sgi-xp: isolate xpc_vars_part structure to sn2 only
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / misc / sgi-xp / xpc_partition.c
blobbe5b7547dab45a2ab9d27b7ca9bf3f3920b65e57
1 /*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
6 * Copyright (c) 2004-2008 Silicon Graphics, Inc. All Rights Reserved.
7 */
9 /*
10 * Cross Partition Communication (XPC) partition support.
12 * This is the part of XPC that detects the presence/absence of
13 * other partitions. It provides a heartbeat and monitors the
14 * heartbeats of other partitions.
18 #include <linux/kernel.h>
19 #include <linux/sysctl.h>
20 #include <linux/cache.h>
21 #include <linux/mmzone.h>
22 #include <linux/nodemask.h>
23 #include <asm/sn/intr.h>
24 #include <asm/sn/sn_sal.h>
25 #include <asm/sn/nodepda.h>
26 #include <asm/sn/addrs.h>
27 #include "xpc.h"
29 /* XPC is exiting flag */
30 int xpc_exiting;
32 /* SH_IPI_ACCESS shub register value on startup */
33 static u64 xpc_sh1_IPI_access;
34 static u64 xpc_sh2_IPI_access0;
35 static u64 xpc_sh2_IPI_access1;
36 static u64 xpc_sh2_IPI_access2;
37 static u64 xpc_sh2_IPI_access3;
39 /* original protection values for each node */
40 u64 xpc_prot_vec[MAX_NUMNODES];
42 /* this partition's reserved page pointers */
43 struct xpc_rsvd_page *xpc_rsvd_page;
44 static u64 *xpc_part_nasids;
45 static u64 *xpc_mach_nasids;
47 /* >>> next two variables should be 'xpc_' if they remain here */
48 static int xp_sizeof_nasid_mask; /* actual size in bytes of nasid mask */
49 int xp_nasid_mask_words; /* actual size in words of nasid mask */
51 struct xpc_partition *xpc_partitions;
54 * Generic buffer used to store a local copy of portions of a remote
55 * partition's reserved page (either its header and part_nasids mask,
56 * or its vars).
58 char *xpc_remote_copy_buffer;
59 void *xpc_remote_copy_buffer_base;
62 * Guarantee that the kmalloc'd memory is cacheline aligned.
64 void *
65 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
67 /* see if kmalloc will give us cachline aligned memory by default */
68 *base = kmalloc(size, flags);
69 if (*base == NULL)
70 return NULL;
72 if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
73 return *base;
75 kfree(*base);
77 /* nope, we'll have to do it ourselves */
78 *base = kmalloc(size + L1_CACHE_BYTES, flags);
79 if (*base == NULL)
80 return NULL;
82 return (void *)L1_CACHE_ALIGN((u64)*base);
86 * Given a nasid, get the physical address of the partition's reserved page
87 * for that nasid. This function returns 0 on any error.
89 static u64
90 xpc_get_rsvd_page_pa(int nasid)
92 enum xp_retval ret;
93 s64 status;
94 u64 cookie = 0;
95 u64 rp_pa = nasid; /* seed with nasid */
96 u64 len = 0;
97 u64 buf = buf;
98 u64 buf_len = 0;
99 void *buf_base = NULL;
101 while (1) {
103 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
104 &len);
106 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
107 "0x%016lx, address=0x%016lx, len=0x%016lx\n",
108 status, cookie, rp_pa, len);
110 if (status != SALRET_MORE_PASSES)
111 break;
113 /* >>> L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
114 if (L1_CACHE_ALIGN(len) > buf_len) {
115 kfree(buf_base);
116 buf_len = L1_CACHE_ALIGN(len);
117 buf = (u64)xpc_kmalloc_cacheline_aligned(buf_len,
118 GFP_KERNEL,
119 &buf_base);
120 if (buf_base == NULL) {
121 dev_err(xpc_part, "unable to kmalloc "
122 "len=0x%016lx\n", buf_len);
123 status = SALRET_ERROR;
124 break;
128 ret = xp_remote_memcpy((void *)buf, (void *)rp_pa, buf_len);
129 if (ret != xpSuccess) {
130 dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
131 status = SALRET_ERROR;
132 break;
136 kfree(buf_base);
138 if (status != SALRET_OK)
139 rp_pa = 0;
141 dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
142 return rp_pa;
146 * Fill the partition reserved page with the information needed by
147 * other partitions to discover we are alive and establish initial
148 * communications.
