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 <linux/slab.h>
21 #include "xpc.h"
22 #include <asm/uv/uv_hub.h>
23 
24 /* XPC is exiting flag */
25 int xpc_exiting;
26 
27 /* this partition's reserved page pointers */
28 struct xpc_rsvd_page *xpc_rsvd_page;
29 static unsigned long *xpc_part_nasids;
30 unsigned long *xpc_mach_nasids;
31 
32 static int xpc_nasid_mask_nbytes;	/* #of bytes in nasid mask */
33 int xpc_nasid_mask_nlongs;	/* #of longs in nasid mask */
34 
35 struct xpc_partition *xpc_partitions;
36 
37 /*
38  * Guarantee that the kmalloc'd memory is cacheline aligned.
39  */
40 void *
41 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
42 {
43 	/* see if kmalloc will give us cachline aligned memory by default */
44 	*base = kmalloc(size, flags);
45 	if (*base == NULL)
46 		return NULL;
47 
48 	if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
49 		return *base;
50 
51 	kfree(*base);
52 
53 	/* nope, we'll have to do it ourselves */
54 	*base = kmalloc(size + L1_CACHE_BYTES, flags);
55 	if (*base == NULL)
56 		return NULL;
57 
58 	return (void *)L1_CACHE_ALIGN((u64)*base);
59 }
60 
61 /*
62  * Given a nasid, get the physical address of the  partition's reserved page
63  * for that nasid. This function returns 0 on any error.
64  */
65 static unsigned long
66 xpc_get_rsvd_page_pa(int nasid)
67 {
68 	enum xp_retval ret;
69 	u64 cookie = 0;
70 	unsigned long rp_pa = nasid;	/* seed with nasid */
71 	size_t len = 0;
72 	size_t buf_len = 0;
73 	void *buf = NULL;
74 	void *buf_base = NULL;
75 	enum xp_retval (*get_partition_rsvd_page_pa)
76 		(void *, u64 *, unsigned long *, size_t *) =
77 		xpc_arch_ops.get_partition_rsvd_page_pa;
78 
79 	while (1) {
80 
81 		/* !!! rp_pa will need to be _gpa on UV.
82 		 * ??? So do we save it into the architecture specific parts
83 		 * ??? of the xpc_partition structure? Do we rename this
84 		 * ??? function or have two versions? Rename rp_pa for UV to
85 		 * ??? rp_gpa?
86 		 */
87 		ret = get_partition_rsvd_page_pa(buf, &cookie, &rp_pa, &len);
88 
89 		dev_dbg(xpc_part, "SAL returned with ret=%d, cookie=0x%016lx, "
90 			"address=0x%016lx, len=0x%016lx\n", ret,
91 			(unsigned long)cookie, rp_pa, len);
92 
93 		if (ret != xpNeedMoreInfo)
94 			break;
95 
96 		/* !!! L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
97 		if (is_shub())
98 			len = L1_CACHE_ALIGN(len);
99 
100 		if (len > buf_len) {
101 			kfree(buf_base);
102 			buf_len = L1_CACHE_ALIGN(len);
103 			buf = xpc_kmalloc_cacheline_aligned(buf_len, GFP_KERNEL,
104 							    &buf_base);
105 			if (buf_base == NULL) {
106 				dev_err(xpc_part, "unable to kmalloc "
107 					"len=0x%016lx\n", buf_len);
108 				ret = xpNoMemory;
109 				break;
110 			}
111 		}
112 
113 		ret = xp_remote_memcpy(xp_pa(buf), rp_pa, len);
114 		if (ret != xpSuccess) {
115 			dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
116 			break;
117 		}
118 	}
119 
120 	kfree(buf_base);
121 
122 	if (ret != xpSuccess)
123 		rp_pa = 0;
124 
125 	dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
126 	return rp_pa;
127 }
128 
129 /*
130  * Fill the partition reserved page with the information needed by
131  * other partitions to discover we are alive and establish initial
132  * communications.
