[IA64] run some drivers/misc/sgi-xp through scripts/Lindent
[linux-2.6] / drivers / misc / sgi-xp / xpc_partition.c
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.
5  *
6  * Copyright (c) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
7  */
8
9 /*
10  * Cross Partition Communication (XPC) partition support.
11  *
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.
15  *
16  */
17
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/uncached.h>
24 #include <asm/sn/bte.h>
25 #include <asm/sn/intr.h>
26 #include <asm/sn/sn_sal.h>
27 #include <asm/sn/nodepda.h>
28 #include <asm/sn/addrs.h>
29 #include "xpc.h"
30
31 /* XPC is exiting flag */
32 int xpc_exiting;
33
34 /* SH_IPI_ACCESS shub register value on startup */
35 static u64 xpc_sh1_IPI_access;
36 static u64 xpc_sh2_IPI_access0;
37 static u64 xpc_sh2_IPI_access1;
38 static u64 xpc_sh2_IPI_access2;
39 static u64 xpc_sh2_IPI_access3;
40
41 /* original protection values for each node */
42 u64 xpc_prot_vec[MAX_NUMNODES];
43
44 /* this partition's reserved page pointers */
45 struct xpc_rsvd_page *xpc_rsvd_page;
46 static u64 *xpc_part_nasids;
47 static u64 *xpc_mach_nasids;
48 struct xpc_vars *xpc_vars;
49 struct xpc_vars_part *xpc_vars_part;
50
51 static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
52 static int xp_nasid_mask_words; /* actual size in words of nasid mask */
53
54 /*
55  * For performance reasons, each entry of xpc_partitions[] is cacheline
56  * aligned. And xpc_partitions[] is padded with an additional entry at the
57  * end so that the last legitimate entry doesn't share its cacheline with
58  * another variable.
59  */
60 struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
61
62 /*
63  * Generic buffer used to store a local copy of portions of a remote
64  * partition's reserved page (either its header and part_nasids mask,
65  * or its vars).
66  */
67 char *xpc_remote_copy_buffer;
68 void *xpc_remote_copy_buffer_base;
69
70 /*
71  * Guarantee that the kmalloc'd memory is cacheline aligned.
72  */
73 void *
74 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
75 {
76         /* see if kmalloc will give us cachline aligned memory by default */
77         *base = kmalloc(size, flags);
78         if (*base == NULL) {
79                 return NULL;
80         }
81         if ((u64)*base == L1_CACHE_ALIGN((u64)*base)) {
82                 return *base;
83         }
84         kfree(*base);
85
86         /* nope, we'll have to do it ourselves */
87         *base = kmalloc(size + L1_CACHE_BYTES, flags);
88         if (*base == NULL) {
89                 return NULL;
90         }
91         return (void *)L1_CACHE_ALIGN((u64)*base);
92 }
93
94 /*
95  * Given a nasid, get the physical address of the  partition's reserved page
96  * for that nasid. This function returns 0 on any error.
97  */
98 static u64
99 xpc_get_rsvd_page_pa(int nasid)
100 {
101         bte_result_t bte_res;
102         s64 status;
103         u64 cookie = 0;
104         u64 rp_pa = nasid;      /* seed with nasid */
105         u64 len = 0;
106         u64 buf = buf;
107         u64 buf_len = 0;
108         void *buf_base = NULL;
109
110         while (1) {
111
112                 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
113                                                        &len);
114
115                 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
116                         "0x%016lx, address=0x%016lx, len=0x%016lx\n",
117                         status, cookie, rp_pa, len);
118
119                 if (status != SALRET_MORE_PASSES) {
120                         break;
121                 }
122
123                 if (L1_CACHE_ALIGN(len) > buf_len) {
124                         kfree(buf_base);
125                         buf_len = L1_CACHE_ALIGN(len);
126                         buf = (u64)xpc_kmalloc_cacheline_aligned(buf_len,
127                                                                  GFP_KERNEL,
128                                                                  &buf_base);
129                         if (buf_base == NULL) {
130                                 dev_err(xpc_part, "unable to kmalloc "
131                                         "len=0x%016lx\n", buf_len);
132                                 status = SALRET_ERROR;
133                                 break;
134                         }
135                 }
136
137                 bte_res = xp_bte_copy(rp_pa, buf, buf_len,
138                                       (BTE_NOTIFY | BTE_WACQUIRE), NULL);
139                 if (bte_res != BTE_SUCCESS) {
140                         dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
141                         status = SALRET_ERROR;
142                         break;
143                 }
144         }
145
146         kfree(buf_base);
147
148         if (status != SALRET_OK) {
149                 rp_pa = 0;
150         }
151         dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
152         return rp_pa;
153 }
154
155 /*
156  * Fill the partition reserved page with the information needed by
157  * other partitions to discover we are alive and establish initial
158  * communications.
