Loading...
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 */
265static int __xpc_partition_disengaged(struct xpc_partition *part,
266 bool from_timer)
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 if not called from it */
293 if (!from_timer)
294 del_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
306int xpc_partition_disengaged(struct xpc_partition *part)
307{
308 return __xpc_partition_disengaged(part, false);
309}
310
311int xpc_partition_disengaged_from_timer(struct xpc_partition *part)
312{
313 return __xpc_partition_disengaged(part, true);
314}
315
316/*
317 * Mark specified partition as active.
318 */
319enum xp_retval
320xpc_mark_partition_active(struct xpc_partition *part)
321{
322 unsigned long irq_flags;
323 enum xp_retval ret;
324
325 dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
326
327 spin_lock_irqsave(&part->act_lock, irq_flags);
328 if (part->act_state == XPC_P_AS_ACTIVATING) {
329 part->act_state = XPC_P_AS_ACTIVE;
330 ret = xpSuccess;
331 } else {
332 DBUG_ON(part->reason == xpSuccess);
333 ret = part->reason;
334 }
335 spin_unlock_irqrestore(&part->act_lock, irq_flags);
336
337 return ret;
338}
339
340/*
341 * Start the process of deactivating the specified partition.
342 */
343void
344xpc_deactivate_partition(const int line, struct xpc_partition *part,
345 enum xp_retval reason)
346{
347 unsigned long irq_flags;
348
349 spin_lock_irqsave(&part->act_lock, irq_flags);
350
351 if (part->act_state == XPC_P_AS_INACTIVE) {
352 XPC_SET_REASON(part, reason, line);
353 spin_unlock_irqrestore(&part->act_lock, irq_flags);
354 if (reason == xpReactivating) {
355 /* we interrupt ourselves to reactivate partition */
356 xpc_arch_ops.request_partition_reactivation(part);
357 }
358 return;
359 }
360 if (part->act_state == XPC_P_AS_DEACTIVATING) {
361 if ((part->reason == xpUnloading && reason != xpUnloading) ||
362 reason == xpReactivating) {
363 XPC_SET_REASON(part, reason, line);
364 }
365 spin_unlock_irqrestore(&part->act_lock, irq_flags);
366 return;
367 }
368
369 part->act_state = XPC_P_AS_DEACTIVATING;
370 XPC_SET_REASON(part, reason, line);
371
372 spin_unlock_irqrestore(&part->act_lock, irq_flags);
373
374 /* ask remote partition to deactivate with regard to us */
375 xpc_arch_ops.request_partition_deactivation(part);
376
377 /* set a timelimit on the disengage phase of the deactivation request */
378 part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
379 part->disengage_timer.expires = part->disengage_timeout;
380 add_timer(&part->disengage_timer);
381
382 dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
383 XPC_PARTID(part), reason);
384
385 xpc_partition_going_down(part, reason);
386}
387
388/*
389 * Mark specified partition as inactive.
390 */
391void
392xpc_mark_partition_inactive(struct xpc_partition *part)
393{
394 unsigned long irq_flags;
395
396 dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
397 XPC_PARTID(part));
398
399 spin_lock_irqsave(&part->act_lock, irq_flags);
400 part->act_state = XPC_P_AS_INACTIVE;
401 spin_unlock_irqrestore(&part->act_lock, irq_flags);
402 part->remote_rp_pa = 0;
403}
404
405/*
406 * SAL has provided a partition and machine mask. The partition mask
407 * contains a bit for each even nasid in our partition. The machine
408 * mask contains a bit for each even nasid in the entire machine.
409 *
410 * Using those two bit arrays, we can determine which nasids are
411 * known in the machine. Each should also have a reserved page
412 * initialized if they are available for partitioning.
413 */
414void
415xpc_discovery(void)
416{
417 void *remote_rp_base;
418 struct xpc_rsvd_page *remote_rp;
419 unsigned long remote_rp_pa;
420 int region;
421 int region_size;
422 int max_regions;
423 int nasid;
424 unsigned long *discovered_nasids;
425 enum xp_retval ret;
426
427 remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
428 xpc_nasid_mask_nbytes,
429 GFP_KERNEL, &remote_rp_base);
430 if (remote_rp == NULL)
431 return;
432
433 discovered_nasids = kcalloc(xpc_nasid_mask_nlongs, sizeof(long),
434 GFP_KERNEL);
435 if (discovered_nasids == NULL) {
436 kfree(remote_rp_base);
437 return;
438 }
439
440 /*
441 * The term 'region' in this context refers to the minimum number of
442 * nodes that can comprise an access protection grouping. The access
443 * protection is in regards to memory, IOI and IPI.
