Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6.git] / drivers / misc / sgi-xp / xpc_sn2.c
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) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
7  */
8
9 /*
10  * Cross Partition Communication (XPC) sn2-based functions.
11  *
12  *     Architecture specific implementation of common functions.
13  *
14  */
15
16 #include <linux/delay.h>
17 #include <asm/uncached.h>
18 #include <asm/sn/mspec.h>
19 #include <asm/sn/sn_sal.h>
20 #include "xpc.h"
21
22 /*
23  * Define the number of u64s required to represent all the C-brick nasids
24  * as a bitmap.  The cross-partition kernel modules deal only with
25  * C-brick nasids, thus the need for bitmaps which don't account for
26  * odd-numbered (non C-brick) nasids.
27  */
28 #define XPC_MAX_PHYSNODES_SN2   (MAX_NUMALINK_NODES / 2)
29 #define XP_NASID_MASK_BYTES_SN2 ((XPC_MAX_PHYSNODES_SN2 + 7) / 8)
30 #define XP_NASID_MASK_WORDS_SN2 ((XPC_MAX_PHYSNODES_SN2 + 63) / 64)
31
32 /*
33  * Memory for XPC's amo variables is allocated by the MSPEC driver. These
34  * pages are located in the lowest granule. The lowest granule uses 4k pages
35  * for cached references and an alternate TLB handler to never provide a
36  * cacheable mapping for the entire region. This will prevent speculative
37  * reading of cached copies of our lines from being issued which will cause
38  * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
39  * amo variables (based on XP_MAX_NPARTITIONS_SN2) to identify the senders of
40  * NOTIFY IRQs, 128 amo variables (based on XP_NASID_MASK_WORDS_SN2) to identify
41  * the senders of ACTIVATE IRQs, 1 amo variable to identify which remote
42  * partitions (i.e., XPCs) consider themselves currently engaged with the
43  * local XPC and 1 amo variable to request partition deactivation.
44  */
45 #define XPC_NOTIFY_IRQ_AMOS_SN2         0
46 #define XPC_ACTIVATE_IRQ_AMOS_SN2       (XPC_NOTIFY_IRQ_AMOS_SN2 + \
47                                          XP_MAX_NPARTITIONS_SN2)
48 #define XPC_ENGAGED_PARTITIONS_AMO_SN2  (XPC_ACTIVATE_IRQ_AMOS_SN2 + \
49                                          XP_NASID_MASK_WORDS_SN2)
50 #define XPC_DEACTIVATE_REQUEST_AMO_SN2  (XPC_ENGAGED_PARTITIONS_AMO_SN2 + 1)
51
52 /*
53  * Buffer used to store a local copy of portions of a remote partition's
54  * reserved page (either its header and part_nasids mask, or its vars).
55  */
56 static void *xpc_remote_copy_buffer_base_sn2;
57 static char *xpc_remote_copy_buffer_sn2;
58
59 static struct xpc_vars_sn2 *xpc_vars_sn2;
60 static struct xpc_vars_part_sn2 *xpc_vars_part_sn2;
61
62 static int
63 xpc_setup_partitions_sn2(void)
64 {
65         /* nothing needs to be done */
66         return 0;
67 }
68
69 static void
70 xpc_teardown_partitions_sn2(void)
71 {
72         /* nothing needs to be done */
73 }
74
75 /* SH_IPI_ACCESS shub register value on startup */
76 static u64 xpc_sh1_IPI_access_sn2;
77 static u64 xpc_sh2_IPI_access0_sn2;
78 static u64 xpc_sh2_IPI_access1_sn2;
79 static u64 xpc_sh2_IPI_access2_sn2;
80 static u64 xpc_sh2_IPI_access3_sn2;
81
82 /*
83  * Change protections to allow IPI operations.
84  */
85 static void
86 xpc_allow_IPI_ops_sn2(void)
87 {
88         int node;
89         int nasid;
90
91         /* !!! The following should get moved into SAL. */
92         if (is_shub2()) {
93                 xpc_sh2_IPI_access0_sn2 =
94                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
95                 xpc_sh2_IPI_access1_sn2 =
96                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
97                 xpc_sh2_IPI_access2_sn2 =
98                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
99                 xpc_sh2_IPI_access3_sn2 =
100                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
101
102                 for_each_online_node(node) {
103                         nasid = cnodeid_to_nasid(node);
104                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
105                               -1UL);
106                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
107                               -1UL);
108                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
109                               -1UL);
110                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
111                               -1UL);
112                 }
113         } else {
114                 xpc_sh1_IPI_access_sn2 =
115                     (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
116
117                 for_each_online_node(node) {
118                         nasid = cnodeid_to_nasid(node);
119                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
120                               -1UL);
121                 }
122         }
123 }
124
125 /*
126  * Restrict protections to disallow IPI operations.
127  */
128 static void
129 xpc_disallow_IPI_ops_sn2(void)
130 {
131         int node;
132         int nasid;
133
134         /* !!! The following should get moved into SAL. */
135         if (is_shub2()) {
136                 for_each_online_node(node) {
137                         nasid = cnodeid_to_nasid(node);
138                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
139                               xpc_sh2_IPI_access0_sn2);
140                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
141                               xpc_sh2_IPI_access1_sn2);
142                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
143                               xpc_sh2_IPI_access2_sn2);
144                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
145                               xpc_sh2_IPI_access3_sn2);
146                 }
147         } else {
148                 for_each_online_node(node) {
149                         nasid = cnodeid_to_nasid(node);
150                         HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
151                               xpc_sh1_IPI_access_sn2);
152                 }
153         }
154 }
155
156 /*
157  * The following set of functions are used for the sending and receiving of
158  * IRQs (also known as IPIs). There are two flavors of IRQs, one that is
159  * associated with partition activity (SGI_XPC_ACTIVATE) and the other that
160  * is associated with channel activity (SGI_XPC_NOTIFY).
161  */
162
163 static u64
164 xpc_receive_IRQ_amo_sn2(struct amo *amo)
165 {
166         return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_CLEAR);
167 }
168
169 static enum xp_retval
170 xpc_send_IRQ_sn2(struct amo *amo, u64 flag, int nasid, int phys_cpuid,
171                  int vector)
172 {
173         int ret = 0;
174         unsigned long irq_flags;
175
176         local_irq_save(irq_flags);
177
178         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR, flag);
179         sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
180
181         /*
182          * We must always use the nofault function regardless of whether we
183          * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
184          * didn't, we'd never know that the other partition is down and would
185          * keep sending IRQs and amos to it until the heartbeat times out.
186          */
187         ret = xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
188                                                      xp_nofault_PIOR_target));
189
190         local_irq_restore(irq_flags);
191
192         return (ret == 0) ? xpSuccess : xpPioReadError;
193 }
194
195 static struct amo *
196 xpc_init_IRQ_amo_sn2(int index)
197 {
198         struct amo *amo = xpc_vars_sn2->amos_page + index;
199
200         (void)xpc_receive_IRQ_amo_sn2(amo);     /* clear amo variable */
201         return amo;
202 }
203
204 /*
205  * Functions associated with SGI_XPC_ACTIVATE IRQ.
206  */
207
208 /*
209  * Notify the heartbeat check thread that an activate IRQ has been received.
210  */
211 static irqreturn_t
212 xpc_handle_activate_IRQ_sn2(int irq, void *dev_id)
213 {
214         unsigned long irq_flags;
215
216         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
217         xpc_activate_IRQ_rcvd++;
218         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
219
220         wake_up_interruptible(&xpc_activate_IRQ_wq);
221         return IRQ_HANDLED;
222 }
223
224 /*
225  * Flag the appropriate amo variable and send an IRQ to the specified node.
226  */
227 static void
228 xpc_send_activate_IRQ_sn2(unsigned long amos_page_pa, int from_nasid,
229                           int to_nasid, int to_phys_cpuid)
230 {
231         struct amo *amos = (struct amo *)__va(amos_page_pa +
232                                               (XPC_ACTIVATE_IRQ_AMOS_SN2 *
233                                               sizeof(struct amo)));
234
235         (void)xpc_send_IRQ_sn2(&amos[BIT_WORD(from_nasid / 2)],
236                                BIT_MASK(from_nasid / 2), to_nasid,
237                                to_phys_cpuid, SGI_XPC_ACTIVATE);
238 }
239
240 static void
241 xpc_send_local_activate_IRQ_sn2(int from_nasid)
242 {
243         unsigned long irq_flags;
244         struct amo *amos = (struct amo *)__va(xpc_vars_sn2->amos_page_pa +
245                                               (XPC_ACTIVATE_IRQ_AMOS_SN2 *
246                                               sizeof(struct amo)));
247
248         /* fake the sending and receipt of an activate IRQ from remote nasid */
249         FETCHOP_STORE_OP(TO_AMO((u64)&amos[BIT_WORD(from_nasid / 2)].variable),
250                          FETCHOP_OR, BIT_MASK(from_nasid / 2));
251
252         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
253         xpc_activate_IRQ_rcvd++;
254         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
255
256         wake_up_interruptible(&xpc_activate_IRQ_wq);
257 }
258
259 /*
260  * Functions associated with SGI_XPC_NOTIFY IRQ.
261  */
262
263 /*
264  * Check to see if any chctl flags were sent from the specified partition.
265  */
266 static void
267 xpc_check_for_sent_chctl_flags_sn2(struct xpc_partition *part)
268 {
269         union xpc_channel_ctl_flags chctl;
270         unsigned long irq_flags;
271
272         chctl.all_flags = xpc_receive_IRQ_amo_sn2(part->sn.sn2.
273                                                   local_chctl_amo_va);
274         if (chctl.all_flags == 0)
275                 return;
276
277         spin_lock_irqsave(&part->chctl_lock, irq_flags);
278         part->chctl.all_flags |= chctl.all_flags;
279         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
280
281         dev_dbg(xpc_chan, "received notify IRQ from partid=%d, chctl.all_flags="
282                 "0x%llx\n", XPC_PARTID(part), chctl.all_flags);
283
284         xpc_wakeup_channel_mgr(part);
285 }
286
287 /*
288  * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified
289  * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more
290  * than one partition, we use an amo structure per partition to indicate
291  * whether a partition has sent an IRQ or not.  If it has, then wake up the
292  * associated kthread to handle it.
293  *
294  * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IRQs sent by XPC
295  * running on other partitions.
296  *
297  * Noteworthy Arguments:
298  *
299  *      irq - Interrupt ReQuest number. NOT USED.
300  *
301  *      dev_id - partid of IRQ's potential sender.
