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) partition support.
13 * This is the part of XPC that detects the presence/absence of
14 * other partitions. It provides a heartbeat and monitors the
15 * heartbeats of other partitions.
20 #include <linux/kernel.h>
21 #include <linux/sysctl.h>
22 #include <linux/cache.h>
23 #include <linux/mmzone.h>
24 #include <linux/nodemask.h>
25 #include <asm/uncached.h>
26 #include <asm/sn/bte.h>
27 #include <asm/sn/intr.h>
28 #include <asm/sn/sn_sal.h>
29 #include <asm/sn/nodepda.h>
30 #include <asm/sn/addrs.h>
31 #include <asm/sn/xpc.h>
34 /* XPC is exiting flag */
38 /* SH_IPI_ACCESS shub register value on startup */
39 static u64 xpc_sh1_IPI_access;
40 static u64 xpc_sh2_IPI_access0;
41 static u64 xpc_sh2_IPI_access1;
42 static u64 xpc_sh2_IPI_access2;
43 static u64 xpc_sh2_IPI_access3;
46 /* original protection values for each node */
47 u64 xpc_prot_vec[MAX_NUMNODES];
50 /* this partition's reserved page pointers */
51 struct xpc_rsvd_page *xpc_rsvd_page;
52 static u64 *xpc_part_nasids;
53 static u64 *xpc_mach_nasids;
54 struct xpc_vars *xpc_vars;
55 struct xpc_vars_part *xpc_vars_part;
57 static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
58 static int xp_nasid_mask_words; /* actual size in words of nasid mask */
62 * For performance reasons, each entry of xpc_partitions[] is cacheline
63 * aligned. And xpc_partitions[] is padded with an additional entry at the
64 * end so that the last legitimate entry doesn't share its cacheline with
67 struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
71 * Generic buffer used to store a local copy of portions of a remote
72 * partition's reserved page (either its header and part_nasids mask,
75 * xpc_discovery runs only once and is a seperate thread that is
76 * very likely going to be processing in parallel with receiving
79 char ____cacheline_aligned xpc_remote_copy_buffer[XPC_RP_HEADER_SIZE +
84 * Guarantee that the kmalloc'd memory is cacheline aligned.
87 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
89 /* see if kmalloc will give us cachline aligned memory by default */
90 *base = kmalloc(size, flags);
94 if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
99 /* nope, we'll have to do it ourselves */
100 *base = kmalloc(size + L1_CACHE_BYTES, flags);
104 return (void *) L1_CACHE_ALIGN((u64) *base);
109 * Given a nasid, get the physical address of the partition's reserved page
110 * for that nasid. This function returns 0 on any error.
113 xpc_get_rsvd_page_pa(int nasid)
115 bte_result_t bte_res;
118 u64 rp_pa = nasid; /* seed with nasid */
122 void *buf_base = NULL;
127 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
130 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
131 "0x%016lx, address=0x%016lx, len=0x%016lx\n",
132 status, cookie, rp_pa, len);
134 if (status != SALRET_MORE_PASSES) {
138 if (L1_CACHE_ALIGN(len) > buf_len) {
139 if (buf_base != NULL) {
142 buf_len = L1_CACHE_ALIGN(len);
143 buf = (u64) xpc_kmalloc_cacheline_aligned(buf_len,
144 GFP_KERNEL, &buf_base);
145 if (buf_base == NULL) {
146 dev_err(xpc_part, "unable to kmalloc "
147 "len=0x%016lx\n", buf_len);
148 status = SALRET_ERROR;
153 bte_res = xp_bte_copy(rp_pa, ia64_tpa(buf), buf_len,
154 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
155 if (bte_res != BTE_SUCCESS) {
156 dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
157 status = SALRET_ERROR;
162 if (buf_base != NULL) {
166 if (status != SALRET_OK) {
169 dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
175 * Fill the partition reserved page with the information needed by
176 * other partitions to discover we are alive and establish initial
179 struct xpc_rsvd_page *
180 xpc_rsvd_page_init(void)
182 struct xpc_rsvd_page *rp;
184 u64 rp_pa, nasid_array = 0;
188 /* get the local reserved page's address */
191 rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
194 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
197 rp = (struct xpc_rsvd_page *) __va(rp_pa);
199 if (rp->partid != sn_partition_id) {
200 dev_err(xpc_part, "the reserved page's partid of %d should be "
201 "%d\n", rp->partid, sn_partition_id);
205 rp->version = XPC_RP_VERSION;
207 /* establish the actual sizes of the nasid masks */
208 if (rp->SAL_version == 1) {
209 /* SAL_version 1 didn't set the nasids_size field */
210 rp->nasids_size = 128;
212 xp_nasid_mask_bytes = rp->nasids_size;
213 xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
215 /* setup the pointers to the various items in the reserved page */
216 xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
217 xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
218 xpc_vars = XPC_RP_VARS(rp);
219 xpc_vars_part = XPC_RP_VARS_PART(rp);
222 * Before clearing xpc_vars, see if a page of AMOs had been previously
223 * allocated. If not we'll need to allocate one and set permissions
224 * so that cross-partition AMOs are allowed.
