xref: /linux/drivers/misc/sgi-xp/xpc_partition.c (revision bcefe12eff5dca6fdfa94ed85e5bee66380d5cd9)
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/device.h>
19 #include <linux/hardirq.h>
20 #include "xpc.h"
21 
22 /* XPC is exiting flag */
23 int xpc_exiting;
24 
25 /* this partition's reserved page pointers */
26 struct xpc_rsvd_page *xpc_rsvd_page;
27 static unsigned long *xpc_part_nasids;
28 unsigned long *xpc_mach_nasids;
29 
30 static int xpc_nasid_mask_nbytes;	/* #of bytes in nasid mask */
31 int xpc_nasid_mask_nlongs;	/* #of longs in nasid mask */
32 
33 struct xpc_partition *xpc_partitions;
34 
35 /*
36  * Guarantee that the kmalloc'd memory is cacheline aligned.
37  */
38 void *
39 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
40 {
41 	/* see if kmalloc will give us cachline aligned memory by default */
42 	*base = kmalloc(size, flags);
43 	if (*base == NULL)
44 		return NULL;
45 
46 	if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
47 		return *base;
48 
49 	kfree(*base);
50 
51 	/* nope, we'll have to do it ourselves */
52 	*base = kmalloc(size + L1_CACHE_BYTES, flags);
53 	if (*base == NULL)
54 		return NULL;
55 
56 	return (void *)L1_CACHE_ALIGN((u64)*base);
57 }
58 
59 /*
60  * Given a nasid, get the physical address of the  partition's reserved page
61  * for that nasid. This function returns 0 on any error.
62  */
63 static unsigned long
64 xpc_get_rsvd_page_pa(int nasid)
65 {
66 	enum xp_retval ret;
67 	u64 cookie = 0;
68 	unsigned long rp_pa = nasid;	/* seed with nasid */
69 	size_t len = 0;
70 	size_t buf_len = 0;
71 	void *buf = buf;
72 	void *buf_base = NULL;
73 	enum xp_retval (*get_partition_rsvd_page_pa)
74 		(void *, u64 *, unsigned long *, size_t *) =
75 		xpc_arch_ops.get_partition_rsvd_page_pa;
76 
77 	while (1) {
78 
79 		/* !!! rp_pa will need to be _gpa on UV.
80 		 * ??? So do we save it into the architecture specific parts
81 		 * ??? of the xpc_partition structure? Do we rename this
82 		 * ??? function or have two versions? Rename rp_pa for UV to
83 		 * ??? rp_gpa?
84 		 */
85 		ret = get_partition_rsvd_page_pa(buf, &cookie, &rp_pa, &len);
86 
87 		dev_dbg(xpc_part, "SAL returned with ret=%d, cookie=0x%016lx, "
88 			"address=0x%016lx, len=0x%016lx\n", ret,
89 			(unsigned long)cookie, rp_pa, len);
90 
91 		if (ret != xpNeedMoreInfo)
92 			break;
93 
94 		/* !!! L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
95 		if (L1_CACHE_ALIGN(len) > buf_len) {
96 			kfree(buf_base);
97 			buf_len = L1_CACHE_ALIGN(len);
98 			buf = xpc_kmalloc_cacheline_aligned(buf_len, GFP_KERNEL,
99 							    &buf_base);
100 			if (buf_base == NULL) {
101 				dev_err(xpc_part, "unable to kmalloc "
102 					"len=0x%016lx\n", buf_len);
103 				ret = xpNoMemory;
104 				break;
105 			}
106 		}
107 
108 		ret = xp_remote_memcpy(xp_pa(buf), rp_pa, buf_len);
109 		if (ret != xpSuccess) {
110 			dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
111 			break;
112 		}
113 	}
114 
115 	kfree(buf_base);
116 
117 	if (ret != xpSuccess)
118 		rp_pa = 0;
119 
120 	dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
121 	return rp_pa;
122 }
123 
124 /*
125  * Fill the partition reserved page with the information needed by
126  * other partitions to discover we are alive and establish initial
127  * communications.
