]>
Commit | Line | Data |
---|---|---|
89eb8eb9 DN |
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 | * | |
a374c57b | 6 | * Copyright (c) 2004-2009 Silicon Graphics, Inc. All Rights Reserved. |
89eb8eb9 DN |
7 | */ |
8 | ||
89eb8eb9 DN |
9 | /* |
10 | * Cross Partition Communication (XPC) support - standard version. | |
11 | * | |
12 | * XPC provides a message passing capability that crosses partition | |
13 | * boundaries. This module is made up of two parts: | |
14 | * | |
15 | * partition This part detects the presence/absence of other | |
16 | * partitions. It provides a heartbeat and monitors | |
17 | * the heartbeats of other partitions. | |
18 | * | |
19 | * channel This part manages the channels and sends/receives | |
20 | * messages across them to/from other partitions. | |
21 | * | |
22 | * There are a couple of additional functions residing in XP, which | |
23 | * provide an interface to XPC for its users. | |
24 | * | |
25 | * | |
26 | * Caveats: | |
27 | * | |
7fb5e59d | 28 | * . Currently on sn2, we have no way to determine which nasid an IRQ |
c39838ce DN |
29 | * came from. Thus, xpc_send_IRQ_sn2() does a remote amo write |
30 | * followed by an IPI. The amo indicates where data is to be pulled | |
31 | * from, so after the IPI arrives, the remote partition checks the amo | |
32 | * word. The IPI can actually arrive before the amo however, so other | |
33 | * code must periodically check for this case. Also, remote amo | |
7fb5e59d DN |
34 | * operations do not reliably time out. Thus we do a remote PIO read |
35 | * solely to know whether the remote partition is down and whether we | |
36 | * should stop sending IPIs to it. This remote PIO read operation is | |
37 | * set up in a special nofault region so SAL knows to ignore (and | |
c39838ce | 38 | * cleanup) any errors due to the remote amo write, PIO read, and/or |
7fb5e59d | 39 | * PIO write operations. |
89eb8eb9 DN |
40 | * |
41 | * If/when new hardware solves this IPI problem, we should abandon | |
42 | * the current approach. | |
43 | * | |
44 | */ | |
45 | ||
89eb8eb9 | 46 | #include <linux/module.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
261f3b49 DN |
48 | #include <linux/sysctl.h> |
49 | #include <linux/device.h> | |
69913927 | 50 | #include <linux/delay.h> |
a607c389 | 51 | #include <linux/reboot.h> |
1eeb66a1 | 52 | #include <linux/kdebug.h> |
2c2b94f9 | 53 | #include <linux/kthread.h> |
45d9ca49 | 54 | #include "xpc.h" |
89eb8eb9 | 55 | |
89eb8eb9 DN |
56 | /* define two XPC debug device structures to be used with dev_dbg() et al */ |
57 | ||
58 | struct device_driver xpc_dbg_name = { | |
59 | .name = "xpc" | |
60 | }; | |
61 | ||
62 | struct device xpc_part_dbg_subname = { | |
bb0dc43e | 63 | .init_name = "", /* set to "part" at xpc_init() time */ |
89eb8eb9 DN |
64 | .driver = &xpc_dbg_name |
65 | }; | |
66 | ||
67 | struct device xpc_chan_dbg_subname = { | |
bb0dc43e | 68 | .init_name = "", /* set to "chan" at xpc_init() time */ |
89eb8eb9 DN |
69 | .driver = &xpc_dbg_name |
70 | }; | |
71 | ||
72 | struct device *xpc_part = &xpc_part_dbg_subname; | |
73 | struct device *xpc_chan = &xpc_chan_dbg_subname; | |
74 | ||
1f4674b2 DN |
75 | static int xpc_kdebug_ignore; |
76 | ||
89eb8eb9 DN |
77 | /* systune related variables for /proc/sys directories */ |
78 | ||
a607c389 DN |
79 | static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL; |
80 | static int xpc_hb_min_interval = 1; | |
81 | static int xpc_hb_max_interval = 10; | |
89eb8eb9 | 82 | |
a607c389 DN |
83 | static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL; |
84 | static int xpc_hb_check_min_interval = 10; | |
85 | static int xpc_hb_check_max_interval = 120; | |
89eb8eb9 | 86 | |
a47d5dac DN |
87 | int xpc_disengage_timelimit = XPC_DISENGAGE_DEFAULT_TIMELIMIT; |
88 | static int xpc_disengage_min_timelimit; /* = 0 */ | |
89 | static int xpc_disengage_max_timelimit = 120; | |
89eb8eb9 DN |
90 | |
91 | static ctl_table xpc_sys_xpc_hb_dir[] = { | |
92 | { | |
35190506 DN |
93 | .procname = "hb_interval", |
94 | .data = &xpc_hb_interval, | |
95 | .maxlen = sizeof(int), | |
96 | .mode = 0644, | |
6d456111 | 97 | .proc_handler = proc_dointvec_minmax, |
35190506 DN |
98 | .extra1 = &xpc_hb_min_interval, |
99 | .extra2 = &xpc_hb_max_interval}, | |
89eb8eb9 | 100 | { |
35190506 DN |
101 | .procname = "hb_check_interval", |
102 | .data = &xpc_hb_check_interval, | |
103 | .maxlen = sizeof(int), | |
104 | .mode = 0644, | |
6d456111 | 105 | .proc_handler = proc_dointvec_minmax, |
35190506 DN |
106 | .extra1 = &xpc_hb_check_min_interval, |
107 | .extra2 = &xpc_hb_check_max_interval}, | |
68cbf075 | 108 | {} |
89eb8eb9 DN |
109 | }; |
110 | static ctl_table xpc_sys_xpc_dir[] = { | |
111 | { | |
35190506 DN |
112 | .procname = "hb", |
113 | .mode = 0555, | |
114 | .child = xpc_sys_xpc_hb_dir}, | |
e54af724 | 115 | { |
a47d5dac DN |
116 | .procname = "disengage_timelimit", |
117 | .data = &xpc_disengage_timelimit, | |
35190506 DN |
118 | .maxlen = sizeof(int), |
119 | .mode = 0644, | |
6d456111 | 120 | .proc_handler = proc_dointvec_minmax, |
a47d5dac DN |
121 | .extra1 = &xpc_disengage_min_timelimit, |
122 | .extra2 = &xpc_disengage_max_timelimit}, | |
68cbf075 | 123 | {} |
89eb8eb9 DN |
124 | }; |
125 | static ctl_table xpc_sys_dir[] = { | |
126 | { | |
35190506 DN |
127 | .procname = "xpc", |
128 | .mode = 0555, | |
129 | .child = xpc_sys_xpc_dir}, | |
68cbf075 | 130 | {} |
89eb8eb9 DN |
131 | }; |
132 | static struct ctl_table_header *xpc_sysctl; | |
133 | ||
a47d5dac DN |
134 | /* non-zero if any remote partition disengage was timed out */ |
135 | int xpc_disengage_timedout; | |
89eb8eb9 | 136 | |
5b8669df DN |
137 | /* #of activate IRQs received and not yet processed */ |
138 | int xpc_activate_IRQ_rcvd; | |
139 | DEFINE_SPINLOCK(xpc_activate_IRQ_rcvd_lock); | |
89eb8eb9 DN |
140 | |
141 | /* IRQ handler notifies this wait queue on receipt of an IRQ */ | |
6e41017a | 142 | DECLARE_WAIT_QUEUE_HEAD(xpc_activate_IRQ_wq); |
89eb8eb9 DN |
143 | |
144 | static unsigned long xpc_hb_check_timeout; | |
33ba3c77 | 145 | static struct timer_list xpc_hb_timer; |
89eb8eb9 | 146 | |
e54af724 | 147 | /* notification that the xpc_hb_checker thread has exited */ |
f9e505a9 | 148 | static DECLARE_COMPLETION(xpc_hb_checker_exited); |
89eb8eb9 | 149 | |
e54af724 | 150 | /* notification that the xpc_discovery thread has exited */ |
f9e505a9 | 151 | static DECLARE_COMPLETION(xpc_discovery_exited); |
89eb8eb9 | 152 | |
89eb8eb9 DN |
153 | static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *); |
154 | ||
a607c389 DN |
155 | static int xpc_system_reboot(struct notifier_block *, unsigned long, void *); |
156 | static struct notifier_block xpc_reboot_notifier = { | |
