]>
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> | |
55 | #include <linux/slab.h> | |
69913927 | 56 | #include <linux/delay.h> |
a607c389 | 57 | #include <linux/reboot.h> |
f9e505a9 | 58 | #include <linux/completion.h> |
89eb8eb9 DN |
59 | #include <asm/sn/intr.h> |
60 | #include <asm/sn/sn_sal.h> | |
780d09e8 | 61 | #include <asm/kdebug.h> |
89eb8eb9 | 62 | #include <asm/uaccess.h> |
9335d48e | 63 | #include <asm/sn/xpc.h> |
89eb8eb9 DN |
64 | |
65 | ||
66 | /* define two XPC debug device structures to be used with dev_dbg() et al */ | |
67 | ||
68 | struct device_driver xpc_dbg_name = { | |
69 | .name = "xpc" | |
70 | }; | |
71 | ||
72 | struct device xpc_part_dbg_subname = { | |
73 | .bus_id = {0}, /* set to "part" at xpc_init() time */ | |
74 | .driver = &xpc_dbg_name | |
75 | }; | |
76 | ||
77 | struct device xpc_chan_dbg_subname = { | |
78 | .bus_id = {0}, /* set to "chan" at xpc_init() time */ | |
79 | .driver = &xpc_dbg_name | |
80 | }; | |
81 | ||
82 | struct device *xpc_part = &xpc_part_dbg_subname; | |
83 | struct device *xpc_chan = &xpc_chan_dbg_subname; | |
84 | ||
85 | ||
1f4674b2 DN |
86 | static int xpc_kdebug_ignore; |
87 | ||
88 | ||
89eb8eb9 DN |
89 | /* systune related variables for /proc/sys directories */ |
90 | ||
a607c389 DN |
91 | static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL; |
92 | static int xpc_hb_min_interval = 1; | |
93 | static int xpc_hb_max_interval = 10; | |
89eb8eb9 | 94 | |
a607c389 DN |
95 | static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL; |
96 | static int xpc_hb_check_min_interval = 10; | |
97 | static int xpc_hb_check_max_interval = 120; | |
89eb8eb9 | 98 | |
e54af724 DN |
99 | int xpc_disengage_request_timelimit = XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT; |
100 | static int xpc_disengage_request_min_timelimit = 0; | |
101 | static int xpc_disengage_request_max_timelimit = 120; | |
89eb8eb9 DN |
102 | |
103 | static ctl_table xpc_sys_xpc_hb_dir[] = { | |
104 | { | |
105 | 1, | |
106 | "hb_interval", | |
107 | &xpc_hb_interval, | |
108 | sizeof(int), | |
109 | 0644, | |
110 | NULL, | |
111 | &proc_dointvec_minmax, | |
112 | &sysctl_intvec, | |
113 | NULL, | |
a607c389 DN |
114 | &xpc_hb_min_interval, |
115 | &xpc_hb_max_interval | |
89eb8eb9 DN |
116 | }, |
117 | { | |
118 | 2, | |
119 | "hb_check_interval", | |
120 | &xpc_hb_check_interval, | |
121 | sizeof(int), | |
122 | 0644, | |
123 | NULL, | |
124 | &proc_dointvec_minmax, | |
125 | &sysctl_intvec, | |
126 | NULL, | |
a607c389 DN |
127 | &xpc_hb_check_min_interval, |
128 | &xpc_hb_check_max_interval | |
89eb8eb9 DN |
129 | }, |
130 | {0} | |
131 | }; | |
132 | static ctl_table xpc_sys_xpc_dir[] = { | |
133 | { | |
134 | 1, | |
135 | "hb", | |
136 | NULL, | |
137 | 0, | |
138 | 0555, | |
139 | xpc_sys_xpc_hb_dir | |
140 | }, | |
e54af724 DN |
141 | { |
142 | 2, | |
143 | "disengage_request_timelimit", | |
144 | &xpc_disengage_request_timelimit, | |
145 | sizeof(int), | |
146 | 0644, | |
147 | NULL, | |
148 | &proc_dointvec_minmax, | |
149 | &sysctl_intvec, | |
150 | NULL, | |
151 | &xpc_disengage_request_min_timelimit, | |
152 | &xpc_disengage_request_max_timelimit | |
153 | }, | |
89eb8eb9 DN |
154 | {0} |
155 | }; | |
156 | static ctl_table xpc_sys_dir[] = { | |
157 | { | |
158 | 1, | |
159 | "xpc", | |
160 | NULL, | |
161 | 0, | |
162 | 0555, | |
163 | xpc_sys_xpc_dir | |
164 | }, | |
165 | {0} | |
166 | }; | |
167 | static struct ctl_table_header *xpc_sysctl; | |
168 | ||
1ecaded8 DN |
169 | /* non-zero if any remote partition disengage request was timed out */ |
170 | int xpc_disengage_request_timedout; | |
89eb8eb9 DN |
171 | |
172 | /* #of IRQs received */ | |
173 | static atomic_t xpc_act_IRQ_rcvd; | |
174 | ||
175 | /* IRQ handler notifies this wait queue on receipt of an IRQ */ | |
176 | static DECLARE_WAIT_QUEUE_HEAD(xpc_act_IRQ_wq); | |
177 | ||
178 | static unsigned long xpc_hb_check_timeout; | |
179 | ||
e54af724 | 180 | /* notification that the xpc_hb_checker thread has exited */ |
f9e505a9 | 181 | static DECLARE_COMPLETION(xpc_hb_checker_exited); |
89eb8eb9 | 182 | |
e54af724 | 183 | /* notification that the xpc_discovery thread has exited */ |
f9e505a9 | 184 | static DECLARE_COMPLETION(xpc_discovery_exited); |
89eb8eb9 DN |
185 | |
186 | ||
187 | static struct timer_list xpc_hb_timer; | |
188 | ||
189 | ||
190 | static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *); | |
191 | ||
192 | ||
a607c389 DN |
193 | static int xpc_system_reboot(struct notifier_block *, unsigned long, void *); |
194 | static struct notifier_block xpc_reboot_notifier = { | |
195 | .notifier_call = xpc_system_reboot, | |
196 | }; | |
197 | ||
780d09e8 DN |
198 | static int xpc_system_die(struct notifier_block *, unsigned long, void *); |
199 | static struct notifier_block xpc_die_notifier = { | |
200 | .notifier_call = xpc_system_die, | |
201 | }; | |
202 | ||
a607c389 DN |
203 | |
204 | /* | |
205 | * Timer function to enforce the timelimit on the partition disengage request. | |
206 | */ | |
207 | static void | |
208 | xpc_timeout_partition_disengage_request(unsigned long data) | |
209 | { | |
210 | struct xpc_partition *part = (struct xpc_partition *) data; | |
211 | ||
212 | ||
e54af724 | 213 | DBUG_ON(jiffies < part->disengage_request_timeout); |
a607c389 DN |
214 | |
215 | (void) xpc_partition_disengaged(part); | |
216 | ||
217 | DBUG_ON(part->disengage_request_timeout != 0); | |
218 | DBUG_ON(xpc_partition_engaged(1UL << XPC_PARTID(part)) != 0); | |
219 | } | |
220 | ||
221 | ||
89eb8eb9 DN |
222 | /* |
223 | * Notify the heartbeat check thread that an IRQ has been received. | |
224 | */ | |
225 | static irqreturn_t | |
226 | xpc_act_IRQ_handler(int irq, void *dev_id, struct pt_regs *regs) | |
227 | { | |
228 | atomic_inc(&xpc_act_IRQ_rcvd); | |
229 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
230 | return IRQ_HANDLED; | |
231 | } | |
232 | ||
233 | ||
234 | /* | |
235 | * Timer to produce the heartbeat. The timer structures function is | |
236 | * already set when this is initially called. A tunable is used to | |
237 | * specify when the next timeout should occur. | |