226 * The allocated AMO page needs MCA reporting to remain disabled after
227 * XPC has unloaded. To make this work, we keep a copy of the pointer
228 * to this page (i.e., amos_page) in the struct xpc_vars structure,
229 * which is pointed to by the reserved page, and re-use that saved copy
230 * on subsequent loads of XPC. This AMO page is never freed, and its
231 * memory protections are never restricted.
233 if ((amos_page = xpc_vars->amos_page) == NULL) {
234 amos_page = (AMO_t *) TO_AMO(uncached_alloc_page(0));
235 if (amos_page == NULL) {
236 dev_err(xpc_part, "can't allocate page of AMOs\n");
241 * Open up AMO-R/W to cpu. This is done for Shub 1.1 systems
242 * when xpc_allow_IPI_ops() is called via xpc_hb_init().
244 if (!enable_shub_wars_1_1()) {
245 ret = sn_change_memprotect(ia64_tpa((u64) amos_page),
246 PAGE_SIZE, SN_MEMPROT_ACCESS_CLASS_1,
249 dev_err(xpc_part, "can't change memory "
251 uncached_free_page(__IA64_UNCACHED_OFFSET |
252 TO_PHYS((u64) amos_page));
256 } else if (!IS_AMO_ADDRESS((u64) amos_page)) {
258 * EFI's XPBOOT can also set amos_page in the reserved page,
259 * but it happens to leave it as an uncached physical address
260 * and we need it to be an uncached virtual, so we'll have to
263 if (!IS_AMO_PHYS_ADDRESS((u64) amos_page)) {
264 dev_err(xpc_part, "previously used amos_page address "
265 "is bad = 0x%p\n", (void *) amos_page);
268 amos_page = (AMO_t *) TO_AMO((u64) amos_page);
272 memset(xpc_vars, 0, sizeof(struct xpc_vars));
274 xpc_vars->version = XPC_V_VERSION;
275 xpc_vars->act_nasid = cpuid_to_nasid(0);
276 xpc_vars->act_phys_cpuid = cpu_physical_id(0);
277 xpc_vars->vars_part_pa = __pa(xpc_vars_part);
278 xpc_vars->amos_page_pa = ia64_tpa((u64) amos_page);
279 xpc_vars->amos_page = amos_page; /* save for next load of XPC */
282 /* clear xpc_vars_part */
283 memset((u64 *) xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
286 /* initialize the activate IRQ related AMO variables */
287 for (i = 0; i < xp_nasid_mask_words; i++) {
288 (void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
291 /* initialize the engaged remote partitions related AMO variables */
292 (void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
293 (void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
295 /* timestamp of when reserved page was setup by XPC */
296 rp->stamp = CURRENT_TIME;
299 * This signifies to the remote partition that our reserved
300 * page is initialized.
302 rp->vars_pa = __pa(xpc_vars);
309 * Change protections to allow IPI operations (and AMO operations on
313 xpc_allow_IPI_ops(void)
319 // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
322 xpc_sh2_IPI_access0 =
323 (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
324 xpc_sh2_IPI_access1 =
325 (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
326 xpc_sh2_IPI_access2 =
327 (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
328 xpc_sh2_IPI_access3 =
329 (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
331 for_each_online_node(node) {
332 nasid = cnodeid_to_nasid(node);
333 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
335 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
337 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
339 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
345 (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
347 for_each_online_node(node) {
348 nasid = cnodeid_to_nasid(node);
349 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
353 * Since the BIST collides with memory operations on
354 * SHUB 1.1 sn_change_memprotect() cannot be used.