157 | .notifier_call = xpc_system_reboot, | |
158 | }; | |
159 | ||
780d09e8 DN |
160 | static int xpc_system_die(struct notifier_block *, unsigned long, void *); |
161 | static struct notifier_block xpc_die_notifier = { | |
162 | .notifier_call = xpc_system_die, | |
163 | }; | |
164 | ||
a7665b0a | 165 | struct xpc_arch_operations xpc_arch_ops; |
94bd2708 | 166 | |
a607c389 | 167 | /* |
a47d5dac | 168 | * Timer function to enforce the timelimit on the partition disengage. |
a607c389 DN |
169 | */ |
170 | static void | |
a47d5dac | 171 | xpc_timeout_partition_disengage(unsigned long data) |
a607c389 | 172 | { |
35190506 | 173 | struct xpc_partition *part = (struct xpc_partition *)data; |
a607c389 | 174 | |
a47d5dac | 175 | DBUG_ON(time_is_after_jiffies(part->disengage_timeout)); |
a607c389 | 176 | |
35190506 | 177 | (void)xpc_partition_disengaged(part); |
a607c389 | 178 | |
a47d5dac | 179 | DBUG_ON(part->disengage_timeout != 0); |
a7665b0a | 180 | DBUG_ON(xpc_arch_ops.partition_engaged(XPC_PARTID(part))); |
a607c389 DN |
181 | } |
182 | ||
89eb8eb9 DN |
183 | /* |
184 | * Timer to produce the heartbeat. The timer structures function is | |
185 | * already set when this is initially called. A tunable is used to | |
186 | * specify when the next timeout should occur. | |
187 | */ | |
188 | static void | |
189 | xpc_hb_beater(unsigned long dummy) | |
190 | { | |
a7665b0a | 191 | xpc_arch_ops.increment_heartbeat(); |
89eb8eb9 | 192 | |
aaa3cd69 | 193 | if (time_is_before_eq_jiffies(xpc_hb_check_timeout)) |
6e41017a | 194 | wake_up_interruptible(&xpc_activate_IRQ_wq); |
89eb8eb9 DN |
195 | |
196 | xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ); | |
197 | add_timer(&xpc_hb_timer); | |
198 | } | |
199 | ||
33ba3c77 DN |
200 | static void |
201 | xpc_start_hb_beater(void) | |
202 | { | |
a7665b0a | 203 | xpc_arch_ops.heartbeat_init(); |
33ba3c77 DN |
204 | init_timer(&xpc_hb_timer); |
205 | xpc_hb_timer.function = xpc_hb_beater; | |
206 | xpc_hb_beater(0); | |
207 | } | |
208 | ||
209 | static void | |
210 | xpc_stop_hb_beater(void) | |
211 | { | |
212 | del_timer_sync(&xpc_hb_timer); | |
a7665b0a | 213 | xpc_arch_ops.heartbeat_exit(); |
33ba3c77 DN |
214 | } |
215 | ||
61deb86e DN |
216 | /* |
217 | * At periodic intervals, scan through all active partitions and ensure | |
218 | * their heartbeat is still active. If not, the partition is deactivated. | |
219 | */ | |
220 | static void | |
221 | xpc_check_remote_hb(void) | |
222 | { | |
223 | struct xpc_partition *part; | |
224 | short partid; | |
225 | enum xp_retval ret; | |
226 | ||
227 | for (partid = 0; partid < xp_max_npartitions; partid++) { | |
228 | ||
229 | if (xpc_exiting) | |
230 | break; | |
231 | ||
232 | if (partid == xp_partition_id) | |
233 | continue; | |
234 | ||
235 | part = &xpc_partitions[partid]; | |
236 | ||
83469b55 DN |
237 | if (part->act_state == XPC_P_AS_INACTIVE || |
238 | part->act_state == XPC_P_AS_DEACTIVATING) { | |
61deb86e DN |
239 | continue; |
240 | } | |
241 | ||
a7665b0a | 242 | ret = xpc_arch_ops.get_remote_heartbeat(part); |
61deb86e DN |
243 | if (ret != xpSuccess) |
244 | XPC_DEACTIVATE_PARTITION(part, ret); | |
245 | } | |
246 | } | |
247 | ||
89eb8eb9 DN |
248 | /* |
249 | * This thread is responsible for nearly all of the partition | |
250 | * activation/deactivation. | |
251 | */ | |
252 | static int | |
253 | xpc_hb_checker(void *ignore) | |
254 | { | |
35190506 | 255 | int force_IRQ = 0; |
89eb8eb9 DN |
256 | |
257 | /* this thread was marked active by xpc_hb_init() */ | |
258 | ||
f7df8ed1 | 259 | set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU)); |
89eb8eb9 | 260 | |
4c013f5c | 261 | /* set our heartbeating to other partitions into motion */ |
89eb8eb9 | 262 | xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ); |
33ba3c77 | 263 | xpc_start_hb_beater(); |
89eb8eb9 | 264 | |
2c2b94f9 | 265 | while (!xpc_exiting) { |
89eb8eb9 | 266 | |
89eb8eb9 DN |
267 | dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have " |
268 | "been received\n", | |
35190506 | 269 | (int)(xpc_hb_check_timeout - jiffies), |
5b8669df | 270 | xpc_activate_IRQ_rcvd); |
89eb8eb9 | 271 | |
89eb8eb9 | 272 | /* checking of remote heartbeats is skewed by IRQ handling */ |
aaa3cd69 | 273 | if (time_is_before_eq_jiffies(xpc_hb_check_timeout)) { |
5b8669df DN |
274 | xpc_hb_check_timeout = jiffies + |
275 | (xpc_hb_check_interval * HZ); | |
276 | ||
89eb8eb9 DN |
277 | dev_dbg(xpc_part, "checking remote heartbeats\n"); |
278 | xpc_check_remote_hb(); | |
279 | ||
280 | /* | |
5b8669df DN |
281 | * On sn2 we need to periodically recheck to ensure no |
282 | * IRQ/amo pairs have been missed. | |
89eb8eb9 | 283 | */ |
5b8669df DN |
284 | if (is_shub()) |
285 | force_IRQ = 1; | |
89eb8eb9 DN |
286 | } |
287 | ||
a607c389 | 288 | /* check for outstanding IRQs */ |
5b8669df | 289 | if (xpc_activate_IRQ_rcvd > 0 || force_IRQ != 0) { |
89eb8eb9 | 290 | force_IRQ = 0; |
5b8669df DN |
291 | dev_dbg(xpc_part, "processing activate IRQs " |
292 | "received\n"); | |
a7665b0a | 293 | xpc_arch_ops.process_activate_IRQ_rcvd(); |
89eb8eb9 | 294 | } |
a607c389 DN |
295 | |
296 | /* wait for IRQ or timeout */ | |
6e41017a | 297 | (void)wait_event_interruptible(xpc_activate_IRQ_wq, |
5b8669df | 298 | (time_is_before_eq_jiffies( |
aaa3cd69 | 299 | xpc_hb_check_timeout) || |
5b8669df | 300 | xpc_activate_IRQ_rcvd > 0 || |
2c2b94f9 | 301 | xpc_exiting)); |
89eb8eb9 DN |
302 | } |
303 | ||
33ba3c77 DN |
304 | xpc_stop_hb_beater(); |
305 | ||
89eb8eb9 DN |
306 | dev_dbg(xpc_part, "heartbeat checker is exiting\n"); |
307 | ||
e54af724 | 308 | /* mark this thread as having exited */ |
f9e505a9 | 309 | complete(&xpc_hb_checker_exited); |
89eb8eb9 DN |
310 | return 0; |
311 | } | |
312 | ||
89eb8eb9 DN |
313 | /* |
314 | * This thread will attempt to discover other partitions to activate | |
315 | * based on info provided by SAL. This new thread is short lived and | |
316 | * will exit once discovery is complete. | |
317 | */ | |
318 | static int | |
319 | xpc_initiate_discovery(void *ignore) | |
320 | { | |
89eb8eb9 DN |
321 | xpc_discovery(); |
322 | ||
323 | dev_dbg(xpc_part, "discovery thread is exiting\n"); | |
324 | ||
e54af724 | 325 | /* mark this thread as having exited */ |
f9e505a9 | 326 | complete(&xpc_discovery_exited); |
89eb8eb9 DN |
327 | return 0; |
328 | } | |
329 | ||
89eb8eb9 DN |
330 | /* |
331 | * The first kthread assigned to a newly activated partition is the one | |
e17d416b | 332 | * created by XPC HB with which it calls xpc_activating(). XPC hangs on to |
89eb8eb9 DN |
333 | * that kthread until the partition is brought down, at which time that kthread |
