diff options
author | Dean Nelson <dcn@sgi.com> | 2008-04-22 15:46:56 -0400 |
---|---|---|
committer | Tony Luck <tony.luck@intel.com> | 2008-04-22 18:08:17 -0400 |
commit | 45d9ca492e4bd1522d1b5bd125c2908f1cee3d4a (patch) | |
tree | dfbe831a5f71159855c3a252856664411ca53f8a /drivers/misc/sgi-xp/xpc_main.c | |
parent | 9010eff0eadfe4eb60c3f0c71573f0fc505c31e3 (diff) |
[IA64] move XP and XPC to drivers/misc/sgi-xp
Move XPC and XPNET from arch/ia64/sn/kernel to drivers/misc/sgi-xp.
Signed-off-by: Dean Nelson <dcn@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Diffstat (limited to 'drivers/misc/sgi-xp/xpc_main.c')
-rw-r--r-- | drivers/misc/sgi-xp/xpc_main.c | 1431 |
1 files changed, 1431 insertions, 0 deletions
diff --git a/drivers/misc/sgi-xp/xpc_main.c b/drivers/misc/sgi-xp/xpc_main.c new file mode 100644 index 000000000000..bdb2cf1fcbcc --- /dev/null +++ b/drivers/misc/sgi-xp/xpc_main.c | |||
@@ -0,0 +1,1431 @@ | |||
1 | /* | ||
2 | * This file is subject to the terms and conditions of the GNU General Public | ||
3 | * License. See the file "COPYING" in the main directory of this archive | ||
4 | * for more details. | ||
5 | * | ||
6 | * Copyright (c) 2004-2008 Silicon Graphics, Inc. All Rights Reserved. | ||
7 | */ | ||
8 | |||
9 | |||
10 | /* | ||
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/delay.h> | ||
56 | #include <linux/reboot.h> | ||
57 | #include <linux/completion.h> | ||
58 | #include <linux/kdebug.h> | ||
59 | #include <asm/sn/intr.h> | ||
60 | #include <asm/sn/sn_sal.h> | ||
61 | #include <asm/uaccess.h> | ||
62 | #include "xpc.h" | ||
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 | |||
85 | static int xpc_kdebug_ignore; | ||
86 | |||
87 | |||
88 | /* systune related variables for /proc/sys directories */ | ||
89 | |||
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; | ||
93 | |||
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; | ||
97 | |||
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; | ||
101 | |||
102 | static ctl_table xpc_sys_xpc_hb_dir[] = { | ||
103 | { | ||
104 | .ctl_name = CTL_UNNUMBERED, | ||
105 | .procname = "hb_interval", | ||
106 | .data = &xpc_hb_interval, | ||
107 | .maxlen = sizeof(int), | ||
108 | .mode = 0644, | ||
109 | .proc_handler = &proc_dointvec_minmax, | ||
110 | .strategy = &sysctl_intvec, | ||
111 | .extra1 = &xpc_hb_min_interval, | ||
112 | .extra2 = &xpc_hb_max_interval | ||
113 | }, | ||
114 | { | ||
115 | .ctl_name = CTL_UNNUMBERED, | ||
116 | .procname = "hb_check_interval", | ||
117 | .data = &xpc_hb_check_interval, | ||
118 | .maxlen = sizeof(int), | ||
119 | .mode = 0644, | ||
120 | .proc_handler = &proc_dointvec_minmax, | ||
121 | .strategy = &sysctl_intvec, | ||
122 | .extra1 = &xpc_hb_check_min_interval, | ||
123 | .extra2 = &xpc_hb_check_max_interval | ||
124 | }, | ||
125 | {} | ||
126 | }; | ||
127 | static ctl_table xpc_sys_xpc_dir[] = { | ||
128 | { | ||
129 | .ctl_name = CTL_UNNUMBERED, | ||
130 | .procname = "hb", | ||
131 | .mode = 0555, | ||
132 | .child = xpc_sys_xpc_hb_dir | ||
133 | }, | ||
134 | { | ||
135 | .ctl_name = CTL_UNNUMBERED, | ||
136 | .procname = "disengage_request_timelimit", | ||
137 | .data = &xpc_disengage_request_timelimit, | ||
138 | .maxlen = sizeof(int), | ||
139 | .mode = 0644, | ||
140 | .proc_handler = &proc_dointvec_minmax, | ||
141 | .strategy = &sysctl_intvec, | ||
142 | .extra1 = &xpc_disengage_request_min_timelimit, | ||
143 | .extra2 = &xpc_disengage_request_max_timelimit | ||
144 | }, | ||
145 | {} | ||
146 | }; | ||
147 | static ctl_table xpc_sys_dir[] = { | ||
148 | { | ||
149 | .ctl_name = CTL_UNNUMBERED, | ||
150 | .procname = "xpc", | ||
151 | .mode = 0555, | ||
152 | .child = xpc_sys_xpc_dir | ||
153 | }, | ||
154 | {} | ||
155 | }; | ||
156 | static struct ctl_table_header *xpc_sysctl; | ||
157 | |||
158 | /* non-zero if any remote partition disengage request was timed out */ | ||
159 | int xpc_disengage_request_timedout; | ||
160 | |||
161 | /* #of IRQs received */ | ||
162 | static atomic_t xpc_act_IRQ_rcvd; | ||
163 | |||
164 | /* IRQ handler notifies this wait queue on receipt of an IRQ */ | ||
165 | static DECLARE_WAIT_QUEUE_HEAD(xpc_act_IRQ_wq); | ||
166 | |||
167 | static unsigned long xpc_hb_check_timeout; | ||
168 | |||
169 | /* notification that the xpc_hb_checker thread has exited */ | ||
170 | static DECLARE_COMPLETION(xpc_hb_checker_exited); | ||
171 | |||
172 | /* notification that the xpc_discovery thread has exited */ | ||
173 | static DECLARE_COMPLETION(xpc_discovery_exited); | ||
174 | |||
175 | |||
176 | static struct timer_list xpc_hb_timer; | ||
177 | |||
178 | |||
179 | static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *); | ||
180 | |||
181 | |||
182 | static int xpc_system_reboot(struct notifier_block *, unsigned long, void *); | ||
183 | static struct notifier_block xpc_reboot_notifier = { | ||
184 | .notifier_call = xpc_system_reboot, | ||
185 | }; | ||
186 | |||
187 | static int xpc_system_die(struct notifier_block *, unsigned long, void *); | ||
188 | static struct notifier_block xpc_die_notifier = { | ||
189 | .notifier_call = xpc_system_die, | ||
190 | }; | ||
191 | |||
192 | |||
193 | /* | ||
194 | * Timer function to enforce the timelimit on the partition disengage request. | ||
195 | */ | ||
196 | static void | ||
197 | xpc_timeout_partition_disengage_request(unsigned long data) | ||
198 | { | ||
199 | struct xpc_partition *part = (struct xpc_partition *) data; | ||
200 | |||
201 | |||
202 | DBUG_ON(time_before(jiffies, part->disengage_request_timeout)); | ||
203 | |||
204 | (void) xpc_partition_disengaged(part); | ||
205 | |||
206 | DBUG_ON(part->disengage_request_timeout != 0); | ||
207 | DBUG_ON(xpc_partition_engaged(1UL << XPC_PARTID(part)) != 0); | ||
208 | } | ||
209 | |||
210 | |||
211 | /* | ||
212 | * Notify the heartbeat check thread that an IRQ has been received. | ||
213 | */ | ||
214 | static irqreturn_t | ||
215 | xpc_act_IRQ_handler(int irq, void *dev_id) | ||
216 | { | ||
217 | atomic_inc(&xpc_act_IRQ_rcvd); | ||
218 | wake_up_interruptible(&xpc_act_IRQ_wq); | ||
219 | return IRQ_HANDLED; | ||
220 | } | ||
221 | |||
222 | |||
223 | /* | ||
