (remote_entry_pa &
1184 (L1_CACHE_BYTES - 1)));
1186 ret = xpc_pull_remote_cachelines_sn2(part, pulled_entry_cacheline,
1187 (void *)remote_entry_cacheline_pa,
1189 if (ret != xpSuccess) {
1190 dev_dbg(xpc_chan, "failed to pull XPC vars_part from "
1191 "partition %d, ret=%d\n", partid, ret);
1195 /* see if they've been set up yet */
1197 if (pulled_entry->magic != XPC_VP_MAGIC1 &&
1198 pulled_entry->magic != XPC_VP_MAGIC2) {
1200 if (pulled_entry->magic != 0) {
1201 dev_dbg(xpc_chan, "partition %d's XPC vars_part for "
1202 "partition %d has bad magic value (=0x%lx)\n",
1203 partid, sn_partition_id, pulled_entry->magic);
1207 /* they've not been initialized yet */
1211 if (xpc_vars_part[partid].magic == XPC_VP_MAGIC1) {
1213 /* validate the variables */
1215 if (pulled_entry->GPs_pa == 0 ||
1216 pulled_entry->openclose_args_pa == 0 ||
1217 pulled_entry->IPI_amo_pa == 0) {
1219 dev_err(xpc_chan, "partition %d's XPC vars_part for "
1220 "partition %d are not valid\n", partid,
1222 return xpInvalidAddress;
1225 /* the variables we imported look to be valid */
1227 part->remote_GPs_pa = pulled_entry->GPs_pa;
1228 part->remote_openclose_args_pa =
1229 pulled_entry->openclose_args_pa;
1230 part->remote_IPI_amo_va =
1231 (AMO_t *)__va(pulled_entry->IPI_amo_pa);
1232 part->remote_IPI_nasid = pulled_entry->IPI_nasid;
1233 part->remote_IPI_phys_cpuid = pulled_entry->IPI_phys_cpuid;
1235 if (part->nchannels > pulled_entry->nchannels)
1236 part->nchannels = pulled_entry->nchannels;
1238 /* let the other side know that we've pulled their variables */
1240 xpc_vars_part[partid].magic = XPC_VP_MAGIC2;
1243 if (pulled_entry->magic == XPC_VP_MAGIC1)
1250 * Establish first contact with the remote partititon. This involves pulling
1251 * the XPC per partition variables from the remote partition and waiting for
1252 * the remote partition to pull ours.
1254 static enum xp_retval
1255 xpc_make_first_contact_sn2(struct xpc_partition *part)
1260 * Register the remote partition's AMOs with SAL so it can handle
1261 * and cleanup errors within that address range should the remote
1262 * partition go down. We don't unregister this range because it is
1263 * difficult to tell when outstanding writes to the remote partition
1264 * are finished and thus when it is safe to unregister. This should
1265 * not result in wasted space in the SAL xp_addr_region table because
1266 * we should get the same page for remote_amos_page_pa after module
1267 * reloads and system reboots.
1269 if (sn_register_xp_addr_region(part->remote_amos_page_pa,
1270 PAGE_SIZE, 1) < 0) {
1271 dev_warn(xpc_part, "xpc_activating(%d) failed to register "
1272 "xp_addr region\n", XPC_PARTID(part));
1274 ret = xpPhysAddrRegFailed;
1275 XPC_DEACTIVATE_PARTITION(part, ret);
1279 xpc_IPI_send_activated(part);
1281 while ((ret = xpc_pull_remote_vars_part_sn2(part)) != xpSuccess) {
1282 if (ret != xpRetry) {
1283 XPC_DEACTIVATE_PARTITION(part, ret);
1287 dev_dbg(xpc_part, "waiting to make first contact with "
1288 "partition %d\n", XPC_PARTID(part));
1290 /* wait a 1/4 of a second or so */
1291 (void)msleep_interruptible(250);
1293 if (part->act_state == XPC_P_DEACTIVATING)
1294 return part->reason;
1301 * Get the IPI flags and pull the openclose args and/or remote GPs as needed.
