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 | * | |
9335d48e | 6 | * Copyright (c) 2004-2006 Silicon Graphics, Inc. All Rights Reserved. |
89eb8eb9 DN |
7 | */ |
8 | ||
9 | ||
10 | /* | |
11 | * Cross Partition Communication (XPC) support - standard version. | |
12 | * | |
13 | * XPC provides a message passing capability that crosses partition | |
14 | * boundaries. This module is made up of two parts: | |
15 | * | |
16 | * partition This part detects the presence/absence of other | |
17 | * partitions. It provides a heartbeat and monitors | |
18 | * the heartbeats of other partitions. | |
19 | * | |
20 | * channel This part manages the channels and sends/receives | |
21 | * messages across them to/from other partitions. | |
22 | * | |
23 | * There are a couple of additional functions residing in XP, which | |
24 | * provide an interface to XPC for its users. | |
25 | * | |
26 | * | |
27 | * Caveats: | |
28 | * | |
29 | * . We currently have no way to determine which nasid an IPI came | |
30 | * from. Thus, xpc_IPI_send() does a remote AMO write followed by | |
31 | * an IPI. The AMO indicates where data is to be pulled from, so | |
32 | * after the IPI arrives, the remote partition checks the AMO word. | |
33 | * The IPI can actually arrive before the AMO however, so other code | |
34 | * must periodically check for this case. Also, remote AMO operations | |
35 | * do not reliably time out. Thus we do a remote PIO read solely to | |
36 | * know whether the remote partition is down and whether we should | |
37 | * stop sending IPIs to it. This remote PIO read operation is set up | |
38 | * in a special nofault region so SAL knows to ignore (and cleanup) | |
39 | * any errors due to the remote AMO write, PIO read, and/or PIO | |
40 | * write operations. | |
41 | * | |
42 | * If/when new hardware solves this IPI problem, we should abandon | |
43 | * the current approach. | |
44 | * | |
45 | */ | |
46 | ||
47 | ||
48 | #include <linux/kernel.h> | |
49 | #include <linux/module.h> | |
50 | #include <linux/init.h> | |
51 | #include <linux/sched.h> | |
52 | #include <linux/syscalls.h> | |
53 | #include <linux/cache.h> | |
54 | #include <linux/interrupt.h> | |
69913927 | 55 | #include <linux/delay.h> |
a607c389 | 56 | #include <linux/reboot.h> |
f9e505a9 | 57 | #include <linux/completion.h> |
89eb8eb9 DN |
58 | #include <asm/sn/intr.h> |
59 | #include <asm/sn/sn_sal.h> | |
780d09e8 | 60 | #include <asm/kdebug.h> |
89eb8eb9 | 61 | #include <asm/uaccess.h> |
9335d48e | 62 | #include <asm/sn/xpc.h> |
89eb8eb9 DN |
63 | |
64 | ||
65 | /* define two XPC debug device structures to be used with dev_dbg() et al */ | |
66 | ||
67 | struct device_driver xpc_dbg_name = { | |
68 | .name = "xpc" | |
69 | }; | |
70 | ||
71 | struct device xpc_part_dbg_subname = { | |
72 | .bus_id = {0}, /* set to "part" at xpc_init() time */ | |
73 | .driver = &xpc_dbg_name | |
74 | }; | |
75 | ||
76 | struct device xpc_chan_dbg_subname = { | |
77 | .bus_id = {0}, /* set to "chan" at xpc_init() time */ | |
78 | .driver = &xpc_dbg_name | |
79 | }; | |
80 | ||
81 | struct device *xpc_part = &xpc_part_dbg_subname; | |
82 | struct device *xpc_chan = &xpc_chan_dbg_subname; | |
83 | ||
84 | ||
1f4674b2 DN |
85 | static int xpc_kdebug_ignore; |
86 | ||
87 | ||
89eb8eb9 DN |
88 | /* systune related variables for /proc/sys directories */ |
89 | ||
a607c389 DN |
90 | static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL; |
91 | static int xpc_hb_min_interval = 1; | |
92 | static int xpc_hb_max_interval = 10; | |
89eb8eb9 | 93 | |
a607c389 DN |
94 | static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL; |
95 | static int xpc_hb_check_min_interval = 10; | |
96 | static int xpc_hb_check_max_interval = 120; | |
89eb8eb9 | 97 | |
e54af724 DN |
98 | int xpc_disengage_request_timelimit = XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT; |
99 | static int xpc_disengage_request_min_timelimit = 0; | |
100 | static int xpc_disengage_request_max_timelimit = 120; | |
89eb8eb9 DN |
101 | |
102 | static ctl_table xpc_sys_xpc_hb_dir[] = { | |
103 | { | |
104 | 1, | |
105 | "hb_interval", | |
106 | &xpc_hb_interval, | |
107 | sizeof(int), | |
108 | 0644, | |
109 | NULL, | |
110 | &proc_dointvec_minmax, | |
111 | &sysctl_intvec, | |
112 | NULL, | |
a607c389 DN |
113 | &xpc_hb_min_interval, |
114 | &xpc_hb_max_interval | |
89eb8eb9 DN |
115 | }, |
116 | { | |
117 | 2, | |
118 | "hb_check_interval", | |
119 | &xpc_hb_check_interval, | |
120 | sizeof(int), | |
121 | 0644, | |
122 | NULL, | |
123 | &proc_dointvec_minmax, | |
124 | &sysctl_intvec, | |
125 | NULL, | |
a607c389 DN |
126 | &xpc_hb_check_min_interval, |
127 | &xpc_hb_check_max_interval | |
89eb8eb9 DN |
128 | }, |
129 | {0} | |
130 | }; | |
131 | static ctl_table xpc_sys_xpc_dir[] = { | |
132 | { | |
133 | 1, | |
134 | "hb", | |
135 | NULL, | |
136 | 0, | |
137 | 0555, | |
138 | xpc_sys_xpc_hb_dir | |
139 | }, | |
e54af724 DN |
140 | { |
141 | 2, | |
142 | "disengage_request_timelimit", | |
143 | &xpc_disengage_request_timelimit, | |
144 | sizeof(int), | |
145 | 0644, | |
146 | NULL, | |
147 | &proc_dointvec_minmax, | |
148 | &sysctl_intvec, | |
149 | NULL, | |
150 | &xpc_disengage_request_min_timelimit, | |
151 | &xpc_disengage_request_max_timelimit | |
152 | }, | |
89eb8eb9 DN |
153 | {0} |
154 | }; | |
155 | static ctl_table xpc_sys_dir[] = { | |
156 | { | |
157 | 1, | |
158 | "xpc", | |
159 | NULL, | |
160 | 0, | |
161 | 0555, | |
162 | xpc_sys_xpc_dir | |
163 | }, | |
164 | {0} | |
165 | }; | |
166 | static struct ctl_table_header *xpc_sysctl; | |
167 | ||
1ecaded8 DN |
168 | /* non-zero if any remote partition disengage request was timed out */ |
169 | int xpc_disengage_request_timedout; | |
89eb8eb9 DN |
170 | |
171 | /* #of IRQs received */ | |
172 | static atomic_t xpc_act_IRQ_rcvd; | |
173 | ||
174 | /* IRQ handler notifies this wait queue on receipt of an IRQ */ | |
175 | static DECLARE_WAIT_QUEUE_HEAD(xpc_act_IRQ_wq); | |
176 | ||
177 | static unsigned long xpc_hb_check_timeout; | |
178 | ||
e54af724 | 179 | /* notification that the xpc_hb_checker thread has exited */ |
f9e505a9 | 180 | static DECLARE_COMPLETION(xpc_hb_checker_exited); |
89eb8eb9 | 181 | |
e54af724 | 182 | /* notification that the xpc_discovery thread has exited */ |
f9e505a9 | 183 | static DECLARE_COMPLETION(xpc_discovery_exited); |
89eb8eb9 DN |
184 | |
185 | ||
186 | static struct timer_list xpc_hb_timer; | |
187 | ||
188 | ||