302  */
303 static irqreturn_t
304 xpc_handle_notify_IRQ_sn2(int irq, void *dev_id)
305 {
306         short partid = (short)(u64)dev_id;
307         struct xpc_partition *part = &xpc_partitions[partid];
308
309         DBUG_ON(partid < 0 || partid >= XP_MAX_NPARTITIONS_SN2);
310
311         if (xpc_part_ref(part)) {
312                 xpc_check_for_sent_chctl_flags_sn2(part);
313
314                 xpc_part_deref(part);
315         }
316         return IRQ_HANDLED;
317 }
318
319 /*
320  * Check to see if xpc_handle_notify_IRQ_sn2() dropped any IRQs on the floor
321  * because the write to their associated amo variable completed after the IRQ
322  * was received.
323  */
324 static void
325 xpc_check_for_dropped_notify_IRQ_sn2(struct xpc_partition *part)
326 {
327         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
328
329         if (xpc_part_ref(part)) {
330                 xpc_check_for_sent_chctl_flags_sn2(part);
331
332                 part_sn2->dropped_notify_IRQ_timer.expires = jiffies +
333                     XPC_DROPPED_NOTIFY_IRQ_WAIT_INTERVAL;
334                 add_timer(&part_sn2->dropped_notify_IRQ_timer);
335                 xpc_part_deref(part);
336         }
337 }
338
339 /*
340  * Send a notify IRQ to the remote partition that is associated with the
341  * specified channel.
342  */
343 static void
344 xpc_send_notify_IRQ_sn2(struct xpc_channel *ch, u8 chctl_flag,
345                         char *chctl_flag_string, unsigned long *irq_flags)
346 {
347         struct xpc_partition *part = &xpc_partitions[ch->partid];
348         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
349         union xpc_channel_ctl_flags chctl = { 0 };
350         enum xp_retval ret;
351
352         if (likely(part->act_state != XPC_P_AS_DEACTIVATING)) {
353                 chctl.flags[ch->number] = chctl_flag;
354                 ret = xpc_send_IRQ_sn2(part_sn2->remote_chctl_amo_va,
355                                        chctl.all_flags,
356                                        part_sn2->notify_IRQ_nasid,
357                                        part_sn2->notify_IRQ_phys_cpuid,
358                                        SGI_XPC_NOTIFY);
359                 dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
360                         chctl_flag_string, ch->partid, ch->number, ret);
361                 if (unlikely(ret != xpSuccess)) {
362                         if (irq_flags != NULL)
363                                 spin_unlock_irqrestore(&ch->lock, *irq_flags);
364                         XPC_DEACTIVATE_PARTITION(part, ret);
365                         if (irq_flags != NULL)
366                                 spin_lock_irqsave(&ch->lock, *irq_flags);
367                 }
368         }
369 }
370
371 #define XPC_SEND_NOTIFY_IRQ_SN2(_ch, _ipi_f, _irq_f) \
372                 xpc_send_notify_IRQ_sn2(_ch, _ipi_f, #_ipi_f, _irq_f)
373
374 /*
375  * Make it look like the remote partition, which is associated with the
376  * specified channel, sent us a notify IRQ. This faked IRQ will be handled
377  * by xpc_check_for_dropped_notify_IRQ_sn2().
378  */
379 static void
380 xpc_send_local_notify_IRQ_sn2(struct xpc_channel *ch, u8 chctl_flag,
381                               char *chctl_flag_string)
382 {
383         struct xpc_partition *part = &xpc_partitions[ch->partid];
384         union xpc_channel_ctl_flags chctl = { 0 };
385
386         chctl.flags[ch->number] = chctl_flag;
387         FETCHOP_STORE_OP(TO_AMO((u64)&part->sn.sn2.local_chctl_amo_va->
388                                 variable), FETCHOP_OR, chctl.all_flags);
389         dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
390                 chctl_flag_string, ch->partid, ch->number);
391 }
392
393 #define XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(_ch, _ipi_f) \
394                 xpc_send_local_notify_IRQ_sn2(_ch, _ipi_f, #_ipi_f)
395
396 static void
397 xpc_send_chctl_closerequest_sn2(struct xpc_channel *ch,
398                                 unsigned long *irq_flags)
399 {
400         struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
401
402         args->reason = ch->reason;
403         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_CLOSEREQUEST, irq_flags);
404 }
405
406 static void
407 xpc_send_chctl_closereply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
408 {
409         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_CLOSEREPLY, irq_flags);
410 }
411
412 static void
413 xpc_send_chctl_openrequest_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
414 {
415         struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
416
417         args->entry_size = ch->entry_size;
418         args->local_nentries = ch->local_nentries;
419         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENREQUEST, irq_flags);
420 }
421
422 static void
423 xpc_send_chctl_openreply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
424 {
425         struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
426
427         args->remote_nentries = ch->remote_nentries;
428         args->local_nentries = ch->local_nentries;
429         args->local_msgqueue_pa = xp_pa(ch->sn.sn2.local_msgqueue);
430         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENREPLY, irq_flags);
431 }
432
433 static void
434 xpc_send_chctl_opencomplete_sn2(struct xpc_channel *ch,
435                                 unsigned long *irq_flags)
436 {
437         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENCOMPLETE, irq_flags);
438 }
439
440 static void
441 xpc_send_chctl_msgrequest_sn2(struct xpc_channel *ch)
442 {
443         XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_MSGREQUEST, NULL);
444 }
445
446 static void
447 xpc_send_chctl_local_msgrequest_sn2(struct xpc_channel *ch)
448 {
449         XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_MSGREQUEST);
450 }
451
452 static enum xp_retval
453 xpc_save_remote_msgqueue_pa_sn2(struct xpc_channel *ch,
454                                 unsigned long msgqueue_pa)
455 {
456         ch->sn.sn2.remote_msgqueue_pa = msgqueue_pa;
457         return xpSuccess;
458 }
459
460 /*
461  * This next set of functions are used to keep track of when a partition is
462  * potentially engaged in accessing memory belonging to another partition.
463  */
464
465 static void
466 xpc_indicate_partition_engaged_sn2(struct xpc_partition *part)
467 {
468         unsigned long irq_flags;
469         struct amo *amo = (struct amo *)__va(part->sn.sn2.remote_amos_page_pa +
470                                              (XPC_ENGAGED_PARTITIONS_AMO_SN2 *
471                                              sizeof(struct amo)));
472
473         local_irq_save(irq_flags);
474
475         /* set bit corresponding to our partid in remote partition's amo */
476         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
477                          BIT(sn_partition_id));
478
479         /*
480          * We must always use the nofault function regardless of whether we
481          * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
482          * didn't, we'd never know that the other partition is down and would
483          * keep sending IRQs and amos to it until the heartbeat times out.
484          */
485         (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
486                                                                variable),
487                                                      xp_nofault_PIOR_target));
488
489         local_irq_restore(irq_flags);
490 }
491
492 static void
493 xpc_indicate_partition_disengaged_sn2(struct xpc_partition *part)
494 {
495         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
496         unsigned long irq_flags;
497         struct amo *amo = (struct amo *)__va(part_sn2->remote_amos_page_pa +
498                                              (XPC_ENGAGED_PARTITIONS_AMO_SN2 *
499                                              sizeof(struct amo)));
500
501         local_irq_save(irq_flags);
502
503         /* clear bit corresponding to our partid in remote partition's amo */
504         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
505                          ~BIT(sn_partition_id));
506
507         /*
508          * We must always use the nofault function regardless of whether we
509          * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
510          * didn't, we'd never know that the other partition is down and would
511          * keep sending IRQs and amos to it until the heartbeat times out.
512          */
513         (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
514                                                                variable),
515                                                      xp_nofault_PIOR_target));
516
517         local_irq_restore(irq_flags);
518
519         /*
520          * Send activate IRQ to get other side to see that we've cleared our
521          * bit in their engaged partitions amo.
522          */
523         xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
524                                   cnodeid_to_nasid(0),
525                                   part_sn2->activate_IRQ_nasid,
526                                   part_sn2->activate_IRQ_phys_cpuid);
527 }
528
529 static void
530 xpc_assume_partition_disengaged_sn2(short partid)
531 {
532         struct amo *amo = xpc_vars_sn2->amos_page +
533                           XPC_ENGAGED_PARTITIONS_AMO_SN2;
534
535         /* clear bit(s) based on partid mask in our partition's amo */
536         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
537                          ~BIT(partid));
538 }
539
540 static int
541 xpc_partition_engaged_sn2(short partid)
542 {
543         struct amo *amo = xpc_vars_sn2->amos_page +
544                           XPC_ENGAGED_PARTITIONS_AMO_SN2;
545
546         /* our partition's amo variable ANDed with partid mask */
547         return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
548                 BIT(partid)) != 0;
549 }
550
551 static int
552 xpc_any_partition_engaged_sn2(void)
553 {
554         struct amo *amo = xpc_vars_sn2->amos_page +
555                           XPC_ENGAGED_PARTITIONS_AMO_SN2;
556
557         /* our partition's amo variable */
558         return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) != 0;
559 }
560
561 /* original protection values for each node */
562 static u64 xpc_prot_vec_sn2[MAX_NUMNODES];
563
564 /*
565  * Change protections to allow amo operations on non-Shub 1.1 systems.
566  */
567 static enum xp_retval
568 xpc_allow_amo_ops_sn2(struct amo *amos_page)
569 {
570         enum xp_retval ret = xpSuccess;
571
572         /*
573          * On SHUB 1.1, we cannot call sn_change_memprotect() since the BIST
574          * collides with memory operations. On those systems we call
575          * xpc_allow_amo_ops_shub_wars_1_1_sn2() instead.
576          */
577         if (!enable_shub_wars_1_1())
578                 ret = xp_expand_memprotect(ia64_tpa((u64)amos_page), PAGE_SIZE);
579
580         return ret;
581 }
582
583 /*
584  * Change protections to allow amo operations on Shub 1.1 systems.
585  */
586 static void
587 xpc_allow_amo_ops_shub_wars_1_1_sn2(void)
588 {
589         int node;
590         int nasid;
591
592         if (!enable_shub_wars_1_1())
593                 return;
594
595         for_each_online_node(node) {
596                 nasid = cnodeid_to_nasid(node);
597                 /* save current protection values */
598                 xpc_prot_vec_sn2[node] =
599                     (u64)HUB_L((u64 *)GLOBAL_MMR_ADDR(nasid,
600                                                   SH1_MD_DQLP_MMR_DIR_PRIVEC0));
601                 /* open up everything */
602                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
603                                              SH1_MD_DQLP_MMR_DIR_PRIVEC0),
604                       -1UL);
605                 HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
606                                              SH1_MD_DQRP_MMR_DIR_PRIVEC0),
607                       -1UL);
608         }
609 }
610
611 static enum xp_retval
612 xpc_get_partition_rsvd_page_pa_sn2(void *buf, u64 *cookie, unsigned long *rp_pa,
613                                    size_t *len)
614 {
615         s64 status;
616         enum xp_retval ret;
617
618         status = sn_partition_reserved_page_pa((u64)buf, cookie,
619                         (u64 *)rp_pa, (u64 *)len);
620         if (status == SALRET_OK)
621                 ret = xpSuccess;
622         else if (status == SALRET_MORE_PASSES)
623                 ret = xpNeedMoreInfo;
624         else
625                 ret = xpSalError;
626
627         return ret;
628 }
629
630
631 static int
632 xpc_setup_rsvd_page_sn2(struct xpc_rsvd_page *rp)
633 {
634         struct amo *amos_page;
635         int i;
636         int ret;
637
638         xpc_vars_sn2 = XPC_RP_VARS(rp);
639
640         rp->sn.sn2.vars_pa = xp_pa(xpc_vars_sn2);
641
642         /* vars_part array follows immediately after vars */
643         xpc_vars_part_sn2 = (struct xpc_vars_part_sn2 *)((u8 *)XPC_RP_VARS(rp) +
644                                                          XPC_RP_VARS_SIZE);
645
646         /*
647          * Before clearing xpc_vars_sn2, see if a page of amos had been
648          * previously allocated. If not we'll need to allocate one and set
649          * permissions so that cross-partition amos are allowed.