238 | */ | |
239 | static void | |
240 | xpc_hb_beater(unsigned long dummy) | |
241 | { | |
242 | xpc_vars->heartbeat++; | |
243 | ||
244 | if (jiffies >= xpc_hb_check_timeout) { | |
245 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
246 | } | |
247 | ||
248 | xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ); | |
249 | add_timer(&xpc_hb_timer); | |
250 | } | |
251 | ||
252 | ||
253 | /* | |
254 | * This thread is responsible for nearly all of the partition | |
255 | * activation/deactivation. | |
256 | */ | |
257 | static int | |
258 | xpc_hb_checker(void *ignore) | |
259 | { | |
260 | int last_IRQ_count = 0; | |
261 | int new_IRQ_count; | |
262 | int force_IRQ=0; | |
263 | ||
264 | ||
265 | /* this thread was marked active by xpc_hb_init() */ | |
266 | ||
267 | daemonize(XPC_HB_CHECK_THREAD_NAME); | |
268 | ||
269 | set_cpus_allowed(current, cpumask_of_cpu(XPC_HB_CHECK_CPU)); | |
270 | ||
271 | xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ); | |
272 | ||
273 | while (!(volatile int) xpc_exiting) { | |
274 | ||
89eb8eb9 DN |
275 | dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have " |
276 | "been received\n", | |
277 | (int) (xpc_hb_check_timeout - jiffies), | |
278 | atomic_read(&xpc_act_IRQ_rcvd) - last_IRQ_count); | |
279 | ||
280 | ||
281 | /* checking of remote heartbeats is skewed by IRQ handling */ | |
282 | if (jiffies >= xpc_hb_check_timeout) { | |
283 | dev_dbg(xpc_part, "checking remote heartbeats\n"); | |
284 | xpc_check_remote_hb(); | |
285 | ||
286 | /* | |
287 | * We need to periodically recheck to ensure no | |
288 | * IPI/AMO pairs have been missed. That check | |
289 | * must always reset xpc_hb_check_timeout. | |
290 | */ | |
291 | force_IRQ = 1; | |
292 | } | |
293 | ||
294 | ||
a607c389 | 295 | /* check for outstanding IRQs */ |
89eb8eb9 DN |
296 | new_IRQ_count = atomic_read(&xpc_act_IRQ_rcvd); |
297 | if (last_IRQ_count < new_IRQ_count || force_IRQ != 0) { | |
298 | force_IRQ = 0; | |
299 | ||
300 | dev_dbg(xpc_part, "found an IRQ to process; will be " | |
301 | "resetting xpc_hb_check_timeout\n"); | |
302 | ||
303 | last_IRQ_count += xpc_identify_act_IRQ_sender(); | |
304 | if (last_IRQ_count < new_IRQ_count) { | |
305 | /* retry once to help avoid missing AMO */ | |
306 | (void) xpc_identify_act_IRQ_sender(); | |
307 | } | |
308 | last_IRQ_count = new_IRQ_count; | |
309 | ||
310 | xpc_hb_check_timeout = jiffies + | |
311 | (xpc_hb_check_interval * HZ); | |
312 | } | |
a607c389 DN |
313 | |
314 | /* wait for IRQ or timeout */ | |
315 | (void) wait_event_interruptible(xpc_act_IRQ_wq, | |
316 | (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) || | |
317 | jiffies >= xpc_hb_check_timeout || | |
318 | (volatile int) xpc_exiting)); | |
89eb8eb9 DN |
319 | } |
320 | ||
321 | dev_dbg(xpc_part, "heartbeat checker is exiting\n"); | |
322 | ||
323 | ||
e54af724 | 324 | /* mark this thread as having exited */ |
f9e505a9 | 325 | complete(&xpc_hb_checker_exited); |
89eb8eb9 DN |
326 | return 0; |
327 | } | |
328 | ||
329 | ||
330 | /* | |
331 | * This thread will attempt to discover other partitions to activate | |
332 | * based on info provided by SAL. This new thread is short lived and | |
333 | * will exit once discovery is complete. | |
334 | */ | |
335 | static int | |
336 | xpc_initiate_discovery(void *ignore) | |
337 | { | |
338 | daemonize(XPC_DISCOVERY_THREAD_NAME); | |
339 | ||
340 | xpc_discovery(); | |
341 | ||
342 | dev_dbg(xpc_part, "discovery thread is exiting\n"); | |
343 | ||
e54af724 | 344 | /* mark this thread as having exited */ |
f9e505a9 | 345 | complete(&xpc_discovery_exited); |
89eb8eb9 DN |
346 | return 0; |
347 | } | |
348 | ||
349 | ||
350 | /* | |
351 | * Establish first contact with the remote partititon. This involves pulling | |
352 | * the XPC per partition variables from the remote partition and waiting for | |
353 | * the remote partition to pull ours. | |
354 | */ | |
355 | static enum xpc_retval | |
356 | xpc_make_first_contact(struct xpc_partition *part) | |
357 | { | |
358 | enum xpc_retval ret; | |
359 | ||
360 | ||
361 | while ((ret = xpc_pull_remote_vars_part(part)) != xpcSuccess) { | |
362 | if (ret != xpcRetry) { | |
363 | XPC_DEACTIVATE_PARTITION(part, ret); | |
364 | return ret; | |
365 | } | |
366 | ||
367 | dev_dbg(xpc_chan, "waiting to make first contact with " | |
368 | "partition %d\n", XPC_PARTID(part)); | |
369 | ||
370 | /* wait a 1/4 of a second or so */ | |
a607c389 | 371 | (void) msleep_interruptible(250); |
89eb8eb9 DN |
372 | |
373 | if (part->act_state == XPC_P_DEACTIVATING) { | |
374 | return part->reason; | |
375 | } | |
376 | } | |
377 | ||
378 | return xpc_mark_partition_active(part); | |
379 | } | |
380 | ||
381 | ||
382 | /* | |
383 | * The first kthread assigned to a newly activated partition is the one | |
384 | * created by XPC HB with which it calls xpc_partition_up(). XPC hangs on to | |
385 | * that kthread until the partition is brought down, at which time that kthread | |
386 | * returns back to XPC HB. (The return of that kthread will signify to XPC HB | |
387 | * that XPC has dismantled all communication infrastructure for the associated | |
388 | * partition.) This kthread becomes the channel manager for that partition. | |
389 | * | |
390 | * Each active partition has a channel manager, who, besides connecting and | |
391 | * disconnecting channels, will ensure that each of the partition's connected | |
392 | * channels has the required number of assigned kthreads to get the work done. | |
393 | */ | |
394 | static void | |
395 | xpc_channel_mgr(struct xpc_partition *part) | |
396 | { | |
397 | while (part->act_state != XPC_P_DEACTIVATING || | |
a607c389 DN |
398 | atomic_read(&part->nchannels_active) > 0 || |
399 | !xpc_partition_disengaged(part)) { | |
89eb8eb9 DN |
400 | |
401 | xpc_process_channel_activity(part); | |
402 | ||
403 | ||
404 | /* | |
405 | * Wait until we've been requested to activate kthreads or | |
406 | * all of the channel's message queues have been torn down or | |
407 | * a signal is pending. | |
408 | * | |
409 | * The channel_mgr_requests is set to 1 after being awakened, | |
