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 * (C) Copyright 2020 Hewlett Packard Enterprise Development LP
7 * Copyright (c) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
8 */
9
10/*
11 * Cross Partition Communication (XPC) partition support.
12 *
13 *	This is the part of XPC that detects the presence/absence of
14 *	other partitions. It provides a heartbeat and monitors the
15 *	heartbeats of other partitions.
16 *
17 */
18
19#include <linux/device.h>
20#include <linux/hardirq.h>
21#include <linux/slab.h>
22#include "xpc.h"
23#include <asm/uv/uv_hub.h>
24
25/* XPC is exiting flag */
26int xpc_exiting;
27
28/* this partition's reserved page pointers */
29struct xpc_rsvd_page *xpc_rsvd_page;
30static unsigned long *xpc_part_nasids;
31unsigned long *xpc_mach_nasids;
32
33static int xpc_nasid_mask_nbytes;	/* #of bytes in nasid mask */
34int xpc_nasid_mask_nlongs;	/* #of longs in nasid mask */
35
36struct xpc_partition *xpc_partitions;
37
38/*
39 * Guarantee that the kmalloc'd memory is cacheline aligned.
40 */
41void *
42xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
43{
44	/* see if kmalloc will give us cachline aligned memory by default */
45	*base = kmalloc(size, flags);
46	if (*base == NULL)
47		return NULL;
48
49	if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
50		return *base;
51
52	kfree(*base);
53
54	/* nope, we'll have to do it ourselves */
55	*base = kmalloc(size + L1_CACHE_BYTES, flags);
56	if (*base == NULL)
57		return NULL;
58
59	return (void *)L1_CACHE_ALIGN((u64)*base);
60}
61
62/*
63 * Given a nasid, get the physical address of the  partition's reserved page
64 * for that nasid. This function returns 0 on any error.
65 */
66static unsigned long
67xpc_get_rsvd_page_pa(int nasid)
68{
69	enum xp_retval ret;
70	u64 cookie = 0;
71	unsigned long rp_pa = nasid;	/* seed with nasid */
72	size_t len = 0;
73	size_t buf_len = 0;
74	void *buf = NULL;
75	void *buf_base = NULL;
76	enum xp_retval (*get_partition_rsvd_page_pa)
77		(void *, u64 *, unsigned long *, size_t *) =
78		xpc_arch_ops.get_partition_rsvd_page_pa;
79
80	while (1) {
81
82		/* !!! rp_pa will need to be _gpa on UV.
83		 * ??? So do we save it into the architecture specific parts
84		 * ??? of the xpc_partition structure? Do we rename this
85		 * ??? function or have two versions? Rename rp_pa for UV to
86		 * ??? rp_gpa?
87		 */
88		ret = get_partition_rsvd_page_pa(buf, &cookie, &rp_pa, &len);
89
90		dev_dbg(xpc_part, "SAL returned with ret=%d, cookie=0x%016lx, "
91			"address=0x%016lx, len=0x%016lx\n", ret,
92			(unsigned long)cookie, rp_pa, len);
93
94		if (ret != xpNeedMoreInfo)
95			break;
96
97		if (len > buf_len) {
98			kfree(buf_base);
99			buf_len = L1_CACHE_ALIGN(len);
100			buf = xpc_kmalloc_cacheline_aligned(buf_len, GFP_KERNEL,
101							    &buf_base);
102			if (buf_base == NULL) {
103				dev_err(xpc_part, "unable to kmalloc "
104					"len=0x%016lx\n", buf_len);
105				ret = xpNoMemory;
106				break;
107			}
108		}
109
110		ret = xp_remote_memcpy(xp_pa(buf), rp_pa, len);
111		if (ret != xpSuccess) {
112			dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
113			break;
114		}
115	}
116
117	kfree(buf_base);
118
119	if (ret != xpSuccess)
120		rp_pa = 0;
121
122	dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
123	return rp_pa;
124}
125
126/*
127 * Fill the partition reserved page with the information needed by
128 * other partitions to discover we are alive and establish initial
129 * communications.