189 | static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *); | |
190 | ||
191 | ||
a607c389 DN |
192 | static int xpc_system_reboot(struct notifier_block *, unsigned long, void *); |
193 | static struct notifier_block xpc_reboot_notifier = { | |
194 | .notifier_call = xpc_system_reboot, | |
195 | }; | |
196 | ||
780d09e8 DN |
197 | static int xpc_system_die(struct notifier_block *, unsigned long, void *); |
198 | static struct notifier_block xpc_die_notifier = { | |
199 | .notifier_call = xpc_system_die, | |
200 | }; | |
201 | ||
a607c389 DN |
202 | |
203 | /* | |
204 | * Timer function to enforce the timelimit on the partition disengage request. | |
205 | */ | |
206 | static void | |
207 | xpc_timeout_partition_disengage_request(unsigned long data) | |
208 | { | |
209 | struct xpc_partition *part = (struct xpc_partition *) data; | |
210 | ||
211 | ||
e54af724 | 212 | DBUG_ON(jiffies < part->disengage_request_timeout); |
a607c389 DN |
213 | |
214 | (void) xpc_partition_disengaged(part); | |
215 | ||
216 | DBUG_ON(part->disengage_request_timeout != 0); | |
217 | DBUG_ON(xpc_partition_engaged(1UL << XPC_PARTID(part)) != 0); | |
218 | } | |
219 | ||
220 | ||
89eb8eb9 DN |
221 | /* |
222 | * Notify the heartbeat check thread that an IRQ has been received. | |
223 | */ | |
224 | static irqreturn_t | |
225 | xpc_act_IRQ_handler(int irq, void *dev_id, struct pt_regs *regs) | |
226 | { | |
227 | atomic_inc(&xpc_act_IRQ_rcvd); | |
228 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
229 | return IRQ_HANDLED; | |
230 | } | |
231 | ||
232 | ||
233 | /* | |
234 | * Timer to produce the heartbeat. The timer structures function is | |
235 | * already set when this is initially called. A tunable is used to | |
236 | * specify when the next timeout should occur. | |
237 | */ | |
238 | static void | |
239 | xpc_hb_beater(unsigned long dummy) | |
240 | { | |
241 | xpc_vars->heartbeat++; | |
242 | ||
243 | if (jiffies >= xpc_hb_check_timeout) { | |
244 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
245 | } | |
246 | ||
247 | xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ); | |
248 | add_timer(&xpc_hb_timer); | |
249 | } | |
250 | ||
251 | ||
252 | /* | |
253 | * This thread is responsible for nearly all of the partition | |
254 | * activation/deactivation. | |
255 | */ | |
256 | static int | |
257 | xpc_hb_checker(void *ignore) | |
258 | { | |
259 | int last_IRQ_count = 0; | |
260 | int new_IRQ_count; | |
261 | int force_IRQ=0; | |
262 | ||
263 | ||
264 | /* this thread was marked active by xpc_hb_init() */ | |
265 | ||
266 | daemonize(XPC_HB_CHECK_THREAD_NAME); | |
267 | ||
268 | set_cpus_allowed(current, cpumask_of_cpu(XPC_HB_CHECK_CPU)); | |
269 | ||
270 | xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ); | |
271 | ||
272 | while (!(volatile int) xpc_exiting) { | |
273 | ||
89eb8eb9 DN |
274 | dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have " |
275 | "been received\n", | |
276 | (int) (xpc_hb_check_timeout - jiffies), | |
277 | atomic_read(&xpc_act_IRQ_rcvd) - last_IRQ_count); | |
278 | ||
279 | ||
280 | /* checking of remote heartbeats is skewed by IRQ handling */ | |
281 | if (jiffies >= xpc_hb_check_timeout) { | |
282 | dev_dbg(xpc_part, "checking remote heartbeats\n"); | |
283 | xpc_check_remote_hb(); | |
284 | ||
285 | /* | |
286 | * We need to periodically recheck to ensure no | |
287 | * IPI/AMO pairs have been missed. That check | |
288 | * must always reset xpc_hb_check_timeout. | |
289 | */ | |
290 | force_IRQ = 1; | |
291 | } | |
292 | ||
293 | ||
a607c389 | 294 | /* check for outstanding IRQs */ |
89eb8eb9 DN |
295 | new_IRQ_count = atomic_read(&xpc_act_IRQ_rcvd); |
296 | if (last_IRQ_count < new_IRQ_count || force_IRQ != 0) { | |
297 | force_IRQ = 0; | |
298 | ||
299 | dev_dbg(xpc_part, "found an IRQ to process; will be " | |
300 | "resetting xpc_hb_check_timeout\n"); | |
301 | ||
302 | last_IRQ_count += xpc_identify_act_IRQ_sender(); | |
303 | if (last_IRQ_count < new_IRQ_count) { | |
304 | /* retry once to help avoid missing AMO */ | |
305 | (void) xpc_identify_act_IRQ_sender(); | |
306 | } | |
307 | last_IRQ_count = new_IRQ_count; | |
308 | ||
309 | xpc_hb_check_timeout = jiffies + | |
310 | (xpc_hb_check_interval * HZ); | |
311 | } | |
a607c389 DN |
312 | |
313 | /* wait for IRQ or timeout */ | |
314 | (void) wait_event_interruptible(xpc_act_IRQ_wq, | |
315 | (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) || | |
316 | jiffies >= xpc_hb_check_timeout || | |
317 | (volatile int) xpc_exiting)); | |
89eb8eb9 DN |
318 | } |
319 | ||
320 | dev_dbg(xpc_part, "heartbeat checker is exiting\n"); | |
321 | ||
322 | ||
e54af724 | 323 | /* mark this thread as having exited */ |
f9e505a9 | 324 | complete(&xpc_hb_checker_exited); |
89eb8eb9 DN |
325 | return 0; |
326 | } | |
327 | ||
328 | ||
329 | /* | |
330 | * This thread will attempt to discover other partitions to activate | |
331 | * based on info provided by SAL. This new thread is short lived and | |
332 | * will exit once discovery is complete. | |
333 | */ | |
334 | static int | |
335 | xpc_initiate_discovery(void *ignore) | |
336 | { | |
337 | daemonize(XPC_DISCOVERY_THREAD_NAME); | |
338 | ||
339 | xpc_discovery(); | |
340 | ||
341 | dev_dbg(xpc_part, "discovery thread is exiting\n"); | |
342 | ||
e54af724 | 343 | /* mark this thread as having exited */ |
f9e505a9 | 344 | complete(&xpc_discovery_exited); |
89eb8eb9 DN |
345 | return 0; |
346 | } | |
347 | ||
348 | ||
349 | /* | |
350 | * Establish first contact with the remote partititon. This involves pulling | |
351 | * the XPC per partition variables from the remote partition and waiting for | |
352 | * the remote partition to pull ours. | |
353 | */ | |
354 | static enum xpc_retval | |
355 | xpc_make_first_contact(struct xpc_partition *part) | |
356 | { | |
357 | enum xpc_retval ret; | |
358 | ||
359 | ||
360 | while ((ret = xpc_pull_remote_vars_part(part)) != xpcSuccess) { | |
361 | if (ret != xpcRetry) { | |
362 | XPC_DEACTIVATE_PARTITION(part, ret); | |
363 | return ret; | |
364 | } | |
365 | ||
366 | dev_dbg(xpc_chan, "waiting to make first contact with " | |
367 | "partition %d\n", XPC_PARTID(part)); | |
368 | ||
369 | /* wait a 1/4 of a second or so */ | |
a607c389 | 370 | (void) msleep_interruptible(250); |
89eb8eb9 DN |
371 | |
372 | if (part->act_state == XPC_P_DEACTIVATING) { | |
373 | return part->reason; | |
374 | } | |
375 | } | |
376 | ||