128  */
129 int
130 xpc_setup_rsvd_page(void)
131 {
132 	int ret;
133 	struct xpc_rsvd_page *rp;
134 	unsigned long rp_pa;
135 	unsigned long new_ts_jiffies;
136 
137 	/* get the local reserved page's address */
138 
139 	preempt_disable();
140 	rp_pa = xpc_get_rsvd_page_pa(xp_cpu_to_nasid(smp_processor_id()));
141 	preempt_enable();
142 	if (rp_pa == 0) {
143 		dev_err(xpc_part, "SAL failed to locate the reserved page\n");
144 		return -ESRCH;
145 	}
146 	rp = (struct xpc_rsvd_page *)__va(rp_pa);
147 
148 	if (rp->SAL_version < 3) {
149 		/* SAL_versions < 3 had a SAL_partid defined as a u8 */
150 		rp->SAL_partid &= 0xff;
151 	}
152 	BUG_ON(rp->SAL_partid != xp_partition_id);
153 
154 	if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
155 		dev_err(xpc_part, "the reserved page's partid of %d is outside "
156 			"supported range (< 0 || >= %d)\n", rp->SAL_partid,
157 			xp_max_npartitions);
158 		return -EINVAL;
159 	}
160 
161 	rp->version = XPC_RP_VERSION;
162 	rp->max_npartitions = xp_max_npartitions;
163 
164 	/* establish the actual sizes of the nasid masks */
165 	if (rp->SAL_version == 1) {
166 		/* SAL_version 1 didn't set the nasids_size field */
167 		rp->SAL_nasids_size = 128;
168 	}
169 	xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
170 	xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
171 					      BITS_PER_BYTE);
172 
173 	/* setup the pointers to the various items in the reserved page */
174 	xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
175 	xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
176 
177 	ret = xpc_arch_ops.setup_rsvd_page(rp);
178 	if (ret != 0)
179 		return ret;
180 
181 	/*
182 	 * Set timestamp of when reserved page was setup by XPC.
183 	 * This signifies to the remote partition that our reserved
184 	 * page is initialized.
185 	 */
186 	new_ts_jiffies = jiffies;
187 	if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
188 		new_ts_jiffies++;
189 	rp->ts_jiffies = new_ts_jiffies;
190 
191 	xpc_rsvd_page = rp;
192 	return 0;
193 }
194 
195 void
196 xpc_teardown_rsvd_page(void)
197 {
198 	/* a zero timestamp indicates our rsvd page is not initialized */
199 	xpc_rsvd_page->ts_jiffies = 0;
200 }
201 
202 /*
203  * Get a copy of a portion of the remote partition's rsvd page.
204  *
205  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
206  * is large enough to contain a copy of their reserved page header and
207  * part_nasids mask.
208  */
209 enum xp_retval
210 xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
211 		  struct xpc_rsvd_page *remote_rp, unsigned long *remote_rp_pa)
212 {
213 	int l;
214 	enum xp_retval ret;
215 
216 	/* get the reserved page's physical address */
217 
218 	*remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
219 	if (*remote_rp_pa == 0)
220 		return xpNoRsvdPageAddr;
221 
222 	/* pull over the reserved page header and part_nasids mask */
223 	ret = xp_remote_memcpy(xp_pa(remote_rp), *remote_rp_pa,
224 			       XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
225 	if (ret != xpSuccess)
226 		return ret;
227 
228 	if (discovered_nasids != NULL) {
229 		unsigned long *remote_part_nasids =
230 		    XPC_RP_PART_NASIDS(remote_rp);
231 
232 		for (l = 0; l < xpc_nasid_mask_nlongs; l++)
233 			discovered_nasids[l] |= remote_part_nasids[l];
234 	}
235 
236 	/* zero timestamp indicates the reserved page has not been setup */
237 	if (remote_rp->ts_jiffies == 0)
238 		return xpRsvdPageNotSet;
239 
240 	if (XPC_VERSION_MAJOR(remote_rp->version) !=
241 	    XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
242 		return xpBadVersion;
243 	}
244 
245 	/* check that both remote and local partids are valid for each side */
246 	if (remote_rp->SAL_partid < 0 ||
247 	    remote_rp->SAL_partid >= xp_max_npartitions ||
248 	    remote_rp->max_npartitions <= xp_partition_id) {
249 		return xpInvalidPartid;
250 	}
251 
252 	if (remote_rp->SAL_partid == xp_partition_id)
253 		return xpLocalPartid;
254 
255 	return xpSuccess;
256 }
257 
258 /*
259  * See if the other side has responded to a partition deactivate request
260  * from us. Though we requested the remote partition to deactivate with regard
261  * to us, we really only need to wait for the other side to disengage from us.