150 struct xpc_rsvd_page *
151 xpc_setup_rsvd_page(void)
153 struct xpc_rsvd_page *rp;
154 u64 rp_pa;
156 /* get the local reserved page's address */
158 preempt_disable();
159 rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
160 preempt_enable();
161 if (rp_pa == 0) {
162 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
163 return NULL;
165 rp = (struct xpc_rsvd_page *)__va(rp_pa);
167 if (rp->SAL_version < 3) {
168 /* SAL_versions < 3 had a SAL_partid defined as a u8 */
169 rp->SAL_partid &= 0xff;
171 BUG_ON(rp->SAL_partid != sn_partition_id);
173 if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
174 dev_err(xpc_part, "the reserved page's partid of %d is outside "
175 "supported range (< 0 || >= %d)\n", rp->SAL_partid,
176 xp_max_npartitions);
177 return NULL;
180 rp->version = XPC_RP_VERSION;
181 rp->max_npartitions = xp_max_npartitions;
183 /* establish the actual sizes of the nasid masks */
184 if (rp->SAL_version == 1) {
185 /* SAL_version 1 didn't set the nasids_size field */
186 rp->SAL_nasids_size = 128;
188 xp_sizeof_nasid_mask = rp->SAL_nasids_size;
189 xp_nasid_mask_words = DIV_ROUND_UP(xp_sizeof_nasid_mask,
190 BYTES_PER_WORD);
192 /* setup the pointers to the various items in the reserved page */
193 xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
194 xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
196 if (xpc_rsvd_page_init(rp) != xpSuccess)
197 return NULL;
200 * Set timestamp of when reserved page was setup by XPC.
201 * This signifies to the remote partition that our reserved
202 * page is initialized.
204 rp->stamp = CURRENT_TIME;
206 return rp;
210 * Change protections to allow IPI operations (and AMO operations on
211 * Shub 1.1 systems).
213 void
214 xpc_allow_IPI_ops(void)
216 int node;
217 int nasid;
219 /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
221 if (is_shub2()) {
222 xpc_sh2_IPI_access0 =
223 (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
224 xpc_sh2_IPI_access1 =
225 (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
226 xpc_sh2_IPI_access2 =
227 (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
228 xpc_sh2_IPI_access3 =
229 (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
231 for_each_online_node(node) {
232 nasid = cnodeid_to_nasid(node);
233 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
234 -1UL);
235 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
236 -1UL);
237 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
238 -1UL);
239 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
240 -1UL);
243 } else {
244 xpc_sh1_IPI_access =
245 (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
247 for_each_online_node(node) {
248 nasid = cnodeid_to_nasid(node);
249 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
250 -1UL);
253 * Since the BIST collides with memory operations on
254 * SHUB 1.1 sn_change_memprotect() cannot be used.
256 if (enable_shub_wars_1_1()) {
257 /* open up everything */
258 xpc_prot_vec[node] = (u64)HUB_L((u64 *)
259 GLOBAL_MMR_ADDR
260 (nasid,
261 SH1_MD_DQLP_MMR_DIR_PRIVEC0));
262 HUB_S((u64 *)
263 GLOBAL_MMR_ADDR(nasid,
264 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
265 -1UL);
266 HUB_S((u64 *)
267 GLOBAL_MMR_ADDR(nasid,
268 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
269 -1UL);
276 * Restrict protections to disallow IPI operations (and AMO operations on
277 * Shub 1.1 systems).