650          *
651          * The allocated amo page needs MCA reporting to remain disabled after
652          * XPC has unloaded.  To make this work, we keep a copy of the pointer
653          * to this page (i.e., amos_page) in the struct xpc_vars_sn2 structure,
654          * which is pointed to by the reserved page, and re-use that saved copy
655          * on subsequent loads of XPC. This amo page is never freed, and its
656          * memory protections are never restricted.
657          */
658         amos_page = xpc_vars_sn2->amos_page;
659         if (amos_page == NULL) {
660                 amos_page = (struct amo *)TO_AMO(uncached_alloc_page(0, 1));
661                 if (amos_page == NULL) {
662                         dev_err(xpc_part, "can't allocate page of amos\n");
663                         return -ENOMEM;
664                 }
665
666                 /*
667                  * Open up amo-R/W to cpu.  This is done on Shub 1.1 systems
668                  * when xpc_allow_amo_ops_shub_wars_1_1_sn2() is called.
669                  */
670                 ret = xpc_allow_amo_ops_sn2(amos_page);
671                 if (ret != xpSuccess) {
672                         dev_err(xpc_part, "can't allow amo operations\n");
673                         uncached_free_page(__IA64_UNCACHED_OFFSET |
674                                            TO_PHYS((u64)amos_page), 1);
675                         return -EPERM;
676                 }
677         }
678
679         /* clear xpc_vars_sn2 */
680         memset(xpc_vars_sn2, 0, sizeof(struct xpc_vars_sn2));
681
682         xpc_vars_sn2->version = XPC_V_VERSION;
683         xpc_vars_sn2->activate_IRQ_nasid = cpuid_to_nasid(0);
684         xpc_vars_sn2->activate_IRQ_phys_cpuid = cpu_physical_id(0);
685         xpc_vars_sn2->vars_part_pa = xp_pa(xpc_vars_part_sn2);
686         xpc_vars_sn2->amos_page_pa = ia64_tpa((u64)amos_page);
687         xpc_vars_sn2->amos_page = amos_page;    /* save for next load of XPC */
688
689         /* clear xpc_vars_part_sn2 */
690         memset((u64 *)xpc_vars_part_sn2, 0, sizeof(struct xpc_vars_part_sn2) *
691                XP_MAX_NPARTITIONS_SN2);
692
693         /* initialize the activate IRQ related amo variables */
694         for (i = 0; i < xpc_nasid_mask_nlongs; i++)
695                 (void)xpc_init_IRQ_amo_sn2(XPC_ACTIVATE_IRQ_AMOS_SN2 + i);
696
697         /* initialize the engaged remote partitions related amo variables */
698         (void)xpc_init_IRQ_amo_sn2(XPC_ENGAGED_PARTITIONS_AMO_SN2);
699         (void)xpc_init_IRQ_amo_sn2(XPC_DEACTIVATE_REQUEST_AMO_SN2);
700
701         return 0;
702 }
703
704 static int
705 xpc_hb_allowed_sn2(short partid, void *heartbeating_to_mask)
706 {
707         return test_bit(partid, heartbeating_to_mask);
708 }
709
710 static void
711 xpc_allow_hb_sn2(short partid)
712 {
713         DBUG_ON(xpc_vars_sn2 == NULL);
714         set_bit(partid, xpc_vars_sn2->heartbeating_to_mask);
715 }
716
717 static void
718 xpc_disallow_hb_sn2(short partid)
719 {
720         DBUG_ON(xpc_vars_sn2 == NULL);
721         clear_bit(partid, xpc_vars_sn2->heartbeating_to_mask);
722 }
723
724 static void
725 xpc_disallow_all_hbs_sn2(void)
726 {
727         DBUG_ON(xpc_vars_sn2 == NULL);
728         bitmap_zero(xpc_vars_sn2->heartbeating_to_mask, xp_max_npartitions);
729 }
730
731 static void
732 xpc_increment_heartbeat_sn2(void)
733 {
734         xpc_vars_sn2->heartbeat++;
735 }
736
737 static void
738 xpc_offline_heartbeat_sn2(void)
739 {
740         xpc_increment_heartbeat_sn2();
741         xpc_vars_sn2->heartbeat_offline = 1;
742 }
743
744 static void
745 xpc_online_heartbeat_sn2(void)
746 {
747         xpc_increment_heartbeat_sn2();
748         xpc_vars_sn2->heartbeat_offline = 0;
749 }
750
751 static void
752 xpc_heartbeat_init_sn2(void)
753 {
754         DBUG_ON(xpc_vars_sn2 == NULL);
755
756         bitmap_zero(xpc_vars_sn2->heartbeating_to_mask, XP_MAX_NPARTITIONS_SN2);
757         xpc_online_heartbeat_sn2();
758 }
759
760 static void
761 xpc_heartbeat_exit_sn2(void)
762 {
763         xpc_offline_heartbeat_sn2();
764 }
765
766 static enum xp_retval
767 xpc_get_remote_heartbeat_sn2(struct xpc_partition *part)
768 {
769         struct xpc_vars_sn2 *remote_vars;
770         enum xp_retval ret;
771
772         remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer_sn2;
773
774         /* pull the remote vars structure that contains the heartbeat */
775         ret = xp_remote_memcpy(xp_pa(remote_vars),
776                                part->sn.sn2.remote_vars_pa,
777                                XPC_RP_VARS_SIZE);
778         if (ret != xpSuccess)
779                 return ret;
780
781         dev_dbg(xpc_part, "partid=%d, heartbeat=%lld, last_heartbeat=%lld, "
782                 "heartbeat_offline=%lld, HB_mask[0]=0x%lx\n", XPC_PARTID(part),
783                 remote_vars->heartbeat, part->last_heartbeat,
784                 remote_vars->heartbeat_offline,
785                 remote_vars->heartbeating_to_mask[0]);
786
787         if ((remote_vars->heartbeat == part->last_heartbeat &&
788             !remote_vars->heartbeat_offline) ||
789             !xpc_hb_allowed_sn2(sn_partition_id,
790                                 remote_vars->heartbeating_to_mask)) {
791                 ret = xpNoHeartbeat;
792         } else {
793                 part->last_heartbeat = remote_vars->heartbeat;
794         }
795
796         return ret;
797 }
798
799 /*
800  * Get a copy of the remote partition's XPC variables from the reserved page.
801  *
802  * remote_vars points to a buffer that is cacheline aligned for BTE copies and
803  * assumed to be of size XPC_RP_VARS_SIZE.
804  */
805 static enum xp_retval
806 xpc_get_remote_vars_sn2(unsigned long remote_vars_pa,
807                         struct xpc_vars_sn2 *remote_vars)
808 {
809         enum xp_retval ret;
810
811         if (remote_vars_pa == 0)
812                 return xpVarsNotSet;
813
814         /* pull over the cross partition variables */
815         ret = xp_remote_memcpy(xp_pa(remote_vars), remote_vars_pa,
816                                XPC_RP_VARS_SIZE);
817         if (ret != xpSuccess)
818                 return ret;
819
820         if (XPC_VERSION_MAJOR(remote_vars->version) !=
821             XPC_VERSION_MAJOR(XPC_V_VERSION)) {
822                 return xpBadVersion;
823         }
824
825         return xpSuccess;
826 }
827
828 static void
829 xpc_request_partition_activation_sn2(struct xpc_rsvd_page *remote_rp,
830                                      unsigned long remote_rp_pa, int nasid)
831 {
832         xpc_send_local_activate_IRQ_sn2(nasid);
833 }
834
835 static void
836 xpc_request_partition_reactivation_sn2(struct xpc_partition *part)
837 {
838         xpc_send_local_activate_IRQ_sn2(part->sn.sn2.activate_IRQ_nasid);
839 }
840
841 static void
842 xpc_request_partition_deactivation_sn2(struct xpc_partition *part)
843 {
844         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
845         unsigned long irq_flags;
846         struct amo *amo = (struct amo *)__va(part_sn2->remote_amos_page_pa +
847                                              (XPC_DEACTIVATE_REQUEST_AMO_SN2 *
848                                              sizeof(struct amo)));
849
850         local_irq_save(irq_flags);
851
852         /* set bit corresponding to our partid in remote partition's amo */
853         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
854                          BIT(sn_partition_id));
855
856         /*
857          * We must always use the nofault function regardless of whether we
858          * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
859          * didn't, we'd never know that the other partition is down and would
860          * keep sending IRQs and amos to it until the heartbeat times out.
861          */
862         (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
863                                                                variable),
864                                                      xp_nofault_PIOR_target));
865
866         local_irq_restore(irq_flags);
867
868         /*
869          * Send activate IRQ to get other side to see that we've set our
870          * bit in their deactivate request amo.
871          */
872         xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
873                                   cnodeid_to_nasid(0),
874                                   part_sn2->activate_IRQ_nasid,
875                                   part_sn2->activate_IRQ_phys_cpuid);
876 }
877
878 static void
879 xpc_cancel_partition_deactivation_request_sn2(struct xpc_partition *part)
880 {
881         unsigned long irq_flags;
882         struct amo *amo = (struct amo *)__va(part->sn.sn2.remote_amos_page_pa +
883                                              (XPC_DEACTIVATE_REQUEST_AMO_SN2 *
884                                              sizeof(struct amo)));
885
886         local_irq_save(irq_flags);
887
888         /* clear bit corresponding to our partid in remote partition's amo */
889         FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
890                          ~BIT(sn_partition_id));
891
892         /*
893          * We must always use the nofault function regardless of whether we
894          * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
895          * didn't, we'd never know that the other partition is down and would
896          * keep sending IRQs and amos to it until the heartbeat times out.
