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