334 | * returns back to XPC HB. (The return of that kthread will signify to XPC HB | |
335 | * that XPC has dismantled all communication infrastructure for the associated | |
336 | * partition.) This kthread becomes the channel manager for that partition. | |
337 | * | |
338 | * Each active partition has a channel manager, who, besides connecting and | |
339 | * disconnecting channels, will ensure that each of the partition's connected | |
340 | * channels has the required number of assigned kthreads to get the work done. | |
341 | */ | |
342 | static void | |
343 | xpc_channel_mgr(struct xpc_partition *part) | |
344 | { | |
83469b55 | 345 | while (part->act_state != XPC_P_AS_DEACTIVATING || |
35190506 DN |
346 | atomic_read(&part->nchannels_active) > 0 || |
347 | !xpc_partition_disengaged(part)) { | |
89eb8eb9 | 348 | |
7fb5e59d | 349 | xpc_process_sent_chctl_flags(part); |
89eb8eb9 | 350 | |
89eb8eb9 DN |
351 | /* |
352 | * Wait until we've been requested to activate kthreads or | |
353 | * all of the channel's message queues have been torn down or | |
354 | * a signal is pending. | |
355 | * | |
356 | * The channel_mgr_requests is set to 1 after being awakened, | |
357 | * This is done to prevent the channel mgr from making one pass | |
358 | * through the loop for each request, since he will | |
359 | * be servicing all the requests in one pass. The reason it's | |
360 | * set to 1 instead of 0 is so that other kthreads will know | |
361 | * that the channel mgr is running and won't bother trying to | |
362 | * wake him up. | |
363 | */ | |
364 | atomic_dec(&part->channel_mgr_requests); | |
35190506 | 365 | (void)wait_event_interruptible(part->channel_mgr_wq, |
2c2b94f9 | 366 | (atomic_read(&part->channel_mgr_requests) > 0 || |
7fb5e59d | 367 | part->chctl.all_flags != 0 || |
83469b55 | 368 | (part->act_state == XPC_P_AS_DEACTIVATING && |
2c2b94f9 DN |
369 | atomic_read(&part->nchannels_active) == 0 && |
370 | xpc_partition_disengaged(part)))); | |
89eb8eb9 | 371 | atomic_set(&part->channel_mgr_requests, 1); |
89eb8eb9 DN |
372 | } |
373 | } | |
374 | ||
5b8669df DN |
375 | /* |
376 | * Guarantee that the kzalloc'd memory is cacheline aligned. | |
377 | */ | |
378 | void * | |
379 | xpc_kzalloc_cacheline_aligned(size_t size, gfp_t flags, void **base) | |
380 | { | |
381 | /* see if kzalloc will give us cachline aligned memory by default */ | |
382 | *base = kzalloc(size, flags); | |
383 | if (*base == NULL) | |
384 | return NULL; | |
385 | ||
386 | if ((u64)*base == L1_CACHE_ALIGN((u64)*base)) | |
387 | return *base; | |
388 | ||
389 | kfree(*base); | |
390 | ||
391 | /* nope, we'll have to do it ourselves */ | |
392 | *base = kzalloc(size + L1_CACHE_BYTES, flags); | |
393 | if (*base == NULL) | |
394 | return NULL; | |
395 | ||
396 | return (void *)L1_CACHE_ALIGN((u64)*base); | |
397 | } | |
398 | ||
399 | /* | |
400 | * Setup the channel structures necessary to support XPartition Communication | |
401 | * between the specified remote partition and the local one. | |
402 | */ | |
403 | static enum xp_retval | |
404 | xpc_setup_ch_structures(struct xpc_partition *part) | |
405 | { | |
406 | enum xp_retval ret; | |
407 | int ch_number; | |
408 | struct xpc_channel *ch; | |
409 | short partid = XPC_PARTID(part); | |
410 | ||
411 | /* | |
412 | * Allocate all of the channel structures as a contiguous chunk of | |
413 | * memory. | |
414 | */ | |
415 | DBUG_ON(part->channels != NULL); | |
416 | part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_MAX_NCHANNELS, | |
417 | GFP_KERNEL); | |
418 | if (part->channels == NULL) { | |
419 | dev_err(xpc_chan, "can't get memory for channels\n"); | |
420 | return xpNoMemory; | |
421 | } | |
422 | ||
423 | /* allocate the remote open and close args */ | |
424 | ||
425 | part->remote_openclose_args = | |
426 | xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE, | |
427 | GFP_KERNEL, &part-> | |
428 | remote_openclose_args_base); | |
429 | if (part->remote_openclose_args == NULL) { | |
430 | dev_err(xpc_chan, "can't get memory for remote connect args\n"); | |
431 | ret = xpNoMemory; | |
432 | goto out_1; | |
433 | } | |
434 | ||
435 | part->chctl.all_flags = 0; | |
436 | spin_lock_init(&part->chctl_lock); | |
437 | ||
438 | atomic_set(&part->channel_mgr_requests, 1); | |
439 | init_waitqueue_head(&part->channel_mgr_wq); | |
440 | ||
441 | part->nchannels = XPC_MAX_NCHANNELS; | |
442 | ||
443 | atomic_set(&part->nchannels_active, 0); | |
444 | atomic_set(&part->nchannels_engaged, 0); | |
445 | ||
446 | for (ch_number = 0; ch_number < part->nchannels; ch_number++) { | |
447 | ch = &part->channels[ch_number]; | |
448 | ||
449 | ch->partid = partid; | |
450 | ch->number = ch_number; | |
451 | ch->flags = XPC_C_DISCONNECTED; | |
452 | ||
453 | atomic_set(&ch->kthreads_assigned, 0); | |
454 | atomic_set(&ch->kthreads_idle, 0); | |
455 | atomic_set(&ch->kthreads_active, 0); | |
456 | ||
457 | atomic_set(&ch->references, 0); | |
458 | atomic_set(&ch->n_to_notify, 0); | |
459 | ||
460 | spin_lock_init(&ch->lock); | |
461 | init_completion(&ch->wdisconnect_wait); | |
462 | ||
463 | atomic_set(&ch->n_on_msg_allocate_wq, 0); | |
464 | init_waitqueue_head(&ch->msg_allocate_wq); | |
465 | init_waitqueue_head(&ch->idle_wq); | |
466 | } | |
467 | ||
a7665b0a | 468 | ret = xpc_arch_ops.setup_ch_structures(part); |
5b8669df DN |
469 | if (ret != xpSuccess) |
470 | goto out_2; | |
471 | ||
472 | /* | |
473 | * With the setting of the partition setup_state to XPC_P_SS_SETUP, | |
474 | * we're declaring that this partition is ready to go. | |
475 | */ | |
476 | part->setup_state = XPC_P_SS_SETUP; | |
477 | ||
478 | return xpSuccess; | |
479 | ||
480 | /* setup of ch structures failed */ | |
481 | out_2: | |
482 | kfree(part->remote_openclose_args_base); | |
483 | part->remote_openclose_args = NULL; | |
484 | out_1: | |
485 | kfree(part->channels); | |
486 | part->channels = NULL; | |
487 | return ret; | |
488 | } | |
489 | ||
490 | /* | |
491 | * Teardown the channel structures necessary to support XPartition Communication | |
492 | * between the specified remote partition and the local one. | |
493 | */ | |
494 | static void | |
495 | xpc_teardown_ch_structures(struct xpc_partition *part) | |
496 | { | |
497 | DBUG_ON(atomic_read(&part->nchannels_engaged) != 0); | |
498 | DBUG_ON(atomic_read(&part->nchannels_active) != 0); | |
499 | ||
500 | /* | |
501 | * Make this partition inaccessible to local processes by marking it | |
502 | * as no longer setup. Then wait before proceeding with the teardown | |
503 | * until all existing references cease. | |
504 | */ | |
505 | DBUG_ON(part->setup_state != XPC_P_SS_SETUP); | |
506 | part->setup_state = XPC_P_SS_WTEARDOWN; | |
507 | ||