897          */
898         (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
899                                                                variable),
900                                                      xp_nofault_PIOR_target));
901
902         local_irq_restore(irq_flags);
903 }
904
905 static int
906 xpc_partition_deactivation_requested_sn2(short partid)
907 {
908         struct amo *amo = xpc_vars_sn2->amos_page +
909                           XPC_DEACTIVATE_REQUEST_AMO_SN2;
910
911         /* our partition's amo variable ANDed with partid mask */
912         return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
913                 BIT(partid)) != 0;
914 }
915
916 /*
917  * Update the remote partition's info.
918  */
919 static void
920 xpc_update_partition_info_sn2(struct xpc_partition *part, u8 remote_rp_version,
921                               unsigned long *remote_rp_ts_jiffies,
922                               unsigned long remote_rp_pa,
923                               unsigned long remote_vars_pa,
924                               struct xpc_vars_sn2 *remote_vars)
925 {
926         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
927
928         part->remote_rp_version = remote_rp_version;
929         dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
930                 part->remote_rp_version);
931
932         part->remote_rp_ts_jiffies = *remote_rp_ts_jiffies;
933         dev_dbg(xpc_part, "  remote_rp_ts_jiffies = 0x%016lx\n",
934                 part->remote_rp_ts_jiffies);
935
936         part->remote_rp_pa = remote_rp_pa;
937         dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
938
939         part_sn2->remote_vars_pa = remote_vars_pa;
940         dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
941                 part_sn2->remote_vars_pa);
942
943         part->last_heartbeat = remote_vars->heartbeat - 1;
944         dev_dbg(xpc_part, "  last_heartbeat = 0x%016llx\n",
945                 part->last_heartbeat);
946
947         part_sn2->remote_vars_part_pa = remote_vars->vars_part_pa;
948         dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
949                 part_sn2->remote_vars_part_pa);
950
951         part_sn2->activate_IRQ_nasid = remote_vars->activate_IRQ_nasid;
952         dev_dbg(xpc_part, "  activate_IRQ_nasid = 0x%x\n",
953                 part_sn2->activate_IRQ_nasid);
954
955         part_sn2->activate_IRQ_phys_cpuid =
956             remote_vars->activate_IRQ_phys_cpuid;
957         dev_dbg(xpc_part, "  activate_IRQ_phys_cpuid = 0x%x\n",
958                 part_sn2->activate_IRQ_phys_cpuid);
959
960         part_sn2->remote_amos_page_pa = remote_vars->amos_page_pa;
961         dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
962                 part_sn2->remote_amos_page_pa);
963
964         part_sn2->remote_vars_version = remote_vars->version;
965         dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
966                 part_sn2->remote_vars_version);
967 }
968
969 /*
970  * Prior code has determined the nasid which generated a activate IRQ.
971  * Inspect that nasid to determine if its partition needs to be activated
972  * or deactivated.
973  *
974  * A partition is considered "awaiting activation" if our partition
975  * flags indicate it is not active and it has a heartbeat.  A
976  * partition is considered "awaiting deactivation" if our partition
977  * flags indicate it is active but it has no heartbeat or it is not
978  * sending its heartbeat to us.
979  *
980  * To determine the heartbeat, the remote nasid must have a properly
981  * initialized reserved page.
982  */
983 static void
984 xpc_identify_activate_IRQ_req_sn2(int nasid)
985 {
986         struct xpc_rsvd_page *remote_rp;
987         struct xpc_vars_sn2 *remote_vars;
988         unsigned long remote_rp_pa;
989         unsigned long remote_vars_pa;
990         int remote_rp_version;
991         int reactivate = 0;
992         unsigned long remote_rp_ts_jiffies = 0;
993         short partid;
994         struct xpc_partition *part;
995         struct xpc_partition_sn2 *part_sn2;
996         enum xp_retval ret;
997
998         /* pull over the reserved page structure */
999
1000         remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer_sn2;
1001
1002         ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
1003         if (ret != xpSuccess) {
1004                 dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
1005                          "which sent interrupt, reason=%d\n", nasid, ret);
1006                 return;
1007         }
1008
1009         remote_vars_pa = remote_rp->sn.sn2.vars_pa;
1010         remote_rp_version = remote_rp->version;
1011         remote_rp_ts_jiffies = remote_rp->ts_jiffies;
1012
1013         partid = remote_rp->SAL_partid;
1014         part = &xpc_partitions[partid];
1015         part_sn2 = &part->sn.sn2;
1016
1017         /* pull over the cross partition variables */
1018
1019         remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer_sn2;
1020
1021         ret = xpc_get_remote_vars_sn2(remote_vars_pa, remote_vars);
1022         if (ret != xpSuccess) {
1023                 dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
1024                          "which sent interrupt, reason=%d\n", nasid, ret);
1025
1026                 XPC_DEACTIVATE_PARTITION(part, ret);
1027                 return;
1028         }
1029
1030         part->activate_IRQ_rcvd++;
1031
1032         dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
1033                 "%lld:0x%lx\n", (int)nasid, (int)partid,
1034                 part->activate_IRQ_rcvd,
1035                 remote_vars->heartbeat, remote_vars->heartbeating_to_mask[0]);
1036
1037         if (xpc_partition_disengaged(part) &&
1038             part->act_state == XPC_P_AS_INACTIVE) {
1039
1040                 xpc_update_partition_info_sn2(part, remote_rp_version,
1041                                               &remote_rp_ts_jiffies,
1042                                               remote_rp_pa, remote_vars_pa,
1043                                               remote_vars);
1044
1045                 if (xpc_partition_deactivation_requested_sn2(partid)) {
1046                         /*
1047                          * Other side is waiting on us to deactivate even though
1048                          * we already have.
1049                          */
1050                         return;
1051                 }
1052
1053                 xpc_activate_partition(part);
1054                 return;
1055         }
1056
1057         DBUG_ON(part->remote_rp_version == 0);
1058         DBUG_ON(part_sn2->remote_vars_version == 0);
1059
1060         if (remote_rp_ts_jiffies != part->remote_rp_ts_jiffies) {
1061
1062                 /* the other side rebooted */
1063
1064                 DBUG_ON(xpc_partition_engaged_sn2(partid));
1065                 DBUG_ON(xpc_partition_deactivation_requested_sn2(partid));
1066
1067                 xpc_update_partition_info_sn2(part, remote_rp_version,
1068                                               &remote_rp_ts_jiffies,
1069                                               remote_rp_pa, remote_vars_pa,
1070                                               remote_vars);
1071                 reactivate = 1;
1072         }
1073
1074         if (part->disengage_timeout > 0 && !xpc_partition_disengaged(part)) {
1075                 /* still waiting on other side to disengage from us */
1076                 return;
1077         }
1078
1079         if (reactivate)
1080                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
1081         else if (xpc_partition_deactivation_requested_sn2(partid))
1082                 XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
1083 }
1084
1085 /*
1086  * Loop through the activation amo variables and process any bits
1087  * which are set.  Each bit indicates a nasid sending a partition
1088  * activation or deactivation request.
1089  *
1090  * Return #of IRQs detected.
1091  */
1092 int
1093 xpc_identify_activate_IRQ_sender_sn2(void)
1094 {
1095         int l;
1096         int b;
1097         unsigned long nasid_mask_long;
1098         u64 nasid;              /* remote nasid */
1099         int n_IRQs_detected = 0;
1100         struct amo *act_amos;
1101
1102         act_amos = xpc_vars_sn2->amos_page + XPC_ACTIVATE_IRQ_AMOS_SN2;
1103
1104         /* scan through activate amo variables looking for non-zero entries */
1105         for (l = 0; l < xpc_nasid_mask_nlongs; l++) {
1106
1107                 if (xpc_exiting)
1108                         break;
1109
1110                 nasid_mask_long = xpc_receive_IRQ_amo_sn2(&act_amos[l]);
1111
1112                 b = find_first_bit(&nasid_mask_long, BITS_PER_LONG);
1113                 if (b >= BITS_PER_LONG) {
1114                         /* no IRQs from nasids in this amo variable */
1115                         continue;
1116                 }
1117
1118                 dev_dbg(xpc_part, "amo[%d] gave back 0x%lx\n", l,
1119                         nasid_mask_long);
1120
1121                 /*
1122                  * If this nasid has been added to the machine since
1123                  * our partition was reset, this will retain the
1124                  * remote nasid in our reserved pages machine mask.
1125                  * This is used in the event of module reload.
1126                  */
1127                 xpc_mach_nasids[l] |= nasid_mask_long;
1128
1129                 /* locate the nasid(s) which sent interrupts */
1130
1131                 do {
1132                         n_IRQs_detected++;
1133                         nasid = (l * BITS_PER_LONG + b) * 2;
1134                         dev_dbg(xpc_part, "interrupt from nasid %lld\n", nasid);
1135                         xpc_identify_activate_IRQ_req_sn2(nasid);
1136
1137                         b = find_next_bit(&nasid_mask_long, BITS_PER_LONG,
1138                                           b + 1);
1139                 } while (b < BITS_PER_LONG);
1140         }
1141         return n_IRQs_detected;
1142 }
1143
1144 static void
1145 xpc_process_activate_IRQ_rcvd_sn2(void)
1146 {
1147         unsigned long irq_flags;
1148         int n_IRQs_expected;
1149         int n_IRQs_detected;
1150
1151         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1152         n_IRQs_expected = xpc_activate_IRQ_rcvd;
1153         xpc_activate_IRQ_rcvd = 0;
1154         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1155
1156         n_IRQs_detected = xpc_identify_activate_IRQ_sender_sn2();
1157         if (n_IRQs_detected < n_IRQs_expected) {
1158                 /* retry once to help avoid missing amo */
1159                 (void)xpc_identify_activate_IRQ_sender_sn2();
1160         }
1161 }
1162
1163 /*
1164  * Setup the channel structures that are sn2 specific.