377 | return xpc_mark_partition_active(part); | |
378 | } | |
379 | ||
380 | ||
381 | /* | |
382 | * The first kthread assigned to a newly activated partition is the one | |
383 | * created by XPC HB with which it calls xpc_partition_up(). XPC hangs on to | |
384 | * that kthread until the partition is brought down, at which time that kthread | |
385 | * returns back to XPC HB. (The return of that kthread will signify to XPC HB | |
386 | * that XPC has dismantled all communication infrastructure for the associated | |
387 | * partition.) This kthread becomes the channel manager for that partition. | |
388 | * | |
389 | * Each active partition has a channel manager, who, besides connecting and | |
390 | * disconnecting channels, will ensure that each of the partition's connected | |
391 | * channels has the required number of assigned kthreads to get the work done. | |
392 | */ | |
393 | static void | |
394 | xpc_channel_mgr(struct xpc_partition *part) | |
395 | { | |
396 | while (part->act_state != XPC_P_DEACTIVATING || | |
a607c389 DN |
397 | atomic_read(&part->nchannels_active) > 0 || |
398 | !xpc_partition_disengaged(part)) { | |
89eb8eb9 DN |
399 | |
400 | xpc_process_channel_activity(part); | |
401 | ||
402 | ||
403 | /* | |
404 | * Wait until we've been requested to activate kthreads or | |
405 | * all of the channel's message queues have been torn down or | |
406 | * a signal is pending. | |
407 | * | |
408 | * The channel_mgr_requests is set to 1 after being awakened, | |
409 | * This is done to prevent the channel mgr from making one pass | |
410 | * through the loop for each request, since he will | |
411 | * be servicing all the requests in one pass. The reason it's | |
412 | * set to 1 instead of 0 is so that other kthreads will know | |
413 | * that the channel mgr is running and won't bother trying to | |
414 | * wake him up. | |
415 | */ | |
416 | atomic_dec(&part->channel_mgr_requests); | |
417 | (void) wait_event_interruptible(part->channel_mgr_wq, | |
418 | (atomic_read(&part->channel_mgr_requests) > 0 || | |
419 | (volatile u64) part->local_IPI_amo != 0 || | |
420 | ((volatile u8) part->act_state == | |
421 | XPC_P_DEACTIVATING && | |
a607c389 DN |
422 | atomic_read(&part->nchannels_active) == 0 && |
423 | xpc_partition_disengaged(part)))); | |
89eb8eb9 DN |
424 | atomic_set(&part->channel_mgr_requests, 1); |
425 | ||
426 | // >>> Does it need to wakeup periodically as well? In case we | |
427 | // >>> miscalculated the #of kthreads to wakeup or create? | |
428 | } | |
429 | } | |
430 | ||
431 | ||
432 | /* | |
433 | * When XPC HB determines that a partition has come up, it will create a new | |
434 | * kthread and that kthread will call this function to attempt to set up the | |
435 | * basic infrastructure used for Cross Partition Communication with the newly | |
436 | * upped partition. | |
437 | * | |
438 | * The kthread that was created by XPC HB and which setup the XPC | |
439 | * infrastructure will remain assigned to the partition until the partition | |
440 | * goes down. At which time the kthread will teardown the XPC infrastructure | |
441 | * and then exit. | |
442 | * | |
443 | * XPC HB will put the remote partition's XPC per partition specific variables | |
444 | * physical address into xpc_partitions[partid].remote_vars_part_pa prior to | |
445 | * calling xpc_partition_up(). | |
446 | */ | |
447 | static void | |
448 | xpc_partition_up(struct xpc_partition *part) | |
449 | { | |
450 | DBUG_ON(part->channels != NULL); | |
451 | ||
452 | dev_dbg(xpc_chan, "activating partition %d\n", XPC_PARTID(part)); | |
453 | ||
454 | if (xpc_setup_infrastructure(part) != xpcSuccess) { | |
455 | return; | |
456 | } | |
457 | ||
458 | /* | |
459 | * The kthread that XPC HB called us with will become the | |
460 | * channel manager for this partition. It will not return | |
461 | * back to XPC HB until the partition's XPC infrastructure | |
462 | * has been dismantled. | |
463 | */ | |
464 | ||
465 | (void) xpc_part_ref(part); /* this will always succeed */ | |
466 | ||
467 | if (xpc_make_first_contact(part) == xpcSuccess) { | |
468 | xpc_channel_mgr(part); | |
469 | } | |
470 | ||
471 | xpc_part_deref(part); | |
472 | ||
473 | xpc_teardown_infrastructure(part); | |
474 | } | |
475 | ||
476 | ||
477 | static int | |
478 | xpc_activating(void *__partid) | |
479 | { | |
480 | partid_t partid = (u64) __partid; | |
481 | struct xpc_partition *part = &xpc_partitions[partid]; | |
482 | unsigned long irq_flags; | |
e037cda5 | 483 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
89eb8eb9 DN |
484 | int ret; |
485 | ||
486 | ||
487 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | |
488 | ||
489 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
490 | ||
491 | if (part->act_state == XPC_P_DEACTIVATING) { | |
492 | part->act_state = XPC_P_INACTIVE; | |
493 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
494 | part->remote_rp_pa = 0; | |
495 | return 0; | |
496 | } | |
497 | ||
498 | /* indicate the thread is activating */ | |
499 | DBUG_ON(part->act_state != XPC_P_ACTIVATION_REQ); | |
500 | part->act_state = XPC_P_ACTIVATING; | |
501 | ||
502 | XPC_SET_REASON(part, 0, 0); | |
503 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
504 | ||
505 | dev_dbg(xpc_part, "bringing partition %d up\n", partid); | |
506 | ||
507 | daemonize("xpc%02d", partid); | |
508 | ||
509 | /* | |
510 | * This thread needs to run at a realtime priority to prevent a | |
511 | * significant performance degradation. | |
512 | */ | |
513 | ret = sched_setscheduler(current, SCHED_FIFO, ¶m); | |
514 | if (ret != 0) { | |
515 | dev_warn(xpc_part, "unable to set pid %d to a realtime " | |
516 | "priority, ret=%d\n", current->pid, ret); | |
517 | } | |
518 | ||
519 | /* allow this thread and its children to run on any CPU */ | |
520 | set_cpus_allowed(current, CPU_MASK_ALL); | |
521 | ||
522 | /* | |
523 | * Register the remote partition's AMOs with SAL so it can handle | |
524 | * and cleanup errors within that address range should the remote | |
525 | * partition go down. We don't unregister this range because it is | |
526 | * difficult to tell when outstanding writes to the remote partition | |
527 | * are finished and thus when it is safe to unregister. This should | |
528 | * not result in wasted space in the SAL xp_addr_region table because | |
529 | * we should get the same page for remote_amos_page_pa after module | |
530 | * reloads and system reboots. | |