356 if (enable_shub_wars_1_1()) {
357 /* open up everything */
358 xpc_prot_vec[node] = (u64) HUB_L((u64 *)
359 GLOBAL_MMR_ADDR(nasid,
360 SH1_MD_DQLP_MMR_DIR_PRIVEC0));
361 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
362 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
364 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
365 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
374 * Restrict protections to disallow IPI operations (and AMO operations on
378 xpc_restrict_IPI_ops(void)
384 // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
388 for_each_online_node(node) {
389 nasid = cnodeid_to_nasid(node);
390 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
391 xpc_sh2_IPI_access0);
392 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
393 xpc_sh2_IPI_access1);
394 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
395 xpc_sh2_IPI_access2);
396 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
397 xpc_sh2_IPI_access3);
402 for_each_online_node(node) {
403 nasid = cnodeid_to_nasid(node);
404 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
407 if (enable_shub_wars_1_1()) {
408 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
409 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
411 HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
412 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
421 * At periodic intervals, scan through all active partitions and ensure
422 * their heartbeat is still active. If not, the partition is deactivated.
425 xpc_check_remote_hb(void)
427 struct xpc_vars *remote_vars;
428 struct xpc_partition *part;
433 remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
435 for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
441 if (partid == sn_partition_id) {
445 part = &xpc_partitions[partid];
447 if (part->act_state == XPC_P_INACTIVE ||
448 part->act_state == XPC_P_DEACTIVATING) {
452 /* pull the remote_hb cache line */
453 bres = xp_bte_copy(part->remote_vars_pa,
454 ia64_tpa((u64) remote_vars),
456 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
457 if (bres != BTE_SUCCESS) {
458 XPC_DEACTIVATE_PARTITION(part,
459 xpc_map_bte_errors(bres));
463 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
464 " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
465 partid, remote_vars->heartbeat, part->last_heartbeat,
466 remote_vars->heartbeat_offline,
467 remote_vars->heartbeating_to_mask);
469 if (((remote_vars->heartbeat == part->last_heartbeat) &&
470 (remote_vars->heartbeat_offline == 0)) ||
471 !xpc_hb_allowed(sn_partition_id, remote_vars)) {
473 XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
477 part->last_heartbeat = remote_vars->heartbeat;
483 * Get a copy of a portion of the remote partition's rsvd page.
485 * remote_rp points to a buffer that is cacheline aligned for BTE copies and
486 * is large enough to contain a copy of their reserved page header and
489 static enum xpc_retval
490 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
491 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
496 /* get the reserved page's physical address */
498 *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
499 if (*remote_rp_pa == 0) {
500 return xpcNoRsvdPageAddr;
504 /* pull over the reserved page header and part_nasids mask */
506 bres = xp_bte_copy(*remote_rp_pa, ia64_tpa((u64) remote_rp),
507 XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
508 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
509 if (bres != BTE_SUCCESS) {
510 return xpc_map_bte_errors(bres);
514 if (discovered_nasids != NULL) {
515 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
518 for (i = 0; i < xp_nasid_mask_words; i++) {
519 discovered_nasids[i] |= remote_part_nasids[i];
524 /* check that the partid is for another partition */
526 if (remote_rp->partid < 1 ||
527 remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
528 return xpcInvalidPartid;
531 if (remote_rp->partid == sn_partition_id) {
532 return xpcLocalPartid;
536 if (XPC_VERSION_MAJOR(remote_rp->version) !=
537 XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
538 return xpcBadVersion;
546 * Get a copy of the remote partition's XPC variables from the reserved page.
548 * remote_vars points to a buffer that is cacheline aligned for BTE copies and
549 * assumed to be of size XPC_RP_VARS_SIZE.