133  */
134 int
135 xpc_setup_rsvd_page(void)
136 {
137 	int ret;
138 	struct xpc_rsvd_page *rp;
139 	unsigned long rp_pa;
140 	unsigned long new_ts_jiffies;
141 
142 	/* get the local reserved page's address */
143 
144 	preempt_disable();
145 	rp_pa = xpc_get_rsvd_page_pa(xp_cpu_to_nasid(smp_processor_id()));
146 	preempt_enable();
147 	if (rp_pa == 0) {
148 		dev_err(xpc_part, "SAL failed to locate the reserved page\n");
149 		return -ESRCH;
150 	}
151 	rp = (struct xpc_rsvd_page *)__va(xp_socket_pa(rp_pa));
152 
153 	if (rp->SAL_version < 3) {
154 		/* SAL_versions < 3 had a SAL_partid defined as a u8 */
155 		rp->SAL_partid &= 0xff;
156 	}
157 	BUG_ON(rp->SAL_partid != xp_partition_id);
158 
159 	if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
160 		dev_err(xpc_part, "the reserved page's partid of %d is outside "
161 			"supported range (< 0 || >= %d)\n", rp->SAL_partid,
162 			xp_max_npartitions);
163 		return -EINVAL;
164 	}
165 
166 	rp->version = XPC_RP_VERSION;
167 	rp->max_npartitions = xp_max_npartitions;
168 
169 	/* establish the actual sizes of the nasid masks */
170 	if (rp->SAL_version == 1) {
171 		/* SAL_version 1 didn't set the nasids_size field */
172 		rp->SAL_nasids_size = 128;
173 	}
174 	xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
175 	xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
176 					      BITS_PER_BYTE);
177 
178 	/* setup the pointers to the various items in the reserved page */
179 	xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
180 	xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
181 
182 	ret = xpc_arch_ops.setup_rsvd_page(rp);
183 	if (ret != 0)
184 		return ret;
185 
186 	/*
187 	 * Set timestamp of when reserved page was setup by XPC.
188 	 * This signifies to the remote partition that our reserved
189 	 * page is initialized.
190 	 */
191 	new_ts_jiffies = jiffies;
192 	if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
193 		new_ts_jiffies++;
194 	rp->ts_jiffies = new_ts_jiffies;
195 
196 	xpc_rsvd_page = rp;
197 	return 0;
198 }
199 
200 void
201 xpc_teardown_rsvd_page(void)
202 {
203 	/* a zero timestamp indicates our rsvd page is not initialized */
204 	xpc_rsvd_page->ts_jiffies = 0;
205 }
206 
207 /*
208  * Get a copy of a portion of the remote partition's rsvd page.
209  *
210  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
211  * is large enough to contain a copy of their reserved page header and
212  * part_nasids mask.
213  */
214 enum xp_retval
215 xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
216 		  struct xpc_rsvd_page *remote_rp, unsigned long *remote_rp_pa)
217 {
218 	int l;
219 	enum xp_retval ret;
220 
221 	/* get the reserved page's physical address */
222 
223 	*remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
224 	if (*remote_rp_pa == 0)
225 		return xpNoRsvdPageAddr;
226 
227 	/* pull over the reserved page header and part_nasids mask */
228 	ret = xp_remote_memcpy(xp_pa(remote_rp), *remote_rp_pa,
229 			       XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
230 	if (ret != xpSuccess)
231 		return ret;
232 
233 	if (discovered_nasids != NULL) {
234 		unsigned long *remote_part_nasids =
235 		    XPC_RP_PART_NASIDS(remote_rp);
236 
237 		for (l = 0; l < xpc_nasid_mask_nlongs; l++)
238 			discovered_nasids[l] |= remote_part_nasids[l];
239 	}
240 
241 	/* zero timestamp indicates the reserved page has not been setup */
242 	if (remote_rp->ts_jiffies == 0)
243 		return xpRsvdPageNotSet;
244 
245 	if (XPC_VERSION_MAJOR(remote_rp->version) !=
246 	    XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
247 		return xpBadVersion;
248 	}
249 
250 	/* check that both remote and local partids are valid for each side */
251 	if (remote_rp->SAL_partid < 0 ||
252 	    remote_rp->SAL_partid >= xp_max_npartitions ||
253 	    remote_rp->max_npartitions <= xp_partition_id) {
254 		return xpInvalidPartid;
255 	}
256 
257 	if (remote_rp->SAL_partid == xp_partition_id)
258 		return xpLocalPartid;
259 
260 	return xpSuccess;
261 }
262 
263 /*
264  * See if the other side has responded to a partition deactivate request
265  * from us. Though we requested the remote partition to deactivate with regard
266  * to us, we really only need to wait for the other side to disengage from us.