224 | * Timer to produce the heartbeat. The timer structures function is | ||
225 | * already set when this is initially called. A tunable is used to | ||
226 | * specify when the next timeout should occur. | ||
227 | */ | ||
228 | static void | ||
229 | xpc_hb_beater(unsigned long dummy) | ||
230 | { | ||
231 | xpc_vars->heartbeat++; | ||
232 | |||
233 | if (time_after_eq(jiffies, xpc_hb_check_timeout)) { | ||
234 | wake_up_interruptible(&xpc_act_IRQ_wq); | ||
235 | } | ||
236 | |||
237 | xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ); | ||
238 | add_timer(&xpc_hb_timer); | ||
239 | } | ||
240 | |||
241 | |||
242 | /* | ||
243 | * This thread is responsible for nearly all of the partition | ||
244 | * activation/deactivation. | ||
245 | */ | ||
246 | static int | ||
247 | xpc_hb_checker(void *ignore) | ||
248 | { | ||
249 | int last_IRQ_count = 0; | ||
250 | int new_IRQ_count; | ||
251 | int force_IRQ=0; | ||
252 | |||
253 | |||
254 | /* this thread was marked active by xpc_hb_init() */ | ||
255 | |||
256 | daemonize(XPC_HB_CHECK_THREAD_NAME); | ||
257 | |||
258 | set_cpus_allowed(current, cpumask_of_cpu(XPC_HB_CHECK_CPU)); | ||
259 | |||
260 | /* set our heartbeating to other partitions into motion */ | ||
261 | xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ); | ||
262 | xpc_hb_beater(0); | ||
263 | |||
264 | while (!(volatile int) xpc_exiting) { | ||
265 | |||
266 | dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have " | ||
267 | "been received\n", | ||
268 | (int) (xpc_hb_check_timeout - jiffies), | ||
269 | atomic_read(&xpc_act_IRQ_rcvd) - last_IRQ_count); | ||
270 | |||
271 | |||
272 | /* checking of remote heartbeats is skewed by IRQ handling */ | ||
273 | if (time_after_eq(jiffies, xpc_hb_check_timeout)) { | ||
274 | dev_dbg(xpc_part, "checking remote heartbeats\n"); | ||
275 | xpc_check_remote_hb(); | ||
276 | |||
277 | /* | ||
278 | * We need to periodically recheck to ensure no | ||
279 | * IPI/AMO pairs have been missed. That check | ||
280 | * must always reset xpc_hb_check_timeout. | ||
281 | */ | ||
282 | force_IRQ = 1; | ||
283 | } | ||
284 | |||
285 | |||
286 | /* check for outstanding IRQs */ | ||
287 | new_IRQ_count = atomic_read(&xpc_act_IRQ_rcvd); | ||
288 | if (last_IRQ_count < new_IRQ_count || force_IRQ != 0) { | ||
289 | force_IRQ = 0; | ||
290 | |||
291 | dev_dbg(xpc_part, "found an IRQ to process; will be " | ||
292 | "resetting xpc_hb_check_timeout\n"); | ||
293 | |||
294 | last_IRQ_count += xpc_identify_act_IRQ_sender(); | ||
295 | if (last_IRQ_count < new_IRQ_count) { | ||
296 | /* retry once to help avoid missing AMO */ | ||
297 | (void) xpc_identify_act_IRQ_sender(); | ||
298 | } | ||
299 | last_IRQ_count = new_IRQ_count; | ||
300 | |||
301 | xpc_hb_check_timeout = jiffies + | ||
302 | (xpc_hb_check_interval * HZ); | ||
303 | } | ||
304 | |||
305 | /* wait for IRQ or timeout */ | ||
306 | (void) wait_event_interruptible(xpc_act_IRQ_wq, | ||
307 | (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) || | ||
308 | time_after_eq(jiffies, xpc_hb_check_timeout) || | ||
309 | (volatile int) xpc_exiting)); | ||
310 | } | ||
311 | |||
312 | dev_dbg(xpc_part, "heartbeat checker is exiting\n"); | ||
313 | |||
314 | |||
315 | /* mark this thread as having exited */ | ||
316 | complete(&xpc_hb_checker_exited); | ||
317 | return 0; | ||
318 | } | ||
319 | |||
320 | |||
321 | /* | ||
322 | * This thread will attempt to discover other partitions to activate | ||
323 | * based on info provided by SAL. This new thread is short lived and | ||
324 | * will exit once discovery is complete. | ||
325 | */ | ||
326 | static int | ||
327 | xpc_initiate_discovery(void *ignore) | ||
328 | { | ||
329 | daemonize(XPC_DISCOVERY_THREAD_NAME); | ||
330 | |||
331 | xpc_discovery(); | ||
332 | |||
333 | dev_dbg(xpc_part, "discovery thread is exiting\n"); | ||
334 | |||
335 | /* mark this thread as having exited */ | ||
336 | complete(&xpc_discovery_exited); | ||
337 | return 0; | ||
338 | } | ||
339 | |||
340 | |||
341 | /* | ||
342 | * Establish first contact with the remote partititon. This involves pulling | ||
343 | * the XPC per partition variables from the remote partition and waiting for | ||
344 | * the remote partition to pull ours. | ||
345 | */ | ||
346 | static enum xpc_retval | ||
347 | xpc_make_first_contact(struct xpc_partition *part) | ||
348 | { | ||
349 | enum xpc_retval ret; | ||
350 | |||
351 | |||
352 | while ((ret = xpc_pull_remote_vars_part(part)) != xpcSuccess) { | ||
353 | if (ret != xpcRetry) { | ||
354 | XPC_DEACTIVATE_PARTITION(part, ret); | ||
355 | return ret; | ||
356 | } | ||
357 | |||
358 | dev_dbg(xpc_chan, "waiting to make first contact with " | ||
359 | "partition %d\n", XPC_PARTID(part)); | ||
360 | |||
361 | /* wait a 1/4 of a second or so */ | ||
362 | (void) msleep_interruptible(250); | ||
363 | |||
364 | if (part->act_state == XPC_P_DEACTIVATING) { | ||
365 | return part->reason; | ||
366 | } | ||
367 | } | ||
368 | |||
369 | return xpc_mark_partition_active(part); | ||
370 | } | ||
371 | |||
372 | |||
373 | /* | ||
374 | * The first kthread assigned to a newly activated partition is the one | ||
375 | * created by XPC HB with which it calls xpc_partition_up(). XPC hangs on to | ||
376 | * that kthread until the partition is brought down, at which time that kthread | ||
377 | * returns back to XPC HB. (The return of that kthread will signify to XPC HB | ||
378 | * that XPC has dismantled all communication infrastructure for the associated | ||
379 | * partition.) This kthread becomes the channel manager for that partition. | ||
380 | * | ||
381 | * Each active partition has a channel manager, who, besides connecting and | ||
382 | * disconnecting channels, will ensure that each of the partition's connected | ||
383 | * channels has the required number of assigned kthreads to get the work done. | ||
384 | */ | ||
385 | static void | ||
386 | xpc_channel_mgr(struct xpc_partition *part) | ||
387 | { | ||
388 | while (part->act_state != XPC_P_DEACTIVATING || | ||
389 | atomic_read(&part->nchannels_active) > 0 || | ||
390 | !xpc_partition_disengaged(part)) { | ||
391 | |||
392 | xpc_process_channel_activity(part); | ||
393 | |||
394 | |||
395 | /* | ||
