xref: /openbmc/linux/drivers/misc/sgi-xp/xpc_partition.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
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 
74 	while (1) {
75 
76 		/* !!! rp_pa will need to be _gpa on UV.
77 		 * ??? So do we save it into the architecture specific parts
78 		 * ??? of the xpc_partition structure? Do we rename this
79 		 * ??? function or have two versions? Rename rp_pa for UV to
80 		 * ??? rp_gpa?
81 		 */
82 		ret = xpc_get_partition_rsvd_page_pa(buf, &cookie, &rp_pa,
83 						     &len);
84 
85 		dev_dbg(xpc_part, "SAL returned with ret=%d, cookie=0x%016lx, "
86 			"address=0x%016lx, len=0x%016lx\n", ret,
87 			(unsigned long)cookie, rp_pa, len);
88 
89 		if (ret != xpNeedMoreInfo)
90 			break;
91 
92 		/* !!! L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
93 		if (L1_CACHE_ALIGN(len) > buf_len) {
94 			kfree(buf_base);
95 			buf_len = L1_CACHE_ALIGN(len);
96 			buf = xpc_kmalloc_cacheline_aligned(buf_len, GFP_KERNEL,
97 							    &buf_base);
98 			if (buf_base == NULL) {
99 				dev_err(xpc_part, "unable to kmalloc "
100 					"len=0x%016lx\n", buf_len);
101 				ret = xpNoMemory;
102 				break;
103 			}
104 		}
105 
106 		ret = xp_remote_memcpy(xp_pa(buf), rp_pa, buf_len);
107 		if (ret != xpSuccess) {
108 			dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
109 			break;
110 		}
111 	}
112 
113 	kfree(buf_base);
114 
115 	if (ret != xpSuccess)
116 		rp_pa = 0;
117 
118 	dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
119 	return rp_pa;
120 }
121 
122 /*
123  * Fill the partition reserved page with the information needed by
124  * other partitions to discover we are alive and establish initial
125  * communications.
126  */
127 int
128 xpc_setup_rsvd_page(void)
129 {
130 	int ret;
131 	struct xpc_rsvd_page *rp;
132 	unsigned long rp_pa;
133 	unsigned long new_ts_jiffies;
134 
135 	/* get the local reserved page's address */
136 
137 	preempt_disable();
138 	rp_pa = xpc_get_rsvd_page_pa(xp_cpu_to_nasid(smp_processor_id()));
139 	preempt_enable();
140 	if (rp_pa == 0) {
141 		dev_err(xpc_part, "SAL failed to locate the reserved page\n");
142 		return -ESRCH;
143 	}
144 	rp = (struct xpc_rsvd_page *)__va(rp_pa);
145 
146 	if (rp->SAL_version < 3) {
147 		/* SAL_versions < 3 had a SAL_partid defined as a u8 */
148 		rp->SAL_partid &= 0xff;
149 	}
150 	BUG_ON(rp->SAL_partid != xp_partition_id);
151 
152 	if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
153 		dev_err(xpc_part, "the reserved page's partid of %d is outside "
154 			"supported range (< 0 || >= %d)\n", rp->SAL_partid,
155 			xp_max_npartitions);
156 		return -EINVAL;
157 	}
158 
159 	rp->version = XPC_RP_VERSION;
160 	rp->max_npartitions = xp_max_npartitions;
161 
162 	/* establish the actual sizes of the nasid masks */
163 	if (rp->SAL_version == 1) {
164 		/* SAL_version 1 didn't set the nasids_size field */
165 		rp->SAL_nasids_size = 128;
166 	}
167 	xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
168 	xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
169 					      BITS_PER_BYTE);
170 
171 	/* setup the pointers to the various items in the reserved page */
172 	xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
173 	xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
174 
175 	ret = xpc_setup_rsvd_page_sn(rp);
176 	if (ret != 0)
177 		return ret;
178 
179 	/*
180 	 * Set timestamp of when reserved page was setup by XPC.
181 	 * This signifies to the remote partition that our reserved
182 	 * page is initialized.
183 	 */
184 	new_ts_jiffies = jiffies;
185 	if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
186 		new_ts_jiffies++;
187 	rp->ts_jiffies = new_ts_jiffies;
188 
189 	xpc_rsvd_page = rp;
190 	return 0;
191 }
192 
193 void
194 xpc_teardown_rsvd_page(void)
195 {
196 	/* a zero timestamp indicates our rsvd page is not initialized */
197 	xpc_rsvd_page->ts_jiffies = 0;
198 }
199 
200 /*
201  * Get a copy of a portion of the remote partition's rsvd page.
202  *
203  * remote_rp points to a buffer that is cacheline aligned for BTE copies and
204  * is large enough to contain a copy of their reserved page header and
205  * part_nasids mask.
