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