159  */
160 struct xpc_rsvd_page *
161 xpc_rsvd_page_init(void)
162 {
163         struct xpc_rsvd_page *rp;
164         AMO_t *amos_page;
165         u64 rp_pa, nasid_array = 0;
166         int i, ret;
167
168         /* get the local reserved page's address */
169
170         preempt_disable();
171         rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
172         preempt_enable();
173         if (rp_pa == 0) {
174                 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
175                 return NULL;
176         }
177         rp = (struct xpc_rsvd_page *)__va(rp_pa);
178
179         if (rp->partid != sn_partition_id) {
180                 dev_err(xpc_part, "the reserved page's partid of %d should be "
181                         "%d\n", rp->partid, sn_partition_id);
182                 return NULL;
183         }
184
185         rp->version = XPC_RP_VERSION;
186
187         /* establish the actual sizes of the nasid masks */
188         if (rp->SAL_version == 1) {
189                 /* SAL_version 1 didn't set the nasids_size field */
190                 rp->nasids_size = 128;
191         }
192         xp_nasid_mask_bytes = rp->nasids_size;
193         xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
194
195         /* setup the pointers to the various items in the reserved page */
196         xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
197         xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
198         xpc_vars = XPC_RP_VARS(rp);
199         xpc_vars_part = XPC_RP_VARS_PART(rp);
200
201         /*
202          * Before clearing xpc_vars, see if a page of AMOs had been previously
203          * allocated. If not we'll need to allocate one and set permissions
204          * so that cross-partition AMOs are allowed.
205          *
206          * The allocated AMO page needs MCA reporting to remain disabled after
207          * XPC has unloaded.  To make this work, we keep a copy of the pointer
208          * to this page (i.e., amos_page) in the struct xpc_vars structure,
209          * which is pointed to by the reserved page, and re-use that saved copy
210          * on subsequent loads of XPC. This AMO page is never freed, and its
211          * memory protections are never restricted.
212          */
213         if ((amos_page = xpc_vars->amos_page) == NULL) {
214                 amos_page = (AMO_t *)TO_AMO(uncached_alloc_page(0));
215                 if (amos_page == NULL) {
216                         dev_err(xpc_part, "can't allocate page of AMOs\n");
217                         return NULL;
218                 }
219
220                 /*
221                  * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
222                  * when xpc_allow_IPI_ops() is called via xpc_hb_init().
223                  */
224                 if (!enable_shub_wars_1_1()) {
225                         ret = sn_change_memprotect(ia64_tpa((u64)amos_page),
226                                                    PAGE_SIZE,
227                                                    SN_MEMPROT_ACCESS_CLASS_1,
228                                                    &nasid_array);
229                         if (ret != 0) {
230                                 dev_err(xpc_part, "can't change memory "
231                                         "protections\n");
232                                 uncached_free_page(__IA64_UNCACHED_OFFSET |
233                                                    TO_PHYS((u64)amos_page));
234                                 return NULL;
235                         }
236                 }
237         } else if (!IS_AMO_ADDRESS((u64)amos_page)) {
238                 /*
239                  * EFI's XPBOOT can also set amos_page in the reserved page,
240                  * but it happens to leave it as an uncached physical address
241                  * and we need it to be an uncached virtual, so we'll have to
242                  * convert it.
243                  */
244                 if (!IS_AMO_PHYS_ADDRESS((u64)amos_page)) {
245                         dev_err(xpc_part, "previously used amos_page address "
246                                 "is bad = 0x%p\n", (void *)amos_page);
247                         return NULL;
248                 }
249                 amos_page = (AMO_t *)TO_AMO((u64)amos_page);
250         }
251
252         /* clear xpc_vars */
253         memset(xpc_vars, 0, sizeof(struct xpc_vars));
254
255         xpc_vars->version = XPC_V_VERSION;
256         xpc_vars->act_nasid = cpuid_to_nasid(0);
257         xpc_vars->act_phys_cpuid = cpu_physical_id(0);
258         xpc_vars->vars_part_pa = __pa(xpc_vars_part);
259         xpc_vars->amos_page_pa = ia64_tpa((u64)amos_page);
260         xpc_vars->amos_page = amos_page;        /* save for next load of XPC */
261
262         /* clear xpc_vars_part */
263         memset((u64 *)xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
264                XP_MAX_PARTITIONS);
265
266         /* initialize the activate IRQ related AMO variables */
267         for (i = 0; i < xp_nasid_mask_words; i++) {
268                 (void)xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
269         }
270
271         /* initialize the engaged remote partitions related AMO variables */
272         (void)xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
273         (void)xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
274
275         /* timestamp of when reserved page was setup by XPC */
276         rp->stamp = CURRENT_TIME;
277
278         /*
279          * This signifies to the remote partition that our reserved
280          * page is initialized.