551 static enum xpc_retval
552 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
557 if (remote_vars_pa == 0) {
558 return xpcVarsNotSet;
562 /* pull over the cross partition variables */
564 bres = xp_bte_copy(remote_vars_pa, ia64_tpa((u64) remote_vars),
566 (BTE_NOTIFY | BTE_WACQUIRE), NULL);
567 if (bres != BTE_SUCCESS) {
568 return xpc_map_bte_errors(bres);
571 if (XPC_VERSION_MAJOR(remote_vars->version) !=
572 XPC_VERSION_MAJOR(XPC_V_VERSION)) {
573 return xpcBadVersion;
581 * Update the remote partition's info.
584 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
585 struct timespec *remote_rp_stamp, u64 remote_rp_pa,
586 u64 remote_vars_pa, struct xpc_vars *remote_vars)
588 part->remote_rp_version = remote_rp_version;
589 dev_dbg(xpc_part, " remote_rp_version = 0x%016lx\n",
590 part->remote_rp_version);
592 part->remote_rp_stamp = *remote_rp_stamp;
593 dev_dbg(xpc_part, " remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
594 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
596 part->remote_rp_pa = remote_rp_pa;
597 dev_dbg(xpc_part, " remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
599 part->remote_vars_pa = remote_vars_pa;
600 dev_dbg(xpc_part, " remote_vars_pa = 0x%016lx\n",
601 part->remote_vars_pa);
603 part->last_heartbeat = remote_vars->heartbeat;
604 dev_dbg(xpc_part, " last_heartbeat = 0x%016lx\n",
605 part->last_heartbeat);
607 part->remote_vars_part_pa = remote_vars->vars_part_pa;
608 dev_dbg(xpc_part, " remote_vars_part_pa = 0x%016lx\n",
609 part->remote_vars_part_pa);
611 part->remote_act_nasid = remote_vars->act_nasid;
612 dev_dbg(xpc_part, " remote_act_nasid = 0x%x\n",
613 part->remote_act_nasid);
615 part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
616 dev_dbg(xpc_part, " remote_act_phys_cpuid = 0x%x\n",
617 part->remote_act_phys_cpuid);
619 part->remote_amos_page_pa = remote_vars->amos_page_pa;
620 dev_dbg(xpc_part, " remote_amos_page_pa = 0x%lx\n",
621 part->remote_amos_page_pa);
623 part->remote_vars_version = remote_vars->version;
624 dev_dbg(xpc_part, " remote_vars_version = 0x%x\n",
625 part->remote_vars_version);
630 * Prior code has determined the nasid which generated an IPI. Inspect
631 * that nasid to determine if its partition needs to be activated or
634 * A partition is consider "awaiting activation" if our partition
635 * flags indicate it is not active and it has a heartbeat. A
636 * partition is considered "awaiting deactivation" if our partition
637 * flags indicate it is active but it has no heartbeat or it is not
638 * sending its heartbeat to us.
640 * To determine the heartbeat, the remote nasid must have a properly
641 * initialized reserved page.
644 xpc_identify_act_IRQ_req(int nasid)
646 struct xpc_rsvd_page *remote_rp;
647 struct xpc_vars *remote_vars;
650 int remote_rp_version;
653 struct timespec remote_rp_stamp = { 0, 0 };
655 struct xpc_partition *part;
659 /* pull over the reserved page structure */
661 remote_rp = (struct xpc_rsvd_page *) xpc_remote_copy_buffer;
663 ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
664 if (ret != xpcSuccess) {
665 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
666 "which sent interrupt, reason=%d\n", nasid, ret);
670 remote_vars_pa = remote_rp->vars_pa;
671 remote_rp_version = remote_rp->version;
672 if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
673 remote_rp_stamp = remote_rp->stamp;
675 partid = remote_rp->partid;
676 part = &xpc_partitions[partid];
679 /* pull over the cross partition variables */
681 remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
683 ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
684 if (ret != xpcSuccess) {
686 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
687 "which sent interrupt, reason=%d\n", nasid, ret);
689 XPC_DEACTIVATE_PARTITION(part, ret);
694 part->act_IRQ_rcvd++;
696 dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
697 "%ld:0x%lx\n", (int) nasid, (int) partid, part->act_IRQ_rcvd,
698 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
700 if (xpc_partition_disengaged(part) &&
701 part->act_state == XPC_P_INACTIVE) {
703 xpc_update_partition_info(part, remote_rp_version,
704 &remote_rp_stamp, remote_rp_pa,
705 remote_vars_pa, remote_vars);
707 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
708 if (xpc_partition_disengage_requested(1UL << partid)) {
710 * Other side is waiting on us to disengage,
711 * even though we already have.