1165  */
1166 static enum xp_retval
1167 xpc_setup_ch_structures_sn2(struct xpc_partition *part)
1168 {
1169         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1170         struct xpc_channel_sn2 *ch_sn2;
1171         enum xp_retval retval;
1172         int ret;
1173         int cpuid;
1174         int ch_number;
1175         struct timer_list *timer;
1176         short partid = XPC_PARTID(part);
1177
1178         /* allocate all the required GET/PUT values */
1179
1180         part_sn2->local_GPs =
1181             xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE, GFP_KERNEL,
1182                                           &part_sn2->local_GPs_base);
1183         if (part_sn2->local_GPs == NULL) {
1184                 dev_err(xpc_chan, "can't get memory for local get/put "
1185                         "values\n");
1186                 return xpNoMemory;
1187         }
1188
1189         part_sn2->remote_GPs =
1190             xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE, GFP_KERNEL,
1191                                           &part_sn2->remote_GPs_base);
1192         if (part_sn2->remote_GPs == NULL) {
1193                 dev_err(xpc_chan, "can't get memory for remote get/put "
1194                         "values\n");
1195                 retval = xpNoMemory;
1196                 goto out_1;
1197         }
1198
1199         part_sn2->remote_GPs_pa = 0;
1200
1201         /* allocate all the required open and close args */
1202
1203         part_sn2->local_openclose_args =
1204             xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE,
1205                                           GFP_KERNEL, &part_sn2->
1206                                           local_openclose_args_base);
1207         if (part_sn2->local_openclose_args == NULL) {
1208                 dev_err(xpc_chan, "can't get memory for local connect args\n");
1209                 retval = xpNoMemory;
1210                 goto out_2;
1211         }
1212
1213         part_sn2->remote_openclose_args_pa = 0;
1214
1215         part_sn2->local_chctl_amo_va = xpc_init_IRQ_amo_sn2(partid);
1216
1217         part_sn2->notify_IRQ_nasid = 0;
1218         part_sn2->notify_IRQ_phys_cpuid = 0;
1219         part_sn2->remote_chctl_amo_va = NULL;
1220
1221         sprintf(part_sn2->notify_IRQ_owner, "xpc%02d", partid);
1222         ret = request_irq(SGI_XPC_NOTIFY, xpc_handle_notify_IRQ_sn2,
1223                           IRQF_SHARED, part_sn2->notify_IRQ_owner,
1224                           (void *)(u64)partid);
1225         if (ret != 0) {
1226                 dev_err(xpc_chan, "can't register NOTIFY IRQ handler, "
1227                         "errno=%d\n", -ret);
1228                 retval = xpLackOfResources;
1229                 goto out_3;
1230         }
1231
1232         /* Setup a timer to check for dropped notify IRQs */
1233         timer = &part_sn2->dropped_notify_IRQ_timer;
1234         init_timer(timer);
1235         timer->function =
1236             (void (*)(unsigned long))xpc_check_for_dropped_notify_IRQ_sn2;
1237         timer->data = (unsigned long)part;
1238         timer->expires = jiffies + XPC_DROPPED_NOTIFY_IRQ_WAIT_INTERVAL;
1239         add_timer(timer);
1240
1241         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1242                 ch_sn2 = &part->channels[ch_number].sn.sn2;
1243
1244                 ch_sn2->local_GP = &part_sn2->local_GPs[ch_number];
1245                 ch_sn2->local_openclose_args =
1246                     &part_sn2->local_openclose_args[ch_number];
1247
1248                 mutex_init(&ch_sn2->msg_to_pull_mutex);
1249         }
1250
1251         /*
1252          * Setup the per partition specific variables required by the
1253          * remote partition to establish channel connections with us.
1254          *
1255          * The setting of the magic # indicates that these per partition
1256          * specific variables are ready to be used.
1257          */
1258         xpc_vars_part_sn2[partid].GPs_pa = xp_pa(part_sn2->local_GPs);
1259         xpc_vars_part_sn2[partid].openclose_args_pa =
1260             xp_pa(part_sn2->local_openclose_args);
1261         xpc_vars_part_sn2[partid].chctl_amo_pa =
1262             xp_pa(part_sn2->local_chctl_amo_va);
1263         cpuid = raw_smp_processor_id(); /* any CPU in this partition will do */
1264         xpc_vars_part_sn2[partid].notify_IRQ_nasid = cpuid_to_nasid(cpuid);
1265         xpc_vars_part_sn2[partid].notify_IRQ_phys_cpuid =
1266             cpu_physical_id(cpuid);
1267         xpc_vars_part_sn2[partid].nchannels = part->nchannels;
1268         xpc_vars_part_sn2[partid].magic = XPC_VP_MAGIC1_SN2;
1269
1270         return xpSuccess;
1271
1272         /* setup of ch structures failed */
1273 out_3:
1274         kfree(part_sn2->local_openclose_args_base);
1275         part_sn2->local_openclose_args = NULL;
1276 out_2:
1277         kfree(part_sn2->remote_GPs_base);
1278         part_sn2->remote_GPs = NULL;
1279 out_1:
1280         kfree(part_sn2->local_GPs_base);
1281         part_sn2->local_GPs = NULL;
1282         return retval;
1283 }
1284
1285 /*
1286  * Teardown the channel structures that are sn2 specific.
1287  */
1288 static void
1289 xpc_teardown_ch_structures_sn2(struct xpc_partition *part)
1290 {
1291         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1292         short partid = XPC_PARTID(part);
1293
1294         /*
1295          * Indicate that the variables specific to the remote partition are no
1296          * longer available for its use.
1297          */
1298         xpc_vars_part_sn2[partid].magic = 0;
1299
1300         /* in case we've still got outstanding timers registered... */
1301         del_timer_sync(&part_sn2->dropped_notify_IRQ_timer);
1302         free_irq(SGI_XPC_NOTIFY, (void *)(u64)partid);
1303
1304         kfree(part_sn2->local_openclose_args_base);
1305         part_sn2->local_openclose_args = NULL;
1306         kfree(part_sn2->remote_GPs_base);
1307         part_sn2->remote_GPs = NULL;
1308         kfree(part_sn2->local_GPs_base);
1309         part_sn2->local_GPs = NULL;
1310         part_sn2->local_chctl_amo_va = NULL;
1311 }
1312
1313 /*
1314  * Create a wrapper that hides the underlying mechanism for pulling a cacheline
1315  * (or multiple cachelines) from a remote partition.
1316  *
1317  * src_pa must be a cacheline aligned physical address on the remote partition.
1318  * dst must be a cacheline aligned virtual address on this partition.
1319  * cnt must be cacheline sized
1320  */
1321 /* ??? Replace this function by call to xp_remote_memcpy() or bte_copy()? */
1322 static enum xp_retval
1323 xpc_pull_remote_cachelines_sn2(struct xpc_partition *part, void *dst,
1324                                const unsigned long src_pa, size_t cnt)
1325 {
1326         enum xp_retval ret;
1327
1328         DBUG_ON(src_pa != L1_CACHE_ALIGN(src_pa));
1329         DBUG_ON((unsigned long)dst != L1_CACHE_ALIGN((unsigned long)dst));
1330         DBUG_ON(cnt != L1_CACHE_ALIGN(cnt));
1331
1332         if (part->act_state == XPC_P_AS_DEACTIVATING)
1333                 return part->reason;
1334
1335         ret = xp_remote_memcpy(xp_pa(dst), src_pa, cnt);
1336         if (ret != xpSuccess) {
1337                 dev_dbg(xpc_chan, "xp_remote_memcpy() from partition %d failed,"
1338                         " ret=%d\n", XPC_PARTID(part), ret);
1339         }
1340         return ret;
1341 }
1342
1343 /*
1344  * Pull the remote per partition specific variables from the specified
1345  * partition.
1346  */
1347 static enum xp_retval
1348 xpc_pull_remote_vars_part_sn2(struct xpc_partition *part)
1349 {
1350         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1351         u8 buffer[L1_CACHE_BYTES * 2];
1352         struct xpc_vars_part_sn2 *pulled_entry_cacheline =
1353             (struct xpc_vars_part_sn2 *)L1_CACHE_ALIGN((u64)buffer);
1354         struct xpc_vars_part_sn2 *pulled_entry;
1355         unsigned long remote_entry_cacheline_pa;
1356         unsigned long remote_entry_pa;
1357         short partid = XPC_PARTID(part);
1358         enum xp_retval ret;
1359
1360         /* pull the cacheline that contains the variables we're interested in */
1361
1362         DBUG_ON(part_sn2->remote_vars_part_pa !=
1363                 L1_CACHE_ALIGN(part_sn2->remote_vars_part_pa));
1364         DBUG_ON(sizeof(struct xpc_vars_part_sn2) != L1_CACHE_BYTES / 2);
1365
1366         remote_entry_pa = part_sn2->remote_vars_part_pa +
1367             sn_partition_id * sizeof(struct xpc_vars_part_sn2);
1368
1369         remote_entry_cacheline_pa = (remote_entry_pa & ~(L1_CACHE_BYTES - 1));
1370
1371         pulled_entry = (struct xpc_vars_part_sn2 *)((u64)pulled_entry_cacheline
1372                                                     + (remote_entry_pa &
1373                                                     (L1_CACHE_BYTES - 1)));
1374
1375         ret = xpc_pull_remote_cachelines_sn2(part, pulled_entry_cacheline,
1376                                              remote_entry_cacheline_pa,
1377                                              L1_CACHE_BYTES);
1378         if (ret != xpSuccess) {
1379                 dev_dbg(xpc_chan, "failed to pull XPC vars_part from "
1380                         "partition %d, ret=%d\n", partid, ret);
1381                 return ret;
1382         }
1383
1384         /* see if they've been set up yet */
1385
1386         if (pulled_entry->magic != XPC_VP_MAGIC1_SN2 &&
1387             pulled_entry->magic != XPC_VP_MAGIC2_SN2) {
1388
1389                 if (pulled_entry->magic != 0) {
1390                         dev_dbg(xpc_chan, "partition %d's XPC vars_part for "
1391                                 "partition %d has bad magic value (=0x%llx)\n",
1392                                 partid, sn_partition_id, pulled_entry->magic);
1393                         return xpBadMagic;
1394                 }
1395
1396                 /* they've not been initialized yet */
1397                 return xpRetry;
1398         }
1399
1400         if (xpc_vars_part_sn2[partid].magic == XPC_VP_MAGIC1_SN2) {
1401
1402                 /* validate the variables */
1403
1404                 if (pulled_entry->GPs_pa == 0 ||
1405                     pulled_entry->openclose_args_pa == 0 ||
1406                     pulled_entry->chctl_amo_pa == 0) {
1407
1408                         dev_err(xpc_chan, "partition %d's XPC vars_part for "
1409                                 "partition %d are not valid\n", partid,
1410                                 sn_partition_id);
1411                         return xpInvalidAddress;
1412                 }
1413
1414                 /* the variables we imported look to be valid */
1415
1416                 part_sn2->remote_GPs_pa = pulled_entry->GPs_pa;
1417                 part_sn2->remote_openclose_args_pa =
1418                     pulled_entry->openclose_args_pa;
1419                 part_sn2->remote_chctl_amo_va =
1420                     (struct amo *)__va(pulled_entry->chctl_amo_pa);
1421                 part_sn2->notify_IRQ_nasid = pulled_entry->notify_IRQ_nasid;
1422                 part_sn2->notify_IRQ_phys_cpuid =
1423                     pulled_entry->notify_IRQ_phys_cpuid;
1424
1425                 if (part->nchannels > pulled_entry->nchannels)
1426                         part->nchannels = pulled_entry->nchannels;
1427
1428                 /* let the other side know that we've pulled their variables */
1429
1430                 xpc_vars_part_sn2[partid].magic = XPC_VP_MAGIC2_SN2;
1431         }
1432
1433         if (pulled_entry->magic == XPC_VP_MAGIC1_SN2)
1434                 return xpRetry;
1435
1436         return xpSuccess;
1437 }
1438
1439 /*
1440  * Establish first contact with the remote partititon. This involves pulling
1441  * the XPC per partition variables from the remote partition and waiting for
1442  * the remote partition to pull ours.