281          */
282         rp->vars_pa = __pa(xpc_vars);
283
284         return rp;
285 }
286
287 /*
288  * Change protections to allow IPI operations (and AMO operations on
289  * Shub 1.1 systems).
290  */
291 void
292 xpc_allow_IPI_ops(void)
293 {
294         int node;
295         int nasid;
296
297         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
298
299         if (is_shub2()) {
300                 xpc_sh2_IPI_access0 =
301                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
302                 xpc_sh2_IPI_access1 =
303                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
304                 xpc_sh2_IPI_access2 =
305                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
306                 xpc_sh2_IPI_access3 =
307                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
308
309                 for_each_online_node(node) {
310                         nasid = cnodeid_to_nasid(node);
311                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
312                               -1UL);
313                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
314                               -1UL);
315                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
316                               -1UL);
317                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
318                               -1UL);
319                 }
320
321         } else {
322                 xpc_sh1_IPI_access =
323                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
324
325                 for_each_online_node(node) {
326                         nasid = cnodeid_to_nasid(node);
327                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
328                               -1UL);
329
330                         /*
331                          * Since the BIST collides with memory operations on
332                          * SHUB 1.1 sn_change_memprotect() cannot be used.
333                          */
334                         if (enable_shub_wars_1_1()) {
335                                 /* open up everything */
336                                 xpc_prot_vec[node] = (u64)HUB_L((u64 *)
337                                                                 GLOBAL_MMR_ADDR
338                                                                 (nasid,
339                                                                  SH1_MD_DQLP_MMR_DIR_PRIVEC0));
340                                 HUB_S((u64 *)
341                                       GLOBAL_MMR_ADDR(nasid,
342                                                       SH1_MD_DQLP_MMR_DIR_PRIVEC0),
343                                       -1UL);
344                                 HUB_S((u64 *)
345                                       GLOBAL_MMR_ADDR(nasid,
346                                                       SH1_MD_DQRP_MMR_DIR_PRIVEC0),
347                                       -1UL);
348                         }
349                 }
350         }
351 }
352
353 /*
354  * Restrict protections to disallow IPI operations (and AMO operations on
355  * Shub 1.1 systems).
356  */
357 void
358 xpc_restrict_IPI_ops(void)
359 {
360         int node;
361         int nasid;
362
363         // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
364
365         if (is_shub2()) {
366
367                 for_each_online_node(node) {
368                         nasid = cnodeid_to_nasid(node);
369                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
370                               xpc_sh2_IPI_access0);
371                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
372                               xpc_sh2_IPI_access1);
373                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
374                               xpc_sh2_IPI_access2);
375                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
376                               xpc_sh2_IPI_access3);
377                 }
378
379         } else {
380
381                 for_each_online_node(node) {
382                         nasid = cnodeid_to_nasid(node);
383                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
384                               xpc_sh1_IPI_access);
385
386                         if (enable_shub_wars_1_1()) {
387                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
388                                                              SH1_MD_DQLP_MMR_DIR_PRIVEC0),
389                                       xpc_prot_vec[node]);
390                                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
391                                                              SH1_MD_DQRP_MMR_DIR_PRIVEC0),
392                                       xpc_prot_vec[node]);
393                         }
394                 }
395         }
396 }
397
398 /*
399  * At periodic intervals, scan through all active partitions and ensure
400  * their heartbeat is still active.  If not, the partition is deactivated.
401  */
402 void
403 xpc_check_remote_hb(void)
404 {
405         struct xpc_vars *remote_vars;
406         struct xpc_partition *part;
407         partid_t partid;
408         bte_result_t bres;
409
410         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
411
412         for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
413
414                 if (xpc_exiting) {
415                         break;
416                 }
417
418                 if (partid == sn_partition_id) {
419                         continue;
420                 }
421
422                 part = &xpc_partitions[partid];
423
424                 if (part->act_state == XPC_P_INACTIVE ||
425                     part->act_state == XPC_P_DEACTIVATING) {
426                         continue;
427                 }
428
429                 /* pull the remote_hb cache line */
430                 bres = xp_bte_copy(part->remote_vars_pa,
431                                    (u64)remote_vars,
432                                    XPC_RP_VARS_SIZE,
433                                    (BTE_NOTIFY | BTE_WACQUIRE), NULL);
434                 if (bres != BTE_SUCCESS) {
435                         XPC_DEACTIVATE_PARTITION(part,
436                                                  xpc_map_bte_errors(bres));
437                         continue;
438                 }
439
440                 dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
441                         " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
442                         partid, remote_vars->heartbeat, part->last_heartbeat,
443                         remote_vars->heartbeat_offline,
444                         remote_vars->heartbeating_to_mask);
445
446                 if (((remote_vars->heartbeat == part->last_heartbeat) &&
447                      (remote_vars->heartbeat_offline == 0)) ||
448                     !xpc_hb_allowed(sn_partition_id, remote_vars)) {
449
450                         XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
451                         continue;
452                 }
453
454                 part->last_heartbeat = remote_vars->heartbeat;
455         }
456 }
457
458 /*
459  * Get a copy of a portion of the remote partition's rsvd page.