531 | */ | |
532 | if (sn_register_xp_addr_region(part->remote_amos_page_pa, | |
533 | PAGE_SIZE, 1) < 0) { | |
534 | dev_warn(xpc_part, "xpc_partition_up(%d) failed to register " | |
535 | "xp_addr region\n", partid); | |
536 | ||
537 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
538 | part->act_state = XPC_P_INACTIVE; | |
539 | XPC_SET_REASON(part, xpcPhysAddrRegFailed, __LINE__); | |
540 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
541 | part->remote_rp_pa = 0; | |
542 | return 0; | |
543 | } | |
544 | ||
a607c389 | 545 | xpc_allow_hb(partid, xpc_vars); |
89eb8eb9 DN |
546 | xpc_IPI_send_activated(part); |
547 | ||
548 | ||
549 | /* | |
550 | * xpc_partition_up() holds this thread and marks this partition as | |
551 | * XPC_P_ACTIVE by calling xpc_hb_mark_active(). | |
552 | */ | |
553 | (void) xpc_partition_up(part); | |
554 | ||
a607c389 | 555 | xpc_disallow_hb(partid, xpc_vars); |
89eb8eb9 DN |
556 | xpc_mark_partition_inactive(part); |
557 | ||
558 | if (part->reason == xpcReactivating) { | |
559 | /* interrupting ourselves results in activating partition */ | |
560 | xpc_IPI_send_reactivate(part); | |
561 | } | |
562 | ||
563 | return 0; | |
564 | } | |
565 | ||
566 | ||
567 | void | |
568 | xpc_activate_partition(struct xpc_partition *part) | |
569 | { | |
570 | partid_t partid = XPC_PARTID(part); | |
571 | unsigned long irq_flags; | |
572 | pid_t pid; | |
573 | ||
574 | ||
575 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
576 | ||
89eb8eb9 DN |
577 | DBUG_ON(part->act_state != XPC_P_INACTIVE); |
578 | ||
7c6c6636 RH |
579 | part->act_state = XPC_P_ACTIVATION_REQ; |
580 | XPC_SET_REASON(part, xpcCloneKThread, __LINE__); | |
89eb8eb9 DN |
581 | |
582 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
7c6c6636 RH |
583 | |
584 | pid = kernel_thread(xpc_activating, (void *) ((u64) partid), 0); | |
585 | ||
586 | if (unlikely(pid <= 0)) { | |
587 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
588 | part->act_state = XPC_P_INACTIVE; | |
589 | XPC_SET_REASON(part, xpcCloneKThreadFailed, __LINE__); | |
590 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
591 | } | |
89eb8eb9 DN |
592 | } |
593 | ||
594 | ||
595 | /* | |
596 | * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified | |
597 | * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more | |
598 | * than one partition, we use an AMO_t structure per partition to indicate | |
599 | * whether a partition has sent an IPI or not. >>> If it has, then wake up the | |
600 | * associated kthread to handle it. | |
601 | * | |
602 | * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IPIs sent by XPC | |
603 | * running on other partitions. | |
604 | * | |
605 | * Noteworthy Arguments: | |
606 | * | |
607 | * irq - Interrupt ReQuest number. NOT USED. | |
608 | * | |
609 | * dev_id - partid of IPI's potential sender. | |
610 | * | |
611 | * regs - processor's context before the processor entered | |
612 | * interrupt code. NOT USED. | |
613 | */ | |
614 | irqreturn_t | |
615 | xpc_notify_IRQ_handler(int irq, void *dev_id, struct pt_regs *regs) | |
616 | { | |
617 | partid_t partid = (partid_t) (u64) dev_id; | |
618 | struct xpc_partition *part = &xpc_partitions[partid]; | |
619 | ||
620 | ||
621 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | |
622 | ||
623 | if (xpc_part_ref(part)) { | |
624 | xpc_check_for_channel_activity(part); | |
625 | ||
626 | xpc_part_deref(part); | |
627 | } | |
628 | return IRQ_HANDLED; | |
629 | } | |
630 | ||
631 | ||
632 | /* | |
633 | * Check to see if xpc_notify_IRQ_handler() dropped any IPIs on the floor | |
634 | * because the write to their associated IPI amo completed after the IRQ/IPI | |
635 | * was received. | |
636 | */ | |
637 | void | |
638 | xpc_dropped_IPI_check(struct xpc_partition *part) | |
639 | { | |
640 | if (xpc_part_ref(part)) { | |
641 | xpc_check_for_channel_activity(part); | |
642 | ||
643 | part->dropped_IPI_timer.expires = jiffies + | |
644 | XPC_P_DROPPED_IPI_WAIT; | |
645 | add_timer(&part->dropped_IPI_timer); | |
646 | xpc_part_deref(part); | |
647 | } | |
648 | } | |
649 | ||
650 | ||
651 | void | |
652 | xpc_activate_kthreads(struct xpc_channel *ch, int needed) | |
653 | { | |
654 | int idle = atomic_read(&ch->kthreads_idle); | |
655 | int assigned = atomic_read(&ch->kthreads_assigned); | |
656 | int wakeup; | |
657 | ||
658 | ||
659 | DBUG_ON(needed <= 0); | |
660 | ||
661 | if (idle > 0) { | |
662 | wakeup = (needed > idle) ? idle : needed; | |
663 | needed -= wakeup; | |
664 | ||
665 | dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, " | |
666 | "channel=%d\n", wakeup, ch->partid, ch->number); | |
667 | ||
668 | /* only wakeup the requested number of kthreads */ | |
669 | wake_up_nr(&ch->idle_wq, wakeup); | |
670 | } | |
671 | ||
672 | if (needed <= 0) { | |
673 | return; | |
674 | } | |
675 | ||
676 | if (needed + assigned > ch->kthreads_assigned_limit) { | |
677 | needed = ch->kthreads_assigned_limit - assigned; | |
678 | // >>>should never be less than 0 | |
679 | if (needed <= 0) { | |
680 | return; | |
681 | } | |
682 | } | |
683 | ||
684 | dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n", | |
685 | needed, ch->partid, ch->number); | |
686 | ||
687 | xpc_create_kthreads(ch, needed); | |
688 | } | |
689 | ||
690 | ||
691 | /* | |
692 | * This function is where XPC's kthreads wait for messages to deliver. | |
693 | */ | |
694 | static void | |
695 | xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch) | |
696 | { | |
697 | do { | |
698 | /* deliver messages to their intended recipients */ | |
699 | ||
700 | while ((volatile s64) ch->w_local_GP.get < | |
701 | (volatile s64) ch->w_remote_GP.put && | |
702 | !((volatile u32) ch->flags & | |
703 | XPC_C_DISCONNECTING)) { | |
704 | xpc_deliver_msg(ch); | |
705 | } | |
706 | ||
707 | if (atomic_inc_return(&ch->kthreads_idle) > | |
708 | ch->kthreads_idle_limit) { | |
709 | /* too many idle kthreads on this channel */ | |
710 | atomic_dec(&ch->kthreads_idle); | |
711 | break; | |
712 | } | |
713 | ||
714 | dev_dbg(xpc_chan, "idle kthread calling " | |
715 | "wait_event_interruptible_exclusive()\n"); | |
716 | ||
717 | (void) wait_event_interruptible_exclusive(ch->idle_wq, | |
718 | ((volatile s64) ch->w_local_GP.get < | |
719 | (volatile s64) ch->w_remote_GP.put || | |
720 | ((volatile u32) ch->flags & | |
721 | XPC_C_DISCONNECTING))); | |
722 | ||
723 | atomic_dec(&ch->kthreads_idle); | |