279 void
280 xpc_restrict_IPI_ops(void)
282 int node;
283 int nasid;
285 /* >>> Change SH_IPI_ACCESS code to use SAL call once it is available */
287 if (is_shub2()) {
289 for_each_online_node(node) {
290 nasid = cnodeid_to_nasid(node);
291 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
292 xpc_sh2_IPI_access0);
293 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
294 xpc_sh2_IPI_access1);
295 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
296 xpc_sh2_IPI_access2);
297 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
298 xpc_sh2_IPI_access3);
301 } else {
303 for_each_online_node(node) {
304 nasid = cnodeid_to_nasid(node);
305 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
306 xpc_sh1_IPI_access);
308 if (enable_shub_wars_1_1()) {
309 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
310 SH1_MD_DQLP_MMR_DIR_PRIVEC0),
311 xpc_prot_vec[node]);
312 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
313 SH1_MD_DQRP_MMR_DIR_PRIVEC0),
314 xpc_prot_vec[node]);
321 * At periodic intervals, scan through all active partitions and ensure
322 * their heartbeat is still active. If not, the partition is deactivated.
324 void
325 xpc_check_remote_hb(void)
327 struct xpc_vars *remote_vars;
328 struct xpc_partition *part;
329 short partid;
330 enum xp_retval ret;
332 remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
334 for (partid = 0; partid < xp_max_npartitions; partid++) {
336 if (xpc_exiting)
337 break;
339 if (partid == sn_partition_id)
340 continue;
342 part = &xpc_partitions[partid];
344 if (part->act_state == XPC_P_INACTIVE ||
345 part->act_state == XPC_P_DEACTIVATING) {
346 continue;
349 /* pull the remote_hb cache line */
350 ret = xp_remote_memcpy(remote_vars,
351 (void *)part->remote_vars_pa,
352 XPC_RP_VARS_SIZE);
353 if (ret != xpSuccess) {
354 XPC_DEACTIVATE_PARTITION(part, ret);
355 continue;
358 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
359 " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
360 partid, remote_vars->heartbeat, part->last_heartbeat,
361 remote_vars->heartbeat_offline,
362 remote_vars->heartbeating_to_mask);
364 if (((remote_vars->heartbeat == part->last_heartbeat) &&
365 (remote_vars->heartbeat_offline == 0)) ||
366 !xpc_hb_allowed(sn_partition_id, remote_vars)) {
368 XPC_DEACTIVATE_PARTITION(part, xpNoHeartbeat);
369 continue;
372 part->last_heartbeat = remote_vars->heartbeat;
377 * Get a copy of a portion of the remote partition's rsvd page.
379 * remote_rp points to a buffer that is cacheline aligned for BTE copies and
380 * is large enough to contain a copy of their reserved page header and
381 * part_nasids mask.
383 static enum xp_retval
384 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
385 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
387 int i;
388 enum xp_retval ret;
390 /* get the reserved page's physical address */
392 *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
393 if (*remote_rp_pa == 0)
394 return xpNoRsvdPageAddr;
396 /* pull over the reserved page header and part_nasids mask */
397 ret = xp_remote_memcpy(remote_rp, (void *)*remote_rp_pa,
398 XPC_RP_HEADER_SIZE + xp_sizeof_nasid_mask);
399 if (ret != xpSuccess)
400 return ret;
402 if (discovered_nasids != NULL) {
403 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
405 for (i = 0; i < xp_nasid_mask_words; i++)
406 discovered_nasids[i] |= remote_part_nasids[i];
409 /* check that the partid is valid and is for another partition */
411 if (remote_rp->SAL_partid < 0 ||
412 remote_rp->SAL_partid >= xp_max_npartitions) {
413 return xpInvalidPartid;
416 if (remote_rp->SAL_partid == sn_partition_id)
417 return xpLocalPartid;
419 /* see if the rest of the reserved page has been set up by XPC */
420 if (timespec_equal(&remote_rp->stamp, &ZERO_STAMP))
421 return xpRsvdPageNotSet;
423 if (XPC_VERSION_MAJOR(remote_rp->version) !=
424 XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
425 return xpBadVersion;
428 if (remote_rp->max_npartitions <= sn_partition_id)
429 return xpInvalidPartid;
431 return xpSuccess;
435 * Get a copy of the remote partition's XPC variables from the reserved page.