460  *
461  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
462  * is large enough to contain a copy of their reserved page header and
463  * part_nasids mask.
464  */
465 static enum xpc_retval
466 xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
467                   struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
468 {
469         int bres, i;
470
471         /* get the reserved page's physical address */
472
473         *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
474         if (*remote_rp_pa == 0) {
475                 return xpcNoRsvdPageAddr;
476         }
477
478         /* pull over the reserved page header and part_nasids mask */
479         bres = xp_bte_copy(*remote_rp_pa, (u64)remote_rp,
480                            XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
481                            (BTE_NOTIFY | BTE_WACQUIRE), NULL);
482         if (bres != BTE_SUCCESS) {
483                 return xpc_map_bte_errors(bres);
484         }
485
486         if (discovered_nasids != NULL) {
487                 u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
488
489                 for (i = 0; i < xp_nasid_mask_words; i++) {
490                         discovered_nasids[i] |= remote_part_nasids[i];
491                 }
492         }
493
494         /* check that the partid is for another partition */
495
496         if (remote_rp->partid < 1 ||
497             remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
498                 return xpcInvalidPartid;
499         }
500
501         if (remote_rp->partid == sn_partition_id) {
502                 return xpcLocalPartid;
503         }
504
505         if (XPC_VERSION_MAJOR(remote_rp->version) !=
506             XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
507                 return xpcBadVersion;
508         }
509
510         return xpcSuccess;
511 }
512
513 /*
514  * Get a copy of the remote partition's XPC variables from the reserved page.
515  *
516  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
517  * assumed to be of size XPC_RP_VARS_SIZE.
518  */
519 static enum xpc_retval
520 xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
521 {
522         int bres;
523
524         if (remote_vars_pa == 0) {
525                 return xpcVarsNotSet;
526         }
527
528         /* pull over the cross partition variables */
529         bres = xp_bte_copy(remote_vars_pa, (u64)remote_vars, XPC_RP_VARS_SIZE,
530                            (BTE_NOTIFY | BTE_WACQUIRE), NULL);
531         if (bres != BTE_SUCCESS) {
532                 return xpc_map_bte_errors(bres);
533         }
534
535         if (XPC_VERSION_MAJOR(remote_vars->version) !=
536             XPC_VERSION_MAJOR(XPC_V_VERSION)) {
537                 return xpcBadVersion;
538         }
539
540         return xpcSuccess;
541 }
542
543 /*
544  * Update the remote partition's info.
545  */
546 static void
547 xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
548                           struct timespec *remote_rp_stamp, u64 remote_rp_pa,
549                           u64 remote_vars_pa, struct xpc_vars *remote_vars)
550 {
551         part->remote_rp_version = remote_rp_version;
552         dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
553                 part->remote_rp_version);
554
555         part->remote_rp_stamp = *remote_rp_stamp;
556         dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
557                 part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
558
559         part->remote_rp_pa = remote_rp_pa;
560         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
561
562         part->remote_vars_pa = remote_vars_pa;
563         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
564                 part->remote_vars_pa);
565
566         part->last_heartbeat = remote_vars->heartbeat;
567         dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
568                 part->last_heartbeat);
569
570         part->remote_vars_part_pa = remote_vars->vars_part_pa;
571         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
572                 part->remote_vars_part_pa);
573
574         part->remote_act_nasid = remote_vars->act_nasid;
575         dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
576                 part->remote_act_nasid);
577
578         part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
579         dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
580                 part->remote_act_phys_cpuid);
581
582         part->remote_amos_page_pa = remote_vars->amos_page_pa;
583         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
584                 part->remote_amos_page_pa);
585
586         part->remote_vars_version = remote_vars->version;
587         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
588                 part->remote_vars_version);
589 }
590
591 /*
592  * Prior code has determined the nasid which generated an IPI.  Inspect
593  * that nasid to determine if its partition needs to be activated or
594  * deactivated.