130 */
131int
132xpc_setup_rsvd_page(void)
133{
134	int ret;
135	struct xpc_rsvd_page *rp;
136	unsigned long rp_pa;
137	unsigned long new_ts_jiffies;
138
139	/* get the local reserved page's address */
140
141	preempt_disable();
142	rp_pa = xpc_get_rsvd_page_pa(xp_cpu_to_nasid(smp_processor_id()));
143	preempt_enable();
144	if (rp_pa == 0) {
145		dev_err(xpc_part, "SAL failed to locate the reserved page\n");
146		return -ESRCH;
147	}
148	rp = (struct xpc_rsvd_page *)__va(xp_socket_pa(rp_pa));
149
150	if (rp->SAL_version < 3) {
151		/* SAL_versions < 3 had a SAL_partid defined as a u8 */
152		rp->SAL_partid &= 0xff;
153	}
154	BUG_ON(rp->SAL_partid != xp_partition_id);
155
156	if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
157		dev_err(xpc_part, "the reserved page's partid of %d is outside "
158			"supported range (< 0 || >= %d)\n", rp->SAL_partid,
159			xp_max_npartitions);
160		return -EINVAL;
161	}
162
163	rp->version = XPC_RP_VERSION;
164	rp->max_npartitions = xp_max_npartitions;
165
166	/* establish the actual sizes of the nasid masks */
167	if (rp->SAL_version == 1) {
168		/* SAL_version 1 didn't set the nasids_size field */
169		rp->SAL_nasids_size = 128;
170	}
171	xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
172	xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
173					      BITS_PER_BYTE);
174
175	/* setup the pointers to the various items in the reserved page */
176	xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
177	xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
178
179	ret = xpc_arch_ops.setup_rsvd_page(rp);
180	if (ret != 0)
181		return ret;
182
183	/*
184	 * Set timestamp of when reserved page was setup by XPC.
185	 * This signifies to the remote partition that our reserved
186	 * page is initialized.
187	 */
188	new_ts_jiffies = jiffies;
189	if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
190		new_ts_jiffies++;
191	rp->ts_jiffies = new_ts_jiffies;
192
193	xpc_rsvd_page = rp;
194	return 0;
195}
196
197void
198xpc_teardown_rsvd_page(void)
199{
200	/* a zero timestamp indicates our rsvd page is not initialized */
201	xpc_rsvd_page->ts_jiffies = 0;
202}
203
204/*
205 * Get a copy of a portion of the remote partition's rsvd page.
206 *
207 * remote_rp points to a buffer that is cacheline aligned for BTE copies and
208 * is large enough to contain a copy of their reserved page header and
209 * part_nasids mask.
210 */
211enum xp_retval
212xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
213		  struct xpc_rsvd_page *remote_rp, unsigned long *remote_rp_pa)
214{
215	int l;
216	enum xp_retval ret;
217
218	/* get the reserved page's physical address */
219
220	*remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
221	if (*remote_rp_pa == 0)
222		return xpNoRsvdPageAddr;
223
224	/* pull over the reserved page header and part_nasids mask */
225	ret = xp_remote_memcpy(xp_pa(remote_rp), *remote_rp_pa,
226			       XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
227	if (ret != xpSuccess)
228		return ret;
229
230	if (discovered_nasids != NULL) {
231		unsigned long *remote_part_nasids =
232		    XPC_RP_PART_NASIDS(remote_rp);
233
234		for (l = 0; l < xpc_nasid_mask_nlongs; l++)
235			discovered_nasids[l] |= remote_part_nasids[l];
236	}
237
238	/* zero timestamp indicates the reserved page has not been setup */
239	if (remote_rp->ts_jiffies == 0)
240		return xpRsvdPageNotSet;
241
242	if (XPC_VERSION_MAJOR(remote_rp->version) !=
243	    XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
244		return xpBadVersion;
245	}
246
247	/* check that both remote and local partids are valid for each side */
248	if (remote_rp->SAL_partid < 0 ||
249	    remote_rp->SAL_partid >= xp_max_npartitions ||
250	    remote_rp->max_npartitions <= xp_partition_id) {
251		return xpInvalidPartid;
252	}
253
254	if (remote_rp->SAL_partid == xp_partition_id)
255		return xpLocalPartid;
256
257	return xpSuccess;
258}
259
260/*
261 * See if the other side has responded to a partition deactivate request
262 * from us. Though we requested the remote partition to deactivate with regard
263 * to us, we really only need to wait for the other side to disengage from us.