724 | ||
725 | } while (!((volatile u32) ch->flags & XPC_C_DISCONNECTING)); | |
726 | } | |
727 | ||
728 | ||
729 | static int | |
730 | xpc_daemonize_kthread(void *args) | |
731 | { | |
732 | partid_t partid = XPC_UNPACK_ARG1(args); | |
733 | u16 ch_number = XPC_UNPACK_ARG2(args); | |
734 | struct xpc_partition *part = &xpc_partitions[partid]; | |
735 | struct xpc_channel *ch; | |
736 | int n_needed; | |
e54af724 | 737 | unsigned long irq_flags; |
89eb8eb9 DN |
738 | |
739 | ||
740 | daemonize("xpc%02dc%d", partid, ch_number); | |
741 | ||
742 | dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n", | |
743 | partid, ch_number); | |
744 | ||
745 | ch = &part->channels[ch_number]; | |
746 | ||
747 | if (!(ch->flags & XPC_C_DISCONNECTING)) { | |
89eb8eb9 DN |
748 | |
749 | /* let registerer know that connection has been established */ | |
750 | ||
e54af724 | 751 | spin_lock_irqsave(&ch->lock, irq_flags); |
4c2cd966 DN |
752 | if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) { |
753 | ch->flags |= XPC_C_CONNECTEDCALLOUT; | |
e54af724 DN |
754 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
755 | ||
89eb8eb9 DN |
756 | xpc_connected_callout(ch); |
757 | ||
4c2cd966 DN |
758 | spin_lock_irqsave(&ch->lock, irq_flags); |
759 | ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE; | |
760 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
761 | ||
89eb8eb9 DN |
762 | /* |
763 | * It is possible that while the callout was being | |
764 | * made that the remote partition sent some messages. | |
765 | * If that is the case, we may need to activate | |
766 | * additional kthreads to help deliver them. We only | |
767 | * need one less than total #of messages to deliver. | |
768 | */ | |
769 | n_needed = ch->w_remote_GP.put - ch->w_local_GP.get - 1; | |
770 | if (n_needed > 0 && | |
771 | !(ch->flags & XPC_C_DISCONNECTING)) { | |
772 | xpc_activate_kthreads(ch, n_needed); | |
773 | } | |
e54af724 DN |
774 | } else { |
775 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
89eb8eb9 DN |
776 | } |
777 | ||
778 | xpc_kthread_waitmsgs(part, ch); | |
779 | } | |
780 | ||
a607c389 | 781 | if (atomic_dec_return(&ch->kthreads_assigned) == 0) { |
e54af724 | 782 | spin_lock_irqsave(&ch->lock, irq_flags); |
4c2cd966 DN |
783 | if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && |
784 | !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { | |
785 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT; | |
e54af724 DN |
786 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
787 | ||
246c7e33 | 788 | xpc_disconnect_callout(ch, xpcDisconnecting); |
4c2cd966 DN |
789 | |
790 | spin_lock_irqsave(&ch->lock, irq_flags); | |
791 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE; | |
a607c389 | 792 | } |
4c2cd966 | 793 | spin_unlock_irqrestore(&ch->lock, irq_flags); |
a607c389 DN |
794 | if (atomic_dec_return(&part->nchannels_engaged) == 0) { |
795 | xpc_mark_partition_disengaged(part); | |
796 | xpc_IPI_send_disengage(part); | |
797 | } | |
89eb8eb9 DN |
798 | } |
799 | ||
800 | ||
801 | xpc_msgqueue_deref(ch); | |
802 | ||
803 | dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n", | |
804 | partid, ch_number); | |
805 | ||
806 | xpc_part_deref(part); | |
807 | return 0; | |
808 | } | |
809 | ||
810 | ||
811 | /* | |
812 | * For each partition that XPC has established communications with, there is | |
813 | * a minimum of one kernel thread assigned to perform any operation that | |
814 | * may potentially sleep or block (basically the callouts to the asynchronous | |
815 | * functions registered via xpc_connect()). | |
816 | * | |
817 | * Additional kthreads are created and destroyed by XPC as the workload | |
818 | * demands. | |
819 | * | |
820 | * A kthread is assigned to one of the active channels that exists for a given | |
821 | * partition. | |
822 | */ | |
823 | void | |
824 | xpc_create_kthreads(struct xpc_channel *ch, int needed) | |
825 | { | |
826 | unsigned long irq_flags; | |
827 | pid_t pid; | |
828 | u64 args = XPC_PACK_ARGS(ch->partid, ch->number); | |
a607c389 | 829 | struct xpc_partition *part = &xpc_partitions[ch->partid]; |
89eb8eb9 DN |
830 | |
831 | ||
832 | while (needed-- > 0) { | |
e54af724 DN |
833 | |
834 | /* | |
835 | * The following is done on behalf of the newly created | |
836 | * kthread. That kthread is responsible for doing the | |
837 | * counterpart to the following before it exits. | |
838 | */ | |
839 | (void) xpc_part_ref(part); | |
840 | xpc_msgqueue_ref(ch); | |
841 | if (atomic_inc_return(&ch->kthreads_assigned) == 1 && | |
842 | atomic_inc_return(&part->nchannels_engaged) == 1) { | |
843 | xpc_mark_partition_engaged(part); | |
844 | } | |
845 | ||
89eb8eb9 DN |
846 | pid = kernel_thread(xpc_daemonize_kthread, (void *) args, 0); |
847 | if (pid < 0) { | |
848 | /* the fork failed */ | |
e54af724 DN |
849 | if (atomic_dec_return(&ch->kthreads_assigned) == 0 && |
850 | atomic_dec_return(&part->nchannels_engaged) == 0) { | |
851 | xpc_mark_partition_disengaged(part); | |
852 | xpc_IPI_send_disengage(part); | |
853 | } | |
854 | xpc_msgqueue_deref(ch); | |
855 | xpc_part_deref(part); | |
89eb8eb9 DN |
856 | |
857 | if (atomic_read(&ch->kthreads_assigned) < | |
858 | ch->kthreads_idle_limit) { | |
859 | /* | |
860 | * Flag this as an error only if we have an | |
861 | * insufficient #of kthreads for the channel | |
862 | * to function. | |
863 | * | |
864 | * No xpc_msgqueue_ref() is needed here since | |
865 | * the channel mgr is doing this. | |
866 | */ | |
867 | spin_lock_irqsave(&ch->lock, irq_flags); | |
868 | XPC_DISCONNECT_CHANNEL(ch, xpcLackOfResources, | |
869 | &irq_flags); | |
870 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
871 | } | |
872 | break; | |
873 | } | |
874 | ||
89eb8eb9 DN |
875 | ch->kthreads_created++; // >>> temporary debug only!!! |
876 | } | |
877 | } | |
878 | ||
879 | ||
880 | void | |
881 | xpc_disconnect_wait(int ch_number) | |
882 | { | |
a607c389 | 883 | unsigned long irq_flags; |
89eb8eb9 DN |
884 | partid_t partid; |
885 | struct xpc_partition *part; | |
886 | struct xpc_channel *ch; | |
e54af724 | 887 | int wakeup_channel_mgr; |
89eb8eb9 DN |
888 | |
889 | ||
890 | /* now wait for all callouts to the caller's function to cease */ | |
891 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
892 | part = &xpc_partitions[partid]; | |
893 | ||
e54af724 DN |
894 | if (!xpc_part_ref(part)) { |
895 | continue; | |
896 | } | |