595  *
596  * A partition is consider "awaiting activation" if our partition
597  * flags indicate it is not active and it has a heartbeat.  A
598  * partition is considered "awaiting deactivation" if our partition
599  * flags indicate it is active but it has no heartbeat or it is not
600  * sending its heartbeat to us.
601  *
602  * To determine the heartbeat, the remote nasid must have a properly
603  * initialized reserved page.
604  */
605 static void
606 xpc_identify_act_IRQ_req(int nasid)
607 {
608         struct xpc_rsvd_page *remote_rp;
609         struct xpc_vars *remote_vars;
610         u64 remote_rp_pa;
611         u64 remote_vars_pa;
612         int remote_rp_version;
613         int reactivate = 0;
614         int stamp_diff;
615         struct timespec remote_rp_stamp = { 0, 0 };
616         partid_t partid;
617         struct xpc_partition *part;
618         enum xpc_retval ret;
619
620         /* pull over the reserved page structure */
621
622         remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer;
623
624         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
625         if (ret != xpcSuccess) {
626                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
627                          "which sent interrupt, reason=%d\n", nasid, ret);
628                 return;
629         }
630
631         remote_vars_pa = remote_rp->vars_pa;
632         remote_rp_version = remote_rp->version;
633         if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
634                 remote_rp_stamp = remote_rp->stamp;
635         }
636         partid = remote_rp->partid;
637         part = &xpc_partitions[partid];
638
639         /* pull over the cross partition variables */
640
641         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
642
643         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
644         if (ret != xpcSuccess) {
645
646                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
647                          "which sent interrupt, reason=%d\n", nasid, ret);
648
649                 XPC_DEACTIVATE_PARTITION(part, ret);
650                 return;
651         }
652
653         part->act_IRQ_rcvd++;
654
655         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
656                 "%ld:0x%lx\n", (int)nasid, (int)partid, part->act_IRQ_rcvd,
657                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
658
659         if (xpc_partition_disengaged(part) && part->act_state == XPC_P_INACTIVE) {
660
661                 xpc_update_partition_info(part, remote_rp_version,
662                                           &remote_rp_stamp, remote_rp_pa,
663                                           remote_vars_pa, remote_vars);
664
665                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
666                         if (xpc_partition_disengage_requested(1UL << partid)) {
667                                 /*
668                                  * Other side is waiting on us to disengage,
669                                  * even though we already have.
670                                  */
671                                 return;
672                         }
673                 } else {
674                         /* other side doesn't support disengage requests */
675                         xpc_clear_partition_disengage_request(1UL << partid);
676                 }
677
678                 xpc_activate_partition(part);
679                 return;
680         }
681
682         DBUG_ON(part->remote_rp_version == 0);
683         DBUG_ON(part->remote_vars_version == 0);
684
685         if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
686                 DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
687                                                        remote_vars_version));
688
689                 if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
690                         DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
691                                                                version));
692                         /* see if the other side rebooted */
693                         if (part->remote_amos_page_pa ==
694                             remote_vars->amos_page_pa &&
695                             xpc_hb_allowed(sn_partition_id, remote_vars)) {
696                                 /* doesn't look that way, so ignore the IPI */
697                                 return;
698                         }
699                 }
700
701                 /*
702                  * Other side rebooted and previous XPC didn't support the
703                  * disengage request, so we don't need to do anything special.
704                  */
705
706                 xpc_update_partition_info(part, remote_rp_version,
707                                           &remote_rp_stamp, remote_rp_pa,
708                                           remote_vars_pa, remote_vars);
709                 part->reactivate_nasid = nasid;
710                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
711                 return;
712         }
713
714         DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
715
716         if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
717                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
718
719                 /*
720                  * Other side rebooted and previous XPC did support the
721                  * disengage request, but the new one doesn't.
722                  */
723
724                 xpc_clear_partition_engaged(1UL << partid);
725                 xpc_clear_partition_disengage_request(1UL << partid);
726
727                 xpc_update_partition_info(part, remote_rp_version,
728                                           &remote_rp_stamp, remote_rp_pa,
729                                           remote_vars_pa, remote_vars);
730                 reactivate = 1;
731
732         } else {
733                 DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
734
735                 stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
736                                                 &remote_rp_stamp);
737                 if (stamp_diff != 0) {
738                         DBUG_ON(stamp_diff >= 0);
739
740                         /*
741                          * Other side rebooted and the previous XPC did support
742                          * the disengage request, as does the new one.