1304 xpc_get_IPI_flags_sn2(struct xpc_partition *part)
1306 unsigned long irq_flags;
1311 * See if there are any IPI flags to be handled.
1314 spin_lock_irqsave(&part->IPI_lock, irq_flags);
1315 IPI_amo = part->local_IPI_amo;
1317 part->local_IPI_amo = 0;
1319 spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
1321 if (XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(IPI_amo)) {
1322 ret = xpc_pull_remote_cachelines_sn2(part,
1323 part->remote_openclose_args,
1325 remote_openclose_args_pa,
1326 XPC_OPENCLOSE_ARGS_SIZE);
1327 if (ret != xpSuccess) {
1328 XPC_DEACTIVATE_PARTITION(part, ret);
1330 dev_dbg(xpc_chan, "failed to pull openclose args from "
1331 "partition %d, ret=%d\n", XPC_PARTID(part),
1334 /* don't bother processing IPIs anymore */
1339 if (XPC_ANY_MSG_IPI_FLAGS_SET(IPI_amo)) {
1340 ret = xpc_pull_remote_cachelines_sn2(part, part->remote_GPs,
1341 (void *)part->remote_GPs_pa,
1343 if (ret != xpSuccess) {
1344 XPC_DEACTIVATE_PARTITION(part, ret);
1346 dev_dbg(xpc_chan, "failed to pull GPs from partition "
1347 "%d, ret=%d\n", XPC_PARTID(part), ret);
1349 /* don't bother processing IPIs anymore */
1357 static struct xpc_msg *
1358 xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
1360 struct xpc_partition *part = &xpc_partitions[ch->partid];
1361 struct xpc_msg *remote_msg, *msg;
1362 u32 msg_index, nmsgs;
1366 if (mutex_lock_interruptible(&ch->msg_to_pull_mutex) != 0) {
1367 /* we were interrupted by a signal */
1371 while (get >= ch->next_msg_to_pull) {
1373 /* pull as many messages as are ready and able to be pulled */
1375 msg_index = ch->next_msg_to_pull % ch->remote_nentries;
1377 DBUG_ON(ch->next_msg_to_pull >= ch->w_remote_GP.put);
1378 nmsgs = ch->w_remote_GP.put - ch->next_msg_to_pull;
1379 if (msg_index + nmsgs > ch->remote_nentries) {
1380 /* ignore the ones that wrap the msg queue for now */
1381 nmsgs = ch->remote_nentries - msg_index;
1384 msg_offset = msg_index * ch->msg_size;
1385 msg = (struct xpc_msg *)((u64)ch->remote_msgqueue + msg_offset);
1386 remote_msg = (struct xpc_msg *)(ch->remote_msgqueue_pa +
1389 ret = xpc_pull_remote_cachelines_sn2(part, msg, remote_msg,
1390 nmsgs * ch->msg_size);
1391 if (ret != xpSuccess) {
1393 dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
1394 " msg %ld from partition %d, channel=%d, "
1395 "ret=%d\n", nmsgs, ch->next_msg_to_pull,
1396 ch->partid, ch->number, ret);
1398 XPC_DEACTIVATE_PARTITION(part, ret);
1400 mutex_unlock(&ch->msg_to_pull_mutex);
1404 ch->next_msg_to_pull += nmsgs;
1407 mutex_unlock(&ch->msg_to_pull_mutex);
1409 /* return the message we were looking for */
1410 msg_offset = (get % ch->remote_nentries) * ch->msg_size;
1411 msg = (struct xpc_msg *)((u64)ch->remote_msgqueue + msg_offset);
1417 * Get a message to be delivered.
1419 static struct xpc_msg *
1420 xpc_get_deliverable_msg_sn2(struct xpc_channel *ch)
1422 struct xpc_msg *msg = NULL;
1426 if (ch->flags & XPC_C_DISCONNECTING)
1429 get = ch->w_local_GP.get;
1430 rmb(); /* guarantee that .get loads before .put */
1431 if (get == ch->w_remote_GP.put)
1434 /* There are messages waiting to be pulled and delivered.