1443  */
1444 static enum xp_retval
1445 xpc_make_first_contact_sn2(struct xpc_partition *part)
1446 {
1447         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1448         enum xp_retval ret;
1449
1450         /*
1451          * Register the remote partition's amos with SAL so it can handle
1452          * and cleanup errors within that address range should the remote
1453          * partition go down. We don't unregister this range because it is
1454          * difficult to tell when outstanding writes to the remote partition
1455          * are finished and thus when it is safe to unregister. This should
1456          * not result in wasted space in the SAL xp_addr_region table because
1457          * we should get the same page for remote_amos_page_pa after module
1458          * reloads and system reboots.
1459          */
1460         if (sn_register_xp_addr_region(part_sn2->remote_amos_page_pa,
1461                                        PAGE_SIZE, 1) < 0) {
1462                 dev_warn(xpc_part, "xpc_activating(%d) failed to register "
1463                          "xp_addr region\n", XPC_PARTID(part));
1464
1465                 ret = xpPhysAddrRegFailed;
1466                 XPC_DEACTIVATE_PARTITION(part, ret);
1467                 return ret;
1468         }
1469
1470         /*
1471          * Send activate IRQ to get other side to activate if they've not
1472          * already begun to do so.
1473          */
1474         xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
1475                                   cnodeid_to_nasid(0),
1476                                   part_sn2->activate_IRQ_nasid,
1477                                   part_sn2->activate_IRQ_phys_cpuid);
1478
1479         while ((ret = xpc_pull_remote_vars_part_sn2(part)) != xpSuccess) {
1480                 if (ret != xpRetry) {
1481                         XPC_DEACTIVATE_PARTITION(part, ret);
1482                         return ret;
1483                 }
1484
1485                 dev_dbg(xpc_part, "waiting to make first contact with "
1486                         "partition %d\n", XPC_PARTID(part));
1487
1488                 /* wait a 1/4 of a second or so */
1489                 (void)msleep_interruptible(250);
1490
1491                 if (part->act_state == XPC_P_AS_DEACTIVATING)
1492                         return part->reason;
1493         }
1494
1495         return xpSuccess;
1496 }
1497
1498 /*
1499  * Get the chctl flags and pull the openclose args and/or remote GPs as needed.
1500  */
1501 static u64
1502 xpc_get_chctl_all_flags_sn2(struct xpc_partition *part)
1503 {
1504         struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1505         unsigned long irq_flags;
1506         union xpc_channel_ctl_flags chctl;
1507         enum xp_retval ret;
1508
1509         /*
1510          * See if there are any chctl flags to be handled.
1511          */
1512
1513         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1514         chctl = part->chctl;
1515         if (chctl.all_flags != 0)
1516                 part->chctl.all_flags = 0;
1517
1518         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1519
1520         if (xpc_any_openclose_chctl_flags_set(&chctl)) {
1521                 ret = xpc_pull_remote_cachelines_sn2(part, part->
1522                                                      remote_openclose_args,
1523                                                      part_sn2->
1524                                                      remote_openclose_args_pa,
1525                                                      XPC_OPENCLOSE_ARGS_SIZE);
1526                 if (ret != xpSuccess) {
1527                         XPC_DEACTIVATE_PARTITION(part, ret);
1528
1529                         dev_dbg(xpc_chan, "failed to pull openclose args from "
1530                                 "partition %d, ret=%d\n", XPC_PARTID(part),
1531                                 ret);
1532
1533                         /* don't bother processing chctl flags anymore */
1534                         chctl.all_flags = 0;
1535                 }
1536         }
1537
1538         if (xpc_any_msg_chctl_flags_set(&chctl)) {
1539                 ret = xpc_pull_remote_cachelines_sn2(part, part_sn2->remote_GPs,
1540                                                      part_sn2->remote_GPs_pa,
1541                                                      XPC_GP_SIZE);
1542                 if (ret != xpSuccess) {
1543                         XPC_DEACTIVATE_PARTITION(part, ret);
1544
1545                         dev_dbg(xpc_chan, "failed to pull GPs from partition "
1546                                 "%d, ret=%d\n", XPC_PARTID(part), ret);
1547
1548                         /* don't bother processing chctl flags anymore */
1549                         chctl.all_flags = 0;
1550                 }
1551         }
1552
1553         return chctl.all_flags;
1554 }
1555
1556 /*
1557  * Allocate the local message queue and the notify queue.
1558  */
1559 static enum xp_retval
1560 xpc_allocate_local_msgqueue_sn2(struct xpc_channel *ch)
1561 {
1562         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1563         unsigned long irq_flags;
1564         int nentries;
1565         size_t nbytes;
1566
1567         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1568
1569                 nbytes = nentries * ch->entry_size;
1570                 ch_sn2->local_msgqueue =
1571                     xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL,
1572                                                   &ch_sn2->local_msgqueue_base);
1573                 if (ch_sn2->local_msgqueue == NULL)
1574                         continue;
1575
1576                 nbytes = nentries * sizeof(struct xpc_notify_sn2);
1577                 ch_sn2->notify_queue = kzalloc(nbytes, GFP_KERNEL);
1578                 if (ch_sn2->notify_queue == NULL) {
1579                         kfree(ch_sn2->local_msgqueue_base);
1580                         ch_sn2->local_msgqueue = NULL;
1581                         continue;
1582                 }
1583
1584                 spin_lock_irqsave(&ch->lock, irq_flags);
1585                 if (nentries < ch->local_nentries) {
1586                         dev_dbg(xpc_chan, "nentries=%d local_nentries=%d, "
1587                                 "partid=%d, channel=%d\n", nentries,
1588                                 ch->local_nentries, ch->partid, ch->number);
1589
1590                         ch->local_nentries = nentries;
1591                 }
1592                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1593                 return xpSuccess;
1594         }
1595
1596         dev_dbg(xpc_chan, "can't get memory for local message queue and notify "
1597                 "queue, partid=%d, channel=%d\n", ch->partid, ch->number);
1598         return xpNoMemory;
1599 }
1600
1601 /*
1602  * Allocate the cached remote message queue.
1603  */
1604 static enum xp_retval
1605 xpc_allocate_remote_msgqueue_sn2(struct xpc_channel *ch)
1606 {
1607         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1608         unsigned long irq_flags;
1609         int nentries;
1610         size_t nbytes;
1611
1612         DBUG_ON(ch->remote_nentries <= 0);
1613
1614         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1615
1616                 nbytes = nentries * ch->entry_size;
1617                 ch_sn2->remote_msgqueue =
1618                     xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL, &ch_sn2->
1619                                                   remote_msgqueue_base);
1620                 if (ch_sn2->remote_msgqueue == NULL)
1621                         continue;
1622
1623                 spin_lock_irqsave(&ch->lock, irq_flags);
1624                 if (nentries < ch->remote_nentries) {
1625                         dev_dbg(xpc_chan, "nentries=%d remote_nentries=%d, "
1626                                 "partid=%d, channel=%d\n", nentries,
1627                                 ch->remote_nentries, ch->partid, ch->number);
1628
1629                         ch->remote_nentries = nentries;
1630                 }
1631                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1632                 return xpSuccess;
1633         }
1634
1635         dev_dbg(xpc_chan, "can't get memory for cached remote message queue, "
1636                 "partid=%d, channel=%d\n", ch->partid, ch->number);
1637         return xpNoMemory;
1638 }
1639
1640 /*
1641  * Allocate message queues and other stuff associated with a channel.
1642  *
1643  * Note: Assumes all of the channel sizes are filled in.
1644  */
1645 static enum xp_retval
1646 xpc_setup_msg_structures_sn2(struct xpc_channel *ch)
1647 {
1648         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1649         enum xp_retval ret;
1650
1651         DBUG_ON(ch->flags & XPC_C_SETUP);
1652
1653         ret = xpc_allocate_local_msgqueue_sn2(ch);
1654         if (ret == xpSuccess) {
1655
1656                 ret = xpc_allocate_remote_msgqueue_sn2(ch);
1657                 if (ret != xpSuccess) {
1658                         kfree(ch_sn2->local_msgqueue_base);
1659                         ch_sn2->local_msgqueue = NULL;
1660                         kfree(ch_sn2->notify_queue);
1661                         ch_sn2->notify_queue = NULL;
1662                 }
1663         }
1664         return ret;
1665 }
1666
1667 /*
1668  * Free up message queues and other stuff that were allocated for the specified
1669  * channel.
1670  */
1671 static void
1672 xpc_teardown_msg_structures_sn2(struct xpc_channel *ch)
1673 {
1674         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1675
1676         DBUG_ON(!spin_is_locked(&ch->lock));
1677
1678         ch_sn2->remote_msgqueue_pa = 0;
1679
1680         ch_sn2->local_GP->get = 0;
1681         ch_sn2->local_GP->put = 0;
1682         ch_sn2->remote_GP.get = 0;
1683         ch_sn2->remote_GP.put = 0;
1684         ch_sn2->w_local_GP.get = 0;
1685         ch_sn2->w_local_GP.put = 0;
1686         ch_sn2->w_remote_GP.get = 0;
1687         ch_sn2->w_remote_GP.put = 0;
1688         ch_sn2->next_msg_to_pull = 0;
1689
1690         if (ch->flags & XPC_C_SETUP) {
1691                 dev_dbg(xpc_chan, "ch->flags=0x%x, partid=%d, channel=%d\n",
1692                         ch->flags, ch->partid, ch->number);
1693
1694                 kfree(ch_sn2->local_msgqueue_base);
1695                 ch_sn2->local_msgqueue = NULL;
1696                 kfree(ch_sn2->remote_msgqueue_base);
1697                 ch_sn2->remote_msgqueue = NULL;
1698                 kfree(ch_sn2->notify_queue);
1699                 ch_sn2->notify_queue = NULL;
1700         }
1701 }
1702
1703 /*
1704  * Notify those who wanted to be notified upon delivery of their message.
1705  */
1706 static void
1707 xpc_notify_senders_sn2(struct xpc_channel *ch, enum xp_retval reason, s64 put)
1708 {
1709         struct xpc_notify_sn2 *notify;
1710         u8 notify_type;
1711         s64 get = ch->sn.sn2.w_remote_GP.get - 1;
1712
1713         while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
1714
1715                 notify = &ch->sn.sn2.notify_queue[get % ch->local_nentries];
1716
1717                 /*
1718                  * See if the notify entry indicates it was associated with
1719                  * a message who's sender wants to be notified. It is possible
1720                  * that it is, but someone else is doing or has done the
1721                  * notification.