743                          */
744
745                         DBUG_ON(xpc_partition_engaged(1UL << partid));
746                         DBUG_ON(xpc_partition_disengage_requested(1UL <<
747                                                                   partid));
748
749                         xpc_update_partition_info(part, remote_rp_version,
750                                                   &remote_rp_stamp,
751                                                   remote_rp_pa, remote_vars_pa,
752                                                   remote_vars);
753                         reactivate = 1;
754                 }
755         }
756
757         if (part->disengage_request_timeout > 0 &&
758             !xpc_partition_disengaged(part)) {
759                 /* still waiting on other side to disengage from us */
760                 return;
761         }
762
763         if (reactivate) {
764                 part->reactivate_nasid = nasid;
765                 XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
766
767         } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
768                    xpc_partition_disengage_requested(1UL << partid)) {
769                 XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
770         }
771 }
772
773 /*
774  * Loop through the activation AMO variables and process any bits
775  * which are set.  Each bit indicates a nasid sending a partition
776  * activation or deactivation request.
777  *
778  * Return #of IRQs detected.
779  */
780 int
781 xpc_identify_act_IRQ_sender(void)
782 {
783         int word, bit;
784         u64 nasid_mask;
785         u64 nasid;              /* remote nasid */
786         int n_IRQs_detected = 0;
787         AMO_t *act_amos;
788
789         act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
790
791         /* scan through act AMO variable looking for non-zero entries */
792         for (word = 0; word < xp_nasid_mask_words; word++) {
793
794                 if (xpc_exiting) {
795                         break;
796                 }
797
798                 nasid_mask = xpc_IPI_receive(&act_amos[word]);
799                 if (nasid_mask == 0) {
800                         /* no IRQs from nasids in this variable */
801                         continue;
802                 }
803
804                 dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
805                         nasid_mask);
806
807                 /*
808                  * If this nasid has been added to the machine since
809                  * our partition was reset, this will retain the
810                  * remote nasid in our reserved pages machine mask.
811                  * This is used in the event of module reload.
812                  */
813                 xpc_mach_nasids[word] |= nasid_mask;
814
815                 /* locate the nasid(s) which sent interrupts */
816
817                 for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
818                         if (nasid_mask & (1UL << bit)) {
819                                 n_IRQs_detected++;
820                                 nasid = XPC_NASID_FROM_W_B(word, bit);
821                                 dev_dbg(xpc_part, "interrupt from nasid %ld\n",
822                                         nasid);
823                                 xpc_identify_act_IRQ_req(nasid);
824                         }
825                 }
826         }
827         return n_IRQs_detected;
828 }
829
830 /*
831  * See if the other side has responded to a partition disengage request
832  * from us.
833  */
834 int
835 xpc_partition_disengaged(struct xpc_partition *part)
836 {
837         partid_t partid = XPC_PARTID(part);
838         int disengaged;
839
840         disengaged = (xpc_partition_engaged(1UL << partid) == 0);
841         if (part->disengage_request_timeout) {
842                 if (!disengaged) {
843                         if (time_before(jiffies, part->disengage_request_timeout)) {
844                                 /* timelimit hasn't been reached yet */
845                                 return 0;
846                         }
847
848                         /*
849                          * Other side hasn't responded to our disengage
850                          * request in a timely fashion, so assume it's dead.
851                          */
852
853                         dev_info(xpc_part, "disengage from remote partition %d "
854                                  "timed out\n", partid);
855                         xpc_disengage_request_timedout = 1;
856                         xpc_clear_partition_engaged(1UL << partid);
857                         disengaged = 1;
858                 }
859                 part->disengage_request_timeout = 0;
860
861                 /* cancel the timer function, provided it's not us */
862                 if (!in_interrupt()) {
863                         del_singleshot_timer_sync(&part->
864                                                   disengage_request_timer);
865                 }
866
867                 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
868                         part->act_state != XPC_P_INACTIVE);
869                 if (part->act_state != XPC_P_INACTIVE) {
870                         xpc_wakeup_channel_mgr(part);
871                 }
872
873                 if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
874                         xpc_cancel_partition_disengage_request(part);
875                 }
876         }
877         return disengaged;
878 }
879
880 /*
881  * Mark specified partition as active.
882  */
883 enum xpc_retval
884 xpc_mark_partition_active(struct xpc_partition *part)
885 {
886         unsigned long irq_flags;
887         enum xpc_retval ret;
888
889         dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
890
891         spin_lock_irqsave(&part->act_lock, irq_flags);
892         if (part->act_state == XPC_P_ACTIVATING) {
893                 part->act_state = XPC_P_ACTIVE;
894                 ret = xpcSuccess;
895         } else {
896                 DBUG_ON(part->reason == xpcSuccess);
897                 ret = part->reason;
898         }
899         spin_unlock_irqrestore(&part->act_lock, irq_flags);
900
901         return ret;
902 }
903
904 /*
905  * Notify XPC that the partition is down.