206  */
207 enum xp_retval
208 xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
209 		  struct xpc_rsvd_page *remote_rp, unsigned long *remote_rp_pa)
210 {
211 	int l;
212 	enum xp_retval ret;
213 
214 	/* get the reserved page's physical address */
215 
216 	*remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
217 	if (*remote_rp_pa == 0)
218 		return xpNoRsvdPageAddr;
219 
220 	/* pull over the reserved page header and part_nasids mask */
221 	ret = xp_remote_memcpy(xp_pa(remote_rp), *remote_rp_pa,
222 			       XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
223 	if (ret != xpSuccess)
224 		return ret;
225 
226 	if (discovered_nasids != NULL) {
227 		unsigned long *remote_part_nasids =
228 		    XPC_RP_PART_NASIDS(remote_rp);
229 
230 		for (l = 0; l < xpc_nasid_mask_nlongs; l++)
231 			discovered_nasids[l] |= remote_part_nasids[l];
232 	}
233 
234 	/* zero timestamp indicates the reserved page has not been setup */
235 	if (remote_rp->ts_jiffies == 0)
236 		return xpRsvdPageNotSet;
237 
238 	if (XPC_VERSION_MAJOR(remote_rp->version) !=
239 	    XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
240 		return xpBadVersion;
241 	}
242 
243 	/* check that both remote and local partids are valid for each side */
244 	if (remote_rp->SAL_partid < 0 ||
245 	    remote_rp->SAL_partid >= xp_max_npartitions ||
246 	    remote_rp->max_npartitions <= xp_partition_id) {
247 		return xpInvalidPartid;
248 	}
249 
250 	if (remote_rp->SAL_partid == xp_partition_id)
251 		return xpLocalPartid;
252 
253 	return xpSuccess;
254 }
255 
256 /*
257  * See if the other side has responded to a partition deactivate request
258  * from us. Though we requested the remote partition to deactivate with regard
259  * to us, we really only need to wait for the other side to disengage from us.
260  */
261 int
262 xpc_partition_disengaged(struct xpc_partition *part)
263 {
264 	short partid = XPC_PARTID(part);
265 	int disengaged;
266 
267 	disengaged = !xpc_partition_engaged(partid);
268 	if (part->disengage_timeout) {
269 		if (!disengaged) {
270 			if (time_is_after_jiffies(part->disengage_timeout)) {
271 				/* timelimit hasn't been reached yet */
272 				return 0;
273 			}
274 
275 			/*
276 			 * Other side hasn't responded to our deactivate
277 			 * request in a timely fashion, so assume it's dead.
278 			 */
279 
280 			dev_info(xpc_part, "deactivate request to remote "
281 				 "partition %d timed out\n", partid);
282 			xpc_disengage_timedout = 1;
283 			xpc_assume_partition_disengaged(partid);
284 			disengaged = 1;
285 		}
286 		part->disengage_timeout = 0;
287 
288 		/* cancel the timer function, provided it's not us */
289 		if (!in_interrupt())
290 			del_singleshot_timer_sync(&part->disengage_timer);
291 
292 		DBUG_ON(part->act_state != XPC_P_AS_DEACTIVATING &&
293 			part->act_state != XPC_P_AS_INACTIVE);
294 		if (part->act_state != XPC_P_AS_INACTIVE)
295 			xpc_wakeup_channel_mgr(part);
296 
297 		xpc_cancel_partition_deactivation_request(part);
298 	}
299 	return disengaged;
300 }
301 
302 /*
303  * Mark specified partition as active.
304  */
305 enum xp_retval
306 xpc_mark_partition_active(struct xpc_partition *part)
307 {
308 	unsigned long irq_flags;
309 	enum xp_retval ret;
310 
311 	dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
312 
313 	spin_lock_irqsave(&part->act_lock, irq_flags);
314 	if (part->act_state == XPC_P_AS_ACTIVATING) {
315 		part->act_state = XPC_P_AS_ACTIVE;
316 		ret = xpSuccess;
317 	} else {
318 		DBUG_ON(part->reason == xpSuccess);
319 		ret = part->reason;
320 	}
321 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
322 
323 	return ret;
324 }
325 
326 /*
327  * Start the process of deactivating the specified partition.
328  */
329 void
330 xpc_deactivate_partition(const int line, struct xpc_partition *part,
331 			 enum xp_retval reason)
332 {
333 	unsigned long irq_flags;
334 
335 	spin_lock_irqsave(&part->act_lock, irq_flags);
336 
337 	if (part->act_state == XPC_P_AS_INACTIVE) {
338 		XPC_SET_REASON(part, reason, line);
339 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
340 		if (reason == xpReactivating) {
341 			/* we interrupt ourselves to reactivate partition */
342 			xpc_request_partition_reactivation(part);
343 		}
344 		return;
345 	}
346 	if (part->act_state == XPC_P_AS_DEACTIVATING) {
347 		if ((part->reason == xpUnloading && reason != xpUnloading) ||
348 		    reason == xpReactivating) {
349 			XPC_SET_REASON(part, reason, line);
350 		}
351 		spin_unlock_irqrestore(&part->act_lock, irq_flags);
352 		return;
353 	}
354 
355 	part->act_state = XPC_P_AS_DEACTIVATING;
356 	XPC_SET_REASON(part, reason, line);
357 
358 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
359 
360 	/* ask remote partition to deactivate with regard to us */
361 	xpc_request_partition_deactivation(part);
362 
363 	/* set a timelimit on the disengage phase of the deactivation request */
364 	part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
365 	part->disengage_timer.expires = part->disengage_timeout;
366 	add_timer(&part->disengage_timer);
367 
368 	dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
369 		XPC_PARTID(part), reason);
370 
371 	xpc_partition_going_down(part, reason);
372 }
373 
374 /*
375  * Mark specified partition as inactive.