437 * remote_vars points to a buffer that is cacheline aligned for BTE copies and
438 * assumed to be of size XPC_RP_VARS_SIZE.
440 static enum xp_retval
441 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
443 enum xp_retval ret;
445 if (remote_vars_pa == 0)
446 return xpVarsNotSet;
448 /* pull over the cross partition variables */
449 ret = xp_remote_memcpy(remote_vars, (void *)remote_vars_pa,
450 XPC_RP_VARS_SIZE);
451 if (ret != xpSuccess)
452 return ret;
454 if (XPC_VERSION_MAJOR(remote_vars->version) !=
455 XPC_VERSION_MAJOR(XPC_V_VERSION)) {
456 return xpBadVersion;
459 return xpSuccess;
463 * Update the remote partition's info.
465 static void
466 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
467 struct timespec *remote_rp_stamp, u64 remote_rp_pa,
468 u64 remote_vars_pa, struct xpc_vars *remote_vars)
470 part->remote_rp_version = remote_rp_version;
471 dev_dbg(xpc_part, " remote_rp_version = 0x%016x\n",
472 part->remote_rp_version);
474 part->remote_rp_stamp = *remote_rp_stamp;
475 dev_dbg(xpc_part, " remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
476 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
478 part->remote_rp_pa = remote_rp_pa;
479 dev_dbg(xpc_part, " remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
481 part->remote_vars_pa = remote_vars_pa;
482 dev_dbg(xpc_part, " remote_vars_pa = 0x%016lx\n",
483 part->remote_vars_pa);
485 part->last_heartbeat = remote_vars->heartbeat;
486 dev_dbg(xpc_part, " last_heartbeat = 0x%016lx\n",
487 part->last_heartbeat);
489 /* >>> remote_vars_part_pa and vars_part_pa are sn2 only!!! */
490 part->remote_vars_part_pa = remote_vars->vars_part_pa;
491 dev_dbg(xpc_part, " remote_vars_part_pa = 0x%016lx\n",
492 part->remote_vars_part_pa);
494 part->remote_act_nasid = remote_vars->act_nasid;
495 dev_dbg(xpc_part, " remote_act_nasid = 0x%x\n",
496 part->remote_act_nasid);
498 part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
499 dev_dbg(xpc_part, " remote_act_phys_cpuid = 0x%x\n",
500 part->remote_act_phys_cpuid);
502 part->remote_amos_page_pa = remote_vars->amos_page_pa;
503 dev_dbg(xpc_part, " remote_amos_page_pa = 0x%lx\n",
504 part->remote_amos_page_pa);
506 part->remote_vars_version = remote_vars->version;
507 dev_dbg(xpc_part, " remote_vars_version = 0x%x\n",
508 part->remote_vars_version);
512 * Prior code has determined the nasid which generated an IPI. Inspect
513 * that nasid to determine if its partition needs to be activated or
514 * deactivated.
516 * A partition is consider "awaiting activation" if our partition
517 * flags indicate it is not active and it has a heartbeat. A
518 * partition is considered "awaiting deactivation" if our partition
519 * flags indicate it is active but it has no heartbeat or it is not
520 * sending its heartbeat to us.
522 * To determine the heartbeat, the remote nasid must have a properly
523 * initialized reserved page.