906  */
907 void
908 xpc_deactivate_partition(const int line, struct xpc_partition *part,
909                          enum xpc_retval reason)
910 {
911         unsigned long irq_flags;
912
913         spin_lock_irqsave(&part->act_lock, irq_flags);
914
915         if (part->act_state == XPC_P_INACTIVE) {
916                 XPC_SET_REASON(part, reason, line);
917                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
918                 if (reason == xpcReactivating) {
919                         /* we interrupt ourselves to reactivate partition */
920                         xpc_IPI_send_reactivate(part);
921                 }
922                 return;
923         }
924         if (part->act_state == XPC_P_DEACTIVATING) {
925                 if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
926                     reason == xpcReactivating) {
927                         XPC_SET_REASON(part, reason, line);
928                 }
929                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
930                 return;
931         }
932
933         part->act_state = XPC_P_DEACTIVATING;
934         XPC_SET_REASON(part, reason, line);
935
936         spin_unlock_irqrestore(&part->act_lock, irq_flags);
937
938         if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
939                 xpc_request_partition_disengage(part);
940                 xpc_IPI_send_disengage(part);
941
942                 /* set a timelimit on the disengage request */
943                 part->disengage_request_timeout = jiffies +
944                     (xpc_disengage_request_timelimit * HZ);
945                 part->disengage_request_timer.expires =
946                     part->disengage_request_timeout;
947                 add_timer(&part->disengage_request_timer);
948         }
949
950         dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
951                 XPC_PARTID(part), reason);
952
953         xpc_partition_going_down(part, reason);
954 }
955
956 /*
957  * Mark specified partition as inactive.
958  */
959 void
960 xpc_mark_partition_inactive(struct xpc_partition *part)
961 {
962         unsigned long irq_flags;
963
964         dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
965                 XPC_PARTID(part));
966
967         spin_lock_irqsave(&part->act_lock, irq_flags);
968         part->act_state = XPC_P_INACTIVE;
969         spin_unlock_irqrestore(&part->act_lock, irq_flags);
970         part->remote_rp_pa = 0;
971 }
972
973 /*
974  * SAL has provided a partition and machine mask.  The partition mask
975  * contains a bit for each even nasid in our partition.  The machine
976  * mask contains a bit for each even nasid in the entire machine.
977  *
978  * Using those two bit arrays, we can determine which nasids are
979  * known in the machine.  Each should also have a reserved page
980  * initialized if they are available for partitioning.
981  */
982 void
983 xpc_discovery(void)
984 {
985         void *remote_rp_base;
986         struct xpc_rsvd_page *remote_rp;
987         struct xpc_vars *remote_vars;
988         u64 remote_rp_pa;
989         u64 remote_vars_pa;
990         int region;
991         int region_size;
992         int max_regions;
993         int nasid;
994         struct xpc_rsvd_page *rp;
995         partid_t partid;
996         struct xpc_partition *part;
997         u64 *discovered_nasids;
998         enum xpc_retval ret;
999
1000         remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
1001                                                   xp_nasid_mask_bytes,
1002                                                   GFP_KERNEL, &remote_rp_base);
1003         if (remote_rp == NULL) {
1004                 return;
1005         }
1006         remote_vars = (struct xpc_vars *)remote_rp;
1007
1008         discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
1009                                     GFP_KERNEL);
1010         if (discovered_nasids == NULL) {
1011                 kfree(remote_rp_base);
1012                 return;
1013         }
1014
1015         rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
1016
1017         /*
1018          * The term 'region' in this context refers to the minimum number of
1019          * nodes that can comprise an access protection grouping. The access
1020          * protection is in regards to memory, IOI and IPI.