1722                  */
1723                 notify_type = notify->type;
1724                 if (notify_type == 0 ||
1725                     cmpxchg(&notify->type, notify_type, 0) != notify_type) {
1726                         continue;
1727                 }
1728
1729                 DBUG_ON(notify_type != XPC_N_CALL);
1730
1731                 atomic_dec(&ch->n_to_notify);
1732
1733                 if (notify->func != NULL) {
1734                         dev_dbg(xpc_chan, "notify->func() called, notify=0x%p "
1735                                 "msg_number=%lld partid=%d channel=%d\n",
1736                                 (void *)notify, get, ch->partid, ch->number);
1737
1738                         notify->func(reason, ch->partid, ch->number,
1739                                      notify->key);
1740
1741                         dev_dbg(xpc_chan, "notify->func() returned, notify=0x%p"
1742                                 " msg_number=%lld partid=%d channel=%d\n",
1743                                 (void *)notify, get, ch->partid, ch->number);
1744                 }
1745         }
1746 }
1747
1748 static void
1749 xpc_notify_senders_of_disconnect_sn2(struct xpc_channel *ch)
1750 {
1751         xpc_notify_senders_sn2(ch, ch->reason, ch->sn.sn2.w_local_GP.put);
1752 }
1753
1754 /*
1755  * Clear some of the msg flags in the local message queue.
1756  */
1757 static inline void
1758 xpc_clear_local_msgqueue_flags_sn2(struct xpc_channel *ch)
1759 {
1760         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1761         struct xpc_msg_sn2 *msg;
1762         s64 get;
1763
1764         get = ch_sn2->w_remote_GP.get;
1765         do {
1766                 msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
1767                                              (get % ch->local_nentries) *
1768                                              ch->entry_size);
1769                 DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
1770                 msg->flags = 0;
1771         } while (++get < ch_sn2->remote_GP.get);
1772 }
1773
1774 /*
1775  * Clear some of the msg flags in the remote message queue.
1776  */
1777 static inline void
1778 xpc_clear_remote_msgqueue_flags_sn2(struct xpc_channel *ch)
1779 {
1780         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1781         struct xpc_msg_sn2 *msg;
1782         s64 put, remote_nentries = ch->remote_nentries;
1783
1784         /* flags are zeroed when the buffer is allocated */
1785         if (ch_sn2->remote_GP.put < remote_nentries)
1786                 return;
1787
1788         put = max(ch_sn2->w_remote_GP.put, remote_nentries);
1789         do {
1790                 msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
1791                                              (put % remote_nentries) *
1792                                              ch->entry_size);
1793                 DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
1794                 DBUG_ON(!(msg->flags & XPC_M_SN2_DONE));
1795                 DBUG_ON(msg->number != put - remote_nentries);
1796                 msg->flags = 0;
1797         } while (++put < ch_sn2->remote_GP.put);
1798 }
1799
1800 static int
1801 xpc_n_of_deliverable_payloads_sn2(struct xpc_channel *ch)
1802 {
1803         return ch->sn.sn2.w_remote_GP.put - ch->sn.sn2.w_local_GP.get;
1804 }
1805
1806 static void
1807 xpc_process_msg_chctl_flags_sn2(struct xpc_partition *part, int ch_number)
1808 {
1809         struct xpc_channel *ch = &part->channels[ch_number];
1810         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1811         int npayloads_sent;
1812
1813         ch_sn2->remote_GP = part->sn.sn2.remote_GPs[ch_number];
1814
1815         /* See what, if anything, has changed for each connected channel */
1816
1817         xpc_msgqueue_ref(ch);
1818
1819         if (ch_sn2->w_remote_GP.get == ch_sn2->remote_GP.get &&
1820             ch_sn2->w_remote_GP.put == ch_sn2->remote_GP.put) {
1821                 /* nothing changed since GPs were last pulled */
1822                 xpc_msgqueue_deref(ch);
1823                 return;
1824         }
1825
1826         if (!(ch->flags & XPC_C_CONNECTED)) {
1827                 xpc_msgqueue_deref(ch);
1828                 return;
1829         }
1830
1831         /*
1832          * First check to see if messages recently sent by us have been
1833          * received by the other side. (The remote GET value will have
1834          * changed since we last looked at it.)
1835          */
1836
1837         if (ch_sn2->w_remote_GP.get != ch_sn2->remote_GP.get) {
1838
1839                 /*
1840                  * We need to notify any senders that want to be notified
1841                  * that their sent messages have been received by their
1842                  * intended recipients. We need to do this before updating
1843                  * w_remote_GP.get so that we don't allocate the same message
1844                  * queue entries prematurely (see xpc_allocate_msg()).
1845                  */
1846                 if (atomic_read(&ch->n_to_notify) > 0) {
1847                         /*
1848                          * Notify senders that messages sent have been
1849                          * received and delivered by the other side.
1850                          */
1851                         xpc_notify_senders_sn2(ch, xpMsgDelivered,
1852                                                ch_sn2->remote_GP.get);
1853                 }
1854
1855                 /*
1856                  * Clear msg->flags in previously sent messages, so that
1857                  * they're ready for xpc_allocate_msg().
1858                  */
1859                 xpc_clear_local_msgqueue_flags_sn2(ch);
1860
1861                 ch_sn2->w_remote_GP.get = ch_sn2->remote_GP.get;
1862
1863                 dev_dbg(xpc_chan, "w_remote_GP.get changed to %lld, partid=%d, "
1864                         "channel=%d\n", ch_sn2->w_remote_GP.get, ch->partid,
1865                         ch->number);
1866
1867                 /*
1868                  * If anyone was waiting for message queue entries to become
1869                  * available, wake them up.
1870                  */
1871                 if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1872                         wake_up(&ch->msg_allocate_wq);
1873         }
1874
1875         /*
1876          * Now check for newly sent messages by the other side. (The remote
1877          * PUT value will have changed since we last looked at it.)
1878          */
1879
1880         if (ch_sn2->w_remote_GP.put != ch_sn2->remote_GP.put) {
1881                 /*
1882                  * Clear msg->flags in previously received messages, so that
1883                  * they're ready for xpc_get_deliverable_payload_sn2().
1884                  */
1885                 xpc_clear_remote_msgqueue_flags_sn2(ch);
1886
1887                 smp_wmb(); /* ensure flags have been cleared before bte_copy */
1888                 ch_sn2->w_remote_GP.put = ch_sn2->remote_GP.put;
1889
1890                 dev_dbg(xpc_chan, "w_remote_GP.put changed to %lld, partid=%d, "
1891                         "channel=%d\n", ch_sn2->w_remote_GP.put, ch->partid,
1892                         ch->number);
1893
1894                 npayloads_sent = xpc_n_of_deliverable_payloads_sn2(ch);
1895                 if (npayloads_sent > 0) {
1896                         dev_dbg(xpc_chan, "msgs waiting to be copied and "
1897                                 "delivered=%d, partid=%d, channel=%d\n",
1898                                 npayloads_sent, ch->partid, ch->number);
1899
1900                         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)
1901                                 xpc_activate_kthreads(ch, npayloads_sent);
1902                 }
1903         }
1904
1905         xpc_msgqueue_deref(ch);
1906 }
1907
1908 static struct xpc_msg_sn2 *
1909 xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
1910 {
1911         struct xpc_partition *part = &xpc_partitions[ch->partid];
1912         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1913         unsigned long remote_msg_pa;
1914         struct xpc_msg_sn2 *msg;
1915         u32 msg_index;
1916         u32 nmsgs;
1917         u64 msg_offset;
1918         enum xp_retval ret;
1919
1920         if (mutex_lock_interruptible(&ch_sn2->msg_to_pull_mutex) != 0) {
1921                 /* we were interrupted by a signal */
1922                 return NULL;
1923         }
1924
1925         while (get >= ch_sn2->next_msg_to_pull) {
1926
1927                 /* pull as many messages as are ready and able to be pulled */
1928
1929                 msg_index = ch_sn2->next_msg_to_pull % ch->remote_nentries;
1930
1931                 DBUG_ON(ch_sn2->next_msg_to_pull >= ch_sn2->w_remote_GP.put);
1932                 nmsgs = ch_sn2->w_remote_GP.put - ch_sn2->next_msg_to_pull;
1933                 if (msg_index + nmsgs > ch->remote_nentries) {
1934                         /* ignore the ones that wrap the msg queue for now */
1935                         nmsgs = ch->remote_nentries - msg_index;
1936                 }
1937
1938                 msg_offset = msg_index * ch->entry_size;
1939                 msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
1940                     msg_offset);
1941                 remote_msg_pa = ch_sn2->remote_msgqueue_pa + msg_offset;
1942
1943                 ret = xpc_pull_remote_cachelines_sn2(part, msg, remote_msg_pa,
1944                                                      nmsgs * ch->entry_size);
1945                 if (ret != xpSuccess) {
1946
1947                         dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
1948                                 " msg %lld from partition %d, channel=%d, "
1949                                 "ret=%d\n", nmsgs, ch_sn2->next_msg_to_pull,
1950                                 ch->partid, ch->number, ret);
1951
1952                         XPC_DEACTIVATE_PARTITION(part, ret);
1953
1954                         mutex_unlock(&ch_sn2->msg_to_pull_mutex);
1955                         return NULL;
1956                 }
1957
1958                 ch_sn2->next_msg_to_pull += nmsgs;
1959         }
1960
1961         mutex_unlock(&ch_sn2->msg_to_pull_mutex);
1962
1963         /* return the message we were looking for */
1964         msg_offset = (get % ch->remote_nentries) * ch->entry_size;
1965         msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue + msg_offset);
1966
1967         return msg;
1968 }
1969
1970 /*
1971  * Get the next deliverable message's payload.
1972  */
1973 static void *
1974 xpc_get_deliverable_payload_sn2(struct xpc_channel *ch)
1975 {
1976         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1977         struct xpc_msg_sn2 *msg;
1978         void *payload = NULL;
1979         s64 get;
1980
1981         do {
1982                 if (ch->flags & XPC_C_DISCONNECTING)
1983                         break;
1984
1985                 get = ch_sn2->w_local_GP.get;
1986                 smp_rmb();      /* guarantee that .get loads before .put */
1987                 if (get == ch_sn2->w_remote_GP.put)
1988                         break;
1989
1990                 /* There are messages waiting to be pulled and delivered.
1991                  * We need to try to secure one for ourselves. We'll do this
1992                  * by trying to increment w_local_GP.get and hope that no one
1993                  * else beats us to it. If they do, we'll we'll simply have
1994                  * to try again for the next one.