1435 * We need to try to secure one for ourselves. We'll do this
1436 * by trying to increment w_local_GP.get and hope that no one
1437 * else beats us to it. If they do, we'll we'll simply have
1438 * to try again for the next one.
1441 if (cmpxchg(&ch->w_local_GP.get, get, get + 1) == get) {
1442 /* we got the entry referenced by get */
1444 dev_dbg(xpc_chan, "w_local_GP.get changed to %ld, "
1445 "partid=%d, channel=%d\n", get + 1,
1446 ch->partid, ch->number);
1448 /* pull the message from the remote partition */
1450 msg = xpc_pull_remote_msg_sn2(ch, get);
1452 DBUG_ON(msg != NULL && msg->number != get);
1453 DBUG_ON(msg != NULL && (msg->flags & XPC_M_DONE));
1454 DBUG_ON(msg != NULL && !(msg->flags & XPC_M_READY));
1465 * Now we actually send the messages that are ready to be sent by advancing
1466 * the local message queue's Put value and then send an IPI to the recipient
1470 xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
1472 struct xpc_msg *msg;
1473 s64 put = initial_put + 1;
1479 if (put == ch->w_local_GP.put)
1482 msg = (struct xpc_msg *)((u64)ch->local_msgqueue +
1483 (put % ch->local_nentries) *
1486 if (!(msg->flags & XPC_M_READY))
1492 if (put == initial_put) {
1493 /* nothing's changed */
1497 if (cmpxchg_rel(&ch->local_GP->put, initial_put, put) !=
1499 /* someone else beat us to it */
1500 DBUG_ON(ch->local_GP->put < initial_put);
1504 /* we just set the new value of local_GP->put */
1506 dev_dbg(xpc_chan, "local_GP->put changed to %ld, partid=%d, "
1507 "channel=%d\n", put, ch->partid, ch->number);
1512 * We need to ensure that the message referenced by
1513 * local_GP->put is not XPC_M_READY or that local_GP->put
1514 * equals w_local_GP.put, so we'll go have a look.
1520 xpc_IPI_send_msgrequest_sn2(ch);
1524 * Allocate an entry for a message from the message queue associated with the
1525 * specified channel.
1527 static enum xp_retval
1528 xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
1529 struct xpc_msg **address_of_msg)
1531 struct xpc_msg *msg;
1535 /* this reference will be dropped in xpc_send_msg_sn2() */
1536 xpc_msgqueue_ref(ch);
1538 if (ch->flags & XPC_C_DISCONNECTING) {
1539 xpc_msgqueue_deref(ch);
1542 if (!(ch->flags & XPC_C_CONNECTED)) {
1543 xpc_msgqueue_deref(ch);
1544 return xpNotConnected;
1548 * Get the next available message entry from the local message queue.
1549 * If none are available, we'll make sure that we grab the latest
1556 put = ch->w_local_GP.put;
1557 rmb(); /* guarantee that .put loads before .get */
1558 if (put - ch->w_remote_GP.get < ch->local_nentries) {
1560 /* There are available message entries. We need to try
1561 * to secure one for ourselves. We'll do this by trying
1562 * to increment w_local_GP.put as long as someone else
1563 * doesn't beat us to it. If they do, we'll have to
1566 if (cmpxchg(&ch->w_local_GP.put, put, put + 1) == put) {
1567 /* we got the entry referenced by put */
1570 continue; /* try again */
1574 * There aren't any available msg entries at this time.
1576 * In waiting for a message entry to become available,
1577 * we set a timeout in case the other side is not
1578 * sending completion IPIs. This lets us fake an IPI
1579 * that will cause the IPI handler to fetch the latest
1580 * GP values as if an IPI was sent by the other side.