410 | * This is done to prevent the channel mgr from making one pass | |
411 | * through the loop for each request, since he will | |
412 | * be servicing all the requests in one pass. The reason it's | |
413 | * set to 1 instead of 0 is so that other kthreads will know | |
414 | * that the channel mgr is running and won't bother trying to | |
415 | * wake him up. | |
416 | */ | |
417 | atomic_dec(&part->channel_mgr_requests); | |
418 | (void) wait_event_interruptible(part->channel_mgr_wq, | |
419 | (atomic_read(&part->channel_mgr_requests) > 0 || | |
420 | (volatile u64) part->local_IPI_amo != 0 || | |
421 | ((volatile u8) part->act_state == | |
422 | XPC_P_DEACTIVATING && | |
a607c389 DN |
423 | atomic_read(&part->nchannels_active) == 0 && |
424 | xpc_partition_disengaged(part)))); | |
89eb8eb9 DN |
425 | atomic_set(&part->channel_mgr_requests, 1); |
426 | ||
427 | // >>> Does it need to wakeup periodically as well? In case we | |
428 | // >>> miscalculated the #of kthreads to wakeup or create? | |
429 | } | |
430 | } | |
431 | ||
432 | ||
433 | /* | |
434 | * When XPC HB determines that a partition has come up, it will create a new | |
435 | * kthread and that kthread will call this function to attempt to set up the | |
436 | * basic infrastructure used for Cross Partition Communication with the newly | |
437 | * upped partition. | |
438 | * | |
439 | * The kthread that was created by XPC HB and which setup the XPC | |
440 | * infrastructure will remain assigned to the partition until the partition | |
441 | * goes down. At which time the kthread will teardown the XPC infrastructure | |
442 | * and then exit. | |
443 | * | |
444 | * XPC HB will put the remote partition's XPC per partition specific variables | |
445 | * physical address into xpc_partitions[partid].remote_vars_part_pa prior to | |
446 | * calling xpc_partition_up(). | |
447 | */ | |
448 | static void | |
449 | xpc_partition_up(struct xpc_partition *part) | |
450 | { | |
451 | DBUG_ON(part->channels != NULL); | |
452 | ||
453 | dev_dbg(xpc_chan, "activating partition %d\n", XPC_PARTID(part)); | |
454 | ||
455 | if (xpc_setup_infrastructure(part) != xpcSuccess) { | |
456 | return; | |
457 | } | |
458 | ||
459 | /* | |
460 | * The kthread that XPC HB called us with will become the | |
461 | * channel manager for this partition. It will not return | |
462 | * back to XPC HB until the partition's XPC infrastructure | |
463 | * has been dismantled. | |
464 | */ | |
465 | ||
466 | (void) xpc_part_ref(part); /* this will always succeed */ | |
467 | ||
468 | if (xpc_make_first_contact(part) == xpcSuccess) { | |
469 | xpc_channel_mgr(part); | |
470 | } | |
471 | ||
472 | xpc_part_deref(part); | |
473 | ||
474 | xpc_teardown_infrastructure(part); | |
475 | } | |
476 | ||
477 | ||
478 | static int | |
479 | xpc_activating(void *__partid) | |
480 | { | |
481 | partid_t partid = (u64) __partid; | |
482 | struct xpc_partition *part = &xpc_partitions[partid]; | |
483 | unsigned long irq_flags; | |
d46523ea | 484 | struct sched_param param = { sched_priority: MAX_RT_PRIO - 1 }; |
89eb8eb9 DN |
485 | int ret; |
486 | ||
487 | ||
488 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | |
489 | ||
490 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
491 | ||
492 | if (part->act_state == XPC_P_DEACTIVATING) { | |
493 | part->act_state = XPC_P_INACTIVE; | |
494 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
495 | part->remote_rp_pa = 0; | |
496 | return 0; | |
497 | } | |
498 | ||
499 | /* indicate the thread is activating */ | |
500 | DBUG_ON(part->act_state != XPC_P_ACTIVATION_REQ); | |
501 | part->act_state = XPC_P_ACTIVATING; | |
502 | ||
503 | XPC_SET_REASON(part, 0, 0); | |
504 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
505 | ||
506 | dev_dbg(xpc_part, "bringing partition %d up\n", partid); | |
507 | ||
508 | daemonize("xpc%02d", partid); | |
509 | ||
510 | /* | |
511 | * This thread needs to run at a realtime priority to prevent a | |
512 | * significant performance degradation. | |
513 | */ | |
514 | ret = sched_setscheduler(current, SCHED_FIFO, ¶m); | |
515 | if (ret != 0) { | |
516 | dev_warn(xpc_part, "unable to set pid %d to a realtime " | |
517 | "priority, ret=%d\n", current->pid, ret); | |
518 | } | |
519 | ||
520 | /* allow this thread and its children to run on any CPU */ | |
521 | set_cpus_allowed(current, CPU_MASK_ALL); | |
522 | ||
523 | /* | |
524 | * Register the remote partition's AMOs with SAL so it can handle | |
525 | * and cleanup errors within that address range should the remote | |
526 | * partition go down. We don't unregister this range because it is | |
527 | * difficult to tell when outstanding writes to the remote partition | |
528 | * are finished and thus when it is safe to unregister. This should | |
529 | * not result in wasted space in the SAL xp_addr_region table because | |
530 | * we should get the same page for remote_amos_page_pa after module | |
531 | * reloads and system reboots. | |
532 | */ | |
533 | if (sn_register_xp_addr_region(part->remote_amos_page_pa, | |
534 | PAGE_SIZE, 1) < 0) { | |
535 | dev_warn(xpc_part, "xpc_partition_up(%d) failed to register " | |
536 | "xp_addr region\n", partid); | |
537 | ||
538 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
539 | part->act_state = XPC_P_INACTIVE; | |
540 | XPC_SET_REASON(part, xpcPhysAddrRegFailed, __LINE__); | |
541 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
542 | part->remote_rp_pa = 0; | |
543 | return 0; | |
544 | } | |
545 | ||
a607c389 | 546 | xpc_allow_hb(partid, xpc_vars); |
89eb8eb9 DN |
547 | xpc_IPI_send_activated(part); |
548 | ||
549 | ||
550 | /* | |
551 | * xpc_partition_up() holds this thread and marks this partition as | |
552 | * XPC_P_ACTIVE by calling xpc_hb_mark_active(). | |
553 | */ | |
554 | (void) xpc_partition_up(part); | |
555 | ||
a607c389 | 556 | xpc_disallow_hb(partid, xpc_vars); |
89eb8eb9 DN |
557 | xpc_mark_partition_inactive(part); |
558 | ||
559 | if (part->reason == xpcReactivating) { | |
560 | /* interrupting ourselves results in activating partition */ | |