396 | * Wait until we've been requested to activate kthreads or | ||
397 | * all of the channel's message queues have been torn down or | ||
398 | * a signal is pending. | ||
399 | * | ||
400 | * The channel_mgr_requests is set to 1 after being awakened, | ||
401 | * This is done to prevent the channel mgr from making one pass | ||
402 | * through the loop for each request, since he will | ||
403 | * be servicing all the requests in one pass. The reason it's | ||
404 | * set to 1 instead of 0 is so that other kthreads will know | ||
405 | * that the channel mgr is running and won't bother trying to | ||
406 | * wake him up. | ||
407 | */ | ||
408 | atomic_dec(&part->channel_mgr_requests); | ||
409 | (void) wait_event_interruptible(part->channel_mgr_wq, | ||
410 | (atomic_read(&part->channel_mgr_requests) > 0 || | ||
411 | (volatile u64) part->local_IPI_amo != 0 || | ||
412 | ((volatile u8) part->act_state == | ||
413 | XPC_P_DEACTIVATING && | ||
414 | atomic_read(&part->nchannels_active) == 0 && | ||
415 | xpc_partition_disengaged(part)))); | ||
416 | atomic_set(&part->channel_mgr_requests, 1); | ||
417 | |||
418 | // >>> Does it need to wakeup periodically as well? In case we | ||
419 | // >>> miscalculated the #of kthreads to wakeup or create? | ||
420 | } | ||
421 | } | ||
422 | |||
423 | |||
424 | /* | ||
425 | * When XPC HB determines that a partition has come up, it will create a new | ||
426 | * kthread and that kthread will call this function to attempt to set up the | ||
427 | * basic infrastructure used for Cross Partition Communication with the newly | ||
428 | * upped partition. | ||
429 | * | ||
430 | * The kthread that was created by XPC HB and which setup the XPC | ||
431 | * infrastructure will remain assigned to the partition until the partition | ||
432 | * goes down. At which time the kthread will teardown the XPC infrastructure | ||
433 | * and then exit. | ||
434 | * | ||
435 | * XPC HB will put the remote partition's XPC per partition specific variables | ||
436 | * physical address into xpc_partitions[partid].remote_vars_part_pa prior to | ||
437 | * calling xpc_partition_up(). | ||
438 | */ | ||
439 | static void | ||
440 | xpc_partition_up(struct xpc_partition *part) | ||
441 | { | ||
442 | DBUG_ON(part->channels != NULL); | ||
443 | |||
444 | dev_dbg(xpc_chan, "activating partition %d\n", XPC_PARTID(part)); | ||
445 | |||
446 | if (xpc_setup_infrastructure(part) != xpcSuccess) { | ||
447 | return; | ||
448 | } | ||
449 | |||
450 | /* | ||
451 | * The kthread that XPC HB called us with will become the | ||
452 | * channel manager for this partition. It will not return | ||
453 | * back to XPC HB until the partition's XPC infrastructure | ||
454 | * has been dismantled. | ||
455 | */ | ||
456 | |||
457 | (void) xpc_part_ref(part); /* this will always succeed */ | ||
458 | |||
459 | if (xpc_make_first_contact(part) == xpcSuccess) { | ||
460 | xpc_channel_mgr(part); | ||
461 | } | ||
462 | |||
463 | xpc_part_deref(part); | ||
464 | |||
465 | xpc_teardown_infrastructure(part); | ||
466 | } | ||
467 | |||
468 | |||
469 | static int | ||
470 | xpc_activating(void *__partid) | ||
471 | { | ||
472 | partid_t partid = (u64) __partid; | ||
473 | struct xpc_partition *part = &xpc_partitions[partid]; | ||
474 | unsigned long irq_flags; | ||
475 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | ||
476 | int ret; | ||
477 | |||
478 | |||
479 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | ||
480 | |||
481 | spin_lock_irqsave(&part->act_lock, irq_flags); | ||
482 | |||
483 | if (part->act_state == XPC_P_DEACTIVATING) { | ||
484 | part->act_state = XPC_P_INACTIVE; | ||
485 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | ||
486 | part->remote_rp_pa = 0; | ||
487 | return 0; | ||
488 | } | ||
489 | |||
490 | /* indicate the thread is activating */ | ||
491 | DBUG_ON(part->act_state != XPC_P_ACTIVATION_REQ); | ||
492 | part->act_state = XPC_P_ACTIVATING; | ||
493 | |||
494 | XPC_SET_REASON(part, 0, 0); | ||
495 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | ||
496 | |||
497 | dev_dbg(xpc_part, "bringing partition %d up\n", partid); | ||
498 | |||
499 | daemonize("xpc%02d", partid); | ||
500 | |||
501 | /* | ||
502 | * This thread needs to run at a realtime priority to prevent a | ||
503 | * significant performance degradation. | ||
504 | */ | ||
505 | ret = sched_setscheduler(current, SCHED_FIFO, ¶m); | ||
506 | if (ret != 0) { | ||
507 | dev_warn(xpc_part, "unable to set pid %d to a realtime " | ||
508 | "priority, ret=%d\n", current->pid, ret); | ||
509 | } | ||
510 | |||
511 | /* allow this thread and its children to run on any CPU */ | ||
512 | set_cpus_allowed(current, CPU_MASK_ALL); | ||
513 | |||
514 | /* | ||
515 | * Register the remote partition's AMOs with SAL so it can handle | ||
516 | * and cleanup errors within that address range should the remote | ||
517 | * partition go down. We don't unregister this range because it is | ||
518 | * difficult to tell when outstanding writes to the remote partition | ||
519 | * are finished and thus when it is safe to unregister. This should | ||
520 | * not result in wasted space in the SAL xp_addr_region table because | ||
521 | * we should get the same page for remote_amos_page_pa after module | ||
522 | * reloads and system reboots. | ||
523 | */ | ||
524 | if (sn_register_xp_addr_region(part->remote_amos_page_pa, | ||
525 | PAGE_SIZE, 1) < 0) { | ||
526 | dev_warn(xpc_part, "xpc_partition_up(%d) failed to register " | ||
527 | "xp_addr region\n", partid); | ||
528 | |||
529 | spin_lock_irqsave(&part->act_lock, irq_flags); | ||
530 | part->act_state = XPC_P_INACTIVE; | ||
531 | XPC_SET_REASON(part, xpcPhysAddrRegFailed, __LINE__); | ||
532 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | ||
533 | part->remote_rp_pa = 0; | ||
534 | return 0; | ||
535 | } | ||
536 | |||
537 | xpc_allow_hb(partid, xpc_vars); | ||
538 | xpc_IPI_send_activated(part); | ||
539 | |||
540 | |||
541 | /* | ||
542 | * xpc_partition_up() holds this thread and marks this partition as | ||
543 | * XPC_P_ACTIVE by calling xpc_hb_mark_active(). | ||
544 | */ | ||
545 | (void) xpc_partition_up(part); | ||
546 | |||