1582 if (ret == xpTimeout)
1583 xpc_IPI_send_local_msgrequest_sn2(ch);
1585 if (flags & XPC_NOWAIT) {
1586 xpc_msgqueue_deref(ch);
1590 ret = xpc_allocate_msg_wait(ch);
1591 if (ret != xpInterrupted && ret != xpTimeout) {
1592 xpc_msgqueue_deref(ch);
1597 /* get the message's address and initialize it */
1598 msg = (struct xpc_msg *)((u64)ch->local_msgqueue +
1599 (put % ch->local_nentries) * ch->msg_size);
1601 DBUG_ON(msg->flags != 0);
1604 dev_dbg(xpc_chan, "w_local_GP.put changed to %ld; msg=0x%p, "
1605 "msg_number=%ld, partid=%d, channel=%d\n", put + 1,
1606 (void *)msg, msg->number, ch->partid, ch->number);
1608 *address_of_msg = msg;
1614 * Common code that does the actual sending of the message by advancing the
1615 * local message queue's Put value and sends an IPI to the partition the
1616 * message is being sent to.
1618 static enum xp_retval
1619 xpc_send_msg_sn2(struct xpc_channel *ch, struct xpc_msg *msg, u8 notify_type,
1620 xpc_notify_func func, void *key)
1622 enum xp_retval ret = xpSuccess;
1623 struct xpc_notify *notify = notify;
1624 s64 put, msg_number = msg->number;
1626 DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
1627 DBUG_ON((((u64)msg - (u64)ch->local_msgqueue) / ch->msg_size) !=
1628 msg_number % ch->local_nentries);
1629 DBUG_ON(msg->flags & XPC_M_READY);
1631 if (ch->flags & XPC_C_DISCONNECTING) {
1632 /* drop the reference grabbed in xpc_allocate_msg_sn2() */
1633 xpc_msgqueue_deref(ch);
1637 if (notify_type != 0) {
1639 * Tell the remote side to send an ACK interrupt when the
1640 * message has been delivered.
1642 msg->flags |= XPC_M_INTERRUPT;
1644 atomic_inc(&ch->n_to_notify);
1646 notify = &ch->notify_queue[msg_number % ch->local_nentries];
1647 notify->func = func;
1649 notify->type = notify_type;
1651 /* >>> is a mb() needed here? */
1653 if (ch->flags & XPC_C_DISCONNECTING) {
1655 * An error occurred between our last error check and
1656 * this one. We will try to clear the type field from
1657 * the notify entry. If we succeed then
1658 * xpc_disconnect_channel() didn't already process
1661 if (cmpxchg(¬ify->type, notify_type, 0) ==
1663 atomic_dec(&ch->n_to_notify);
1667 /* drop reference grabbed in xpc_allocate_msg_sn2() */
1668 xpc_msgqueue_deref(ch);
1673 msg->flags |= XPC_M_READY;
1676 * The preceding store of msg->flags must occur before the following
1677 * load of ch->local_GP->put.
1681 /* see if the message is next in line to be sent, if so send it */
1683 put = ch->local_GP->put;
1684 if (put == msg_number)
1685 xpc_send_msgs_sn2(ch, put);
1687 /* drop the reference grabbed in xpc_allocate_msg_sn2() */
1688 xpc_msgqueue_deref(ch);
1693 * Now we actually acknowledge the messages that have been delivered and ack'd
1694 * by advancing the cached remote message queue's Get value and if requested
1695 * send an IPI to the message sender's partition.
1698 xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
1700 struct xpc_msg *msg;
1701 s64 get = initial_get + 1;
1707 if (get == ch->w_local_GP.get)
1710 msg = (struct xpc_msg *)((u64)ch->remote_msgqueue +
1711 (get % ch->remote_nentries) *
1714 if (!(msg->flags & XPC_M_DONE))
1717 msg_flags |= msg->flags;
1721 if (get == initial_get) {
1722 /* nothing's changed */
1726 if (cmpxchg_rel(&ch->local_GP->get, initial_get, get) !=
1728 /* someone else beat us to it */
1729 DBUG_ON(ch->local_GP->get <= initial_get);
1733 /* we just set the new value of local_GP->get */
1735 dev_dbg(xpc_chan, "local_GP->get changed to %ld, partid=%d, "
1736 "channel=%d\n", get, ch->partid, ch->number);
1738 send_IPI = (msg_flags & XPC_M_INTERRUPT);
1741 * We need to ensure that the message referenced by
1742 * local_GP->get is not XPC_M_DONE or that local_GP->get
1743 * equals w_local_GP.get, so we'll go have a look.