561 | xpc_IPI_send_reactivate(part); | |
562 | } | |
563 | ||
564 | return 0; | |
565 | } | |
566 | ||
567 | ||
568 | void | |
569 | xpc_activate_partition(struct xpc_partition *part) | |
570 | { | |
571 | partid_t partid = XPC_PARTID(part); | |
572 | unsigned long irq_flags; | |
573 | pid_t pid; | |
574 | ||
575 | ||
576 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
577 | ||
89eb8eb9 DN |
578 | DBUG_ON(part->act_state != XPC_P_INACTIVE); |
579 | ||
7c6c6636 RH |
580 | part->act_state = XPC_P_ACTIVATION_REQ; |
581 | XPC_SET_REASON(part, xpcCloneKThread, __LINE__); | |
89eb8eb9 DN |
582 | |
583 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
7c6c6636 RH |
584 | |
585 | pid = kernel_thread(xpc_activating, (void *) ((u64) partid), 0); | |
586 | ||
587 | if (unlikely(pid <= 0)) { | |
588 | spin_lock_irqsave(&part->act_lock, irq_flags); | |
589 | part->act_state = XPC_P_INACTIVE; | |
590 | XPC_SET_REASON(part, xpcCloneKThreadFailed, __LINE__); | |
591 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | |
592 | } | |
89eb8eb9 DN |
593 | } |
594 | ||
595 | ||
596 | /* | |
597 | * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified | |
598 | * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more | |
599 | * than one partition, we use an AMO_t structure per partition to indicate | |
600 | * whether a partition has sent an IPI or not. >>> If it has, then wake up the | |
601 | * associated kthread to handle it. | |
602 | * | |
603 | * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IPIs sent by XPC | |
604 | * running on other partitions. | |
605 | * | |
606 | * Noteworthy Arguments: | |
607 | * | |
608 | * irq - Interrupt ReQuest number. NOT USED. | |
609 | * | |
610 | * dev_id - partid of IPI's potential sender. | |
611 | * | |
612 | * regs - processor's context before the processor entered | |
613 | * interrupt code. NOT USED. | |
614 | */ | |
615 | irqreturn_t | |
616 | xpc_notify_IRQ_handler(int irq, void *dev_id, struct pt_regs *regs) | |
617 | { | |
618 | partid_t partid = (partid_t) (u64) dev_id; | |
619 | struct xpc_partition *part = &xpc_partitions[partid]; | |
620 | ||
621 | ||
622 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | |
623 | ||
624 | if (xpc_part_ref(part)) { | |
625 | xpc_check_for_channel_activity(part); | |
626 | ||
627 | xpc_part_deref(part); | |
628 | } | |
629 | return IRQ_HANDLED; | |
630 | } | |
631 | ||
632 | ||
633 | /* | |
634 | * Check to see if xpc_notify_IRQ_handler() dropped any IPIs on the floor | |
635 | * because the write to their associated IPI amo completed after the IRQ/IPI | |
636 | * was received. | |
637 | */ | |
638 | void | |
639 | xpc_dropped_IPI_check(struct xpc_partition *part) | |
640 | { | |
641 | if (xpc_part_ref(part)) { | |
642 | xpc_check_for_channel_activity(part); | |
643 | ||
644 | part->dropped_IPI_timer.expires = jiffies + | |
645 | XPC_P_DROPPED_IPI_WAIT; | |
646 | add_timer(&part->dropped_IPI_timer); | |
647 | xpc_part_deref(part); | |
648 | } | |
649 | } | |
650 | ||
651 | ||
652 | void | |
653 | xpc_activate_kthreads(struct xpc_channel *ch, int needed) | |
654 | { | |
655 | int idle = atomic_read(&ch->kthreads_idle); | |
656 | int assigned = atomic_read(&ch->kthreads_assigned); | |
657 | int wakeup; | |
658 | ||
659 | ||
660 | DBUG_ON(needed <= 0); | |
661 | ||
662 | if (idle > 0) { | |
663 | wakeup = (needed > idle) ? idle : needed; | |
664 | needed -= wakeup; | |
665 | ||
666 | dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, " | |
667 | "channel=%d\n", wakeup, ch->partid, ch->number); | |
668 | ||
669 | /* only wakeup the requested number of kthreads */ | |
670 | wake_up_nr(&ch->idle_wq, wakeup); | |
671 | } | |
672 | ||
673 | if (needed <= 0) { | |
674 | return; | |
675 | } | |
676 | ||
677 | if (needed + assigned > ch->kthreads_assigned_limit) { | |
678 | needed = ch->kthreads_assigned_limit - assigned; | |
679 | // >>>should never be less than 0 | |
680 | if (needed <= 0) { | |
681 | return; | |
682 | } | |
683 | } | |
684 | ||
685 | dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n", | |
686 | needed, ch->partid, ch->number); | |
687 | ||
688 | xpc_create_kthreads(ch, needed); | |
689 | } | |
690 | ||
691 | ||
692 | /* | |
693 | * This function is where XPC's kthreads wait for messages to deliver. | |
694 | */ | |
695 | static void | |
696 | xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch) | |
697 | { | |
698 | do { | |
699 | /* deliver messages to their intended recipients */ | |
700 | ||
701 | while ((volatile s64) ch->w_local_GP.get < | |
702 | (volatile s64) ch->w_remote_GP.put && | |
703 | !((volatile u32) ch->flags & | |
704 | XPC_C_DISCONNECTING)) { | |
705 | xpc_deliver_msg(ch); | |
706 | } | |
707 | ||
708 | if (atomic_inc_return(&ch->kthreads_idle) > | |
709 | ch->kthreads_idle_limit) { | |
710 | /* too many idle kthreads on this channel */ | |
711 | atomic_dec(&ch->kthreads_idle); | |
712 | break; | |
713 | } | |
714 | ||
715 | dev_dbg(xpc_chan, "idle kthread calling " | |
716 | "wait_event_interruptible_exclusive()\n"); | |
717 | ||
718 | (void) wait_event_interruptible_exclusive(ch->idle_wq, | |
719 | ((volatile s64) ch->w_local_GP.get < | |
720 | (volatile s64) ch->w_remote_GP.put || | |
721 | ((volatile u32) ch->flags & | |
722 | XPC_C_DISCONNECTING))); | |
723 | ||
724 | atomic_dec(&ch->kthreads_idle); | |
725 | ||
726 | } while (!((volatile u32) ch->flags & XPC_C_DISCONNECTING)); | |
727 | } | |
728 | ||
729 | ||
730 | static int | |
731 | xpc_daemonize_kthread(void *args) | |
732 | { | |
733 | partid_t partid = XPC_UNPACK_ARG1(args); | |
734 | u16 ch_number = XPC_UNPACK_ARG2(args); | |
735 | struct xpc_partition *part = &xpc_partitions[partid]; | |
736 | struct xpc_channel *ch; | |
737 | int n_needed; | |
e54af724 | 738 | unsigned long irq_flags; |
89eb8eb9 DN |
739 | |
740 | ||
741 | daemonize("xpc%02dc%d", partid, ch_number); | |
742 | ||
743 | dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n", | |
744 | partid, ch_number); | |
745 | ||
746 | ch = &part->channels[ch_number]; | |
747 | ||
748 | if (!(ch->flags & XPC_C_DISCONNECTING)) { | |
89eb8eb9 DN |
749 | |