508 | wait_event(part->teardown_wq, (atomic_read(&part->references) == 0)); | |
509 | ||
510 | /* now we can begin tearing down the infrastructure */ | |
511 | ||
a7665b0a | 512 | xpc_arch_ops.teardown_ch_structures(part); |
5b8669df DN |
513 | |
514 | kfree(part->remote_openclose_args_base); | |
515 | part->remote_openclose_args = NULL; | |
516 | kfree(part->channels); | |
517 | part->channels = NULL; | |
518 | ||
519 | part->setup_state = XPC_P_SS_TORNDOWN; | |
520 | } | |
521 | ||
89eb8eb9 DN |
522 | /* |
523 | * When XPC HB determines that a partition has come up, it will create a new | |
524 | * kthread and that kthread will call this function to attempt to set up the | |
525 | * basic infrastructure used for Cross Partition Communication with the newly | |
526 | * upped partition. | |
527 | * | |
528 | * The kthread that was created by XPC HB and which setup the XPC | |
e17d416b DN |
529 | * infrastructure will remain assigned to the partition becoming the channel |
530 | * manager for that partition until the partition is deactivating, at which | |
531 | * time the kthread will teardown the XPC infrastructure and then exit. | |
89eb8eb9 | 532 | */ |
89eb8eb9 DN |
533 | static int |
534 | xpc_activating(void *__partid) | |
535 | { | |
64d032ba | 536 | short partid = (u64)__partid; |
89eb8eb9 DN |
537 | struct xpc_partition *part = &xpc_partitions[partid]; |
538 | unsigned long irq_flags; | |
89eb8eb9 | 539 | |
bc63d387 | 540 | DBUG_ON(partid < 0 || partid >= xp_max_npartitions); |
89eb8eb9 DN |
541 | |
542 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
543 | ||
83469b55 DN |
544 | if (part->act_state == XPC_P_AS_DEACTIVATING) { |
545 | part->act_state = XPC_P_AS_INACTIVE; | |
89eb8eb9 DN |
546 | spin_unlock_irqrestore(&part->act_lock, irq_flags); |
547 | part->remote_rp_pa = 0; | |
548 | return 0; | |
549 | } | |
550 | ||
551 | /* indicate the thread is activating */ | |
83469b55 DN |
552 | DBUG_ON(part->act_state != XPC_P_AS_ACTIVATION_REQ); |
553 | part->act_state = XPC_P_AS_ACTIVATING; | |
89eb8eb9 DN |
554 | |
555 | XPC_SET_REASON(part, 0, 0); | |
556 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
557 | ||
e17d416b | 558 | dev_dbg(xpc_part, "activating partition %d\n", partid); |
89eb8eb9 | 559 | |
a7665b0a | 560 | xpc_arch_ops.allow_hb(partid); |
89eb8eb9 | 561 | |
5b8669df | 562 | if (xpc_setup_ch_structures(part) == xpSuccess) { |
e17d416b DN |
563 | (void)xpc_part_ref(part); /* this will always succeed */ |
564 | ||
a7665b0a | 565 | if (xpc_arch_ops.make_first_contact(part) == xpSuccess) { |
e17d416b DN |
566 | xpc_mark_partition_active(part); |
567 | xpc_channel_mgr(part); | |
568 | /* won't return until partition is deactivating */ | |
569 | } | |
570 | ||
571 | xpc_part_deref(part); | |
5b8669df | 572 | xpc_teardown_ch_structures(part); |
e17d416b | 573 | } |
89eb8eb9 | 574 | |
a7665b0a | 575 | xpc_arch_ops.disallow_hb(partid); |
89eb8eb9 DN |
576 | xpc_mark_partition_inactive(part); |
577 | ||
65c17b80 | 578 | if (part->reason == xpReactivating) { |
89eb8eb9 | 579 | /* interrupting ourselves results in activating partition */ |
a7665b0a | 580 | xpc_arch_ops.request_partition_reactivation(part); |
89eb8eb9 DN |
581 | } |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
89eb8eb9 DN |
586 | void |
587 | xpc_activate_partition(struct xpc_partition *part) | |
588 | { | |
64d032ba | 589 | short partid = XPC_PARTID(part); |
89eb8eb9 | 590 | unsigned long irq_flags; |
2c2b94f9 | 591 | struct task_struct *kthread; |
89eb8eb9 | 592 | |
89eb8eb9 DN |
593 | spin_lock_irqsave(&part->act_lock, irq_flags); |
594 | ||
83469b55 | 595 | DBUG_ON(part->act_state != XPC_P_AS_INACTIVE); |
89eb8eb9 | 596 | |
83469b55 | 597 | part->act_state = XPC_P_AS_ACTIVATION_REQ; |
65c17b80 | 598 | XPC_SET_REASON(part, xpCloneKThread, __LINE__); |
89eb8eb9 DN |
599 | |
600 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
7c6c6636 | 601 | |
2c2b94f9 DN |
602 | kthread = kthread_run(xpc_activating, (void *)((u64)partid), "xpc%02d", |
603 | partid); | |
604 | if (IS_ERR(kthread)) { | |
7c6c6636 | 605 | spin_lock_irqsave(&part->act_lock, irq_flags); |
83469b55 | 606 | part->act_state = XPC_P_AS_INACTIVE; |
65c17b80 | 607 | XPC_SET_REASON(part, xpCloneKThreadFailed, __LINE__); |
7c6c6636 RH |
608 | spin_unlock_irqrestore(&part->act_lock, irq_flags); |
609 | } | |
89eb8eb9 DN |
610 | } |
611 | ||
89eb8eb9 DN |
612 | void |
613 | xpc_activate_kthreads(struct xpc_channel *ch, int needed) | |
614 | { | |
615 | int idle = atomic_read(&ch->kthreads_idle); | |
616 | int assigned = atomic_read(&ch->kthreads_assigned); | |
617 | int wakeup; | |
618 | ||
89eb8eb9 DN |
619 | DBUG_ON(needed <= 0); |
620 | ||
621 | if (idle > 0) { | |
622 | wakeup = (needed > idle) ? idle : needed; | |
623 | needed -= wakeup; | |
624 | ||
625 | dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, " | |
626 | "channel=%d\n", wakeup, ch->partid, ch->number); | |
627 | ||
628 | /* only wakeup the requested number of kthreads */ | |
629 | wake_up_nr(&ch->idle_wq, wakeup); | |
630 | } | |
631 | ||
2c2b94f9 | 632 | if (needed <= 0) |
89eb8eb9 | 633 | return; |
89eb8eb9 DN |
634 | |
635 | if (needed + assigned > ch->kthreads_assigned_limit) { | |
636 | needed = ch->kthreads_assigned_limit - assigned; | |
2c2b94f9 | 637 | if (needed <= 0) |
89eb8eb9 | 638 | return; |
89eb8eb9 DN |
639 | } |
640 | ||
641 | dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n", | |
642 | needed, ch->partid, ch->number); | |
643 | ||
a460ef8d | 644 | xpc_create_kthreads(ch, needed, 0); |
89eb8eb9 DN |
645 | } |
646 | ||
89eb8eb9 DN |
647 | /* |
648 | * This function is where XPC's kthreads wait for messages to deliver. | |
649 | */ | |
650 | static void | |
651 | xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch) | |
652 | { | |
a7665b0a RH |
653 | int (*n_of_deliverable_payloads) (struct xpc_channel *) = |
654 | xpc_arch_ops.n_of_deliverable_payloads; | |
655 | ||
89eb8eb9 DN |
656 | do { |
657 | /* deliver messages to their intended recipients */ | |
658 | ||
a7665b0a | 659 | while (n_of_deliverable_payloads(ch) > 0 && |
2c2b94f9 | 660 | !(ch->flags & XPC_C_DISCONNECTING)) { |
bd3e64c1 | 661 | xpc_deliver_payload(ch); |
89eb8eb9 DN |
662 | } |
663 | ||
664 | if (atomic_inc_return(&ch->kthreads_idle) > | |
35190506 | 665 | ch->kthreads_idle_limit) { |
89eb8eb9 DN |
666 | /* too many idle kthreads on this channel */ |
667 | atomic_dec(&ch->kthreads_idle); | |
668 | break; | |
669 | } | |
670 | ||
671 | dev_dbg(xpc_chan, "idle kthread calling " | |
672 | "wait_event_interruptible_exclusive()\n"); | |
673 | ||
35190506 | 674 | (void)wait_event_interruptible_exclusive(ch->idle_wq, |
a7665b0a | 675 | (n_of_deliverable_payloads(ch) > 0 || |