444 */
445 region_size = xp_region_size;
446
447 if (is_uv_system())
448 max_regions = 256;
449 else {
450 max_regions = 64;
451
452 switch (region_size) {
453 case 128:
454 max_regions *= 2;
455 fallthrough;
456 case 64:
457 max_regions *= 2;
458 fallthrough;
459 case 32:
460 max_regions *= 2;
461 region_size = 16;
462 }
463 }
464
465 for (region = 0; region < max_regions; region++) {
466
467 if (xpc_exiting)
468 break;
469
470 dev_dbg(xpc_part, "searching region %d\n", region);
471
472 for (nasid = (region * region_size * 2);
473 nasid < ((region + 1) * region_size * 2); nasid += 2) {
474
475 if (xpc_exiting)
476 break;
477
478 dev_dbg(xpc_part, "checking nasid %d\n", nasid);
479
480 if (test_bit(nasid / 2, xpc_part_nasids)) {
481 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
482 "part of the local partition; skipping "
483 "region\n", nasid);
484 break;
485 }
486
487 if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
488 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
489 "not on Numa-Link network at reset\n",
490 nasid);
491 continue;
492 }
493
494 if (test_bit(nasid / 2, discovered_nasids)) {
495 dev_dbg(xpc_part, "Nasid %d is part of a "
496 "partition which was previously "
497 "discovered\n", nasid);
498 continue;
499 }
500
501 /* pull over the rsvd page header & part_nasids mask */
502
503 ret = xpc_get_remote_rp(nasid, discovered_nasids,
504 remote_rp, &remote_rp_pa);
505 if (ret != xpSuccess) {
506 dev_dbg(xpc_part, "unable to get reserved page "
507 "from nasid %d, reason=%d\n", nasid,
508 ret);
509
510 if (ret == xpLocalPartid)
511 break;
512
513 continue;
514 }
515
516 xpc_arch_ops.request_partition_activation(remote_rp,
517 remote_rp_pa, nasid);
518 }
519 }
520
521 kfree(discovered_nasids);
522 kfree(remote_rp_base);
523}
524
525/*
526 * Given a partid, get the nasids owned by that partition from the
527 * remote partition's reserved page.
528 */
529enum xp_retval
530xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
531{
532 struct xpc_partition *part;
533 unsigned long part_nasid_pa;
534
535 part = &xpc_partitions[partid];
536 if (part->remote_rp_pa == 0)
537 return xpPartitionDown;
538
539 memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
540
541 part_nasid_pa = (unsigned long)XPC_RP_PART_NASIDS(part->remote_rp_pa);
542
543 return xp_remote_memcpy(xp_pa(nasid_mask), part_nasid_pa,
544 xpc_nasid_mask_nbytes);
545}
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 */
25int xpc_exiting;
26
27/* this partition's reserved page pointers */
28struct xpc_rsvd_page *xpc_rsvd_page;
29static unsigned long *xpc_part_nasids;
30unsigned long *xpc_mach_nasids;
31
32static int xpc_nasid_mask_nbytes; /* #of bytes in nasid mask */
33int xpc_nasid_mask_nlongs; /* #of longs in nasid mask */
34
35struct xpc_partition *xpc_partitions;
36
37/*
38 * Guarantee that the kmalloc'd memory is cacheline aligned.
39 */
40void *
41xpc_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 */
65static unsigned long
66xpc_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 */
130int
131xpc_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
196void
197xpc_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 */
210enum xp_retval
211xpc_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 */
264int
265xpc_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 */
308enum xp_retval
309xpc_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 */
332void
333xpc_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 */
380void
381xpc_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 */
403void
404xpc_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 fallthrough;
445 case 64:
446 max_regions *= 2;
447 fallthrough;
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 */
518enum xp_retval
519xpc_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}