525 static void
526 xpc_identify_act_IRQ_req(int nasid)
528 struct xpc_rsvd_page *remote_rp;
529 struct xpc_vars *remote_vars;
530 u64 remote_rp_pa;
531 u64 remote_vars_pa;
532 int remote_rp_version;
533 int reactivate = 0;
534 int stamp_diff;
535 struct timespec remote_rp_stamp = { 0, 0 }; /*>>> ZERO_STAMP */
536 short partid;
537 struct xpc_partition *part;
538 enum xp_retval ret;
540 /* pull over the reserved page structure */
542 remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer;
544 ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
545 if (ret != xpSuccess) {
546 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
547 "which sent interrupt, reason=%d\n", nasid, ret);
548 return;
551 remote_vars_pa = remote_rp->sn.vars_pa;
552 remote_rp_version = remote_rp->version;
553 if (XPC_SUPPORTS_RP_STAMP(remote_rp_version))
554 remote_rp_stamp = remote_rp->stamp;
556 partid = remote_rp->SAL_partid;
557 part = &xpc_partitions[partid];
559 /* pull over the cross partition variables */
561 remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
563 ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
564 if (ret != xpSuccess) {
566 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
567 "which sent interrupt, reason=%d\n", nasid, ret);
569 XPC_DEACTIVATE_PARTITION(part, ret);
570 return;
573 part->act_IRQ_rcvd++;
575 dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
576 "%ld:0x%lx\n", (int)nasid, (int)partid, part->act_IRQ_rcvd,
577 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
579 if (xpc_partition_disengaged(part) &&
580 part->act_state == XPC_P_INACTIVE) {
582 xpc_update_partition_info(part, remote_rp_version,
583 &remote_rp_stamp, remote_rp_pa,
584 remote_vars_pa, remote_vars);
586 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
587 if (xpc_partition_disengage_requested(1UL << partid)) {
589 * Other side is waiting on us to disengage,
590 * even though we already have.
592 return;
594 } else {
595 /* other side doesn't support disengage requests */
596 xpc_clear_partition_disengage_request(1UL << partid);
599 xpc_activate_partition(part);
600 return;
603 DBUG_ON(part->remote_rp_version == 0);
604 DBUG_ON(part->remote_vars_version == 0);
606 if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
607 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
608 remote_vars_version));
610 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
611 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
612 version));
613 /* see if the other side rebooted */
614 if (part->remote_amos_page_pa ==
615 remote_vars->amos_page_pa &&
616 xpc_hb_allowed(sn_partition_id, remote_vars)) {
617 /* doesn't look that way, so ignore the IPI */
618 return;
623 * Other side rebooted and previous XPC didn't support the
624 * disengage request, so we don't need to do anything special.
627 xpc_update_partition_info(part, remote_rp_version,
628 &remote_rp_stamp, remote_rp_pa,
629 remote_vars_pa, remote_vars);
630 part->reactivate_nasid = nasid;
631 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
632 return;
635 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
637 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
638 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
641 * Other side rebooted and previous XPC did support the
642 * disengage request, but the new one doesn't.
645 xpc_clear_partition_engaged(1UL << partid);
646 xpc_clear_partition_disengage_request(1UL << partid);
648 xpc_update_partition_info(part, remote_rp_version,
649 &remote_rp_stamp, remote_rp_pa,
650 remote_vars_pa, remote_vars);
651 reactivate = 1;
653 } else {
654 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
656 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
657 &remote_rp_stamp);
658 if (stamp_diff != 0) {
659 DBUG_ON(stamp_diff >= 0);
662 * Other side rebooted and the previous XPC did support
663 * the disengage request, as does the new one.