264 */
265int
266xpc_partition_disengaged(struct xpc_partition *part)
267{
268	short partid = XPC_PARTID(part);
269	int disengaged;
270
271	disengaged = !xpc_arch_ops.partition_engaged(partid);
272	if (part->disengage_timeout) {
273		if (!disengaged) {
274			if (time_is_after_jiffies(part->disengage_timeout)) {
275				/* timelimit hasn't been reached yet */
276				return 0;
277			}
278
279			/*
280			 * Other side hasn't responded to our deactivate
281			 * request in a timely fashion, so assume it's dead.
282			 */
283
284			dev_info(xpc_part, "deactivate request to remote "
285				 "partition %d timed out\n", partid);
286			xpc_disengage_timedout = 1;
287			xpc_arch_ops.assume_partition_disengaged(partid);
288			disengaged = 1;
289		}
290		part->disengage_timeout = 0;
291
292		/* cancel the timer function, provided it's not us */
293		if (!in_interrupt())
294			del_singleshot_timer_sync(&part->disengage_timer);
295
296		DBUG_ON(part->act_state != XPC_P_AS_DEACTIVATING &&
297			part->act_state != XPC_P_AS_INACTIVE);
298		if (part->act_state != XPC_P_AS_INACTIVE)
299			xpc_wakeup_channel_mgr(part);
300
301		xpc_arch_ops.cancel_partition_deactivation_request(part);
302	}
303	return disengaged;
304}
305
306/*
307 * Mark specified partition as active.
308 */
309enum xp_retval
310xpc_mark_partition_active(struct xpc_partition *part)
311{
312	unsigned long irq_flags;
313	enum xp_retval ret;
314
315	dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
316
317	spin_lock_irqsave(&part->act_lock, irq_flags);
318	if (part->act_state == XPC_P_AS_ACTIVATING) {
319		part->act_state = XPC_P_AS_ACTIVE;
320		ret = xpSuccess;
321	} else {
322		DBUG_ON(part->reason == xpSuccess);
323		ret = part->reason;
324	}
325	spin_unlock_irqrestore(&part->act_lock, irq_flags);
326
327	return ret;
328}
329
330/*
331 * Start the process of deactivating the specified partition.
332 */
333void
334xpc_deactivate_partition(const int line, struct xpc_partition *part,
335			 enum xp_retval reason)
336{
337	unsigned long irq_flags;
338
339	spin_lock_irqsave(&part->act_lock, irq_flags);
340
341	if (part->act_state == XPC_P_AS_INACTIVE) {
342		XPC_SET_REASON(part, reason, line);
343		spin_unlock_irqrestore(&part->act_lock, irq_flags);
344		if (reason == xpReactivating) {
345			/* we interrupt ourselves to reactivate partition */
346			xpc_arch_ops.request_partition_reactivation(part);
347		}
348		return;
349	}
350	if (part->act_state == XPC_P_AS_DEACTIVATING) {
351		if ((part->reason == xpUnloading && reason != xpUnloading) ||
352		    reason == xpReactivating) {
353			XPC_SET_REASON(part, reason, line);
354		}
355		spin_unlock_irqrestore(&part->act_lock, irq_flags);
356		return;
357	}
358
359	part->act_state = XPC_P_AS_DEACTIVATING;
360	XPC_SET_REASON(part, reason, line);
361
362	spin_unlock_irqrestore(&part->act_lock, irq_flags);
363
364	/* ask remote partition to deactivate with regard to us */
365	xpc_arch_ops.request_partition_deactivation(part);
366
367	/* set a timelimit on the disengage phase of the deactivation request */
368	part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
369	part->disengage_timer.expires = part->disengage_timeout;
370	add_timer(&part->disengage_timer);
371
372	dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
373		XPC_PARTID(part), reason);
374
375	xpc_partition_going_down(part, reason);
376}
377
378/*
379 * Mark specified partition as inactive.
380 */
381void
382xpc_mark_partition_inactive(struct xpc_partition *part)
383{
384	unsigned long irq_flags;
385
386	dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
387		XPC_PARTID(part));
388
389	spin_lock_irqsave(&part->act_lock, irq_flags);
390	part->act_state = XPC_P_AS_INACTIVE;
391	spin_unlock_irqrestore(&part->act_lock, irq_flags);
392	part->remote_rp_pa = 0;
393}
394
395/*
396 * SAL has provided a partition and machine mask.  The partition mask
397 * contains a bit for each even nasid in our partition.  The machine
398 * mask contains a bit for each even nasid in the entire machine.