547 | xpc_disallow_hb(partid, xpc_vars); | ||
548 | xpc_mark_partition_inactive(part); | ||
549 | |||
550 | if (part->reason == xpcReactivating) { | ||
551 | /* interrupting ourselves results in activating partition */ | ||
552 | xpc_IPI_send_reactivate(part); | ||
553 | } | ||
554 | |||
555 | return 0; | ||
556 | } | ||
557 | |||
558 | |||
559 | void | ||
560 | xpc_activate_partition(struct xpc_partition *part) | ||
561 | { | ||
562 | partid_t partid = XPC_PARTID(part); | ||
563 | unsigned long irq_flags; | ||
564 | pid_t pid; | ||
565 | |||
566 | |||
567 | spin_lock_irqsave(&part->act_lock, irq_flags); | ||
568 | |||
569 | DBUG_ON(part->act_state != XPC_P_INACTIVE); | ||
570 | |||
571 | part->act_state = XPC_P_ACTIVATION_REQ; | ||
572 | XPC_SET_REASON(part, xpcCloneKThread, __LINE__); | ||
573 | |||
574 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | ||
575 | |||
576 | pid = kernel_thread(xpc_activating, (void *) ((u64) partid), 0); | ||
577 | |||
578 | if (unlikely(pid <= 0)) { | ||
579 | spin_lock_irqsave(&part->act_lock, irq_flags); | ||
580 | part->act_state = XPC_P_INACTIVE; | ||
581 | XPC_SET_REASON(part, xpcCloneKThreadFailed, __LINE__); | ||
582 | spin_unlock_irqrestore(&part->act_lock, irq_flags); | ||
583 | } | ||
584 | } | ||
585 | |||
586 | |||
587 | /* | ||
588 | * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified | ||
589 | * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more | ||
590 | * than one partition, we use an AMO_t structure per partition to indicate | ||
591 | * whether a partition has sent an IPI or not. >>> If it has, then wake up the | ||
592 | * associated kthread to handle it. | ||
593 | * | ||
594 | * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IPIs sent by XPC | ||
595 | * running on other partitions. | ||
596 | * | ||
597 | * Noteworthy Arguments: | ||
598 | * | ||
599 | * irq - Interrupt ReQuest number. NOT USED. | ||
600 | * | ||
601 | * dev_id - partid of IPI's potential sender. | ||
602 | */ | ||
603 | irqreturn_t | ||
604 | xpc_notify_IRQ_handler(int irq, void *dev_id) | ||
605 | { | ||
606 | partid_t partid = (partid_t) (u64) dev_id; | ||
607 | struct xpc_partition *part = &xpc_partitions[partid]; | ||
608 | |||
609 | |||
610 | DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS); | ||
611 | |||
612 | if (xpc_part_ref(part)) { | ||
613 | xpc_check_for_channel_activity(part); | ||
614 | |||
615 | xpc_part_deref(part); | ||
616 | } | ||
617 | return IRQ_HANDLED; | ||
618 | } | ||
619 | |||
620 | |||
621 | /* | ||
622 | * Check to see if xpc_notify_IRQ_handler() dropped any IPIs on the floor | ||
623 | * because the write to their associated IPI amo completed after the IRQ/IPI | ||
624 | * was received. | ||
625 | */ | ||
626 | void | ||
627 | xpc_dropped_IPI_check(struct xpc_partition *part) | ||
628 | { | ||
629 | if (xpc_part_ref(part)) { | ||
630 | xpc_check_for_channel_activity(part); | ||
631 | |||
632 | part->dropped_IPI_timer.expires = jiffies + | ||
633 | XPC_P_DROPPED_IPI_WAIT; | ||
634 | add_timer(&part->dropped_IPI_timer); | ||
635 | xpc_part_deref(part); | ||
636 | } | ||
637 | } | ||
638 | |||
639 | |||
640 | void | ||
641 | xpc_activate_kthreads(struct xpc_channel *ch, int needed) | ||
642 | { | ||
643 | int idle = atomic_read(&ch->kthreads_idle); | ||
644 | int assigned = atomic_read(&ch->kthreads_assigned); | ||
645 | int wakeup; | ||
646 | |||
647 | |||
648 | DBUG_ON(needed <= 0); | ||
649 | |||
650 | if (idle > 0) { | ||
651 | wakeup = (needed > idle) ? idle : needed; | ||
652 | needed -= wakeup; | ||
653 | |||
654 | dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, " | ||
655 | "channel=%d\n", wakeup, ch->partid, ch->number); | ||
656 | |||
657 | /* only wakeup the requested number of kthreads */ | ||
658 | wake_up_nr(&ch->idle_wq, wakeup); | ||
659 | } | ||
660 | |||
661 | if (needed <= 0) { | ||
662 | return; | ||
663 | } | ||
664 | |||
665 | if (needed + assigned > ch->kthreads_assigned_limit) { | ||
666 | needed = ch->kthreads_assigned_limit - assigned; | ||
667 | // >>>should never be less than 0 | ||
668 | if (needed <= 0) { | ||
669 | return; | ||
670 | } | ||
671 | } | ||
672 | |||
673 | dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n", | ||
674 | needed, ch->partid, ch->number); | ||
675 | |||
676 | xpc_create_kthreads(ch, needed, 0); | ||
677 | } | ||
678 | |||
679 | |||
680 | /* | ||
681 | * This function is where XPC's kthreads wait for messages to deliver. | ||
682 | */ | ||
683 | static void | ||
684 | xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch) | ||
685 | { | ||
686 | do { | ||
687 | /* deliver messages to their intended recipients */ | ||
688 | |||
689 | while ((volatile s64) ch->w_local_GP.get < | ||
690 | (volatile s64) ch->w_remote_GP.put && | ||
691 | !((volatile u32) ch->flags & | ||
692 | XPC_C_DISCONNECTING)) { | ||
693 | xpc_deliver_msg(ch); | ||
694 | } | ||
695 | |||
696 | if (atomic_inc_return(&ch->kthreads_idle) > | ||
697 | ch->kthreads_idle_limit) { | ||
698 | /* too many idle kthreads on this channel */ | ||
699 | atomic_dec(&ch->kthreads_idle); | ||
700 | break; | ||
701 | } | ||
702 | |||
703 | dev_dbg(xpc_chan, "idle kthread calling " | ||
704 | "wait_event_interruptible_exclusive()\n"); | ||
705 | |||
706 | (void) wait_event_interruptible_exclusive(ch->idle_wq, | ||
707 | ((volatile s64) ch->w_local_GP.get < | ||
708 | (volatile s64) ch->w_remote_GP.put || | ||
709 | ((volatile u32) ch->flags & | ||
710 | XPC_C_DISCONNECTING))); | ||
711 | |||
712 | atomic_dec(&ch->kthreads_idle); | ||
713 | |||
714 | } while (!((volatile u32) ch->flags & XPC_C_DISCONNECTING)); | ||
715 | } | ||
716 | |||
717 | |||
718 | static int | ||
719 | xpc_daemonize_kthread(void *args) | ||
720 | { | ||
721 | partid_t partid = XPC_UNPACK_ARG1(args); | ||
722 | u16 ch_number = XPC_UNPACK_ARG2(args); | ||
723 | struct xpc_partition *part = &xpc_partitions[partid]; | ||
724 | struct xpc_channel *ch; | ||
725 | int n_needed; | ||
726 | unsigned long irq_flags; | ||
727 | |||
728 | |||
729 | daemonize("xpc%02dc%d", partid, ch_number); | ||
730 | |||
731 | dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n", | ||
732 | partid, ch_number); | ||
733 | |||