2c2b94f9 | 676 | (ch->flags & XPC_C_DISCONNECTING))); |
89eb8eb9 DN |
677 | |
678 | atomic_dec(&ch->kthreads_idle); | |
679 | ||
2c2b94f9 | 680 | } while (!(ch->flags & XPC_C_DISCONNECTING)); |
89eb8eb9 DN |
681 | } |
682 | ||
89eb8eb9 | 683 | static int |
2c2b94f9 | 684 | xpc_kthread_start(void *args) |
89eb8eb9 | 685 | { |
64d032ba | 686 | short partid = XPC_UNPACK_ARG1(args); |
89eb8eb9 DN |
687 | u16 ch_number = XPC_UNPACK_ARG2(args); |
688 | struct xpc_partition *part = &xpc_partitions[partid]; | |
689 | struct xpc_channel *ch; | |
690 | int n_needed; | |
e54af724 | 691 | unsigned long irq_flags; |
a7665b0a RH |
692 | int (*n_of_deliverable_payloads) (struct xpc_channel *) = |
693 | xpc_arch_ops.n_of_deliverable_payloads; | |
89eb8eb9 | 694 | |
89eb8eb9 DN |
695 | dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n", |
696 | partid, ch_number); | |
697 | ||
698 | ch = &part->channels[ch_number]; | |
699 | ||
700 | if (!(ch->flags & XPC_C_DISCONNECTING)) { | |
89eb8eb9 DN |
701 | |
702 | /* let registerer know that connection has been established */ | |
703 | ||
e54af724 | 704 | spin_lock_irqsave(&ch->lock, irq_flags); |
4c2cd966 DN |
705 | if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) { |
706 | ch->flags |= XPC_C_CONNECTEDCALLOUT; | |
e54af724 DN |
707 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
708 | ||
89eb8eb9 DN |
709 | xpc_connected_callout(ch); |
710 | ||
4c2cd966 DN |
711 | spin_lock_irqsave(&ch->lock, irq_flags); |
712 | ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE; | |
713 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
714 | ||
89eb8eb9 DN |
715 | /* |
716 | * It is possible that while the callout was being | |
717 | * made that the remote partition sent some messages. | |
718 | * If that is the case, we may need to activate | |
719 | * additional kthreads to help deliver them. We only | |
720 | * need one less than total #of messages to deliver. | |
721 | */ | |
a7665b0a | 722 | n_needed = n_of_deliverable_payloads(ch) - 1; |
2c2b94f9 | 723 | if (n_needed > 0 && !(ch->flags & XPC_C_DISCONNECTING)) |
89eb8eb9 | 724 | xpc_activate_kthreads(ch, n_needed); |
2c2b94f9 | 725 | |
e54af724 DN |
726 | } else { |
727 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
89eb8eb9 DN |
728 | } |
729 | ||
730 | xpc_kthread_waitmsgs(part, ch); | |
731 | } | |
732 | ||
a460ef8d | 733 | /* let registerer know that connection is disconnecting */ |
e54af724 | 734 | |
a460ef8d DN |
735 | spin_lock_irqsave(&ch->lock, irq_flags); |
736 | if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && | |
35190506 | 737 | !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { |
a460ef8d | 738 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT; |
4c2cd966 | 739 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
a460ef8d | 740 | |
65c17b80 | 741 | xpc_disconnect_callout(ch, xpDisconnecting); |
a460ef8d DN |
742 | |
743 | spin_lock_irqsave(&ch->lock, irq_flags); | |
744 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE; | |
745 | } | |
746 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
747 | ||
a47d5dac DN |
748 | if (atomic_dec_return(&ch->kthreads_assigned) == 0 && |
749 | atomic_dec_return(&part->nchannels_engaged) == 0) { | |
a7665b0a | 750 | xpc_arch_ops.indicate_partition_disengaged(part); |
89eb8eb9 DN |
751 | } |
752 | ||
89eb8eb9 DN |
753 | xpc_msgqueue_deref(ch); |
754 | ||
755 | dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n", | |
756 | partid, ch_number); | |
757 | ||
758 | xpc_part_deref(part); | |
759 | return 0; | |
760 | } | |
761 | ||
89eb8eb9 DN |
762 | /* |
763 | * For each partition that XPC has established communications with, there is | |
764 | * a minimum of one kernel thread assigned to perform any operation that | |
765 | * may potentially sleep or block (basically the callouts to the asynchronous | |
766 | * functions registered via xpc_connect()). | |
767 | * | |
768 | * Additional kthreads are created and destroyed by XPC as the workload | |
769 | * demands. | |
770 | * | |
771 | * A kthread is assigned to one of the active channels that exists for a given | |
772 | * partition. | |
773 | */ | |
774 | void | |
a460ef8d | 775 | xpc_create_kthreads(struct xpc_channel *ch, int needed, |
35190506 | 776 | int ignore_disconnecting) |
89eb8eb9 DN |
777 | { |
778 | unsigned long irq_flags; | |
89eb8eb9 | 779 | u64 args = XPC_PACK_ARGS(ch->partid, ch->number); |
a607c389 | 780 | struct xpc_partition *part = &xpc_partitions[ch->partid]; |
2c2b94f9 | 781 | struct task_struct *kthread; |
a7665b0a RH |
782 | void (*indicate_partition_disengaged) (struct xpc_partition *) = |
783 | xpc_arch_ops.indicate_partition_disengaged; | |
89eb8eb9 | 784 | |
89eb8eb9 | 785 | while (needed-- > 0) { |
e54af724 DN |
786 | |
787 | /* | |
788 | * The following is done on behalf of the newly created | |
789 | * kthread. That kthread is responsible for doing the | |
790 | * counterpart to the following before it exits. | |
791 | */ | |
a460ef8d DN |
792 | if (ignore_disconnecting) { |
793 | if (!atomic_inc_not_zero(&ch->kthreads_assigned)) { | |
794 | /* kthreads assigned had gone to zero */ | |
795 | BUG_ON(!(ch->flags & | |
35190506 | 796 | XPC_C_DISCONNECTINGCALLOUT_MADE)); |
a460ef8d DN |
797 | break; |
798 | } | |
799 | ||
800 | } else if (ch->flags & XPC_C_DISCONNECTING) { | |
801 | break; | |
802 | ||
a47d5dac DN |
803 | } else if (atomic_inc_return(&ch->kthreads_assigned) == 1 && |
804 | atomic_inc_return(&part->nchannels_engaged) == 1) { | |
a7665b0a | 805 | xpc_arch_ops.indicate_partition_engaged(part); |
a460ef8d | 806 | } |
35190506 | 807 | (void)xpc_part_ref(part); |
e54af724 | 808 | xpc_msgqueue_ref(ch); |
e54af724 | 809 | |
2c2b94f9 DN |
810 | kthread = kthread_run(xpc_kthread_start, (void *)args, |
811 | "xpc%02dc%d", ch->partid, ch->number); | |
812 | if (IS_ERR(kthread)) { | |
89eb8eb9 | 813 | /* the fork failed */ |
a460ef8d DN |
814 | |
815 | /* | |
816 | * NOTE: if (ignore_disconnecting && | |
817 | * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true, | |
818 | * then we'll deadlock if all other kthreads assigned | |
819 | * to this channel are blocked in the channel's | |
820 | * registerer, because the only thing that will unblock | |
65c17b80 | 821 | * them is the xpDisconnecting callout that this |
2c2b94f9 | 822 | * failed kthread_run() would have made. |
a460ef8d DN |
823 | */ |
824 | ||
e54af724 DN |
825 | if (atomic_dec_return(&ch->kthreads_assigned) == 0 && |
826 | atomic_dec_return(&part->nchannels_engaged) == 0) { | |
a7665b0a | 827 | indicate_partition_disengaged(part); |
e54af724 DN |
828 | } |
829 | xpc_msgqueue_deref(ch); | |
830 | xpc_part_deref(part); | |
89eb8eb9 DN |
831 | |
832 | if (atomic_read(&ch->kthreads_assigned) < | |