666 DBUG_ON(xpc_partition_engaged(1UL << partid));
667 DBUG_ON(xpc_partition_disengage_requested(1UL <<
668 partid));
670 xpc_update_partition_info(part, remote_rp_version,
671 &remote_rp_stamp,
672 remote_rp_pa, remote_vars_pa,
673 remote_vars);
674 reactivate = 1;
678 if (part->disengage_request_timeout > 0 &&
679 !xpc_partition_disengaged(part)) {
680 /* still waiting on other side to disengage from us */
681 return;
684 if (reactivate) {
685 part->reactivate_nasid = nasid;
686 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
688 } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
689 xpc_partition_disengage_requested(1UL << partid)) {
690 XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
695 * Loop through the activation AMO variables and process any bits
696 * which are set. Each bit indicates a nasid sending a partition
697 * activation or deactivation request.
699 * Return #of IRQs detected.
702 xpc_identify_act_IRQ_sender(void)
704 int word, bit;
705 u64 nasid_mask;
706 u64 nasid; /* remote nasid */
707 int n_IRQs_detected = 0;
708 AMO_t *act_amos;
710 act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
712 /* scan through act AMO variable looking for non-zero entries */
713 for (word = 0; word < xp_nasid_mask_words; word++) {
715 if (xpc_exiting)
716 break;
718 nasid_mask = xpc_IPI_receive(&act_amos[word]);
719 if (nasid_mask == 0) {
720 /* no IRQs from nasids in this variable */
721 continue;
724 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
725 nasid_mask);
728 * If this nasid has been added to the machine since
729 * our partition was reset, this will retain the
730 * remote nasid in our reserved pages machine mask.
731 * This is used in the event of module reload.
733 xpc_mach_nasids[word] |= nasid_mask;
735 /* locate the nasid(s) which sent interrupts */
737 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
738 if (nasid_mask & (1UL << bit)) {
739 n_IRQs_detected++;
740 nasid = XPC_NASID_FROM_W_B(word, bit);
741 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
742 nasid);
743 xpc_identify_act_IRQ_req(nasid);
747 return n_IRQs_detected;
751 * See if the other side has responded to a partition disengage request
752 * from us.
755 xpc_partition_disengaged(struct xpc_partition *part)
757 short partid = XPC_PARTID(part);
758 int disengaged;
760 disengaged = (xpc_partition_engaged(1UL << partid) == 0);
761 if (part->disengage_request_timeout) {
762 if (!disengaged) {
763 if (time_before(jiffies,
764 part->disengage_request_timeout)) {
765 /* timelimit hasn't been reached yet */
766 return 0;
770 * Other side hasn't responded to our disengage
771 * request in a timely fashion, so assume it's dead.
774 dev_info(xpc_part, "disengage from remote partition %d "
775 "timed out\n", partid);
776 xpc_disengage_request_timedout = 1;
777 xpc_clear_partition_engaged(1UL << partid);
778 disengaged = 1;
780 part->disengage_request_timeout = 0;
782 /* cancel the timer function, provided it's not us */
783 if (!in_interrupt()) {
784 del_singleshot_timer_sync(&part->
785 disengage_request_timer);
788 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
789 part->act_state != XPC_P_INACTIVE);
790 if (part->act_state != XPC_P_INACTIVE)
791 xpc_wakeup_channel_mgr(part);
793 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version))
794 xpc_cancel_partition_disengage_request(part);
796 return disengaged;
800 * Mark specified partition as active.
802 enum xp_retval
803 xpc_mark_partition_active(struct xpc_partition *part)
805 unsigned long irq_flags;
806 enum xp_retval ret;
808 dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
810 spin_lock_irqsave(&part->act_lock, irq_flags);
811 if (part->act_state == XPC_P_ACTIVATING) {
812 part->act_state = XPC_P_ACTIVE;
813 ret = xpSuccess;
814 } else {
815 DBUG_ON(part->reason == xpSuccess);
816 ret = part->reason;
818 spin_unlock_irqrestore(&part->act_lock, irq_flags);
820 return ret;
824 * Notify XPC that the partition is down.