89eb8eb9 | 897 | |
e54af724 | 898 | ch = &part->channels[ch_number]; |
89eb8eb9 | 899 | |
e54af724 | 900 | if (!(ch->flags & XPC_C_WDISCONNECT)) { |
89eb8eb9 | 901 | xpc_part_deref(part); |
e54af724 | 902 | continue; |
89eb8eb9 | 903 | } |
e54af724 | 904 | |
f9e505a9 | 905 | wait_for_completion(&ch->wdisconnect_wait); |
e54af724 DN |
906 | |
907 | spin_lock_irqsave(&ch->lock, irq_flags); | |
908 | DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED)); | |
909 | wakeup_channel_mgr = 0; | |
910 | ||
911 | if (ch->delayed_IPI_flags) { | |
912 | if (part->act_state != XPC_P_DEACTIVATING) { | |
913 | spin_lock(&part->IPI_lock); | |
914 | XPC_SET_IPI_FLAGS(part->local_IPI_amo, | |
915 | ch->number, ch->delayed_IPI_flags); | |
916 | spin_unlock(&part->IPI_lock); | |
917 | wakeup_channel_mgr = 1; | |
918 | } | |
919 | ch->delayed_IPI_flags = 0; | |
89eb8eb9 | 920 | } |
e54af724 DN |
921 | |
922 | ch->flags &= ~XPC_C_WDISCONNECT; | |
923 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
924 | ||
925 | if (wakeup_channel_mgr) { | |
926 | xpc_wakeup_channel_mgr(part); | |
927 | } | |
928 | ||
929 | xpc_part_deref(part); | |
89eb8eb9 DN |
930 | } |
931 | } | |
932 | ||
933 | ||
934 | static void | |
a607c389 | 935 | xpc_do_exit(enum xpc_retval reason) |
89eb8eb9 DN |
936 | { |
937 | partid_t partid; | |
1ecaded8 | 938 | int active_part_count, printed_waiting_msg = 0; |
89eb8eb9 | 939 | struct xpc_partition *part; |
1ecaded8 | 940 | unsigned long printmsg_time, disengage_request_timeout = 0; |
89eb8eb9 DN |
941 | |
942 | ||
a607c389 DN |
943 | /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */ |
944 | DBUG_ON(xpc_exiting == 1); | |
89eb8eb9 DN |
945 | |
946 | /* | |
a607c389 DN |
947 | * Let the heartbeat checker thread and the discovery thread |
948 | * (if one is running) know that they should exit. Also wake up | |
949 | * the heartbeat checker thread in case it's sleeping. | |
89eb8eb9 DN |
950 | */ |
951 | xpc_exiting = 1; | |
952 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
953 | ||
a607c389 DN |
954 | /* ignore all incoming interrupts */ |
955 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
89eb8eb9 | 956 | |
e54af724 | 957 | /* wait for the discovery thread to exit */ |
f9e505a9 | 958 | wait_for_completion(&xpc_discovery_exited); |
89eb8eb9 | 959 | |
e54af724 | 960 | /* wait for the heartbeat checker thread to exit */ |
f9e505a9 | 961 | wait_for_completion(&xpc_hb_checker_exited); |
89eb8eb9 | 962 | |
89eb8eb9 | 963 | |
a607c389 DN |
964 | /* sleep for a 1/3 of a second or so */ |
965 | (void) msleep_interruptible(300); | |
89eb8eb9 DN |
966 | |
967 | ||
968 | /* wait for all partitions to become inactive */ | |
969 | ||
1ecaded8 DN |
970 | printmsg_time = jiffies + (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); |
971 | xpc_disengage_request_timedout = 0; | |
a607c389 | 972 | |
89eb8eb9 DN |
973 | do { |
974 | active_part_count = 0; | |
975 | ||
976 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
977 | part = &xpc_partitions[partid]; | |
89eb8eb9 | 978 | |
a607c389 DN |
979 | if (xpc_partition_disengaged(part) && |
980 | part->act_state == XPC_P_INACTIVE) { | |
981 | continue; | |
89eb8eb9 | 982 | } |
a607c389 DN |
983 | |
984 | active_part_count++; | |
985 | ||
986 | XPC_DEACTIVATE_PARTITION(part, reason); | |
89eb8eb9 | 987 | |
1ecaded8 DN |
988 | if (part->disengage_request_timeout > |
989 | disengage_request_timeout) { | |
990 | disengage_request_timeout = | |
991 | part->disengage_request_timeout; | |
992 | } | |
a607c389 | 993 | } |
89eb8eb9 | 994 | |
1ecaded8 DN |
995 | if (xpc_partition_engaged(-1UL)) { |
996 | if (time_after(jiffies, printmsg_time)) { | |
997 | dev_info(xpc_part, "waiting for remote " | |
998 | "partitions to disengage, timeout in " | |
999 | "%ld seconds\n", | |
1000 | (disengage_request_timeout - jiffies) | |
1001 | / HZ); | |
1002 | printmsg_time = jiffies + | |
a607c389 | 1003 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); |
1ecaded8 DN |
1004 | printed_waiting_msg = 1; |
1005 | } | |
1006 | ||
1007 | } else if (active_part_count > 0) { | |
1008 | if (printed_waiting_msg) { | |
1009 | dev_info(xpc_part, "waiting for local partition" | |
1010 | " to disengage\n"); | |
1011 | printed_waiting_msg = 0; | |
1012 | } | |
1013 | ||
1014 | } else { | |
1015 | if (!xpc_disengage_request_timedout) { | |
1016 | dev_info(xpc_part, "all partitions have " | |
1017 | "disengaged\n"); | |
1018 | } | |
1019 | break; | |
89eb8eb9 DN |
1020 | } |
1021 | ||
a607c389 DN |
1022 | /* sleep for a 1/3 of a second or so */ |
1023 | (void) msleep_interruptible(300); | |
1024 | ||
1025 | } while (1); | |
1026 | ||
1027 | DBUG_ON(xpc_partition_engaged(-1UL)); | |
1028 | ||
1029 | ||
1030 | /* indicate to others that our reserved page is uninitialized */ | |
1031 | xpc_rsvd_page->vars_pa = 0; | |
1032 | ||
1033 | /* now it's time to eliminate our heartbeat */ | |
1034 | del_timer_sync(&xpc_hb_timer); | |
e54af724 | 1035 | DBUG_ON(xpc_vars->heartbeating_to_mask != 0); |
89eb8eb9 | 1036 | |
0752c670 DN |
1037 | if (reason == xpcUnloading) { |
1038 | /* take ourselves off of the reboot_notifier_list */ | |
1039 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | |
89eb8eb9 | 1040 | |
0752c670 DN |
1041 | /* take ourselves off of the die_notifier list */ |
1042 | (void) unregister_die_notifier(&xpc_die_notifier); | |
1043 | } | |
780d09e8 | 1044 | |
89eb8eb9 DN |
1045 | /* close down protections for IPI operations */ |
1046 | xpc_restrict_IPI_ops(); | |
1047 | ||
1048 | ||
1049 | /* clear the interface to XPC's functions */ | |
1050 | xpc_clear_interface(); | |
1051 | ||
1052 | if (xpc_sysctl) { | |
1053 | unregister_sysctl_table(xpc_sysctl); | |
1054 | } | |
7682a4c6 DN |
1055 | |
1056 | kfree(xpc_remote_copy_buffer_base); | |
89eb8eb9 DN |
1057 | } |
1058 | ||
1059 | ||
780d09e8 | 1060 | /* |
d6ad033a DN |
1061 | * This function is called when the system is being rebooted. |
1062 | */ | |
1063 | static int | |
1064 | xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused) | |
1065 | { | |
1066 | enum xpc_retval reason; | |
1067 | ||
1068 | ||
1069 | switch (event) { | |
1070 | case SYS_RESTART: | |
1071 | reason = xpcSystemReboot; | |
1072 | break; | |
1073 | case SYS_HALT: | |
1074 | reason = xpcSystemHalt; | |
1075 | break; | |
1076 | case SYS_POWER_OFF: | |
1077 | reason = xpcSystemPoweroff; | |
1078 | break; | |
1079 | default: | |
1080 | reason = xpcSystemGoingDown; | |