399 *
400 * Using those two bit arrays, we can determine which nasids are
401 * known in the machine.  Each should also have a reserved page
402 * initialized if they are available for partitioning.
403 */
404void
405xpc_discovery(void)
406{
407	void *remote_rp_base;
408	struct xpc_rsvd_page *remote_rp;
409	unsigned long remote_rp_pa;
410	int region;
411	int region_size;
412	int max_regions;
413	int nasid;
414	unsigned long *discovered_nasids;
415	enum xp_retval ret;
416
417	remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
418						  xpc_nasid_mask_nbytes,
419						  GFP_KERNEL, &remote_rp_base);
420	if (remote_rp == NULL)
421		return;
422
423	discovered_nasids = kcalloc(xpc_nasid_mask_nlongs, sizeof(long),
424				    GFP_KERNEL);
425	if (discovered_nasids == NULL) {
426		kfree(remote_rp_base);
427		return;
428	}
429
430	/*
431	 * The term 'region' in this context refers to the minimum number of
432	 * nodes that can comprise an access protection grouping. The access
433	 * protection is in regards to memory, IOI and IPI.
434	 */
435	region_size = xp_region_size;
436
437	if (is_uv_system())
438		max_regions = 256;
439	else {
440		max_regions = 64;
441
442		switch (region_size) {
443		case 128:
444			max_regions *= 2;
445			fallthrough;
446		case 64:
447			max_regions *= 2;
448			fallthrough;
449		case 32:
450			max_regions *= 2;
451			region_size = 16;
452		}
453	}
454
455	for (region = 0; region < max_regions; region++) {
456
457		if (xpc_exiting)
458			break;
459
460		dev_dbg(xpc_part, "searching region %d\n", region);
461
462		for (nasid = (region * region_size * 2);
463		     nasid < ((region + 1) * region_size * 2); nasid += 2) {
464
465			if (xpc_exiting)
466				break;
467
468			dev_dbg(xpc_part, "checking nasid %d\n", nasid);
469
470			if (test_bit(nasid / 2, xpc_part_nasids)) {
471				dev_dbg(xpc_part, "PROM indicates Nasid %d is "
472					"part of the local partition; skipping "
473					"region\n", nasid);
474				break;
475			}
476
477			if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
478				dev_dbg(xpc_part, "PROM indicates Nasid %d was "
479					"not on Numa-Link network at reset\n",
480					nasid);
481				continue;
482			}
483
484			if (test_bit(nasid / 2, discovered_nasids)) {
485				dev_dbg(xpc_part, "Nasid %d is part of a "
486					"partition which was previously "
487					"discovered\n", nasid);
488				continue;
489			}
490
491			/* pull over the rsvd page header & part_nasids mask */
492
493			ret = xpc_get_remote_rp(nasid, discovered_nasids,
494						remote_rp, &remote_rp_pa);
495			if (ret != xpSuccess) {
496				dev_dbg(xpc_part, "unable to get reserved page "
497					"from nasid %d, reason=%d\n", nasid,
498					ret);
499
500				if (ret == xpLocalPartid)
501					break;
502
503				continue;
504			}
505
506			xpc_arch_ops.request_partition_activation(remote_rp,
507							 remote_rp_pa, nasid);
508		}
509	}
510
511	kfree(discovered_nasids);
512	kfree(remote_rp_base);
513}
514
515/*
516 * Given a partid, get the nasids owned by that partition from the
517 * remote partition's reserved page.
518 */
519enum xp_retval
520xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
521{
522	struct xpc_partition *part;
523	unsigned long part_nasid_pa;
524
525	part = &xpc_partitions[partid];
526	if (part->remote_rp_pa == 0)
527		return xpPartitionDown;
528
529	memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
530
531	part_nasid_pa = (unsigned long)XPC_RP_PART_NASIDS(part->remote_rp_pa);
532
533	return xp_remote_memcpy(xp_pa(nasid_mask), part_nasid_pa,
534				xpc_nasid_mask_nbytes);
535}
536