35190506 | 833 | ch->kthreads_idle_limit) { |
89eb8eb9 DN |
834 | /* |
835 | * Flag this as an error only if we have an | |
836 | * insufficient #of kthreads for the channel | |
837 | * to function. | |
89eb8eb9 DN |
838 | */ |
839 | spin_lock_irqsave(&ch->lock, irq_flags); | |
65c17b80 | 840 | XPC_DISCONNECT_CHANNEL(ch, xpLackOfResources, |
35190506 | 841 | &irq_flags); |
89eb8eb9 DN |
842 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
843 | } | |
844 | break; | |
845 | } | |
89eb8eb9 DN |
846 | } |
847 | } | |
848 | ||
89eb8eb9 DN |
849 | void |
850 | xpc_disconnect_wait(int ch_number) | |
851 | { | |
a607c389 | 852 | unsigned long irq_flags; |
64d032ba | 853 | short partid; |
89eb8eb9 DN |
854 | struct xpc_partition *part; |
855 | struct xpc_channel *ch; | |
e54af724 | 856 | int wakeup_channel_mgr; |
89eb8eb9 | 857 | |
89eb8eb9 | 858 | /* now wait for all callouts to the caller's function to cease */ |
bc63d387 | 859 | for (partid = 0; partid < xp_max_npartitions; partid++) { |
89eb8eb9 DN |
860 | part = &xpc_partitions[partid]; |
861 | ||
2c2b94f9 | 862 | if (!xpc_part_ref(part)) |
e54af724 | 863 | continue; |
89eb8eb9 | 864 | |
e54af724 | 865 | ch = &part->channels[ch_number]; |
89eb8eb9 | 866 | |
e54af724 | 867 | if (!(ch->flags & XPC_C_WDISCONNECT)) { |
89eb8eb9 | 868 | xpc_part_deref(part); |
e54af724 | 869 | continue; |
89eb8eb9 | 870 | } |
e54af724 | 871 | |
f9e505a9 | 872 | wait_for_completion(&ch->wdisconnect_wait); |
e54af724 DN |
873 | |
874 | spin_lock_irqsave(&ch->lock, irq_flags); | |
875 | DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED)); | |
876 | wakeup_channel_mgr = 0; | |
877 | ||
7fb5e59d | 878 | if (ch->delayed_chctl_flags) { |
83469b55 | 879 | if (part->act_state != XPC_P_AS_DEACTIVATING) { |
7fb5e59d DN |
880 | spin_lock(&part->chctl_lock); |
881 | part->chctl.flags[ch->number] |= | |
882 | ch->delayed_chctl_flags; | |
883 | spin_unlock(&part->chctl_lock); | |
e54af724 DN |
884 | wakeup_channel_mgr = 1; |
885 | } | |
7fb5e59d | 886 | ch->delayed_chctl_flags = 0; |
89eb8eb9 | 887 | } |
e54af724 DN |
888 | |
889 | ch->flags &= ~XPC_C_WDISCONNECT; | |
890 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
891 | ||
2c2b94f9 | 892 | if (wakeup_channel_mgr) |
e54af724 | 893 | xpc_wakeup_channel_mgr(part); |
e54af724 DN |
894 | |
895 | xpc_part_deref(part); | |
89eb8eb9 DN |
896 | } |
897 | } | |
898 | ||
5b8669df DN |
899 | static int |
900 | xpc_setup_partitions(void) | |
901 | { | |
902 | short partid; | |
903 | struct xpc_partition *part; | |
904 | ||
905 | xpc_partitions = kzalloc(sizeof(struct xpc_partition) * | |
906 | xp_max_npartitions, GFP_KERNEL); | |
907 | if (xpc_partitions == NULL) { | |
908 | dev_err(xpc_part, "can't get memory for partition structure\n"); | |
909 | return -ENOMEM; | |
910 | } | |
911 | ||
912 | /* | |
913 | * The first few fields of each entry of xpc_partitions[] need to | |
914 | * be initialized now so that calls to xpc_connect() and | |
915 | * xpc_disconnect() can be made prior to the activation of any remote | |
916 | * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE | |
917 | * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING | |
918 | * PARTITION HAS BEEN ACTIVATED. | |
919 | */ | |
920 | for (partid = 0; partid < xp_max_npartitions; partid++) { | |
921 | part = &xpc_partitions[partid]; | |
922 | ||
923 | DBUG_ON((u64)part != L1_CACHE_ALIGN((u64)part)); | |
924 | ||
925 | part->activate_IRQ_rcvd = 0; | |
926 | spin_lock_init(&part->act_lock); | |
927 | part->act_state = XPC_P_AS_INACTIVE; | |
928 | XPC_SET_REASON(part, 0, 0); | |
929 | ||
930 | init_timer(&part->disengage_timer); | |
931 | part->disengage_timer.function = | |
932 | xpc_timeout_partition_disengage; | |
933 | part->disengage_timer.data = (unsigned long)part; | |
934 | ||
935 | part->setup_state = XPC_P_SS_UNSET; | |
936 | init_waitqueue_head(&part->teardown_wq); | |
937 | atomic_set(&part->references, 0); | |
938 | } | |
939 | ||
a7665b0a | 940 | return xpc_arch_ops.setup_partitions(); |
5b8669df DN |
941 | } |
942 | ||
943 | static void | |
944 | xpc_teardown_partitions(void) | |
945 | { | |
a7665b0a | 946 | xpc_arch_ops.teardown_partitions(); |
5b8669df DN |
947 | kfree(xpc_partitions); |
948 | } | |
949 | ||
89eb8eb9 | 950 | static void |
65c17b80 | 951 | xpc_do_exit(enum xp_retval reason) |
89eb8eb9 | 952 | { |
64d032ba | 953 | short partid; |
1ecaded8 | 954 | int active_part_count, printed_waiting_msg = 0; |
89eb8eb9 | 955 | struct xpc_partition *part; |
a47d5dac | 956 | unsigned long printmsg_time, disengage_timeout = 0; |
89eb8eb9 | 957 | |
a607c389 DN |
958 | /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */ |
959 | DBUG_ON(xpc_exiting == 1); | |
89eb8eb9 DN |
960 | |
961 | /* | |
a607c389 DN |
962 | * Let the heartbeat checker thread and the discovery thread |
963 | * (if one is running) know that they should exit. Also wake up | |
964 | * the heartbeat checker thread in case it's sleeping. | |
89eb8eb9 DN |
965 | */ |
966 | xpc_exiting = 1; | |
6e41017a | 967 | wake_up_interruptible(&xpc_activate_IRQ_wq); |
89eb8eb9 | 968 | |
e54af724 | 969 | /* wait for the discovery thread to exit */ |
f9e505a9 | 970 | wait_for_completion(&xpc_discovery_exited); |
89eb8eb9 | 971 | |
e54af724 | 972 | /* wait for the heartbeat checker thread to exit */ |
f9e505a9 | 973 | wait_for_completion(&xpc_hb_checker_exited); |
89eb8eb9 | 974 | |
a607c389 | 975 | /* sleep for a 1/3 of a second or so */ |
35190506 | 976 | (void)msleep_interruptible(300); |
89eb8eb9 DN |
977 | |
978 | /* wait for all partitions to become inactive */ | |
979 | ||
a47d5dac DN |
980 | printmsg_time = jiffies + (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ); |
981 | xpc_disengage_timedout = 0; | |
a607c389 | 982 | |
89eb8eb9 DN |
983 | do { |
984 | active_part_count = 0; | |
985 | ||
bc63d387 | 986 | for (partid = 0; partid < xp_max_npartitions; partid++) { |
89eb8eb9 | 987 | part = &xpc_partitions[partid]; |
89eb8eb9 | 988 | |
a607c389 | 989 | if (xpc_partition_disengaged(part) && |
83469b55 | 990 | part->act_state == XPC_P_AS_INACTIVE) { |
a607c389 | 991 | continue; |
89eb8eb9 | 992 | } |
a607c389 DN |
993 | |
994 | active_part_count++; | |
995 | ||
996 | XPC_DEACTIVATE_PARTITION(part, reason); | |
89eb8eb9 | 997 | |
a47d5dac DN |
998 | if (part->disengage_timeout > disengage_timeout) |
999 | disengage_timeout = part->disengage_timeout; | |
a607c389 | 1000 | } |
89eb8eb9 | 1001 | |
a7665b0a | 1002 | if (xpc_arch_ops.any_partition_engaged()) { |
aaa3cd69 | 1003 | if (time_is_before_jiffies(printmsg_time)) { |
1ecaded8 | 1004 | dev_info(xpc_part, "waiting for remote " |
a47d5dac DN |
1005 | "partitions to deactivate, timeout in " |