750 | /* let registerer know that connection has been established */ | |
751 | ||
e54af724 DN |
752 | spin_lock_irqsave(&ch->lock, irq_flags); |
753 | if (!(ch->flags & XPC_C_CONNECTCALLOUT)) { | |
754 | ch->flags |= XPC_C_CONNECTCALLOUT; | |
755 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
756 | ||
89eb8eb9 DN |
757 | xpc_connected_callout(ch); |
758 | ||
759 | /* | |
760 | * It is possible that while the callout was being | |
761 | * made that the remote partition sent some messages. | |
762 | * If that is the case, we may need to activate | |
763 | * additional kthreads to help deliver them. We only | |
764 | * need one less than total #of messages to deliver. | |
765 | */ | |
766 | n_needed = ch->w_remote_GP.put - ch->w_local_GP.get - 1; | |
767 | if (n_needed > 0 && | |
768 | !(ch->flags & XPC_C_DISCONNECTING)) { | |
769 | xpc_activate_kthreads(ch, n_needed); | |
770 | } | |
e54af724 DN |
771 | } else { |
772 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
89eb8eb9 DN |
773 | } |
774 | ||
775 | xpc_kthread_waitmsgs(part, ch); | |
776 | } | |
777 | ||
a607c389 | 778 | if (atomic_dec_return(&ch->kthreads_assigned) == 0) { |
e54af724 DN |
779 | spin_lock_irqsave(&ch->lock, irq_flags); |
780 | if ((ch->flags & XPC_C_CONNECTCALLOUT) && | |
781 | !(ch->flags & XPC_C_DISCONNECTCALLOUT)) { | |
782 | ch->flags |= XPC_C_DISCONNECTCALLOUT; | |
783 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
784 | ||
246c7e33 | 785 | xpc_disconnect_callout(ch, xpcDisconnecting); |
e54af724 DN |
786 | } else { |
787 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
a607c389 DN |
788 | } |
789 | if (atomic_dec_return(&part->nchannels_engaged) == 0) { | |
790 | xpc_mark_partition_disengaged(part); | |
791 | xpc_IPI_send_disengage(part); | |
792 | } | |
89eb8eb9 DN |
793 | } |
794 | ||
795 | ||
796 | xpc_msgqueue_deref(ch); | |
797 | ||
798 | dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n", | |
799 | partid, ch_number); | |
800 | ||
801 | xpc_part_deref(part); | |
802 | return 0; | |
803 | } | |
804 | ||
805 | ||
806 | /* | |
807 | * For each partition that XPC has established communications with, there is | |
808 | * a minimum of one kernel thread assigned to perform any operation that | |
809 | * may potentially sleep or block (basically the callouts to the asynchronous | |
810 | * functions registered via xpc_connect()). | |
811 | * | |
812 | * Additional kthreads are created and destroyed by XPC as the workload | |
813 | * demands. | |
814 | * | |
815 | * A kthread is assigned to one of the active channels that exists for a given | |
816 | * partition. | |
817 | */ | |
818 | void | |
819 | xpc_create_kthreads(struct xpc_channel *ch, int needed) | |
820 | { | |
821 | unsigned long irq_flags; | |
822 | pid_t pid; | |
823 | u64 args = XPC_PACK_ARGS(ch->partid, ch->number); | |
a607c389 | 824 | struct xpc_partition *part = &xpc_partitions[ch->partid]; |
89eb8eb9 DN |
825 | |
826 | ||
827 | while (needed-- > 0) { | |
e54af724 DN |
828 | |
829 | /* | |
830 | * The following is done on behalf of the newly created | |
831 | * kthread. That kthread is responsible for doing the | |
832 | * counterpart to the following before it exits. | |
833 | */ | |
834 | (void) xpc_part_ref(part); | |
835 | xpc_msgqueue_ref(ch); | |
836 | if (atomic_inc_return(&ch->kthreads_assigned) == 1 && | |
837 | atomic_inc_return(&part->nchannels_engaged) == 1) { | |
838 | xpc_mark_partition_engaged(part); | |
839 | } | |
840 | ||
89eb8eb9 DN |
841 | pid = kernel_thread(xpc_daemonize_kthread, (void *) args, 0); |
842 | if (pid < 0) { | |
843 | /* the fork failed */ | |
e54af724 DN |
844 | if (atomic_dec_return(&ch->kthreads_assigned) == 0 && |
845 | atomic_dec_return(&part->nchannels_engaged) == 0) { | |
846 | xpc_mark_partition_disengaged(part); | |
847 | xpc_IPI_send_disengage(part); | |
848 | } | |
849 | xpc_msgqueue_deref(ch); | |
850 | xpc_part_deref(part); | |
89eb8eb9 DN |
851 | |
852 | if (atomic_read(&ch->kthreads_assigned) < | |
853 | ch->kthreads_idle_limit) { | |
854 | /* | |
855 | * Flag this as an error only if we have an | |
856 | * insufficient #of kthreads for the channel | |
857 | * to function. | |
858 | * | |
859 | * No xpc_msgqueue_ref() is needed here since | |
860 | * the channel mgr is doing this. | |
861 | */ | |
862 | spin_lock_irqsave(&ch->lock, irq_flags); | |
863 | XPC_DISCONNECT_CHANNEL(ch, xpcLackOfResources, | |
864 | &irq_flags); | |
865 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
866 | } | |
867 | break; | |
868 | } | |
869 | ||
89eb8eb9 DN |
870 | ch->kthreads_created++; // >>> temporary debug only!!! |
871 | } | |
872 | } | |
873 | ||
874 | ||
875 | void | |
876 | xpc_disconnect_wait(int ch_number) | |
877 | { | |
a607c389 | 878 | unsigned long irq_flags; |
89eb8eb9 DN |
879 | partid_t partid; |
880 | struct xpc_partition *part; | |
881 | struct xpc_channel *ch; | |
e54af724 | 882 | int wakeup_channel_mgr; |
89eb8eb9 DN |
883 | |
884 | ||
885 | /* now wait for all callouts to the caller's function to cease */ | |
886 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
887 | part = &xpc_partitions[partid]; | |
888 | ||
e54af724 DN |
889 | if (!xpc_part_ref(part)) { |
890 | continue; | |
891 | } | |
89eb8eb9 | 892 | |
e54af724 | 893 | ch = &part->channels[ch_number]; |
89eb8eb9 | 894 | |
e54af724 | 895 | if (!(ch->flags & XPC_C_WDISCONNECT)) { |
89eb8eb9 | 896 | xpc_part_deref(part); |
e54af724 | 897 | continue; |
89eb8eb9 | 898 | } |
e54af724 | 899 | |
f9e505a9 | 900 | wait_for_completion(&ch->wdisconnect_wait); |
e54af724 DN |
901 | |
902 | spin_lock_irqsave(&ch->lock, irq_flags); | |
903 | DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED)); | |
904 | wakeup_channel_mgr = 0; | |
905 | ||
906 | if (ch->delayed_IPI_flags) { | |
907 | if (part->act_state != XPC_P_DEACTIVATING) { | |
908 | spin_lock(&part->IPI_lock); | |
909 | XPC_SET_IPI_FLAGS(part->local_IPI_amo, | |
910 | ch->number, ch->delayed_IPI_flags); | |
911 | spin_unlock(&part->IPI_lock); | |
912 | wakeup_channel_mgr = 1; | |
913 | } | |
914 | ch->delayed_IPI_flags = 0; | |
89eb8eb9 | 915 | } |
e54af724 DN |
916 | |