716 /* other side doesn't support disengage requests */
717 xpc_clear_partition_disengage_request(1UL << partid);
720 xpc_activate_partition(part);
724 DBUG_ON(part->remote_rp_version == 0);
725 DBUG_ON(part->remote_vars_version == 0);
727 if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
728 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
729 remote_vars_version));
731 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
732 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
734 /* see if the other side rebooted */
735 if (part->remote_amos_page_pa ==
736 remote_vars->amos_page_pa &&
737 xpc_hb_allowed(sn_partition_id,
739 /* doesn't look that way, so ignore the IPI */
745 * Other side rebooted and previous XPC didn't support the
746 * disengage request, so we don't need to do anything special.
749 xpc_update_partition_info(part, remote_rp_version,
750 &remote_rp_stamp, remote_rp_pa,
751 remote_vars_pa, remote_vars);
752 part->reactivate_nasid = nasid;
753 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
757 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
759 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
760 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
763 * Other side rebooted and previous XPC did support the
764 * disengage request, but the new one doesn't.
767 xpc_clear_partition_engaged(1UL << partid);
768 xpc_clear_partition_disengage_request(1UL << partid);
770 xpc_update_partition_info(part, remote_rp_version,
771 &remote_rp_stamp, remote_rp_pa,
772 remote_vars_pa, remote_vars);
776 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
778 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
780 if (stamp_diff != 0) {
781 DBUG_ON(stamp_diff >= 0);
784 * Other side rebooted and the previous XPC did support
785 * the disengage request, as does the new one.
788 DBUG_ON(xpc_partition_engaged(1UL << partid));
789 DBUG_ON(xpc_partition_disengage_requested(1UL <<
792 xpc_update_partition_info(part, remote_rp_version,
793 &remote_rp_stamp, remote_rp_pa,
794 remote_vars_pa, remote_vars);
799 if (part->disengage_request_timeout > 0 &&
800 !xpc_partition_disengaged(part)) {
801 /* still waiting on other side to disengage from us */
806 part->reactivate_nasid = nasid;
807 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
809 } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
810 xpc_partition_disengage_requested(1UL << partid)) {
811 XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
817 * Loop through the activation AMO variables and process any bits
818 * which are set. Each bit indicates a nasid sending a partition
819 * activation or deactivation request.
821 * Return #of IRQs detected.
824 xpc_identify_act_IRQ_sender(void)
828 u64 nasid; /* remote nasid */
829 int n_IRQs_detected = 0;
833 act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
836 /* scan through act AMO variable looking for non-zero entries */
837 for (word = 0; word < xp_nasid_mask_words; word++) {
843 nasid_mask = xpc_IPI_receive(&act_amos[word]);
844 if (nasid_mask == 0) {
845 /* no IRQs from nasids in this variable */
849 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
854 * If this nasid has been added to the machine since
855 * our partition was reset, this will retain the
856 * remote nasid in our reserved pages machine mask.
857 * This is used in the event of module reload.
859 xpc_mach_nasids[word] |= nasid_mask;
862 /* locate the nasid(s) which sent interrupts */
864 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
865 if (nasid_mask & (1UL << bit)) {
867 nasid = XPC_NASID_FROM_W_B(word, bit);
868 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
870 xpc_identify_act_IRQ_req(nasid);
874 return n_IRQs_detected;
879 * See if the other side has responded to a partition disengage request
883 xpc_partition_disengaged(struct xpc_partition *part)
885 partid_t partid = XPC_PARTID(part);
889 disengaged = (xpc_partition_engaged(1UL << partid) == 0);
890 if (part->disengage_request_timeout) {
892 if (jiffies < part->disengage_request_timeout) {
893 /* timelimit hasn't been reached yet */
898 * Other side hasn't responded to our disengage
899 * request in a timely fashion, so assume it's dead.