826 void
827 xpc_deactivate_partition(const int line, struct xpc_partition *part,
828 enum xp_retval reason)
830 unsigned long irq_flags;
832 spin_lock_irqsave(&part->act_lock, irq_flags);
834 if (part->act_state == XPC_P_INACTIVE) {
835 XPC_SET_REASON(part, reason, line);
836 spin_unlock_irqrestore(&part->act_lock, irq_flags);
837 if (reason == xpReactivating) {
838 /* we interrupt ourselves to reactivate partition */
839 xpc_IPI_send_reactivate(part);
841 return;
843 if (part->act_state == XPC_P_DEACTIVATING) {
844 if ((part->reason == xpUnloading && reason != xpUnloading) ||
845 reason == xpReactivating) {
846 XPC_SET_REASON(part, reason, line);
848 spin_unlock_irqrestore(&part->act_lock, irq_flags);
849 return;
852 part->act_state = XPC_P_DEACTIVATING;
853 XPC_SET_REASON(part, reason, line);
855 spin_unlock_irqrestore(&part->act_lock, irq_flags);
857 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
858 xpc_request_partition_disengage(part);
859 xpc_IPI_send_disengage(part);
861 /* set a timelimit on the disengage request */
862 part->disengage_request_timeout = jiffies +
863 (xpc_disengage_request_timelimit * HZ);
864 part->disengage_request_timer.expires =
865 part->disengage_request_timeout;
866 add_timer(&part->disengage_request_timer);
869 dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
870 XPC_PARTID(part), reason);
872 xpc_partition_going_down(part, reason);
876 * Mark specified partition as inactive.
878 void
879 xpc_mark_partition_inactive(struct xpc_partition *part)
881 unsigned long irq_flags;
883 dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
884 XPC_PARTID(part));
886 spin_lock_irqsave(&part->act_lock, irq_flags);
887 part->act_state = XPC_P_INACTIVE;
888 spin_unlock_irqrestore(&part->act_lock, irq_flags);
889 part->remote_rp_pa = 0;
893 * SAL has provided a partition and machine mask. The partition mask
894 * contains a bit for each even nasid in our partition. The machine
895 * mask contains a bit for each even nasid in the entire machine.
897 * Using those two bit arrays, we can determine which nasids are
898 * known in the machine. Each should also have a reserved page
899 * initialized if they are available for partitioning.
901 void
902 xpc_discovery(void)
904 void *remote_rp_base;
905 struct xpc_rsvd_page *remote_rp;
906 struct xpc_vars *remote_vars;
907 u64 remote_rp_pa;
908 u64 remote_vars_pa;
909 int region;
910 int region_size;
911 int max_regions;
912 int nasid;
913 struct xpc_rsvd_page *rp;
914 short partid;
915 struct xpc_partition *part;
916 u64 *discovered_nasids;
917 enum xp_retval ret;
919 remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
920 xp_sizeof_nasid_mask,
921 GFP_KERNEL, &remote_rp_base);
922 if (remote_rp == NULL)
923 return;
925 remote_vars = (struct xpc_vars *)remote_rp;
927 discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
928 GFP_KERNEL);
929 if (discovered_nasids == NULL) {
930 kfree(remote_rp_base);
931 return;
934 rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
937 * The term 'region' in this context refers to the minimum number of
938 * nodes that can comprise an access protection grouping. The access
939 * protection is in regards to memory, IOI and IPI.