734 | ch = &part->channels[ch_number]; | ||
735 | |||
736 | if (!(ch->flags & XPC_C_DISCONNECTING)) { | ||
737 | |||
738 | /* let registerer know that connection has been established */ | ||
739 | |||
740 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
741 | if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) { | ||
742 | ch->flags |= XPC_C_CONNECTEDCALLOUT; | ||
743 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
744 | |||
745 | xpc_connected_callout(ch); | ||
746 | |||
747 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
748 | ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE; | ||
749 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
750 | |||
751 | /* | ||
752 | * It is possible that while the callout was being | ||
753 | * made that the remote partition sent some messages. | ||
754 | * If that is the case, we may need to activate | ||
755 | * additional kthreads to help deliver them. We only | ||
756 | * need one less than total #of messages to deliver. | ||
757 | */ | ||
758 | n_needed = ch->w_remote_GP.put - ch->w_local_GP.get - 1; | ||
759 | if (n_needed > 0 && | ||
760 | !(ch->flags & XPC_C_DISCONNECTING)) { | ||
761 | xpc_activate_kthreads(ch, n_needed); | ||
762 | } | ||
763 | } else { | ||
764 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
765 | } | ||
766 | |||
767 | xpc_kthread_waitmsgs(part, ch); | ||
768 | } | ||
769 | |||
770 | /* let registerer know that connection is disconnecting */ | ||
771 | |||
772 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
773 | if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) && | ||
774 | !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) { | ||
775 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT; | ||
776 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
777 | |||
778 | xpc_disconnect_callout(ch, xpcDisconnecting); | ||
779 | |||
780 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
781 | ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE; | ||
782 | } | ||
783 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
784 | |||
785 | if (atomic_dec_return(&ch->kthreads_assigned) == 0) { | ||
786 | if (atomic_dec_return(&part->nchannels_engaged) == 0) { | ||
787 | xpc_mark_partition_disengaged(part); | ||
788 | xpc_IPI_send_disengage(part); | ||
789 | } | ||
790 | } | ||
791 | |||
792 | xpc_msgqueue_deref(ch); | ||
793 | |||
794 | dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n", | ||
795 | partid, ch_number); | ||
796 | |||
797 | xpc_part_deref(part); | ||
798 | return 0; | ||
799 | } | ||
800 | |||
801 | |||
802 | /* | ||
803 | * For each partition that XPC has established communications with, there is | ||
804 | * a minimum of one kernel thread assigned to perform any operation that | ||
805 | * may potentially sleep or block (basically the callouts to the asynchronous | ||
806 | * functions registered via xpc_connect()). | ||
807 | * | ||
808 | * Additional kthreads are created and destroyed by XPC as the workload | ||
809 | * demands. | ||
810 | * | ||
811 | * A kthread is assigned to one of the active channels that exists for a given | ||
812 | * partition. | ||
813 | */ | ||
814 | void | ||
815 | xpc_create_kthreads(struct xpc_channel *ch, int needed, | ||
816 | int ignore_disconnecting) | ||
817 | { | ||
818 | unsigned long irq_flags; | ||
819 | pid_t pid; | ||
820 | u64 args = XPC_PACK_ARGS(ch->partid, ch->number); | ||
821 | struct xpc_partition *part = &xpc_partitions[ch->partid]; | ||
822 | |||
823 | |||
824 | while (needed-- > 0) { | ||
825 | |||
826 | /* | ||
827 | * The following is done on behalf of the newly created | ||
828 | * kthread. That kthread is responsible for doing the | ||
829 | * counterpart to the following before it exits. | ||
830 | */ | ||
831 | if (ignore_disconnecting) { | ||
832 | if (!atomic_inc_not_zero(&ch->kthreads_assigned)) { | ||
833 | /* kthreads assigned had gone to zero */ | ||
834 | BUG_ON(!(ch->flags & | ||
835 | XPC_C_DISCONNECTINGCALLOUT_MADE)); | ||
836 | break; | ||
837 | } | ||
838 | |||
839 | } else if (ch->flags & XPC_C_DISCONNECTING) { | ||
840 | break; | ||
841 | |||
842 | } else if (atomic_inc_return(&ch->kthreads_assigned) == 1) { | ||
843 | if (atomic_inc_return(&part->nchannels_engaged) == 1) | ||
844 | xpc_mark_partition_engaged(part); | ||
845 | } | ||
846 | (void) xpc_part_ref(part); | ||
847 | xpc_msgqueue_ref(ch); | ||
848 | |||
849 | pid = kernel_thread(xpc_daemonize_kthread, (void *) args, 0); | ||
850 | if (pid < 0) { | ||
851 | /* the fork failed */ | ||
852 | |||
853 | /* | ||
854 | * NOTE: if (ignore_disconnecting && | ||
855 | * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true, | ||
856 | * then we'll deadlock if all other kthreads assigned | ||
857 | * to this channel are blocked in the channel's | ||
858 | * registerer, because the only thing that will unblock | ||
859 | * them is the xpcDisconnecting callout that this | ||
860 | * failed kernel_thread would have made. | ||
861 | */ | ||
862 | |||
863 | if (atomic_dec_return(&ch->kthreads_assigned) == 0 && | ||
864 | atomic_dec_return(&part->nchannels_engaged) == 0) { | ||
865 | xpc_mark_partition_disengaged(part); | ||
866 | xpc_IPI_send_disengage(part); | ||
867 | } | ||
868 | xpc_msgqueue_deref(ch); | ||
869 | xpc_part_deref(part); | ||
870 | |||
871 | if (atomic_read(&ch->kthreads_assigned) < | ||
872 | ch->kthreads_idle_limit) { | ||
873 | /* | ||
874 | * Flag this as an error only if we have an | ||
875 | * insufficient #of kthreads for the channel | ||
876 | * to function. | ||
877 | */ | ||
878 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
879 | XPC_DISCONNECT_CHANNEL(ch, xpcLackOfResources, | ||
880 | &irq_flags); | ||
881 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
882 | } | ||
883 | break; | ||
884 | } | ||
885 | |||
886 | ch->kthreads_created++; // >>> temporary debug only!!! | ||
887 | } | ||
888 | } | ||
889 | |||
890 | |||
891 | void | ||
892 | xpc_disconnect_wait(int ch_number) | ||
893 | { | ||
894 | unsigned long irq_flags; | ||
895 | partid_t partid; | ||
896 | struct xpc_partition *part; | ||
897 | struct xpc_channel *ch; | ||
898 | int wakeup_channel_mgr; | ||
899 | |||
900 | |||