262  */
263 int
264 xpc_partition_disengaged(struct xpc_partition *part)
265 {
266 	short partid = XPC_PARTID(part);
267 	int disengaged;
268 
269 	disengaged = !xpc_arch_ops.partition_engaged(partid);
270 	if (part->disengage_timeout) {
271 		if (!disengaged) {
272 			if (time_is_after_jiffies(part->disengage_timeout)) {
273 				/* timelimit hasn't been reached yet */
274 				return 0;
275 			}
276 
277 			/*
278 			 * Other side hasn't responded to our deactivate
279 			 * request in a timely fashion, so assume it's dead.
280 			 */
281 
282 			dev_info(xpc_part, "deactivate request to remote "
283 				 "partition %d timed out\n", partid);
284 			xpc_disengage_timedout = 1;
285 			xpc_arch_ops.assume_partition_disengaged(partid);
286 			disengaged = 1;
287 		}
288 		part->disengage_timeout = 0;
289 
290 		/* cancel the timer function, provided it's not us */
291 		if (!in_interrupt())
292 			del_singleshot_timer_sync(&part->disengage_timer);
293 
294 		DBUG_ON(part->act_state != XPC_P_AS_DEACTIVATING &&
295 			part->act_state != XPC_P_AS_INACTIVE);
296 		if (part->act_state != XPC_P_AS_INACTIVE)
297 			xpc_wakeup_channel_mgr(part);
298 
299 		xpc_arch_ops.cancel_partition_deactivation_request(part);
300 	}
301 	return disengaged;
302 }
303 
304 /*
305  * Mark specified partition as active.
306  */
307 enum xp_retval
308 xpc_mark_partition_active(struct xpc_partition *part)
309 {
310 	unsigned long irq_flags;
311 	enum xp_retval ret;
312 
313 	dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
314 
315 	spin_lock_irqsave(&part->act_lock, irq_flags);
316 	if (part->act_state == XPC_P_AS_ACTIVATING) {
317 		part->act_state = XPC_P_AS_ACTIVE;
318 		ret = xpSuccess;
319 	} else {
320 		DBUG_ON(part->reason == xpSuccess);
321 		ret = part->reason;
322 	}
323 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
324 
325 	return ret;
326 }
327 
328 /*
329  * Start the process of deactivating the specified partition.
330  */
331 void
332 xpc_deactivate_partition(const int line, struct xpc_partition *part,
333 			 enum xp_retval reason)
334 {
335 	unsigned long irq_flags;
336 
337 	spin_lock_irqsave(&part->act_lock, irq_flags);
338 
339 	if (part->act_state == XPC_P_AS_INACTIVE) {
340 		XPC_SET_REASON(part, reason, line);
341 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
342 		if (reason == xpReactivating) {
343 			/* we interrupt ourselves to reactivate partition */
344 			xpc_arch_ops.request_partition_reactivation(part);
345 		}
346 		return;
347 	}
348 	if (part->act_state == XPC_P_AS_DEACTIVATING) {
349 		if ((part->reason == xpUnloading && reason != xpUnloading) ||
350 		    reason == xpReactivating) {
351 			XPC_SET_REASON(part, reason, line);
352 		}
353 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
354 		return;
355 	}
356 
357 	part->act_state = XPC_P_AS_DEACTIVATING;
358 	XPC_SET_REASON(part, reason, line);
359 
360 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
361 
362 	/* ask remote partition to deactivate with regard to us */
363 	xpc_arch_ops.request_partition_deactivation(part);
364 
365 	/* set a timelimit on the disengage phase of the deactivation request */
366 	part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
367 	part->disengage_timer.expires = part->disengage_timeout;
368 	add_timer(&part->disengage_timer);
369 
370 	dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
371 		XPC_PARTID(part), reason);
372 
373 	xpc_partition_going_down(part, reason);
374 }
375 
376 /*
377  * Mark specified partition as inactive.
378  */
379 void
380 xpc_mark_partition_inactive(struct xpc_partition *part)
381 {
382 	unsigned long irq_flags;
383 
384 	dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
385 		XPC_PARTID(part));
386 
387 	spin_lock_irqsave(&part->act_lock, irq_flags);
388 	part->act_state = XPC_P_AS_INACTIVE;
389 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
390 	part->remote_rp_pa = 0;
391 }
392 
393 /*
394  * SAL has provided a partition and machine mask.  The partition mask
395  * contains a bit for each even nasid in our partition.  The machine
396  * mask contains a bit for each even nasid in the entire machine.