941 max_regions = 64;
942 region_size = sn_region_size;
944 switch (region_size) {
945 case 128:
946 max_regions *= 2;
947 case 64:
948 max_regions *= 2;
949 case 32:
950 max_regions *= 2;
951 region_size = 16;
952 DBUG_ON(!is_shub2());
955 for (region = 0; region < max_regions; region++) {
957 if (xpc_exiting)
958 break;
960 dev_dbg(xpc_part, "searching region %d\n", region);
962 for (nasid = (region * region_size * 2);
963 nasid < ((region + 1) * region_size * 2); nasid += 2) {
965 if (xpc_exiting)
966 break;
968 dev_dbg(xpc_part, "checking nasid %d\n", nasid);
970 if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
971 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
972 "part of the local partition; skipping "
973 "region\n", nasid);
974 break;
977 if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
978 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
979 "not on Numa-Link network at reset\n",
980 nasid);
981 continue;
984 if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
985 dev_dbg(xpc_part, "Nasid %d is part of a "
986 "partition which was previously "
987 "discovered\n", nasid);
988 continue;
991 /* pull over the reserved page structure */
993 ret = xpc_get_remote_rp(nasid, discovered_nasids,
994 remote_rp, &remote_rp_pa);
995 if (ret != xpSuccess) {
996 dev_dbg(xpc_part, "unable to get reserved page "
997 "from nasid %d, reason=%d\n", nasid,
998 ret);
1000 if (ret == xpLocalPartid)
1001 break;
1003 continue;
1006 remote_vars_pa = remote_rp->sn.vars_pa;
1008 partid = remote_rp->SAL_partid;
1009 part = &xpc_partitions[partid];
1011 /* pull over the cross partition variables */
1013 ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1014 if (ret != xpSuccess) {
1015 dev_dbg(xpc_part, "unable to get XPC variables "
1016 "from nasid %d, reason=%d\n", nasid,
1017 ret);
1019 XPC_DEACTIVATE_PARTITION(part, ret);
1020 continue;
1023 if (part->act_state != XPC_P_INACTIVE) {
1024 dev_dbg(xpc_part, "partition %d on nasid %d is "
1025 "already activating\n", partid, nasid);
1026 break;
1030 * Register the remote partition's AMOs with SAL so it
1031 * can handle and cleanup errors within that address
1032 * range should the remote partition go down. We don't
1033 * unregister this range because it is difficult to
1034 * tell when outstanding writes to the remote partition
1035 * are finished and thus when it is thus safe to
1036 * unregister. This should not result in wasted space
1037 * in the SAL xp_addr_region table because we should
1038 * get the same page for remote_act_amos_pa after
1039 * module reloads and system reboots.
1041 if (sn_register_xp_addr_region
1042 (remote_vars->amos_page_pa, PAGE_SIZE, 1) < 0) {
1043 dev_dbg(xpc_part,
1044 "partition %d failed to "
1045 "register xp_addr region 0x%016lx\n",
1046 partid, remote_vars->amos_page_pa);
1048 XPC_SET_REASON(part, xpPhysAddrRegFailed,
1049 __LINE__);
1050 break;
1054 * The remote nasid is valid and available.
1055 * Send an interrupt to that nasid to notify
1056 * it that we are ready to begin activation.
1058 dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1059 "nasid %d, phys_cpuid 0x%x\n",
1060 remote_vars->amos_page_pa,
1061 remote_vars->act_nasid,
1062 remote_vars->act_phys_cpuid);
1064 if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1065 version)) {
1066 part->remote_amos_page_pa =
1067 remote_vars->amos_page_pa;
1068 xpc_mark_partition_disengaged(part);
1069 xpc_cancel_partition_disengage_request(part);
1071 xpc_IPI_send_activate(remote_vars);
1075 kfree(discovered_nasids);
1076 kfree(remote_rp_base);
1080 * Given a partid, get the nasids owned by that partition from the
1081 * remote partition's reserved page.
1083 enum xp_retval
1084 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
1086 struct xpc_partition *part;
1087 u64 part_nasid_pa;
1089 part = &xpc_partitions[partid];
1090 if (part->remote_rp_pa == 0)
1091 return xpPartitionDown;
1093 memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1095 part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
1097 return xp_remote_memcpy(nasid_mask, (void *)part_nasid_pa,
1098 xp_sizeof_nasid_mask);