901 | /* now wait for all callouts to the caller's function to cease */ | ||
902 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | ||
903 | part = &xpc_partitions[partid]; | ||
904 | |||
905 | if (!xpc_part_ref(part)) { | ||
906 | continue; | ||
907 | } | ||
908 | |||
909 | ch = &part->channels[ch_number]; | ||
910 | |||
911 | if (!(ch->flags & XPC_C_WDISCONNECT)) { | ||
912 | xpc_part_deref(part); | ||
913 | continue; | ||
914 | } | ||
915 | |||
916 | wait_for_completion(&ch->wdisconnect_wait); | ||
917 | |||
918 | spin_lock_irqsave(&ch->lock, irq_flags); | ||
919 | DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED)); | ||
920 | wakeup_channel_mgr = 0; | ||
921 | |||
922 | if (ch->delayed_IPI_flags) { | ||
923 | if (part->act_state != XPC_P_DEACTIVATING) { | ||
924 | spin_lock(&part->IPI_lock); | ||
925 | XPC_SET_IPI_FLAGS(part->local_IPI_amo, | ||
926 | ch->number, ch->delayed_IPI_flags); | ||
927 | spin_unlock(&part->IPI_lock); | ||
928 | wakeup_channel_mgr = 1; | ||
929 | } | ||
930 | ch->delayed_IPI_flags = 0; | ||
931 | } | ||
932 | |||
933 | ch->flags &= ~XPC_C_WDISCONNECT; | ||
934 | spin_unlock_irqrestore(&ch->lock, irq_flags); | ||
935 | |||
936 | if (wakeup_channel_mgr) { | ||
937 | xpc_wakeup_channel_mgr(part); | ||
938 | } | ||
939 | |||
940 | xpc_part_deref(part); | ||
941 | } | ||
942 | } | ||
943 | |||
944 | |||
945 | static void | ||
946 | xpc_do_exit(enum xpc_retval reason) | ||
947 | { | ||
948 | partid_t partid; | ||
949 | int active_part_count, printed_waiting_msg = 0; | ||
950 | struct xpc_partition *part; | ||
951 | unsigned long printmsg_time, disengage_request_timeout = 0; | ||
952 | |||
953 | |||
954 | /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */ | ||
955 | DBUG_ON(xpc_exiting == 1); | ||
956 | |||
957 | /* | ||
958 | * Let the heartbeat checker thread and the discovery thread | ||
959 | * (if one is running) know that they should exit. Also wake up | ||
960 | * the heartbeat checker thread in case it's sleeping. | ||
961 | */ | ||
962 | xpc_exiting = 1; | ||
963 | wake_up_interruptible(&xpc_act_IRQ_wq); | ||
964 | |||
965 | /* ignore all incoming interrupts */ | ||
966 | free_irq(SGI_XPC_ACTIVATE, NULL); | ||
967 | |||
968 | /* wait for the discovery thread to exit */ | ||
969 | wait_for_completion(&xpc_discovery_exited); | ||
970 | |||
971 | /* wait for the heartbeat checker thread to exit */ | ||
972 | wait_for_completion(&xpc_hb_checker_exited); | ||
973 | |||
974 | |||
975 | /* sleep for a 1/3 of a second or so */ | ||
976 | (void) msleep_interruptible(300); | ||
977 | |||
978 | |||
979 | /* wait for all partitions to become inactive */ | ||
980 | |||
981 | printmsg_time = jiffies + (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); | ||
982 | xpc_disengage_request_timedout = 0; | ||
983 | |||
984 | do { | ||
985 | active_part_count = 0; | ||
986 | |||
987 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | ||
988 | part = &xpc_partitions[partid]; | ||
989 | |||
990 | if (xpc_partition_disengaged(part) && | ||
991 | part->act_state == XPC_P_INACTIVE) { | ||
992 | continue; | ||
993 | } | ||
994 | |||
995 | active_part_count++; | ||
996 | |||
997 | XPC_DEACTIVATE_PARTITION(part, reason); | ||
998 | |||
999 | if (part->disengage_request_timeout > | ||
1000 | disengage_request_timeout) { | ||
1001 | disengage_request_timeout = | ||
1002 | part->disengage_request_timeout; | ||
1003 | } | ||
1004 | } | ||
1005 | |||
1006 | if (xpc_partition_engaged(-1UL)) { | ||
1007 | if (time_after(jiffies, printmsg_time)) { | ||
1008 | dev_info(xpc_part, "waiting for remote " | ||
1009 | "partitions to disengage, timeout in " | ||
1010 | "%ld seconds\n", | ||
1011 | (disengage_request_timeout - jiffies) | ||
1012 | / HZ); | ||
1013 | printmsg_time = jiffies + | ||
1014 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ); | ||
1015 | printed_waiting_msg = 1; | ||
1016 | } | ||
1017 | |||
1018 | } else if (active_part_count > 0) { | ||
1019 | if (printed_waiting_msg) { | ||
1020 | dev_info(xpc_part, "waiting for local partition" | ||
1021 | " to disengage\n"); | ||
1022 | printed_waiting_msg = 0; | ||
1023 | } | ||
1024 | |||
1025 | } else { | ||
1026 | if (!xpc_disengage_request_timedout) { | ||
1027 | dev_info(xpc_part, "all partitions have " | ||
1028 | "disengaged\n"); | ||
1029 | } | ||
1030 | break; | ||
1031 | } | ||
1032 | |||
1033 | /* sleep for a 1/3 of a second or so */ | ||
1034 | (void) msleep_interruptible(300); | ||
1035 | |||
1036 | } while (1); | ||
1037 | |||
1038 | DBUG_ON(xpc_partition_engaged(-1UL)); | ||
1039 | |||
1040 | |||
1041 | /* indicate to others that our reserved page is uninitialized */ | ||
1042 | xpc_rsvd_page->vars_pa = 0; | ||
1043 | |||
1044 | /* now it's time to eliminate our heartbeat */ | ||
1045 | del_timer_sync(&xpc_hb_timer); | ||
1046 | DBUG_ON(xpc_vars->heartbeating_to_mask != 0); | ||
1047 | |||
1048 | if (reason == xpcUnloading) { | ||
1049 | /* take ourselves off of the reboot_notifier_list */ | ||
1050 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | ||
1051 | |||
1052 | /* take ourselves off of the die_notifier list */ | ||
1053 | (void) unregister_die_notifier(&xpc_die_notifier); | ||
1054 | } | ||
1055 | |||
1056 | /* close down protections for IPI operations */ | ||
1057 | xpc_restrict_IPI_ops(); | ||
1058 | |||
1059 | |||
1060 | /* clear the interface to XPC's functions */ | ||
1061 | xpc_clear_interface(); | ||
1062 | |||
1063 | if (xpc_sysctl) { | ||
1064 | unregister_sysctl_table(xpc_sysctl); | ||
1065 | } | ||
1066 | |||
1067 | kfree(xpc_remote_copy_buffer_base); | ||
1068 | } | ||
1069 | |||
1070 | |||
1071 | /* | ||
1072 | * This function is called when the system is being rebooted. | ||
1073 | */ | ||
1074 | static int | ||
1075 | xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused) | ||
1076 | { | ||
1077 | enum xpc_retval reason; | ||
1078 | |||
1079 | |||
1080 | switch (event) { | ||
1081 | case SYS_RESTART: | ||
1082 | reason = xpcSystemReboot; | ||
1083 | break; | ||
1084 | case SYS_HALT: | ||
1085 | reason = xpcSystemHalt; | ||
1086 | break; | ||
1087 | case SYS_POWER_OFF: | ||
1088 | reason = xpcSystemPoweroff; | ||
1089 | break; | ||
1090 | default: | ||