902 dev_info(xpc_part, "disengage from remote partition %d "
903 "timed out\n", partid);
904 xpc_disengage_request_timedout = 1;
905 xpc_clear_partition_engaged(1UL << partid);
908 part->disengage_request_timeout = 0;
910 /* cancel the timer function, provided it's not us */
911 if (!in_interrupt()) {
912 del_singleshot_timer_sync(&part->
913 disengage_request_timer);
916 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
917 part->act_state != XPC_P_INACTIVE);
918 if (part->act_state != XPC_P_INACTIVE) {
919 xpc_wakeup_channel_mgr(part);
922 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
923 xpc_cancel_partition_disengage_request(part);
931 * Mark specified partition as active.
934 xpc_mark_partition_active(struct xpc_partition *part)
936 unsigned long irq_flags;
940 dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
942 spin_lock_irqsave(&part->act_lock, irq_flags);
943 if (part->act_state == XPC_P_ACTIVATING) {
944 part->act_state = XPC_P_ACTIVE;
947 DBUG_ON(part->reason == xpcSuccess);
950 spin_unlock_irqrestore(&part->act_lock, irq_flags);
957 * Notify XPC that the partition is down.
960 xpc_deactivate_partition(const int line, struct xpc_partition *part,
961 enum xpc_retval reason)
963 unsigned long irq_flags;
966 spin_lock_irqsave(&part->act_lock, irq_flags);
968 if (part->act_state == XPC_P_INACTIVE) {
969 XPC_SET_REASON(part, reason, line);
970 spin_unlock_irqrestore(&part->act_lock, irq_flags);
971 if (reason == xpcReactivating) {
972 /* we interrupt ourselves to reactivate partition */
973 xpc_IPI_send_reactivate(part);
977 if (part->act_state == XPC_P_DEACTIVATING) {
978 if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
979 reason == xpcReactivating) {
980 XPC_SET_REASON(part, reason, line);
982 spin_unlock_irqrestore(&part->act_lock, irq_flags);
986 part->act_state = XPC_P_DEACTIVATING;
987 XPC_SET_REASON(part, reason, line);
989 spin_unlock_irqrestore(&part->act_lock, irq_flags);
991 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
992 xpc_request_partition_disengage(part);
993 xpc_IPI_send_disengage(part);
995 /* set a timelimit on the disengage request */
996 part->disengage_request_timeout = jiffies +
997 (xpc_disengage_request_timelimit * HZ);
998 part->disengage_request_timer.expires =
999 part->disengage_request_timeout;
1000 add_timer(&part->disengage_request_timer);
1003 dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
1004 XPC_PARTID(part), reason);
1006 xpc_partition_going_down(part, reason);
1011 * Mark specified partition as inactive.
1014 xpc_mark_partition_inactive(struct xpc_partition *part)
1016 unsigned long irq_flags;
1019 dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
1022 spin_lock_irqsave(&part->act_lock, irq_flags);
1023 part->act_state = XPC_P_INACTIVE;
1024 spin_unlock_irqrestore(&part->act_lock, irq_flags);
1025 part->remote_rp_pa = 0;
1030 * SAL has provided a partition and machine mask. The partition mask
1031 * contains a bit for each even nasid in our partition. The machine
1032 * mask contains a bit for each even nasid in the entire machine.
1034 * Using those two bit arrays, we can determine which nasids are
1035 * known in the machine. Each should also have a reserved page
1036 * initialized if they are available for partitioning.
1041 void *remote_rp_base;
1042 struct xpc_rsvd_page *remote_rp;
1043 struct xpc_vars *remote_vars;
1050 struct xpc_rsvd_page *rp;
1052 struct xpc_partition *part;
1053 u64 *discovered_nasids;
1054 enum xpc_retval ret;
1057 remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
1058 xp_nasid_mask_bytes,
1059 GFP_KERNEL, &remote_rp_base);
1060 if (remote_rp == NULL) {
1063 remote_vars = (struct xpc_vars *) remote_rp;
1066 discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
1068 if (discovered_nasids == NULL) {
1069 kfree(remote_rp_base);
1073 rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
1076 * The term 'region' in this context refers to the minimum number of
1077 * nodes that can comprise an access protection grouping. The access
1078 * protection is in regards to memory, IOI and IPI.