1081 | } | |
1082 | ||
1083 | xpc_do_exit(reason); | |
1084 | return NOTIFY_DONE; | |
1085 | } | |
1086 | ||
1087 | ||
1088 | /* | |
1089 | * Notify other partitions to disengage from all references to our memory. | |
780d09e8 DN |
1090 | */ |
1091 | static void | |
1092 | xpc_die_disengage(void) | |
1093 | { | |
1094 | struct xpc_partition *part; | |
1095 | partid_t partid; | |
1096 | unsigned long engaged; | |
1ecaded8 | 1097 | long time, printmsg_time, disengage_request_timeout; |
780d09e8 DN |
1098 | |
1099 | ||
1100 | /* keep xpc_hb_checker thread from doing anything (just in case) */ | |
1101 | xpc_exiting = 1; | |
1102 | ||
1103 | xpc_vars->heartbeating_to_mask = 0; /* indicate we're deactivated */ | |
1104 | ||
1105 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1106 | part = &xpc_partitions[partid]; | |
1107 | ||
1108 | if (!XPC_SUPPORTS_DISENGAGE_REQUEST(part-> | |
1109 | remote_vars_version)) { | |
1110 | ||
1111 | /* just in case it was left set by an earlier XPC */ | |
1112 | xpc_clear_partition_engaged(1UL << partid); | |
1113 | continue; | |
1114 | } | |
1115 | ||
1116 | if (xpc_partition_engaged(1UL << partid) || | |
1117 | part->act_state != XPC_P_INACTIVE) { | |
1118 | xpc_request_partition_disengage(part); | |
1119 | xpc_mark_partition_disengaged(part); | |
1120 | xpc_IPI_send_disengage(part); | |
1121 | } | |
1122 | } | |
1123 | ||
1ecaded8 DN |
1124 | time = rtc_time(); |
1125 | printmsg_time = time + | |
1126 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * sn_rtc_cycles_per_second); | |
1127 | disengage_request_timeout = time + | |
780d09e8 DN |
1128 | (xpc_disengage_request_timelimit * sn_rtc_cycles_per_second); |
1129 | ||
1130 | /* wait for all other partitions to disengage from us */ | |
1131 | ||
1ecaded8 DN |
1132 | while (1) { |
1133 | engaged = xpc_partition_engaged(-1UL); | |
1134 | if (!engaged) { | |
1135 | dev_info(xpc_part, "all partitions have disengaged\n"); | |
1136 | break; | |
1137 | } | |
780d09e8 | 1138 | |
1ecaded8 DN |
1139 | time = rtc_time(); |
1140 | if (time >= disengage_request_timeout) { | |
1141 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1142 | if (engaged & (1UL << partid)) { | |
1143 | dev_info(xpc_part, "disengage from " | |
1144 | "remote partition %d timed " | |
1145 | "out\n", partid); | |
1146 | } | |
1147 | } | |
1148 | break; | |
1149 | } | |
1150 | ||
1151 | if (time >= printmsg_time) { | |
780d09e8 | 1152 | dev_info(xpc_part, "waiting for remote partitions to " |
1ecaded8 DN |
1153 | "disengage, timeout in %ld seconds\n", |
1154 | (disengage_request_timeout - time) / | |
1155 | sn_rtc_cycles_per_second); | |
1156 | printmsg_time = time + | |
1157 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * | |
780d09e8 DN |
1158 | sn_rtc_cycles_per_second); |
1159 | } | |
1160 | } | |
780d09e8 DN |
1161 | } |
1162 | ||
1163 | ||
780d09e8 | 1164 | /* |
1f4674b2 DN |
1165 | * This function is called when the system is being restarted or halted due |
1166 | * to some sort of system failure. If this is the case we need to notify the | |
1167 | * other partitions to disengage from all references to our memory. | |
1168 | * This function can also be called when our heartbeater could be offlined | |
1169 | * for a time. In this case we need to notify other partitions to not worry | |
1170 | * about the lack of a heartbeat. | |
780d09e8 DN |
1171 | */ |
1172 | static int | |
1173 | xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused) | |
1174 | { | |
1175 | switch (event) { | |
1176 | case DIE_MACHINE_RESTART: | |
1177 | case DIE_MACHINE_HALT: | |
1178 | xpc_die_disengage(); | |
1179 | break; | |
1f4674b2 DN |
1180 | |
1181 | case DIE_KDEBUG_ENTER: | |
1182 | /* Should lack of heartbeat be ignored by other partitions? */ | |
1183 | if (!xpc_kdebug_ignore) { | |
1184 | break; | |
1185 | } | |
1186 | /* fall through */ | |
780d09e8 DN |
1187 | case DIE_MCA_MONARCH_ENTER: |
1188 | case DIE_INIT_MONARCH_ENTER: | |
1189 | xpc_vars->heartbeat++; | |
1190 | xpc_vars->heartbeat_offline = 1; | |
1191 | break; | |
1f4674b2 DN |
1192 | |
1193 | case DIE_KDEBUG_LEAVE: | |
1194 | /* Is lack of heartbeat being ignored by other partitions? */ | |
1195 | if (!xpc_kdebug_ignore) { | |
1196 | break; | |
1197 | } | |
1198 | /* fall through */ | |
780d09e8 DN |
1199 | case DIE_MCA_MONARCH_LEAVE: |
1200 | case DIE_INIT_MONARCH_LEAVE: | |
1201 | xpc_vars->heartbeat++; | |
1202 | xpc_vars->heartbeat_offline = 0; | |
1203 | break; | |
1204 | } | |
1205 | ||
1206 | return NOTIFY_DONE; | |
1207 | } | |
1208 | ||
1209 | ||
89eb8eb9 DN |
1210 | int __init |
1211 | xpc_init(void) | |
1212 | { | |
1213 | int ret; | |
1214 | partid_t partid; | |
1215 | struct xpc_partition *part; | |
1216 | pid_t pid; | |
7682a4c6 | 1217 | size_t buf_size; |
89eb8eb9 DN |
1218 | |
1219 | ||
408865ce DN |
1220 | if (!ia64_platform_is("sn2")) { |
1221 | return -ENODEV; | |
1222 | } | |
1223 | ||
7682a4c6 DN |
1224 | |
1225 | buf_size = max(XPC_RP_VARS_SIZE, | |
1226 | XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES); | |
1227 | xpc_remote_copy_buffer = xpc_kmalloc_cacheline_aligned(buf_size, | |
1228 | GFP_KERNEL, &xpc_remote_copy_buffer_base); | |
1229 | if (xpc_remote_copy_buffer == NULL) | |
1230 | return -ENOMEM; | |
89eb8eb9 DN |
1231 | |
1232 | snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part"); | |
1233 | snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan"); | |
1234 | ||
1235 | xpc_sysctl = register_sysctl_table(xpc_sys_dir, 1); | |
1236 | ||
1237 | /* | |
1238 | * The first few fields of each entry of xpc_partitions[] need to | |
1239 | * be initialized now so that calls to xpc_connect() and | |
1240 | * xpc_disconnect() can be made prior to the activation of any remote | |
1241 | * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE | |
1242 | * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING | |
1243 | * PARTITION HAS BEEN ACTIVATED. | |
1244 | */ | |
1245 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1246 | part = &xpc_partitions[partid]; | |
1247 | ||
1248 | DBUG_ON((u64) part != L1_CACHE_ALIGN((u64) part)); | |
1249 | ||
1250 | part->act_IRQ_rcvd = 0; | |
1251 | spin_lock_init(&part->act_lock); | |
1252 | part->act_state = XPC_P_INACTIVE; | |
1253 | XPC_SET_REASON(part, 0, 0); | |
a607c389 DN |
1254 | |
1255 | init_timer(&part->disengage_request_timer); | |
1256 | part->disengage_request_timer.function = | |
1257 | xpc_timeout_partition_disengage_request; | |