1006 | "%ld seconds\n", (disengage_timeout - | |
1007 | jiffies) / HZ); | |
1ecaded8 | 1008 | printmsg_time = jiffies + |
a47d5dac | 1009 | (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ); |
1ecaded8 DN |
1010 | printed_waiting_msg = 1; |
1011 | } | |
1012 | ||
1013 | } else if (active_part_count > 0) { | |
1014 | if (printed_waiting_msg) { | |
1015 | dev_info(xpc_part, "waiting for local partition" | |
a47d5dac | 1016 | " to deactivate\n"); |
1ecaded8 DN |
1017 | printed_waiting_msg = 0; |
1018 | } | |
1019 | ||
1020 | } else { | |
a47d5dac | 1021 | if (!xpc_disengage_timedout) { |
1ecaded8 | 1022 | dev_info(xpc_part, "all partitions have " |
a47d5dac | 1023 | "deactivated\n"); |
1ecaded8 DN |
1024 | } |
1025 | break; | |
89eb8eb9 DN |
1026 | } |
1027 | ||
a607c389 | 1028 | /* sleep for a 1/3 of a second or so */ |
35190506 | 1029 | (void)msleep_interruptible(300); |
a607c389 DN |
1030 | |
1031 | } while (1); | |
1032 | ||
a7665b0a | 1033 | DBUG_ON(xpc_arch_ops.any_partition_engaged()); |
a607c389 | 1034 | |
5b8669df | 1035 | xpc_teardown_rsvd_page(); |
a607c389 | 1036 | |
65c17b80 | 1037 | if (reason == xpUnloading) { |
35190506 | 1038 | (void)unregister_die_notifier(&xpc_die_notifier); |
bc63d387 | 1039 | (void)unregister_reboot_notifier(&xpc_reboot_notifier); |
0752c670 | 1040 | } |
780d09e8 | 1041 | |
89eb8eb9 DN |
1042 | /* clear the interface to XPC's functions */ |
1043 | xpc_clear_interface(); | |
1044 | ||
2c2b94f9 | 1045 | if (xpc_sysctl) |
89eb8eb9 | 1046 | unregister_sysctl_table(xpc_sysctl); |
7682a4c6 | 1047 | |
5b8669df | 1048 | xpc_teardown_partitions(); |
6e41017a DN |
1049 | |
1050 | if (is_shub()) | |
1051 | xpc_exit_sn2(); | |
b7f7b074 | 1052 | else if (is_uv()) |
6e41017a | 1053 | xpc_exit_uv(); |
89eb8eb9 DN |
1054 | } |
1055 | ||
780d09e8 | 1056 | /* |
d6ad033a DN |
1057 | * This function is called when the system is being rebooted. |
1058 | */ | |
1059 | static int | |
1060 | xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused) | |
1061 | { | |
65c17b80 | 1062 | enum xp_retval reason; |
d6ad033a | 1063 | |
d6ad033a DN |
1064 | switch (event) { |
1065 | case SYS_RESTART: | |
65c17b80 | 1066 | reason = xpSystemReboot; |
d6ad033a DN |
1067 | break; |
1068 | case SYS_HALT: | |
65c17b80 | 1069 | reason = xpSystemHalt; |
d6ad033a DN |
1070 | break; |
1071 | case SYS_POWER_OFF: | |
65c17b80 | 1072 | reason = xpSystemPoweroff; |
d6ad033a DN |
1073 | break; |
1074 | default: | |
65c17b80 | 1075 | reason = xpSystemGoingDown; |
d6ad033a DN |
1076 | } |
1077 | ||
1078 | xpc_do_exit(reason); | |
1079 | return NOTIFY_DONE; | |
1080 | } | |
1081 | ||
d6ad033a | 1082 | /* |
a47d5dac DN |
1083 | * Notify other partitions to deactivate from us by first disengaging from all |
1084 | * references to our memory. | |
780d09e8 DN |
1085 | */ |
1086 | static void | |
a47d5dac | 1087 | xpc_die_deactivate(void) |
780d09e8 DN |
1088 | { |
1089 | struct xpc_partition *part; | |
64d032ba | 1090 | short partid; |
a47d5dac | 1091 | int any_engaged; |
261f3b49 DN |
1092 | long keep_waiting; |
1093 | long wait_to_print; | |
780d09e8 | 1094 | |
780d09e8 DN |
1095 | /* keep xpc_hb_checker thread from doing anything (just in case) */ |
1096 | xpc_exiting = 1; | |
1097 | ||
a7665b0a | 1098 | xpc_arch_ops.disallow_all_hbs(); /*indicate we're deactivated */ |
780d09e8 | 1099 | |
bc63d387 | 1100 | for (partid = 0; partid < xp_max_npartitions; partid++) { |
780d09e8 DN |
1101 | part = &xpc_partitions[partid]; |
1102 | ||
a7665b0a | 1103 | if (xpc_arch_ops.partition_engaged(partid) || |
83469b55 | 1104 | part->act_state != XPC_P_AS_INACTIVE) { |
a7665b0a RH |
1105 | xpc_arch_ops.request_partition_deactivation(part); |
1106 | xpc_arch_ops.indicate_partition_disengaged(part); | |
780d09e8 DN |
1107 | } |
1108 | } | |
1109 | ||
a47d5dac DN |
1110 | /* |
1111 | * Though we requested that all other partitions deactivate from us, | |
261f3b49 DN |
1112 | * we only wait until they've all disengaged or we've reached the |
1113 | * defined timelimit. | |
1114 | * | |
1115 | * Given that one iteration through the following while-loop takes | |
1116 | * approximately 200 microseconds, calculate the #of loops to take | |
1117 | * before bailing and the #of loops before printing a waiting message. | |
a47d5dac | 1118 | */ |
261f3b49 DN |
1119 | keep_waiting = xpc_disengage_timelimit * 1000 * 5; |
1120 | wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL * 1000 * 5; | |
780d09e8 | 1121 | |
1ecaded8 | 1122 | while (1) { |
a7665b0a | 1123 | any_engaged = xpc_arch_ops.any_partition_engaged(); |
a47d5dac DN |
1124 | if (!any_engaged) { |
1125 | dev_info(xpc_part, "all partitions have deactivated\n"); | |
1ecaded8 DN |
1126 | break; |
1127 | } | |
780d09e8 | 1128 | |
261f3b49 | 1129 | if (!keep_waiting--) { |
bc63d387 DN |
1130 | for (partid = 0; partid < xp_max_npartitions; |
1131 | partid++) { | |
a7665b0a | 1132 | if (xpc_arch_ops.partition_engaged(partid)) { |
a47d5dac | 1133 | dev_info(xpc_part, "deactivate from " |
35190506 DN |
1134 | "remote partition %d timed " |
1135 | "out\n", partid); | |
1ecaded8 DN |
1136 | } |
1137 | } | |
1138 | break; | |
1139 | } | |
1140 | ||
261f3b49 | 1141 | if (!wait_to_print--) { |
780d09e8 | 1142 | dev_info(xpc_part, "waiting for remote partitions to " |
a47d5dac | 1143 | "deactivate, timeout in %ld seconds\n", |
261f3b49 DN |
1144 | keep_waiting / (1000 * 5)); |
1145 | wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL * | |
1146 | 1000 * 5; | |
780d09e8 | 1147 | } |
261f3b49 DN |
1148 | |
1149 | udelay(200); | |
780d09e8 | 1150 | } |
780d09e8 DN |
1151 | } |
1152 | ||
780d09e8 | 1153 | /* |
1f4674b2 DN |
1154 | * This function is called when the system is being restarted or halted due |
1155 | * to some sort of system failure. If this is the case we need to notify the | |
1156 | * other partitions to disengage from all references to our memory. | |
1157 | * This function can also be called when our heartbeater could be offlined | |
1158 | * for a time. In this case we need to notify other partitions to not worry | |
1159 | * about the lack of a heartbeat. | |
780d09e8 DN |
1160 | */ |
1161 | static int | |
1162 | xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused) | |
1163 | { | |
261f3b49 | 1164 | #ifdef CONFIG_IA64 /* !!! temporary kludge */ |
780d09e8 DN |
1165 | switch (event) { |
1166 | case DIE_MACHINE_RESTART: | |
1167 | case DIE_MACHINE_HALT: | |
a47d5dac | 1168 | xpc_die_deactivate(); |
780d09e8 | 1169 | break; |
1f4674b2 DN |
1170 | |
1171 | case DIE_KDEBUG_ENTER: | |
1172 | /* Should lack of heartbeat be ignored by other partitions? */ | |
2c2b94f9 | 1173 | if (!xpc_kdebug_ignore) |
1f4674b2 | 1174 | break; |
2c2b94f9 | 1175 | |