1091 | reason = xpcSystemGoingDown; | ||
1092 | } | ||
1093 | |||
1094 | xpc_do_exit(reason); | ||
1095 | return NOTIFY_DONE; | ||
1096 | } | ||
1097 | |||
1098 | |||
1099 | /* | ||
1100 | * Notify other partitions to disengage from all references to our memory. | ||
1101 | */ | ||
1102 | static void | ||
1103 | xpc_die_disengage(void) | ||
1104 | { | ||
1105 | struct xpc_partition *part; | ||
1106 | partid_t partid; | ||
1107 | unsigned long engaged; | ||
1108 | long time, printmsg_time, disengage_request_timeout; | ||
1109 | |||
1110 | |||
1111 | /* keep xpc_hb_checker thread from doing anything (just in case) */ | ||
1112 | xpc_exiting = 1; | ||
1113 | |||
1114 | xpc_vars->heartbeating_to_mask = 0; /* indicate we're deactivated */ | ||
1115 | |||
1116 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | ||
1117 | part = &xpc_partitions[partid]; | ||
1118 | |||
1119 | if (!XPC_SUPPORTS_DISENGAGE_REQUEST(part-> | ||
1120 | remote_vars_version)) { | ||
1121 | |||
1122 | /* just in case it was left set by an earlier XPC */ | ||
1123 | xpc_clear_partition_engaged(1UL << partid); | ||
1124 | continue; | ||
1125 | } | ||
1126 | |||
1127 | if (xpc_partition_engaged(1UL << partid) || | ||
1128 | part->act_state != XPC_P_INACTIVE) { | ||
1129 | xpc_request_partition_disengage(part); | ||
1130 | xpc_mark_partition_disengaged(part); | ||
1131 | xpc_IPI_send_disengage(part); | ||
1132 | } | ||
1133 | } | ||
1134 | |||
1135 | time = rtc_time(); | ||
1136 | printmsg_time = time + | ||
1137 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * sn_rtc_cycles_per_second); | ||
1138 | disengage_request_timeout = time + | ||
1139 | (xpc_disengage_request_timelimit * sn_rtc_cycles_per_second); | ||
1140 | |||
1141 | /* wait for all other partitions to disengage from us */ | ||
1142 | |||
1143 | while (1) { | ||
1144 | engaged = xpc_partition_engaged(-1UL); | ||
1145 | if (!engaged) { | ||
1146 | dev_info(xpc_part, "all partitions have disengaged\n"); | ||
1147 | break; | ||
1148 | } | ||
1149 | |||
1150 | time = rtc_time(); | ||
1151 | if (time >= disengage_request_timeout) { | ||
1152 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | ||
1153 | if (engaged & (1UL << partid)) { | ||
1154 | dev_info(xpc_part, "disengage from " | ||
1155 | "remote partition %d timed " | ||
1156 | "out\n", partid); | ||
1157 | } | ||
1158 | } | ||
1159 | break; | ||
1160 | } | ||
1161 | |||
1162 | if (time >= printmsg_time) { | ||
1163 | dev_info(xpc_part, "waiting for remote partitions to " | ||
1164 | "disengage, timeout in %ld seconds\n", | ||
1165 | (disengage_request_timeout - time) / | ||
1166 | sn_rtc_cycles_per_second); | ||
1167 | printmsg_time = time + | ||
1168 | (XPC_DISENGAGE_PRINTMSG_INTERVAL * | ||
1169 | sn_rtc_cycles_per_second); | ||
1170 | } | ||
1171 | } | ||
1172 | } | ||
1173 | |||
1174 | |||
1175 | /* | ||
1176 | * This function is called when the system is being restarted or halted due | ||
1177 | * to some sort of system failure. If this is the case we need to notify the | ||
1178 | * other partitions to disengage from all references to our memory. | ||
1179 | * This function can also be called when our heartbeater could be offlined | ||
1180 | * for a time. In this case we need to notify other partitions to not worry | ||
1181 | * about the lack of a heartbeat. | ||
1182 | */ | ||
1183 | static int | ||
1184 | xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused) | ||
1185 | { | ||
1186 | switch (event) { | ||
1187 | case DIE_MACHINE_RESTART: | ||
1188 | case DIE_MACHINE_HALT: | ||
1189 | xpc_die_disengage(); | ||
1190 | break; | ||
1191 | |||
1192 | case DIE_KDEBUG_ENTER: | ||
1193 | /* Should lack of heartbeat be ignored by other partitions? */ | ||
1194 | if (!xpc_kdebug_ignore) { | ||
1195 | break; | ||
1196 | } | ||
1197 | /* fall through */ | ||
1198 | case DIE_MCA_MONARCH_ENTER: | ||
1199 | case DIE_INIT_MONARCH_ENTER: | ||
1200 | xpc_vars->heartbeat++; | ||
1201 | xpc_vars->heartbeat_offline = 1; | ||
1202 | break; | ||
1203 | |||
1204 | case DIE_KDEBUG_LEAVE: | ||
1205 | /* Is lack of heartbeat being ignored by other partitions? */ | ||
1206 | if (!xpc_kdebug_ignore) { | ||
1207 | break; | ||
1208 | } | ||
1209 | /* fall through */ | ||
1210 | case DIE_MCA_MONARCH_LEAVE: | ||
1211 | case DIE_INIT_MONARCH_LEAVE: | ||
1212 | xpc_vars->heartbeat++; | ||
1213 | xpc_vars->heartbeat_offline = 0; | ||
1214 | break; | ||
1215 | } | ||
1216 | |||
1217 | return NOTIFY_DONE; | ||
1218 | } | ||
1219 | |||
1220 | |||
1221 | int __init | ||
1222 | xpc_init(void) | ||
1223 | { | ||
1224 | int ret; | ||
1225 | partid_t partid; | ||
1226 | struct xpc_partition *part; | ||
1227 | pid_t pid; | ||
1228 | size_t buf_size; | ||
1229 | |||
1230 | |||
1231 | if (!ia64_platform_is("sn2")) { | ||
1232 | return -ENODEV; | ||
1233 | } | ||
1234 | |||
1235 | |||
1236 | buf_size = max(XPC_RP_VARS_SIZE, | ||
1237 | XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES); | ||
1238 | xpc_remote_copy_buffer = xpc_kmalloc_cacheline_aligned(buf_size, | ||
1239 | GFP_KERNEL, &xpc_remote_copy_buffer_base); | ||
1240 | if (xpc_remote_copy_buffer == NULL) | ||
1241 | return -ENOMEM; | ||
1242 | |||
1243 | snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part"); | ||
1244 | snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan"); | ||
1245 | |||
1246 | xpc_sysctl = register_sysctl_table(xpc_sys_dir); | ||
1247 | |||
1248 | /* | ||
1249 | * The first few fields of each entry of xpc_partitions[] need to | ||
1250 | * be initialized now so that calls to xpc_connect() and | ||
1251 | * xpc_disconnect() can be made prior to the activation of any remote | ||
1252 | * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE | ||
1253 | * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING | ||
1254 | * PARTITION HAS BEEN ACTIVATED. | ||
1255 | */ | ||
1256 | for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) { | ||
1257 | part = &xpc_partitions[partid]; | ||
1258 | |||
1259 | DBUG_ON((u64) part != L1_CACHE_ALIGN((u64) part)); | ||
1260 | |||
1261 | part->act_IRQ_rcvd = 0; | ||
1262 | spin_lock_init(&part->act_lock); | ||
1263 | part->act_state = XPC_P_INACTIVE; | ||
1264 | XPC_SET_REASON(part, 0, 0); | ||