1749 xpc_IPI_send_msgrequest_sn2(ch);
1753 xpc_received_msg_sn2(struct xpc_channel *ch, struct xpc_msg *msg)
1756 s64 msg_number = msg->number;
1758 dev_dbg(xpc_chan, "msg=0x%p, msg_number=%ld, partid=%d, channel=%d\n",
1759 (void *)msg, msg_number, ch->partid, ch->number);
1761 DBUG_ON((((u64)msg - (u64)ch->remote_msgqueue) / ch->msg_size) !=
1762 msg_number % ch->remote_nentries);
1763 DBUG_ON(msg->flags & XPC_M_DONE);
1765 msg->flags |= XPC_M_DONE;
1768 * The preceding store of msg->flags must occur before the following
1769 * load of ch->local_GP->get.
1774 * See if this message is next in line to be acknowledged as having
1777 get = ch->local_GP->get;
1778 if (get == msg_number)
1779 xpc_acknowledge_msgs_sn2(ch, get, msg->flags);
1785 xpc_rsvd_page_init = xpc_rsvd_page_init_sn2;
1786 xpc_increment_heartbeat = xpc_increment_heartbeat_sn2;
1787 xpc_offline_heartbeat = xpc_offline_heartbeat_sn2;
1788 xpc_online_heartbeat = xpc_online_heartbeat_sn2;
1789 xpc_heartbeat_init = xpc_heartbeat_init_sn2;
1790 xpc_heartbeat_exit = xpc_heartbeat_exit_sn2;
1791 xpc_check_remote_hb = xpc_check_remote_hb_sn2;
1793 xpc_initiate_partition_activation =
1794 xpc_initiate_partition_activation_sn2;
1795 xpc_process_act_IRQ_rcvd = xpc_process_act_IRQ_rcvd_sn2;
1796 xpc_setup_infrastructure = xpc_setup_infrastructure_sn2;
1797 xpc_teardown_infrastructure = xpc_teardown_infrastructure_sn2;
1798 xpc_make_first_contact = xpc_make_first_contact_sn2;
1799 xpc_get_IPI_flags = xpc_get_IPI_flags_sn2;
1800 xpc_get_deliverable_msg = xpc_get_deliverable_msg_sn2;
1802 xpc_mark_partition_engaged = xpc_mark_partition_engaged_sn2;
1803 xpc_mark_partition_disengaged = xpc_mark_partition_disengaged_sn2;
1804 xpc_request_partition_disengage = xpc_request_partition_disengage_sn2;
1805 xpc_cancel_partition_disengage_request =
1806 xpc_cancel_partition_disengage_request_sn2;
1807 xpc_partition_engaged = xpc_partition_engaged_sn2;
1808 xpc_partition_disengage_requested =
1809 xpc_partition_disengage_requested_sn2;
1810 xpc_clear_partition_engaged = xpc_clear_partition_engaged_sn2;
1811 xpc_clear_partition_disengage_request =
1812 xpc_clear_partition_disengage_request_sn2;
1814 xpc_IPI_send_local_activate = xpc_IPI_send_local_activate_sn2;
1815 xpc_IPI_send_activated = xpc_IPI_send_activated_sn2;
1816 xpc_IPI_send_local_reactivate = xpc_IPI_send_local_reactivate_sn2;
1817 xpc_IPI_send_disengage = xpc_IPI_send_disengage_sn2;
1819 xpc_IPI_send_closerequest = xpc_IPI_send_closerequest_sn2;
1820 xpc_IPI_send_closereply = xpc_IPI_send_closereply_sn2;
1821 xpc_IPI_send_openrequest = xpc_IPI_send_openrequest_sn2;
1822 xpc_IPI_send_openreply = xpc_IPI_send_openreply_sn2;
1824 xpc_allocate_msg = xpc_allocate_msg_sn2;
1826 xpc_send_msg = xpc_send_msg_sn2;
1827 xpc_received_msg = xpc_received_msg_sn2;