267  */
268 int
269 xpc_partition_disengaged(struct xpc_partition *part)
270 {
271 	short partid = XPC_PARTID(part);
272 	int disengaged;
273 
274 	disengaged = !xpc_arch_ops.partition_engaged(partid);
275 	if (part->disengage_timeout) {
276 		if (!disengaged) {
277 			if (time_is_after_jiffies(part->disengage_timeout)) {
278 				/* timelimit hasn't been reached yet */
279 				return 0;
280 			}
281 
282 			/*
283 			 * Other side hasn't responded to our deactivate
284 			 * request in a timely fashion, so assume it's dead.
285 			 */
286 
287 			dev_info(xpc_part, "deactivate request to remote "
288 				 "partition %d timed out\n", partid);
289 			xpc_disengage_timedout = 1;
290 			xpc_arch_ops.assume_partition_disengaged(partid);
291 			disengaged = 1;
292 		}
293 		part->disengage_timeout = 0;
294 
295 		/* cancel the timer function, provided it's not us */
296 		if (!in_interrupt())
297 			del_singleshot_timer_sync(&part->disengage_timer);
298 
299 		DBUG_ON(part->act_state != XPC_P_AS_DEACTIVATING &&
300 			part->act_state != XPC_P_AS_INACTIVE);
301 		if (part->act_state != XPC_P_AS_INACTIVE)
302 			xpc_wakeup_channel_mgr(part);
303 
304 		xpc_arch_ops.cancel_partition_deactivation_request(part);
305 	}
306 	return disengaged;
307 }
308 
309 /*
310  * Mark specified partition as active.
311  */
312 enum xp_retval
313 xpc_mark_partition_active(struct xpc_partition *part)
314 {
315 	unsigned long irq_flags;
316 	enum xp_retval ret;
317 
318 	dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
319 
320 	spin_lock_irqsave(&part->act_lock, irq_flags);
321 	if (part->act_state == XPC_P_AS_ACTIVATING) {
322 		part->act_state = XPC_P_AS_ACTIVE;
323 		ret = xpSuccess;
324 	} else {
325 		DBUG_ON(part->reason == xpSuccess);
326 		ret = part->reason;
327 	}
328 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
329 
330 	return ret;
331 }
332 
333 /*
334  * Start the process of deactivating the specified partition.
335  */
336 void
337 xpc_deactivate_partition(const int line, struct xpc_partition *part,
338 			 enum xp_retval reason)
339 {
340 	unsigned long irq_flags;
341 
342 	spin_lock_irqsave(&part->act_lock, irq_flags);
343 
344 	if (part->act_state == XPC_P_AS_INACTIVE) {
345 		XPC_SET_REASON(part, reason, line);
346 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
347 		if (reason == xpReactivating) {
348 			/* we interrupt ourselves to reactivate partition */
349 			xpc_arch_ops.request_partition_reactivation(part);
350 		}
351 		return;
352 	}
353 	if (part->act_state == XPC_P_AS_DEACTIVATING) {
354 		if ((part->reason == xpUnloading && reason != xpUnloading) ||
355 		    reason == xpReactivating) {
356 			XPC_SET_REASON(part, reason, line);
357 		}
358 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
359 		return;
360 	}
361 
362 	part->act_state = XPC_P_AS_DEACTIVATING;
363 	XPC_SET_REASON(part, reason, line);
364 
365 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
366 
367 	/* ask remote partition to deactivate with regard to us */
368 	xpc_arch_ops.request_partition_deactivation(part);
369 
370 	/* set a timelimit on the disengage phase of the deactivation request */
371 	part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
372 	part->disengage_timer.expires = part->disengage_timeout;
373 	add_timer(&part->disengage_timer);
374 
375 	dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
376 		XPC_PARTID(part), reason);
377 
378 	xpc_partition_going_down(part, reason);
379 }
380 
381 /*
382  * Mark specified partition as inactive.
383  */
384 void
385 xpc_mark_partition_inactive(struct xpc_partition *part)
386 {
387 	unsigned long irq_flags;
388 
389 	dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
390 		XPC_PARTID(part));
391 
392 	spin_lock_irqsave(&part->act_lock, irq_flags);
393 	part->act_state = XPC_P_AS_INACTIVE;
394 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
395 	part->remote_rp_pa = 0;
396 }
397 
398 /*
399  * SAL has provided a partition and machine mask.  The partition mask
400  * contains a bit for each even nasid in our partition.  The machine
401  * mask contains a bit for each even nasid in the entire machine.