397  *
398  * Using those two bit arrays, we can determine which nasids are
399  * known in the machine.  Each should also have a reserved page
400  * initialized if they are available for partitioning.
401  */
402 void
403 xpc_discovery(void)
404 {
405 	void *remote_rp_base;
406 	struct xpc_rsvd_page *remote_rp;
407 	unsigned long remote_rp_pa;
408 	int region;
409 	int region_size;
410 	int max_regions;
411 	int nasid;
412 	struct xpc_rsvd_page *rp;
413 	unsigned long *discovered_nasids;
414 	enum xp_retval ret;
415 
416 	remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
417 						  xpc_nasid_mask_nbytes,
418 						  GFP_KERNEL, &remote_rp_base);
419 	if (remote_rp == NULL)
420 		return;
421 
422 	discovered_nasids = kzalloc(sizeof(long) * xpc_nasid_mask_nlongs,
423 				    GFP_KERNEL);
424 	if (discovered_nasids == NULL) {
425 		kfree(remote_rp_base);
426 		return;
427 	}
428 
429 	rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
430 
431 	/*
432 	 * The term 'region' in this context refers to the minimum number of
433 	 * nodes that can comprise an access protection grouping. The access
434 	 * protection is in regards to memory, IOI and IPI.
435 	 */
436 	max_regions = 64;
437 	region_size = xp_region_size;
438 
439 	switch (region_size) {
440 	case 128:
441 		max_regions *= 2;
442 	case 64:
443 		max_regions *= 2;
444 	case 32:
445 		max_regions *= 2;
446 		region_size = 16;
447 		DBUG_ON(!is_shub2());
448 	}
449 
450 	for (region = 0; region < max_regions; region++) {
451 
452 		if (xpc_exiting)
453 			break;
454 
455 		dev_dbg(xpc_part, "searching region %d\n", region);
456 
457 		for (nasid = (region * region_size * 2);
458 		     nasid < ((region + 1) * region_size * 2); nasid += 2) {
459 
460 			if (xpc_exiting)
461 				break;
462 
463 			dev_dbg(xpc_part, "checking nasid %d\n", nasid);
464 
465 			if (test_bit(nasid / 2, xpc_part_nasids)) {
466 				dev_dbg(xpc_part, "PROM indicates Nasid %d is "
467 					"part of the local partition; skipping "
468 					"region\n", nasid);
469 				break;
470 			}
471 
472 			if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
473 				dev_dbg(xpc_part, "PROM indicates Nasid %d was "
474 					"not on Numa-Link network at reset\n",
475 					nasid);
476 				continue;
477 			}
478 
479 			if (test_bit(nasid / 2, discovered_nasids)) {
480 				dev_dbg(xpc_part, "Nasid %d is part of a "
481 					"partition which was previously "
482 					"discovered\n", nasid);
483 				continue;
484 			}
485 
486 			/* pull over the rsvd page header & part_nasids mask */
487 
488 			ret = xpc_get_remote_rp(nasid, discovered_nasids,
489 						remote_rp, &remote_rp_pa);
490 			if (ret != xpSuccess) {
491 				dev_dbg(xpc_part, "unable to get reserved page "
492 					"from nasid %d, reason=%d\n", nasid,
493 					ret);
494 
495 				if (ret == xpLocalPartid)
496 					break;
497 
498 				continue;
499 			}
500 
501 			xpc_arch_ops.request_partition_activation(remote_rp,
502 							 remote_rp_pa, nasid);
503 		}
504 	}
505 
506 	kfree(discovered_nasids);
507 	kfree(remote_rp_base);
508 }
509 
510 /*
511  * Given a partid, get the nasids owned by that partition from the
512  * remote partition's reserved page.
513  */
514 enum xp_retval
515 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
516 {
517 	struct xpc_partition *part;
518 	unsigned long part_nasid_pa;
519 
520 	part = &xpc_partitions[partid];
521 	if (part->remote_rp_pa == 0)
522 		return xpPartitionDown;
523 
524 	memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
525 
526 	part_nasid_pa = (unsigned long)XPC_RP_PART_NASIDS(part->remote_rp_pa);
527 
528 	return xp_remote_memcpy(xp_pa(nasid_mask), part_nasid_pa,
529 				xpc_nasid_mask_nbytes);
530 }
531