376  */
377 void
378 xpc_mark_partition_inactive(struct xpc_partition *part)
379 {
380 	unsigned long irq_flags;
381 
382 	dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
383 		XPC_PARTID(part));
384 
385 	spin_lock_irqsave(&part->act_lock, irq_flags);
386 	part->act_state = XPC_P_AS_INACTIVE;
387 	spin_unlock_irqrestore(&part->act_lock, irq_flags);
388 	part->remote_rp_pa = 0;
389 }
390 
391 /*
392  * SAL has provided a partition and machine mask.  The partition mask
393  * contains a bit for each even nasid in our partition.  The machine
394  * mask contains a bit for each even nasid in the entire machine.
395  *
396  * Using those two bit arrays, we can determine which nasids are
397  * known in the machine.  Each should also have a reserved page
398  * initialized if they are available for partitioning.
399  */
400 void
401 xpc_discovery(void)
402 {
403 	void *remote_rp_base;
404 	struct xpc_rsvd_page *remote_rp;
405 	unsigned long remote_rp_pa;
406 	int region;
407 	int region_size;
408 	int max_regions;
409 	int nasid;
410 	struct xpc_rsvd_page *rp;
411 	unsigned long *discovered_nasids;
412 	enum xp_retval ret;
413 
414 	remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
415 						  xpc_nasid_mask_nbytes,
416 						  GFP_KERNEL, &remote_rp_base);
417 	if (remote_rp == NULL)
418 		return;
419 
420 	discovered_nasids = kzalloc(sizeof(long) * xpc_nasid_mask_nlongs,
421 				    GFP_KERNEL);
422 	if (discovered_nasids == NULL) {
423 		kfree(remote_rp_base);
424 		return;
425 	}
426 
427 	rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
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 	max_regions = 64;
435 	region_size = xp_region_size;
436 
437 	switch (region_size) {
438 	case 128:
439 		max_regions *= 2;
440 	case 64:
441 		max_regions *= 2;
442 	case 32:
443 		max_regions *= 2;
444 		region_size = 16;
445 		DBUG_ON(!is_shub2());
446 	}
447 
448 	for (region = 0; region < max_regions; region++) {
449 
450 		if (xpc_exiting)
451 			break;
452 
453 		dev_dbg(xpc_part, "searching region %d\n", region);
454 
455 		for (nasid = (region * region_size * 2);
456 		     nasid < ((region + 1) * region_size * 2); nasid += 2) {
457 
458 			if (xpc_exiting)
459 				break;
460 
461 			dev_dbg(xpc_part, "checking nasid %d\n", nasid);
462 
463 			if (test_bit(nasid / 2, xpc_part_nasids)) {
464 				dev_dbg(xpc_part, "PROM indicates Nasid %d is "
465 					"part of the local partition; skipping "
466 					"region\n", nasid);
467 				break;
468 			}
469 
470 			if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
471 				dev_dbg(xpc_part, "PROM indicates Nasid %d was "
472 					"not on Numa-Link network at reset\n",
473 					nasid);
474 				continue;
475 			}
476 
477 			if (test_bit(nasid / 2, discovered_nasids)) {
478 				dev_dbg(xpc_part, "Nasid %d is part of a "
479 					"partition which was previously "
480 					"discovered\n", nasid);
481 				continue;
482 			}
483 
484 			/* pull over the rsvd page header & part_nasids mask */
485 
486 			ret = xpc_get_remote_rp(nasid, discovered_nasids,
487 						remote_rp, &remote_rp_pa);
488 			if (ret != xpSuccess) {
489 				dev_dbg(xpc_part, "unable to get reserved page "
490 					"from nasid %d, reason=%d\n", nasid,
491 					ret);
492 
493 				if (ret == xpLocalPartid)
494 					break;
495 
496 				continue;
497 			}
498 
499 			xpc_request_partition_activation(remote_rp,
500 							 remote_rp_pa, nasid);
501 		}
502 	}
503 
504 	kfree(discovered_nasids);
505 	kfree(remote_rp_base);
506 }
507 
508 /*
509  * Given a partid, get the nasids owned by that partition from the
510  * remote partition's reserved page.
511  */
512 enum xp_retval
513 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
514 {
515 	struct xpc_partition *part;
516 	unsigned long part_nasid_pa;
517 
518 	part = &xpc_partitions[partid];
519 	if (part->remote_rp_pa == 0)
520 		return xpPartitionDown;
521 
522 	memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
523 
524 	part_nasid_pa = (unsigned long)XPC_RP_PART_NASIDS(part->remote_rp_pa);
525 
526 	return xp_remote_memcpy(xp_pa(nasid_mask), part_nasid_pa,
527 				xpc_nasid_mask_nbytes);
528 }
529