402  *
403  * Using those two bit arrays, we can determine which nasids are
404  * known in the machine.  Each should also have a reserved page
405  * initialized if they are available for partitioning.
406  */
407 void
408 xpc_discovery(void)
409 {
410 	void *remote_rp_base;
411 	struct xpc_rsvd_page *remote_rp;
412 	unsigned long remote_rp_pa;
413 	int region;
414 	int region_size;
415 	int max_regions;
416 	int nasid;
417 	unsigned long *discovered_nasids;
418 	enum xp_retval ret;
419 
420 	remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
421 						  xpc_nasid_mask_nbytes,
422 						  GFP_KERNEL, &remote_rp_base);
423 	if (remote_rp == NULL)
424 		return;
425 
426 	discovered_nasids = kcalloc(xpc_nasid_mask_nlongs, sizeof(long),
427 				    GFP_KERNEL);
428 	if (discovered_nasids == NULL) {
429 		kfree(remote_rp_base);
430 		return;
431 	}
432 
433 	/*
434 	 * The term 'region' in this context refers to the minimum number of
435 	 * nodes that can comprise an access protection grouping. The access
436 	 * protection is in regards to memory, IOI and IPI.
437 	 */
438 	region_size = xp_region_size;
439 
440 	if (is_uv())
441 		max_regions = 256;
442 	else {
443 		max_regions = 64;
444 
445 		switch (region_size) {
446 		case 128:
447 			max_regions *= 2;
448 			/* fall through */
449 		case 64:
450 			max_regions *= 2;
451 			/* fall through */
452 		case 32:
453 			max_regions *= 2;
454 			region_size = 16;
455 			DBUG_ON(!is_shub2());
456 		}
457 	}
458 
459 	for (region = 0; region < max_regions; region++) {
460 
461 		if (xpc_exiting)
462 			break;
463 
464 		dev_dbg(xpc_part, "searching region %d\n", region);
465 
466 		for (nasid = (region * region_size * 2);
467 		     nasid < ((region + 1) * region_size * 2); nasid += 2) {
468 
469 			if (xpc_exiting)
470 				break;
471 
472 			dev_dbg(xpc_part, "checking nasid %d\n", nasid);
473 
474 			if (test_bit(nasid / 2, xpc_part_nasids)) {
475 				dev_dbg(xpc_part, "PROM indicates Nasid %d is "
476 					"part of the local partition; skipping "
477 					"region\n", nasid);
478 				break;
479 			}
480 
481 			if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
482 				dev_dbg(xpc_part, "PROM indicates Nasid %d was "
483 					"not on Numa-Link network at reset\n",
484 					nasid);
485 				continue;
486 			}
487 
488 			if (test_bit(nasid / 2, discovered_nasids)) {
489 				dev_dbg(xpc_part, "Nasid %d is part of a "
490 					"partition which was previously "
491 					"discovered\n", nasid);
492 				continue;
493 			}
494 
495 			/* pull over the rsvd page header & part_nasids mask */
496 
497 			ret = xpc_get_remote_rp(nasid, discovered_nasids,
498 						remote_rp, &remote_rp_pa);
499 			if (ret != xpSuccess) {
500 				dev_dbg(xpc_part, "unable to get reserved page "
501 					"from nasid %d, reason=%d\n", nasid,
502 					ret);
503 
504 				if (ret == xpLocalPartid)
505 					break;
506 
507 				continue;
508 			}
509 
510 			xpc_arch_ops.request_partition_activation(remote_rp,
511 							 remote_rp_pa, nasid);
512 		}
513 	}
514 
515 	kfree(discovered_nasids);
516 	kfree(remote_rp_base);
517 }
518 
519 /*
520  * Given a partid, get the nasids owned by that partition from the
521  * remote partition's reserved page.
522  */
523 enum xp_retval
524 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
525 {
526 	struct xpc_partition *part;
527 	unsigned long part_nasid_pa;
528 
529 	part = &xpc_partitions[partid];
530 	if (part->remote_rp_pa == 0)
531 		return xpPartitionDown;
532 
533 	memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
534 
535 	part_nasid_pa = (unsigned long)XPC_RP_PART_NASIDS(part->remote_rp_pa);
536 
537 	return xp_remote_memcpy(xp_pa(nasid_mask), part_nasid_pa,
538 				xpc_nasid_mask_nbytes);
539 }
540