917 | ch->flags &= ~XPC_C_WDISCONNECT; | |
918 | spin_unlock_irqrestore(&ch->lock, irq_flags); | |
919 | ||
920 | if (wakeup_channel_mgr) { | |
921 | xpc_wakeup_channel_mgr(part); | |
922 | } | |
923 | ||
924 | xpc_part_deref(part); | |
89eb8eb9 DN |
925 | } |
926 | } | |
927 | ||
928 | ||
929 | static void | |
a607c389 | 930 | xpc_do_exit(enum xpc_retval reason) |
89eb8eb9 DN |
931 | { |
932 | partid_t partid; | |
1ecaded8 | 933 | int active_part_count, printed_waiting_msg = 0; |
89eb8eb9 | 934 | struct xpc_partition *part; |
1ecaded8 | 935 | unsigned long printmsg_time, disengage_request_timeout = 0; |
89eb8eb9 DN |
936 | |
937 | ||
a607c389 DN |
938 | /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */ |
939 | DBUG_ON(xpc_exiting == 1); | |
89eb8eb9 DN |
940 | |
941 | /* | |
a607c389 DN |
942 | * Let the heartbeat checker thread and the discovery thread |
943 | * (if one is running) know that they should exit. Also wake up | |
944 | * the heartbeat checker thread in case it's sleeping. | |
89eb8eb9 DN |
945 | */ |
946 | xpc_exiting = 1; | |
947 | wake_up_interruptible(&xpc_act_IRQ_wq); | |
948 | ||
a607c389 DN |
949 | /* ignore all incoming interrupts */ |
950 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
89eb8eb9 | 951 | |
e54af724 | 952 | /* wait for the discovery thread to exit */ |
f9e505a9 | 953 | wait_for_completion(&xpc_discovery_exited); |
89eb8eb9 | 954 | |
e54af724 | 955 | /* wait for the heartbeat checker thread to exit */ |
f9e505a9 | 956 | wait_for_completion(&xpc_hb_checker_exited); |
89eb8eb9 | 957 | |
89eb8eb9 | 958 | |
a607c389 DN |
959 | /* sleep for a 1/3 of a second or so */ |
960 | (void) msleep_interruptible(300); | |
89eb8eb9 DN |
961 | |
962 | ||
963 | /* wait for all partitions to become inactive */ | |
964 | ||
1ecaded8 DN |
965 | printmsg_time = jiffies + (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); |
966 | xpc_disengage_request_timedout = 0; | |
a607c389 | 967 | |
89eb8eb9 DN |
968 | do { |
969 | active_part_count = 0; | |
970 | ||
971 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
972 | part = &xpc_partitions[partid]; | |
89eb8eb9 | 973 | |
a607c389 DN |
974 | if (xpc_partition_disengaged(part) && |
975 | part->act_state == XPC_P_INACTIVE) { | |
976 | continue; | |
89eb8eb9 | 977 | } |
a607c389 DN |
978 | |
979 | active_part_count++; | |
980 | ||
981 | XPC_DEACTIVATE_PARTITION(part, reason); | |
89eb8eb9 | 982 | |
1ecaded8 DN |
983 | if (part->disengage_request_timeout > |
984 | disengage_request_timeout) { | |
985 | disengage_request_timeout = | |
986 | part->disengage_request_timeout; | |
987 | } | |
a607c389 | 988 | } |
89eb8eb9 | 989 | |
1ecaded8 DN |
990 | if (xpc_partition_engaged(-1UL)) { |
991 | if (time_after(jiffies, printmsg_time)) { | |
992 | dev_info(xpc_part, "waiting for remote " | |
993 | "partitions to disengage, timeout in " | |
994 | "%ld seconds\n", | |
995 | (disengage_request_timeout - jiffies) | |
996 | / HZ); | |
997 | printmsg_time = jiffies + | |
a607c389 | 998 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); |
1ecaded8 DN |
999 | printed_waiting_msg = 1; |
1000 | } | |
1001 | ||
1002 | } else if (active_part_count > 0) { | |
1003 | if (printed_waiting_msg) { | |
1004 | dev_info(xpc_part, "waiting for local partition" | |
1005 | " to disengage\n"); | |
1006 | printed_waiting_msg = 0; | |
1007 | } | |
1008 | ||
1009 | } else { | |
1010 | if (!xpc_disengage_request_timedout) { | |
1011 | dev_info(xpc_part, "all partitions have " | |
1012 | "disengaged\n"); | |
1013 | } | |
1014 | break; | |
89eb8eb9 DN |
1015 | } |
1016 | ||
a607c389 DN |
1017 | /* sleep for a 1/3 of a second or so */ |
1018 | (void) msleep_interruptible(300); | |
1019 | ||
1020 | } while (1); | |
1021 | ||
1022 | DBUG_ON(xpc_partition_engaged(-1UL)); | |
1023 | ||
1024 | ||
1025 | /* indicate to others that our reserved page is uninitialized */ | |
1026 | xpc_rsvd_page->vars_pa = 0; | |
1027 | ||
1028 | /* now it's time to eliminate our heartbeat */ | |
1029 | del_timer_sync(&xpc_hb_timer); | |
e54af724 | 1030 | DBUG_ON(xpc_vars->heartbeating_to_mask != 0); |
89eb8eb9 | 1031 | |
0752c670 DN |
1032 | if (reason == xpcUnloading) { |
1033 | /* take ourselves off of the reboot_notifier_list */ | |
1034 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | |
89eb8eb9 | 1035 | |
0752c670 DN |
1036 | /* take ourselves off of the die_notifier list */ |
1037 | (void) unregister_die_notifier(&xpc_die_notifier); | |
1038 | } | |
780d09e8 | 1039 | |
89eb8eb9 DN |
1040 | /* close down protections for IPI operations */ |
1041 | xpc_restrict_IPI_ops(); | |
1042 | ||
1043 | ||
1044 | /* clear the interface to XPC's functions */ | |
1045 | xpc_clear_interface(); | |
1046 | ||
1047 | if (xpc_sysctl) { | |
1048 | unregister_sysctl_table(xpc_sysctl); | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | ||
780d09e8 | 1053 | /* |
d6ad033a DN |
1054 | * This function is called when the system is being rebooted. |
1055 | */ | |
1056 | static int | |
1057 | xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused) | |
1058 | { | |
1059 | enum xpc_retval reason; | |
1060 | ||
1061 | ||
1062 | switch (event) { | |
1063 | case SYS_RESTART: | |
1064 | reason = xpcSystemReboot; | |
1065 | break; | |
1066 | case SYS_HALT: | |
1067 | reason = xpcSystemHalt; | |
1068 | break; | |
1069 | case SYS_POWER_OFF: | |
1070 | reason = xpcSystemPoweroff; | |
1071 | break; | |
1072 | default: | |
1073 | reason = xpcSystemGoingDown; | |
1074 | } | |
1075 | ||
1076 | xpc_do_exit(reason); | |
1077 | return NOTIFY_DONE; | |
1078 | } | |
1079 | ||
1080 | ||
1081 | /* | |
1082 | * Notify other partitions to disengage from all references to our memory. | |
780d09e8 DN |
1083 | */ |
1084 | static void | |
1085 | xpc_die_disengage(void) | |
1086 | { | |
1087 | struct xpc_partition *part; | |
1088 | partid_t partid; | |
1089 | unsigned long engaged; | |
1ecaded8 | 1090 | long time, printmsg_time, disengage_request_timeout; |
780d09e8 DN |
1091 | |
1092 | ||
1093 | /* keep xpc_hb_checker thread from doing anything (just in case) */ | |
1094 | xpc_exiting = 1; | |
1095 | ||
1096 | xpc_vars->heartbeating_to_mask = 0; /* indicate we're deactivated */ | |