1081 region_size = sn_region_size;
1083 switch (region_size) {
1091 DBUG_ON(!is_shub2());
1094 for (region = 0; region < max_regions; region++) {
1096 if ((volatile int) xpc_exiting) {
1100 dev_dbg(xpc_part, "searching region %d\n", region);
1102 for (nasid = (region * region_size * 2);
1103 nasid < ((region + 1) * region_size * 2);
1106 if ((volatile int) xpc_exiting) {
1110 dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1113 if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
1114 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1115 "part of the local partition; skipping "
1120 if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
1121 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1122 "not on Numa-Link network at reset\n",
1127 if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1128 dev_dbg(xpc_part, "Nasid %d is part of a "
1129 "partition which was previously "
1130 "discovered\n", nasid);
1135 /* pull over the reserved page structure */
1137 ret = xpc_get_remote_rp(nasid, discovered_nasids,
1138 remote_rp, &remote_rp_pa);
1139 if (ret != xpcSuccess) {
1140 dev_dbg(xpc_part, "unable to get reserved page "
1141 "from nasid %d, reason=%d\n", nasid,
1144 if (ret == xpcLocalPartid) {
1150 remote_vars_pa = remote_rp->vars_pa;
1152 partid = remote_rp->partid;
1153 part = &xpc_partitions[partid];
1156 /* pull over the cross partition variables */
1158 ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1159 if (ret != xpcSuccess) {
1160 dev_dbg(xpc_part, "unable to get XPC variables "
1161 "from nasid %d, reason=%d\n", nasid,
1164 XPC_DEACTIVATE_PARTITION(part, ret);
1168 if (part->act_state != XPC_P_INACTIVE) {
1169 dev_dbg(xpc_part, "partition %d on nasid %d is "
1170 "already activating\n", partid, nasid);
1175 * Register the remote partition's AMOs with SAL so it
1176 * can handle and cleanup errors within that address
1177 * range should the remote partition go down. We don't
1178 * unregister this range because it is difficult to
1179 * tell when outstanding writes to the remote partition
1180 * are finished and thus when it is thus safe to
1181 * unregister. This should not result in wasted space
1182 * in the SAL xp_addr_region table because we should
1183 * get the same page for remote_act_amos_pa after
1184 * module reloads and system reboots.
1186 if (sn_register_xp_addr_region(
1187 remote_vars->amos_page_pa,
1188 PAGE_SIZE, 1) < 0) {
1189 dev_dbg(xpc_part, "partition %d failed to "
1190 "register xp_addr region 0x%016lx\n",
1191 partid, remote_vars->amos_page_pa);
1193 XPC_SET_REASON(part, xpcPhysAddrRegFailed,
1199 * The remote nasid is valid and available.
1200 * Send an interrupt to that nasid to notify
1201 * it that we are ready to begin activation.
1203 dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1204 "nasid %d, phys_cpuid 0x%x\n",
1205 remote_vars->amos_page_pa,
1206 remote_vars->act_nasid,
1207 remote_vars->act_phys_cpuid);
1209 if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1211 part->remote_amos_page_pa =
1212 remote_vars->amos_page_pa;
1213 xpc_mark_partition_disengaged(part);
1214 xpc_cancel_partition_disengage_request(part);
1216 xpc_IPI_send_activate(remote_vars);
1220 kfree(discovered_nasids);
1221 kfree(remote_rp_base);
1226 * Given a partid, get the nasids owned by that partition from the
1227 * remote partition's reserved page.
1230 xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
1232 struct xpc_partition *part;
1237 part = &xpc_partitions[partid];
1238 if (part->remote_rp_pa == 0) {
1239 return xpcPartitionDown;
1242 memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1244 part_nasid_pa = (u64) XPC_RP_PART_NASIDS(part->remote_rp_pa);
1246 bte_res = xp_bte_copy(part_nasid_pa, ia64_tpa((u64) nasid_mask),
1247 xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE), NULL);
1249 return xpc_map_bte_errors(bte_res);