1258 | part->disengage_request_timer.data = (unsigned long) part; | |
1259 | ||
89eb8eb9 DN |
1260 | part->setup_state = XPC_P_UNSET; |
1261 | init_waitqueue_head(&part->teardown_wq); | |
1262 | atomic_set(&part->references, 0); | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * Open up protections for IPI operations (and AMO operations on | |
1267 | * Shub 1.1 systems). | |
1268 | */ | |
1269 | xpc_allow_IPI_ops(); | |
1270 | ||
1271 | /* | |
1272 | * Interrupts being processed will increment this atomic variable and | |
1273 | * awaken the heartbeat thread which will process the interrupts. | |
1274 | */ | |
1275 | atomic_set(&xpc_act_IRQ_rcvd, 0); | |
1276 | ||
1277 | /* | |
1278 | * This is safe to do before the xpc_hb_checker thread has started | |
1279 | * because the handler releases a wait queue. If an interrupt is | |
1280 | * received before the thread is waiting, it will not go to sleep, | |
1281 | * but rather immediately process the interrupt. | |
1282 | */ | |
1283 | ret = request_irq(SGI_XPC_ACTIVATE, xpc_act_IRQ_handler, 0, | |
1284 | "xpc hb", NULL); | |
1285 | if (ret != 0) { | |
1286 | dev_err(xpc_part, "can't register ACTIVATE IRQ handler, " | |
1287 | "errno=%d\n", -ret); | |
1288 | ||
1289 | xpc_restrict_IPI_ops(); | |
1290 | ||
1291 | if (xpc_sysctl) { | |
1292 | unregister_sysctl_table(xpc_sysctl); | |
1293 | } | |
7682a4c6 DN |
1294 | |
1295 | kfree(xpc_remote_copy_buffer_base); | |
89eb8eb9 DN |
1296 | return -EBUSY; |
1297 | } | |
1298 | ||
1299 | /* | |
1300 | * Fill the partition reserved page with the information needed by | |
1301 | * other partitions to discover we are alive and establish initial | |
1302 | * communications. | |
1303 | */ | |
1304 | xpc_rsvd_page = xpc_rsvd_page_init(); | |
1305 | if (xpc_rsvd_page == NULL) { | |
1306 | dev_err(xpc_part, "could not setup our reserved page\n"); | |
1307 | ||
1308 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
1309 | xpc_restrict_IPI_ops(); | |
1310 | ||
1311 | if (xpc_sysctl) { | |
1312 | unregister_sysctl_table(xpc_sysctl); | |
1313 | } | |
7682a4c6 DN |
1314 | |
1315 | kfree(xpc_remote_copy_buffer_base); | |
89eb8eb9 DN |
1316 | return -EBUSY; |
1317 | } | |
1318 | ||
1319 | ||
a607c389 DN |
1320 | /* add ourselves to the reboot_notifier_list */ |
1321 | ret = register_reboot_notifier(&xpc_reboot_notifier); | |
1322 | if (ret != 0) { | |
1323 | dev_warn(xpc_part, "can't register reboot notifier\n"); | |
1324 | } | |
1325 | ||
780d09e8 DN |
1326 | /* add ourselves to the die_notifier list (i.e., ia64die_chain) */ |
1327 | ret = register_die_notifier(&xpc_die_notifier); | |
1328 | if (ret != 0) { | |
1329 | dev_warn(xpc_part, "can't register die notifier\n"); | |
1330 | } | |
1331 | ||
a607c389 | 1332 | |
89eb8eb9 DN |
1333 | /* |
1334 | * Set the beating to other partitions into motion. This is | |
1335 | * the last requirement for other partitions' discovery to | |
1336 | * initiate communications with us. | |
1337 | */ | |
1338 | init_timer(&xpc_hb_timer); | |
1339 | xpc_hb_timer.function = xpc_hb_beater; | |
1340 | xpc_hb_beater(0); | |
1341 | ||
1342 | ||
1343 | /* | |
1344 | * The real work-horse behind xpc. This processes incoming | |
1345 | * interrupts and monitors remote heartbeats. | |
1346 | */ | |
1347 | pid = kernel_thread(xpc_hb_checker, NULL, 0); | |
1348 | if (pid < 0) { | |
1349 | dev_err(xpc_part, "failed while forking hb check thread\n"); | |
1350 | ||
1351 | /* indicate to others that our reserved page is uninitialized */ | |
1352 | xpc_rsvd_page->vars_pa = 0; | |
1353 | ||
a607c389 DN |
1354 | /* take ourselves off of the reboot_notifier_list */ |
1355 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | |
1356 | ||
780d09e8 DN |
1357 | /* take ourselves off of the die_notifier list */ |
1358 | (void) unregister_die_notifier(&xpc_die_notifier); | |
1359 | ||
89eb8eb9 DN |
1360 | del_timer_sync(&xpc_hb_timer); |
1361 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
1362 | xpc_restrict_IPI_ops(); | |
1363 | ||
1364 | if (xpc_sysctl) { | |
1365 | unregister_sysctl_table(xpc_sysctl); | |
1366 | } | |
7682a4c6 DN |
1367 | |
1368 | kfree(xpc_remote_copy_buffer_base); | |
89eb8eb9 DN |
1369 | return -EBUSY; |
1370 | } | |
1371 | ||
1372 | ||
1373 | /* | |
1374 | * Startup a thread that will attempt to discover other partitions to | |
1375 | * activate based on info provided by SAL. This new thread is short | |
1376 | * lived and will exit once discovery is complete. | |
1377 | */ | |
1378 | pid = kernel_thread(xpc_initiate_discovery, NULL, 0); | |
1379 | if (pid < 0) { | |
1380 | dev_err(xpc_part, "failed while forking discovery thread\n"); | |
1381 | ||
1382 | /* mark this new thread as a non-starter */ | |
f9e505a9 | 1383 | complete(&xpc_discovery_exited); |
89eb8eb9 | 1384 | |
a607c389 | 1385 | xpc_do_exit(xpcUnloading); |
89eb8eb9 DN |
1386 | return -EBUSY; |
1387 | } | |
1388 | ||
1389 | ||
1390 | /* set the interface to point at XPC's functions */ | |
1391 | xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect, | |
1392 | xpc_initiate_allocate, xpc_initiate_send, | |
1393 | xpc_initiate_send_notify, xpc_initiate_received, | |
1394 | xpc_initiate_partid_to_nasids); | |
1395 | ||
1396 | return 0; | |
1397 | } | |
1398 | module_init(xpc_init); | |
1399 | ||
1400 | ||
1401 | void __exit | |
1402 | xpc_exit(void) | |
1403 | { | |
a607c389 | 1404 | xpc_do_exit(xpcUnloading); |
89eb8eb9 DN |
1405 | } |
1406 | module_exit(xpc_exit); | |
1407 | ||
1408 | ||
1409 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
1410 | MODULE_DESCRIPTION("Cross Partition Communication (XPC) support"); | |
1411 | MODULE_LICENSE("GPL"); | |
1412 | ||
1413 | module_param(xpc_hb_interval, int, 0); | |
1414 | MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between " | |
1415 | "heartbeat increments."); | |
1416 | ||
1417 | module_param(xpc_hb_check_interval, int, 0); | |
1418 | MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between " | |
1419 | "heartbeat checks."); | |
1420 | ||
e54af724 DN |
1421 | module_param(xpc_disengage_request_timelimit, int, 0); |
1422 | MODULE_PARM_DESC(xpc_disengage_request_timelimit, "Number of seconds to wait " | |
1423 | "for disengage request to complete."); | |
1424 | ||
1f4674b2 DN |
1425 | module_param(xpc_kdebug_ignore, int, 0); |
1426 | MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by " | |
1427 | "other partitions when dropping into kdebug."); | |
1428 |