1097 | ||
1098 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1099 | part = &xpc_partitions[partid]; | |
1100 | ||
1101 | if (!XPC_SUPPORTS_DISENGAGE_REQUEST(part-> | |
1102 | remote_vars_version)) { | |
1103 | ||
1104 | /* just in case it was left set by an earlier XPC */ | |
1105 | xpc_clear_partition_engaged(1UL << partid); | |
1106 | continue; | |
1107 | } | |
1108 | ||
1109 | if (xpc_partition_engaged(1UL << partid) || | |
1110 | part->act_state != XPC_P_INACTIVE) { | |
1111 | xpc_request_partition_disengage(part); | |
1112 | xpc_mark_partition_disengaged(part); | |
1113 | xpc_IPI_send_disengage(part); | |
1114 | } | |
1115 | } | |
1116 | ||
1ecaded8 DN |
1117 | time = rtc_time(); |
1118 | printmsg_time = time + | |
1119 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * sn_rtc_cycles_per_second); | |
1120 | disengage_request_timeout = time + | |
780d09e8 DN |
1121 | (xpc_disengage_request_timelimit * sn_rtc_cycles_per_second); |
1122 | ||
1123 | /* wait for all other partitions to disengage from us */ | |
1124 | ||
1ecaded8 DN |
1125 | while (1) { |
1126 | engaged = xpc_partition_engaged(-1UL); | |
1127 | if (!engaged) { | |
1128 | dev_info(xpc_part, "all partitions have disengaged\n"); | |
1129 | break; | |
1130 | } | |
780d09e8 | 1131 | |
1ecaded8 DN |
1132 | time = rtc_time(); |
1133 | if (time >= disengage_request_timeout) { | |
1134 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1135 | if (engaged & (1UL << partid)) { | |
1136 | dev_info(xpc_part, "disengage from " | |
1137 | "remote partition %d timed " | |
1138 | "out\n", partid); | |
1139 | } | |
1140 | } | |
1141 | break; | |
1142 | } | |
1143 | ||
1144 | if (time >= printmsg_time) { | |
780d09e8 | 1145 | dev_info(xpc_part, "waiting for remote partitions to " |
1ecaded8 DN |
1146 | "disengage, timeout in %ld seconds\n", |
1147 | (disengage_request_timeout - time) / | |
1148 | sn_rtc_cycles_per_second); | |
1149 | printmsg_time = time + | |
1150 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * | |
780d09e8 DN |
1151 | sn_rtc_cycles_per_second); |
1152 | } | |
1153 | } | |
780d09e8 DN |
1154 | } |
1155 | ||
1156 | ||
780d09e8 | 1157 | /* |
1f4674b2 DN |
1158 | * This function is called when the system is being restarted or halted due |
1159 | * to some sort of system failure. If this is the case we need to notify the | |
1160 | * other partitions to disengage from all references to our memory. | |
1161 | * This function can also be called when our heartbeater could be offlined | |
1162 | * for a time. In this case we need to notify other partitions to not worry | |
1163 | * about the lack of a heartbeat. | |
780d09e8 DN |
1164 | */ |
1165 | static int | |
1166 | xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused) | |
1167 | { | |
1168 | switch (event) { | |
1169 | case DIE_MACHINE_RESTART: | |
1170 | case DIE_MACHINE_HALT: | |
1171 | xpc_die_disengage(); | |
1172 | break; | |
1f4674b2 DN |
1173 | |
1174 | case DIE_KDEBUG_ENTER: | |
1175 | /* Should lack of heartbeat be ignored by other partitions? */ | |
1176 | if (!xpc_kdebug_ignore) { | |
1177 | break; | |
1178 | } | |
1179 | /* fall through */ | |
780d09e8 DN |
1180 | case DIE_MCA_MONARCH_ENTER: |
1181 | case DIE_INIT_MONARCH_ENTER: | |
1182 | xpc_vars->heartbeat++; | |
1183 | xpc_vars->heartbeat_offline = 1; | |
1184 | break; | |
1f4674b2 DN |
1185 | |
1186 | case DIE_KDEBUG_LEAVE: | |
1187 | /* Is lack of heartbeat being ignored by other partitions? */ | |
1188 | if (!xpc_kdebug_ignore) { | |
1189 | break; | |
1190 | } | |
1191 | /* fall through */ | |
780d09e8 DN |
1192 | case DIE_MCA_MONARCH_LEAVE: |
1193 | case DIE_INIT_MONARCH_LEAVE: | |
1194 | xpc_vars->heartbeat++; | |
1195 | xpc_vars->heartbeat_offline = 0; | |
1196 | break; | |
1197 | } | |
1198 | ||
1199 | return NOTIFY_DONE; | |
1200 | } | |
1201 | ||
1202 | ||
89eb8eb9 DN |
1203 | int __init |
1204 | xpc_init(void) | |
1205 | { | |
1206 | int ret; | |
1207 | partid_t partid; | |
1208 | struct xpc_partition *part; | |
1209 | pid_t pid; | |
1210 | ||
1211 | ||
408865ce DN |
1212 | if (!ia64_platform_is("sn2")) { |
1213 | return -ENODEV; | |
1214 | } | |
1215 | ||
89eb8eb9 DN |
1216 | /* |
1217 | * xpc_remote_copy_buffer is used as a temporary buffer for bte_copy'ng | |
4b38fcd4 DN |
1218 | * various portions of a partition's reserved page. Its size is based |
1219 | * on the size of the reserved page header and part_nasids mask. So we | |
1220 | * need to ensure that the other items will fit as well. | |
89eb8eb9 | 1221 | */ |
4b38fcd4 | 1222 | if (XPC_RP_VARS_SIZE > XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES) { |
89eb8eb9 DN |
1223 | dev_err(xpc_part, "xpc_remote_copy_buffer is not big enough\n"); |
1224 | return -EPERM; | |
1225 | } | |
1226 | DBUG_ON((u64) xpc_remote_copy_buffer != | |
1227 | L1_CACHE_ALIGN((u64) xpc_remote_copy_buffer)); | |
1228 | ||
1229 | snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part"); | |
1230 | snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan"); | |
1231 | ||
1232 | xpc_sysctl = register_sysctl_table(xpc_sys_dir, 1); | |
1233 | ||
1234 | /* | |
1235 | * The first few fields of each entry of xpc_partitions[] need to | |
1236 | * be initialized now so that calls to xpc_connect() and | |
1237 | * xpc_disconnect() can be made prior to the activation of any remote | |
1238 | * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE | |
1239 | * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING | |
1240 | * PARTITION HAS BEEN ACTIVATED. | |
1241 | */ | |
1242 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | |
1243 | part = &xpc_partitions[partid]; | |
1244 | ||
1245 | DBUG_ON((u64) part != L1_CACHE_ALIGN((u64) part)); | |
1246 | ||
1247 | part->act_IRQ_rcvd = 0; | |
1248 | spin_lock_init(&part->act_lock); | |
1249 | part->act_state = XPC_P_INACTIVE; | |
1250 | XPC_SET_REASON(part, 0, 0); | |
a607c389 DN |
1251 | |
1252 | init_timer(&part->disengage_request_timer); | |
1253 | part->disengage_request_timer.function = | |
1254 | xpc_timeout_partition_disengage_request; | |