1f4674b2 | 1176 | /* fall through */ |
780d09e8 DN |
1177 | case DIE_MCA_MONARCH_ENTER: |
1178 | case DIE_INIT_MONARCH_ENTER: | |
a7665b0a | 1179 | xpc_arch_ops.offline_heartbeat(); |
780d09e8 | 1180 | break; |
1f4674b2 DN |
1181 | |
1182 | case DIE_KDEBUG_LEAVE: | |
1183 | /* Is lack of heartbeat being ignored by other partitions? */ | |
2c2b94f9 | 1184 | if (!xpc_kdebug_ignore) |
1f4674b2 | 1185 | break; |
2c2b94f9 | 1186 | |
1f4674b2 | 1187 | /* fall through */ |
780d09e8 DN |
1188 | case DIE_MCA_MONARCH_LEAVE: |
1189 | case DIE_INIT_MONARCH_LEAVE: | |
a7665b0a | 1190 | xpc_arch_ops.online_heartbeat(); |
780d09e8 DN |
1191 | break; |
1192 | } | |
261f3b49 DN |
1193 | #else |
1194 | xpc_die_deactivate(); | |
1195 | #endif | |
780d09e8 DN |
1196 | |
1197 | return NOTIFY_DONE; | |
1198 | } | |
1199 | ||
89eb8eb9 DN |
1200 | int __init |
1201 | xpc_init(void) | |
1202 | { | |
1203 | int ret; | |
2c2b94f9 | 1204 | struct task_struct *kthread; |
ee6665e3 | 1205 | |
bb0dc43e KS |
1206 | dev_set_name(xpc_part, "part"); |
1207 | dev_set_name(xpc_chan, "chan"); | |
89eb8eb9 | 1208 | |
94bd2708 DN |
1209 | if (is_shub()) { |
1210 | /* | |
1211 | * The ia64-sn2 architecture supports at most 64 partitions. | |
c39838ce | 1212 | * And the inability to unregister remote amos restricts us |
94bd2708 DN |
1213 | * further to only support exactly 64 partitions on this |
1214 | * architecture, no less. | |
1215 | */ | |
5b8669df DN |
1216 | if (xp_max_npartitions != 64) { |
1217 | dev_err(xpc_part, "max #of partitions not set to 64\n"); | |
1218 | ret = -EINVAL; | |
1219 | } else { | |
1220 | ret = xpc_init_sn2(); | |
1221 | } | |
94bd2708 DN |
1222 | |
1223 | } else if (is_uv()) { | |
5b8669df | 1224 | ret = xpc_init_uv(); |
94bd2708 DN |
1225 | |
1226 | } else { | |
5b8669df | 1227 | ret = -ENODEV; |
94bd2708 | 1228 | } |
408865ce | 1229 | |
5b8669df DN |
1230 | if (ret != 0) |
1231 | return ret; | |
1232 | ||
1233 | ret = xpc_setup_partitions(); | |
1234 | if (ret != 0) { | |
bc63d387 | 1235 | dev_err(xpc_part, "can't get memory for partition structure\n"); |
ee6665e3 | 1236 | goto out_1; |
bc63d387 | 1237 | } |
89eb8eb9 | 1238 | |
bc63d387 DN |
1239 | xpc_sysctl = register_sysctl_table(xpc_sys_dir); |
1240 | ||
89eb8eb9 DN |
1241 | /* |
1242 | * Fill the partition reserved page with the information needed by | |
1243 | * other partitions to discover we are alive and establish initial | |
1244 | * communications. | |
1245 | */ | |
5b8669df DN |
1246 | ret = xpc_setup_rsvd_page(); |
1247 | if (ret != 0) { | |
bc63d387 | 1248 | dev_err(xpc_part, "can't setup our reserved page\n"); |
ee6665e3 | 1249 | goto out_2; |
89eb8eb9 DN |
1250 | } |
1251 | ||
a607c389 DN |
1252 | /* add ourselves to the reboot_notifier_list */ |
1253 | ret = register_reboot_notifier(&xpc_reboot_notifier); | |
2c2b94f9 | 1254 | if (ret != 0) |
a607c389 | 1255 | dev_warn(xpc_part, "can't register reboot notifier\n"); |
a607c389 | 1256 | |
1eeb66a1 | 1257 | /* add ourselves to the die_notifier list */ |
780d09e8 | 1258 | ret = register_die_notifier(&xpc_die_notifier); |
2c2b94f9 | 1259 | if (ret != 0) |
780d09e8 | 1260 | dev_warn(xpc_part, "can't register die notifier\n"); |
780d09e8 | 1261 | |
89eb8eb9 DN |
1262 | /* |
1263 | * The real work-horse behind xpc. This processes incoming | |
1264 | * interrupts and monitors remote heartbeats. | |
1265 | */ | |
2c2b94f9 DN |
1266 | kthread = kthread_run(xpc_hb_checker, NULL, XPC_HB_CHECK_THREAD_NAME); |
1267 | if (IS_ERR(kthread)) { | |
89eb8eb9 | 1268 | dev_err(xpc_part, "failed while forking hb check thread\n"); |
bc63d387 | 1269 | ret = -EBUSY; |
ee6665e3 | 1270 | goto out_3; |
89eb8eb9 DN |
1271 | } |
1272 | ||
89eb8eb9 DN |
1273 | /* |
1274 | * Startup a thread that will attempt to discover other partitions to | |
1275 | * activate based on info provided by SAL. This new thread is short | |
1276 | * lived and will exit once discovery is complete. | |
1277 | */ | |
2c2b94f9 DN |
1278 | kthread = kthread_run(xpc_initiate_discovery, NULL, |
1279 | XPC_DISCOVERY_THREAD_NAME); | |
1280 | if (IS_ERR(kthread)) { | |
89eb8eb9 DN |
1281 | dev_err(xpc_part, "failed while forking discovery thread\n"); |
1282 | ||
1283 | /* mark this new thread as a non-starter */ | |
f9e505a9 | 1284 | complete(&xpc_discovery_exited); |
89eb8eb9 | 1285 | |
65c17b80 | 1286 | xpc_do_exit(xpUnloading); |
89eb8eb9 DN |
1287 | return -EBUSY; |
1288 | } | |
1289 | ||
89eb8eb9 DN |
1290 | /* set the interface to point at XPC's functions */ |
1291 | xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect, | |
97bf1aa1 DN |
1292 | xpc_initiate_send, xpc_initiate_send_notify, |
1293 | xpc_initiate_received, xpc_initiate_partid_to_nasids); | |
89eb8eb9 DN |
1294 | |
1295 | return 0; | |
bc63d387 DN |
1296 | |
1297 | /* initialization was not successful */ | |
ee6665e3 | 1298 | out_3: |
5b8669df | 1299 | xpc_teardown_rsvd_page(); |
94bd2708 | 1300 | |
bc63d387 DN |
1301 | (void)unregister_die_notifier(&xpc_die_notifier); |
1302 | (void)unregister_reboot_notifier(&xpc_reboot_notifier); | |
ee6665e3 | 1303 | out_2: |
bc63d387 DN |
1304 | if (xpc_sysctl) |
1305 | unregister_sysctl_table(xpc_sysctl); | |
5b8669df DN |
1306 | |
1307 | xpc_teardown_partitions(); | |
6e41017a DN |
1308 | out_1: |
1309 | if (is_shub()) | |
1310 | xpc_exit_sn2(); | |
b7f7b074 | 1311 | else if (is_uv()) |
6e41017a | 1312 | xpc_exit_uv(); |
bc63d387 | 1313 | return ret; |
89eb8eb9 | 1314 | } |
89eb8eb9 | 1315 | |
35190506 | 1316 | module_init(xpc_init); |
89eb8eb9 DN |
1317 | |
1318 | void __exit | |
1319 | xpc_exit(void) | |
1320 | { | |
65c17b80 | 1321 | xpc_do_exit(xpUnloading); |
89eb8eb9 | 1322 | } |
89eb8eb9 | 1323 | |
35190506 | 1324 | module_exit(xpc_exit); |
89eb8eb9 DN |
1325 | |
1326 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
1327 | MODULE_DESCRIPTION("Cross Partition Communication (XPC) support"); | |
1328 | MODULE_LICENSE("GPL"); | |
1329 | ||
1330 | module_param(xpc_hb_interval, int, 0); | |
1331 | MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between " | |
35190506 | 1332 | "heartbeat increments."); |
89eb8eb9 DN |
1333 | |
1334 | module_param(xpc_hb_check_interval, int, 0); | |
1335 | MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between " | |
35190506 | 1336 | "heartbeat checks."); |
89eb8eb9 | 1337 | |
a47d5dac DN |
1338 | module_param(xpc_disengage_timelimit, int, 0); |
1339 | MODULE_PARM_DESC(xpc_disengage_timelimit, "Number of seconds to wait " | |
1340 | "for disengage to complete."); | |
e54af724 | 1341 | |
1f4674b2 DN |
1342 | module_param(xpc_kdebug_ignore, int, 0); |
1343 | MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by " | |
35190506 | 1344 | "other partitions when dropping into kdebug."); |