1021          */
1022         max_regions = 64;
1023         region_size = sn_region_size;
1024
1025         switch (region_size) {
1026         case 128:
1027                 max_regions *= 2;
1028         case 64:
1029                 max_regions *= 2;
1030         case 32:
1031                 max_regions *= 2;
1032                 region_size = 16;
1033                 DBUG_ON(!is_shub2());
1034         }
1035
1036         for (region = 0; region < max_regions; region++) {
1037
1038                 if ((volatile int)xpc_exiting) {
1039                         break;
1040                 }
1041
1042                 dev_dbg(xpc_part, "searching region %d\n", region);
1043
1044                 for (nasid = (region * region_size * 2);
1045                      nasid < ((region + 1) * region_size * 2); nasid += 2) {
1046
1047                         if ((volatile int)xpc_exiting) {
1048                                 break;
1049                         }
1050
1051                         dev_dbg(xpc_part, "checking nasid %d\n", nasid);
1052
1053                         if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
1054                                 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
1055                                         "part of the local partition; skipping "
1056                                         "region\n", nasid);
1057                                 break;
1058                         }
1059
1060                         if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
1061                                 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
1062                                         "not on Numa-Link network at reset\n",
1063                                         nasid);
1064                                 continue;
1065                         }
1066
1067                         if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
1068                                 dev_dbg(xpc_part, "Nasid %d is part of a "
1069                                         "partition which was previously "
1070                                         "discovered\n", nasid);
1071                                 continue;
1072                         }
1073
1074                         /* pull over the reserved page structure */
1075
1076                         ret = xpc_get_remote_rp(nasid, discovered_nasids,
1077                                                 remote_rp, &remote_rp_pa);
1078                         if (ret != xpcSuccess) {
1079                                 dev_dbg(xpc_part, "unable to get reserved page "
1080                                         "from nasid %d, reason=%d\n", nasid,
1081                                         ret);
1082
1083                                 if (ret == xpcLocalPartid) {
1084                                         break;
1085                                 }
1086                                 continue;
1087                         }
1088
1089                         remote_vars_pa = remote_rp->vars_pa;
1090
1091                         partid = remote_rp->partid;
1092                         part = &xpc_partitions[partid];
1093
1094                         /* pull over the cross partition variables */
1095
1096                         ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
1097                         if (ret != xpcSuccess) {
1098                                 dev_dbg(xpc_part, "unable to get XPC variables "
1099                                         "from nasid %d, reason=%d\n", nasid,
1100                                         ret);
1101
1102                                 XPC_DEACTIVATE_PARTITION(part, ret);
1103                                 continue;
1104                         }
1105
1106                         if (part->act_state != XPC_P_INACTIVE) {
1107                                 dev_dbg(xpc_part, "partition %d on nasid %d is "
1108                                         "already activating\n", partid, nasid);
1109                                 break;
1110                         }
1111
1112                         /*
1113                          * Register the remote partition's AMOs with SAL so it
1114                          * can handle and cleanup errors within that address
1115                          * range should the remote partition go down. We don't
1116                          * unregister this range because it is difficult to
1117                          * tell when outstanding writes to the remote partition
1118                          * are finished and thus when it is thus safe to
1119                          * unregister. This should not result in wasted space
1120                          * in the SAL xp_addr_region table because we should
1121                          * get the same page for remote_act_amos_pa after
1122                          * module reloads and system reboots.
1123                          */
1124                         if (sn_register_xp_addr_region
1125                             (remote_vars->amos_page_pa, PAGE_SIZE, 1) < 0) {
1126                                 dev_dbg(xpc_part,
1127                                         "partition %d failed to "
1128                                         "register xp_addr region 0x%016lx\n",
1129                                         partid, remote_vars->amos_page_pa);
1130
1131                                 XPC_SET_REASON(part, xpcPhysAddrRegFailed,
1132                                                __LINE__);
1133                                 break;
1134                         }
1135
1136                         /*
1137                          * The remote nasid is valid and available.
1138                          * Send an interrupt to that nasid to notify
1139                          * it that we are ready to begin activation.
1140                          */
1141                         dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
1142                                 "nasid %d, phys_cpuid 0x%x\n",
1143                                 remote_vars->amos_page_pa,
1144                                 remote_vars->act_nasid,
1145                                 remote_vars->act_phys_cpuid);
1146
1147                         if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
1148                                                            version)) {
1149                                 part->remote_amos_page_pa =
1150                                     remote_vars->amos_page_pa;
1151                                 xpc_mark_partition_disengaged(part);
1152                                 xpc_cancel_partition_disengage_request(part);
1153                         }
1154                         xpc_IPI_send_activate(remote_vars);
1155                 }
1156         }
1157
1158         kfree(discovered_nasids);
1159         kfree(remote_rp_base);
1160 }
1161
1162 /*
1163  * Given a partid, get the nasids owned by that partition from the
1164  * remote partition's reserved page.
1165  */
1166 enum xpc_retval
1167 xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
1168 {
1169         struct xpc_partition *part;
1170         u64 part_nasid_pa;
1171         int bte_res;
1172
1173         part = &xpc_partitions[partid];
1174         if (part->remote_rp_pa == 0) {
1175                 return xpcPartitionDown;
1176         }
1177
1178         memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
1179
1180         part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
1181
1182         bte_res = xp_bte_copy(part_nasid_pa, (u64)nasid_mask,
1183                               xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE),
1184                               NULL);
1185
1186         return xpc_map_bte_errors(bte_res);
1187 }