1255 | part->disengage_request_timer.data = (unsigned long) part; | |
1256 | ||
89eb8eb9 DN |
1257 | part->setup_state = XPC_P_UNSET; |
1258 | init_waitqueue_head(&part->teardown_wq); | |
1259 | atomic_set(&part->references, 0); | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Open up protections for IPI operations (and AMO operations on | |
1264 | * Shub 1.1 systems). | |
1265 | */ | |
1266 | xpc_allow_IPI_ops(); | |
1267 | ||
1268 | /* | |
1269 | * Interrupts being processed will increment this atomic variable and | |
1270 | * awaken the heartbeat thread which will process the interrupts. | |
1271 | */ | |
1272 | atomic_set(&xpc_act_IRQ_rcvd, 0); | |
1273 | ||
1274 | /* | |
1275 | * This is safe to do before the xpc_hb_checker thread has started | |
1276 | * because the handler releases a wait queue. If an interrupt is | |
1277 | * received before the thread is waiting, it will not go to sleep, | |
1278 | * but rather immediately process the interrupt. | |
1279 | */ | |
1280 | ret = request_irq(SGI_XPC_ACTIVATE, xpc_act_IRQ_handler, 0, | |
1281 | "xpc hb", NULL); | |
1282 | if (ret != 0) { | |
1283 | dev_err(xpc_part, "can't register ACTIVATE IRQ handler, " | |
1284 | "errno=%d\n", -ret); | |
1285 | ||
1286 | xpc_restrict_IPI_ops(); | |
1287 | ||
1288 | if (xpc_sysctl) { | |
1289 | unregister_sysctl_table(xpc_sysctl); | |
1290 | } | |
1291 | return -EBUSY; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Fill the partition reserved page with the information needed by | |
1296 | * other partitions to discover we are alive and establish initial | |
1297 | * communications. | |
1298 | */ | |
1299 | xpc_rsvd_page = xpc_rsvd_page_init(); | |
1300 | if (xpc_rsvd_page == NULL) { | |
1301 | dev_err(xpc_part, "could not setup our reserved page\n"); | |
1302 | ||
1303 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
1304 | xpc_restrict_IPI_ops(); | |
1305 | ||
1306 | if (xpc_sysctl) { | |
1307 | unregister_sysctl_table(xpc_sysctl); | |
1308 | } | |
1309 | return -EBUSY; | |
1310 | } | |
1311 | ||
1312 | ||
a607c389 DN |
1313 | /* add ourselves to the reboot_notifier_list */ |
1314 | ret = register_reboot_notifier(&xpc_reboot_notifier); | |
1315 | if (ret != 0) { | |
1316 | dev_warn(xpc_part, "can't register reboot notifier\n"); | |
1317 | } | |
1318 | ||
780d09e8 DN |
1319 | /* add ourselves to the die_notifier list (i.e., ia64die_chain) */ |
1320 | ret = register_die_notifier(&xpc_die_notifier); | |
1321 | if (ret != 0) { | |
1322 | dev_warn(xpc_part, "can't register die notifier\n"); | |
1323 | } | |
1324 | ||
a607c389 | 1325 | |
89eb8eb9 DN |
1326 | /* |
1327 | * Set the beating to other partitions into motion. This is | |
1328 | * the last requirement for other partitions' discovery to | |
1329 | * initiate communications with us. | |
1330 | */ | |
1331 | init_timer(&xpc_hb_timer); | |
1332 | xpc_hb_timer.function = xpc_hb_beater; | |
1333 | xpc_hb_beater(0); | |
1334 | ||
1335 | ||
1336 | /* | |
1337 | * The real work-horse behind xpc. This processes incoming | |
1338 | * interrupts and monitors remote heartbeats. | |
1339 | */ | |
1340 | pid = kernel_thread(xpc_hb_checker, NULL, 0); | |
1341 | if (pid < 0) { | |
1342 | dev_err(xpc_part, "failed while forking hb check thread\n"); | |
1343 | ||
1344 | /* indicate to others that our reserved page is uninitialized */ | |
1345 | xpc_rsvd_page->vars_pa = 0; | |
1346 | ||
a607c389 DN |
1347 | /* take ourselves off of the reboot_notifier_list */ |
1348 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | |
1349 | ||
780d09e8 DN |
1350 | /* take ourselves off of the die_notifier list */ |
1351 | (void) unregister_die_notifier(&xpc_die_notifier); | |
1352 | ||
89eb8eb9 DN |
1353 | del_timer_sync(&xpc_hb_timer); |
1354 | free_irq(SGI_XPC_ACTIVATE, NULL); | |
1355 | xpc_restrict_IPI_ops(); | |
1356 | ||
1357 | if (xpc_sysctl) { | |
1358 | unregister_sysctl_table(xpc_sysctl); | |
1359 | } | |
1360 | return -EBUSY; | |
1361 | } | |
1362 | ||
1363 | ||
1364 | /* | |
1365 | * Startup a thread that will attempt to discover other partitions to | |
1366 | * activate based on info provided by SAL. This new thread is short | |
1367 | * lived and will exit once discovery is complete. | |
1368 | */ | |
1369 | pid = kernel_thread(xpc_initiate_discovery, NULL, 0); | |
1370 | if (pid < 0) { | |
1371 | dev_err(xpc_part, "failed while forking discovery thread\n"); | |
1372 | ||
1373 | /* mark this new thread as a non-starter */ | |
f9e505a9 | 1374 | complete(&xpc_discovery_exited); |
89eb8eb9 | 1375 | |
a607c389 | 1376 | xpc_do_exit(xpcUnloading); |
89eb8eb9 DN |
1377 | return -EBUSY; |
1378 | } | |
1379 | ||
1380 | ||
1381 | /* set the interface to point at XPC's functions */ | |
1382 | xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect, | |
1383 | xpc_initiate_allocate, xpc_initiate_send, | |
1384 | xpc_initiate_send_notify, xpc_initiate_received, | |
1385 | xpc_initiate_partid_to_nasids); | |
1386 | ||
1387 | return 0; | |
1388 | } | |
1389 | module_init(xpc_init); | |
1390 | ||
1391 | ||
1392 | void __exit | |
1393 | xpc_exit(void) | |
1394 | { | |
a607c389 | 1395 | xpc_do_exit(xpcUnloading); |
89eb8eb9 DN |
1396 | } |
1397 | module_exit(xpc_exit); | |
1398 | ||
1399 | ||
1400 | MODULE_AUTHOR("Silicon Graphics, Inc."); | |
1401 | MODULE_DESCRIPTION("Cross Partition Communication (XPC) support"); | |
1402 | MODULE_LICENSE("GPL"); | |
1403 | ||
1404 | module_param(xpc_hb_interval, int, 0); | |
1405 | MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between " | |
1406 | "heartbeat increments."); | |
1407 | ||
1408 | module_param(xpc_hb_check_interval, int, 0); | |
1409 | MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between " | |
1410 | "heartbeat checks."); | |
1411 | ||
e54af724 DN |
1412 | module_param(xpc_disengage_request_timelimit, int, 0); |
1413 | MODULE_PARM_DESC(xpc_disengage_request_timelimit, "Number of seconds to wait " | |
1414 | "for disengage request to complete."); | |
1415 | ||
1f4674b2 DN |
1416 | module_param(xpc_kdebug_ignore, int, 0); |
1417 | MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by " | |
1418 | "other partitions when dropping into kdebug."); | |
1419 |