1265 | |||
1266 | init_timer(&part->disengage_request_timer); | ||
1267 | part->disengage_request_timer.function = | ||
1268 | xpc_timeout_partition_disengage_request; | ||
1269 | part->disengage_request_timer.data = (unsigned long) part; | ||
1270 | |||
1271 | part->setup_state = XPC_P_UNSET; | ||
1272 | init_waitqueue_head(&part->teardown_wq); | ||
1273 | atomic_set(&part->references, 0); | ||
1274 | } | ||
1275 | |||
1276 | /* | ||
1277 | * Open up protections for IPI operations (and AMO operations on | ||
1278 | * Shub 1.1 systems). | ||
1279 | */ | ||
1280 | xpc_allow_IPI_ops(); | ||
1281 | |||
1282 | /* | ||
1283 | * Interrupts being processed will increment this atomic variable and | ||
1284 | * awaken the heartbeat thread which will process the interrupts. | ||
1285 | */ | ||
1286 | atomic_set(&xpc_act_IRQ_rcvd, 0); | ||
1287 | |||
1288 | /* | ||
1289 | * This is safe to do before the xpc_hb_checker thread has started | ||
1290 | * because the handler releases a wait queue. If an interrupt is | ||
1291 | * received before the thread is waiting, it will not go to sleep, | ||
1292 | * but rather immediately process the interrupt. | ||
1293 | */ | ||
1294 | ret = request_irq(SGI_XPC_ACTIVATE, xpc_act_IRQ_handler, 0, | ||
1295 | "xpc hb", NULL); | ||
1296 | if (ret != 0) { | ||
1297 | dev_err(xpc_part, "can't register ACTIVATE IRQ handler, " | ||
1298 | "errno=%d\n", -ret); | ||
1299 | |||
1300 | xpc_restrict_IPI_ops(); | ||
1301 | |||
1302 | if (xpc_sysctl) { | ||
1303 | unregister_sysctl_table(xpc_sysctl); | ||
1304 | } | ||
1305 | |||
1306 | kfree(xpc_remote_copy_buffer_base); | ||
1307 | return -EBUSY; | ||
1308 | } | ||
1309 | |||
1310 | /* | ||
1311 | * Fill the partition reserved page with the information needed by | ||
1312 | * other partitions to discover we are alive and establish initial | ||
1313 | * communications. | ||
1314 | */ | ||
1315 | xpc_rsvd_page = xpc_rsvd_page_init(); | ||
1316 | if (xpc_rsvd_page == NULL) { | ||
1317 | dev_err(xpc_part, "could not setup our reserved page\n"); | ||
1318 | |||
1319 | free_irq(SGI_XPC_ACTIVATE, NULL); | ||
1320 | xpc_restrict_IPI_ops(); | ||
1321 | |||
1322 | if (xpc_sysctl) { | ||
1323 | unregister_sysctl_table(xpc_sysctl); | ||
1324 | } | ||
1325 | |||
1326 | kfree(xpc_remote_copy_buffer_base); | ||
1327 | return -EBUSY; | ||
1328 | } | ||
1329 | |||
1330 | |||
1331 | /* add ourselves to the reboot_notifier_list */ | ||
1332 | ret = register_reboot_notifier(&xpc_reboot_notifier); | ||
1333 | if (ret != 0) { | ||
1334 | dev_warn(xpc_part, "can't register reboot notifier\n"); | ||
1335 | } | ||
1336 | |||
1337 | /* add ourselves to the die_notifier list */ | ||
1338 | ret = register_die_notifier(&xpc_die_notifier); | ||
1339 | if (ret != 0) { | ||
1340 | dev_warn(xpc_part, "can't register die notifier\n"); | ||
1341 | } | ||
1342 | |||
1343 | init_timer(&xpc_hb_timer); | ||
1344 | xpc_hb_timer.function = xpc_hb_beater; | ||
1345 | |||
1346 | /* | ||
1347 | * The real work-horse behind xpc. This processes incoming | ||
1348 | * interrupts and monitors remote heartbeats. | ||
1349 | */ | ||
1350 | pid = kernel_thread(xpc_hb_checker, NULL, 0); | ||
1351 | if (pid < 0) { | ||
1352 | dev_err(xpc_part, "failed while forking hb check thread\n"); | ||
1353 | |||
1354 | /* indicate to others that our reserved page is uninitialized */ | ||
1355 | xpc_rsvd_page->vars_pa = 0; | ||
1356 | |||
1357 | /* take ourselves off of the reboot_notifier_list */ | ||
1358 | (void) unregister_reboot_notifier(&xpc_reboot_notifier); | ||
1359 | |||
1360 | /* take ourselves off of the die_notifier list */ | ||
1361 | (void) unregister_die_notifier(&xpc_die_notifier); | ||
1362 | |||
1363 | del_timer_sync(&xpc_hb_timer); | ||
1364 | free_irq(SGI_XPC_ACTIVATE, NULL); | ||
1365 | xpc_restrict_IPI_ops(); | ||
1366 | |||
1367 | if (xpc_sysctl) { | ||
1368 | unregister_sysctl_table(xpc_sysctl); | ||
1369 | } | ||
1370 | |||
1371 | kfree(xpc_remote_copy_buffer_base); | ||
1372 | return -EBUSY; | ||
1373 | } | ||
1374 | |||
1375 | |||
1376 | /* | ||
1377 | * Startup a thread that will attempt to discover other partitions to | ||
1378 | * activate based on info provided by SAL. This new thread is short | ||
1379 | * lived and will exit once discovery is complete. | ||
1380 | */ | ||
1381 | pid = kernel_thread(xpc_initiate_discovery, NULL, 0); | ||
1382 | if (pid < 0) { | ||
1383 | dev_err(xpc_part, "failed while forking discovery thread\n"); | ||
1384 | |||
1385 | /* mark this new thread as a non-starter */ | ||
1386 | complete(&xpc_discovery_exited); | ||
1387 | |||
1388 | xpc_do_exit(xpcUnloading); | ||
1389 | return -EBUSY; | ||
1390 | } | ||
1391 | |||
1392 | |||
1393 | /* set the interface to point at XPC's functions */ | ||
1394 | xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect, | ||
1395 | xpc_initiate_allocate, xpc_initiate_send, | ||
1396 | xpc_initiate_send_notify, xpc_initiate_received, | ||
1397 | xpc_initiate_partid_to_nasids); | ||
1398 | |||
1399 | return 0; | ||
1400 | } | ||
1401 | module_init(xpc_init); | ||
1402 | |||
1403 | |||
1404 | void __exit | ||
1405 | xpc_exit(void) | ||
1406 | { | ||
1407 | xpc_do_exit(xpcUnloading); | ||
1408 | } | ||
1409 | module_exit(xpc_exit); | ||
1410 | |||
1411 | |||
1412 | MODULE_AUTHOR("Silicon Graphics, Inc."); | ||
1413 | MODULE_DESCRIPTION("Cross Partition Communication (XPC) support"); | ||
1414 | MODULE_LICENSE("GPL"); | ||
1415 | |||
1416 | module_param(xpc_hb_interval, int, 0); | ||
1417 | MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between " | ||
1418 | "heartbeat increments."); | ||
1419 | |||
1420 | module_param(xpc_hb_check_interval, int, 0); | ||
1421 | MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between " | ||
1422 | "heartbeat checks."); | ||
1423 | |||
1424 | module_param(xpc_disengage_request_timelimit, int, 0); | ||
1425 | MODULE_PARM_DESC(xpc_disengage_request_timelimit, "Number of seconds to wait " | ||
1426 | "for disengage request to complete."); | ||
1427 | |||
1428 | module_param(xpc_kdebug_ignore, int, 0); | ||
1429 | MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by " | ||
1430 | "other partitions when dropping into kdebug."); | ||
1431 | |||