1995                  */
1996
1997                 if (cmpxchg(&ch_sn2->w_local_GP.get, get, get + 1) == get) {
1998                         /* we got the entry referenced by get */
1999
2000                         dev_dbg(xpc_chan, "w_local_GP.get changed to %lld, "
2001                                 "partid=%d, channel=%d\n", get + 1,
2002                                 ch->partid, ch->number);
2003
2004                         /* pull the message from the remote partition */
2005
2006                         msg = xpc_pull_remote_msg_sn2(ch, get);
2007
2008                         if (msg != NULL) {
2009                                 DBUG_ON(msg->number != get);
2010                                 DBUG_ON(msg->flags & XPC_M_SN2_DONE);
2011                                 DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
2012
2013                                 payload = &msg->payload;
2014                         }
2015                         break;
2016                 }
2017
2018         } while (1);
2019
2020         return payload;
2021 }
2022
2023 /*
2024  * Now we actually send the messages that are ready to be sent by advancing
2025  * the local message queue's Put value and then send a chctl msgrequest to the
2026  * recipient partition.
2027  */
2028 static void
2029 xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
2030 {
2031         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2032         struct xpc_msg_sn2 *msg;
2033         s64 put = initial_put + 1;
2034         int send_msgrequest = 0;
2035
2036         while (1) {
2037
2038                 while (1) {
2039                         if (put == ch_sn2->w_local_GP.put)
2040                                 break;
2041
2042                         msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
2043                                                      local_msgqueue + (put %
2044                                                      ch->local_nentries) *
2045                                                      ch->entry_size);
2046
2047                         if (!(msg->flags & XPC_M_SN2_READY))
2048                                 break;
2049
2050                         put++;
2051                 }
2052
2053                 if (put == initial_put) {
2054                         /* nothing's changed */
2055                         break;
2056                 }
2057
2058                 if (cmpxchg_rel(&ch_sn2->local_GP->put, initial_put, put) !=
2059                     initial_put) {
2060                         /* someone else beat us to it */
2061                         DBUG_ON(ch_sn2->local_GP->put < initial_put);
2062                         break;
2063                 }
2064
2065                 /* we just set the new value of local_GP->put */
2066
2067                 dev_dbg(xpc_chan, "local_GP->put changed to %lld, partid=%d, "
2068                         "channel=%d\n", put, ch->partid, ch->number);
2069
2070                 send_msgrequest = 1;
2071
2072                 /*
2073                  * We need to ensure that the message referenced by
2074                  * local_GP->put is not XPC_M_SN2_READY or that local_GP->put
2075                  * equals w_local_GP.put, so we'll go have a look.
2076                  */
2077                 initial_put = put;
2078         }
2079
2080         if (send_msgrequest)
2081                 xpc_send_chctl_msgrequest_sn2(ch);
2082 }
2083
2084 /*
2085  * Allocate an entry for a message from the message queue associated with the
2086  * specified channel.
2087  */
2088 static enum xp_retval
2089 xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
2090                      struct xpc_msg_sn2 **address_of_msg)
2091 {
2092         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2093         struct xpc_msg_sn2 *msg;
2094         enum xp_retval ret;
2095         s64 put;
2096
2097         /*
2098          * Get the next available message entry from the local message queue.
2099          * If none are available, we'll make sure that we grab the latest
2100          * GP values.
2101          */
2102         ret = xpTimeout;
2103
2104         while (1) {
2105
2106                 put = ch_sn2->w_local_GP.put;
2107                 smp_rmb();      /* guarantee that .put loads before .get */
2108                 if (put - ch_sn2->w_remote_GP.get < ch->local_nentries) {
2109
2110                         /* There are available message entries. We need to try
2111                          * to secure one for ourselves. We'll do this by trying
2112                          * to increment w_local_GP.put as long as someone else
2113                          * doesn't beat us to it. If they do, we'll have to
2114                          * try again.
2115                          */
2116                         if (cmpxchg(&ch_sn2->w_local_GP.put, put, put + 1) ==
2117                             put) {
2118                                 /* we got the entry referenced by put */
2119                                 break;
2120                         }
2121                         continue;       /* try again */
2122                 }
2123
2124                 /*
2125                  * There aren't any available msg entries at this time.
2126                  *
2127                  * In waiting for a message entry to become available,
2128                  * we set a timeout in case the other side is not sending
2129                  * completion interrupts. This lets us fake a notify IRQ
2130                  * that will cause the notify IRQ handler to fetch the latest
2131                  * GP values as if an interrupt was sent by the other side.
2132                  */
2133                 if (ret == xpTimeout)
2134                         xpc_send_chctl_local_msgrequest_sn2(ch);
2135
2136                 if (flags & XPC_NOWAIT)
2137                         return xpNoWait;
2138
2139                 ret = xpc_allocate_msg_wait(ch);
2140                 if (ret != xpInterrupted && ret != xpTimeout)
2141                         return ret;
2142         }
2143
2144         /* get the message's address and initialize it */
2145         msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
2146                                      (put % ch->local_nentries) *
2147                                      ch->entry_size);
2148
2149         DBUG_ON(msg->flags != 0);
2150         msg->number = put;
2151
2152         dev_dbg(xpc_chan, "w_local_GP.put changed to %lld; msg=0x%p, "
2153                 "msg_number=%lld, partid=%d, channel=%d\n", put + 1,
2154                 (void *)msg, msg->number, ch->partid, ch->number);
2155
2156         *address_of_msg = msg;
2157         return xpSuccess;
2158 }
2159
2160 /*
2161  * Common code that does the actual sending of the message by advancing the
2162  * local message queue's Put value and sends a chctl msgrequest to the
2163  * partition the message is being sent to.
2164  */
2165 static enum xp_retval
2166 xpc_send_payload_sn2(struct xpc_channel *ch, u32 flags, void *payload,
2167                      u16 payload_size, u8 notify_type, xpc_notify_func func,
2168                      void *key)
2169 {
2170         enum xp_retval ret = xpSuccess;
2171         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2172         struct xpc_msg_sn2 *msg = msg;
2173         struct xpc_notify_sn2 *notify = notify;
2174         s64 msg_number;
2175         s64 put;
2176
2177         DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
2178
2179         if (XPC_MSG_SIZE(payload_size) > ch->entry_size)
2180                 return xpPayloadTooBig;
2181
2182         xpc_msgqueue_ref(ch);
2183
2184         if (ch->flags & XPC_C_DISCONNECTING) {
2185                 ret = ch->reason;
2186                 goto out_1;
2187         }
2188         if (!(ch->flags & XPC_C_CONNECTED)) {
2189                 ret = xpNotConnected;
2190                 goto out_1;
2191         }
2192
2193         ret = xpc_allocate_msg_sn2(ch, flags, &msg);
2194         if (ret != xpSuccess)
2195                 goto out_1;
2196
2197         msg_number = msg->number;
2198
2199         if (notify_type != 0) {
2200                 /*
2201                  * Tell the remote side to send an ACK interrupt when the
2202                  * message has been delivered.
2203                  */
2204                 msg->flags |= XPC_M_SN2_INTERRUPT;
2205
2206                 atomic_inc(&ch->n_to_notify);
2207
2208                 notify = &ch_sn2->notify_queue[msg_number % ch->local_nentries];
2209                 notify->func = func;
2210                 notify->key = key;
2211                 notify->type = notify_type;
2212
2213                 /* ??? Is a mb() needed here? */
2214
2215                 if (ch->flags & XPC_C_DISCONNECTING) {
2216                         /*
2217                          * An error occurred between our last error check and
2218                          * this one. We will try to clear the type field from
2219                          * the notify entry. If we succeed then
2220                          * xpc_disconnect_channel() didn't already process
2221                          * the notify entry.
2222                          */
2223                         if (cmpxchg(&notify->type, notify_type, 0) ==
2224                             notify_type) {
2225                                 atomic_dec(&ch->n_to_notify);
2226                                 ret = ch->reason;
2227                         }
2228                         goto out_1;
2229                 }
2230         }
2231
2232         memcpy(&msg->payload, payload, payload_size);
2233
2234         msg->flags |= XPC_M_SN2_READY;
2235
2236         /*
2237          * The preceding store of msg->flags must occur before the following
2238          * load of local_GP->put.
2239          */
2240         smp_mb();
2241
2242         /* see if the message is next in line to be sent, if so send it */
2243
2244         put = ch_sn2->local_GP->put;
2245         if (put == msg_number)
2246                 xpc_send_msgs_sn2(ch, put);
2247
2248 out_1:
2249         xpc_msgqueue_deref(ch);
2250         return ret;
2251 }
2252
2253 /*
2254  * Now we actually acknowledge the messages that have been delivered and ack'd
2255  * by advancing the cached remote message queue's Get value and if requested
2256  * send a chctl msgrequest to the message sender's partition.
2257  *
2258  * If a message has XPC_M_SN2_INTERRUPT set, send an interrupt to the partition
2259  * that sent the message.
2260  */
2261 static void
2262 xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
2263 {
2264         struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2265         struct xpc_msg_sn2 *msg;
2266         s64 get = initial_get + 1;
2267         int send_msgrequest = 0;
2268
2269         while (1) {
2270
2271                 while (1) {
2272                         if (get == ch_sn2->w_local_GP.get)
2273                                 break;
2274
2275                         msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
2276                                                      remote_msgqueue + (get %
2277                                                      ch->remote_nentries) *
2278                                                      ch->entry_size);
2279
2280                         if (!(msg->flags & XPC_M_SN2_DONE))
2281                                 break;
2282
2283                         msg_flags |= msg->flags;
2284                         get++;
2285                 }
2286
2287                 if (get == initial_get) {
2288                         /* nothing's changed */
2289                         break;
2290                 }
2291
2292                 if (cmpxchg_rel(&ch_sn2->local_GP->get, initial_get, get) !=
2293                     initial_get) {
2294                         /* someone else beat us to it */
2295                         DBUG_ON(ch_sn2->local_GP->get <= initial_get);
2296                         break;
2297                 }
2298
2299                 /* we just set the new value of local_GP->get */
2300
2301                 dev_dbg(xpc_chan, "local_GP->get changed to %lld, partid=%d, "
2302                         "channel=%d\n", get, ch->partid, ch->number);
2303
2304                 send_msgrequest = (msg_flags & XPC_M_SN2_INTERRUPT);
2305
2306                 /*
2307                  * We need to ensure that the message referenced by
2308                  * local_GP->get is not XPC_M_SN2_DONE or that local_GP->get
2309                  * equals w_local_GP.get, so we'll go have a look.
2310                  */
2311                 initial_get = get;
2312         }
2313
2314         if (send_msgrequest)
2315                 xpc_send_chctl_msgrequest_sn2(ch);
2316 }
2317
2318 static void
2319 xpc_received_payload_sn2(struct xpc_channel *ch, void *payload)
2320 {
2321         struct xpc_msg_sn2 *msg;
2322         s64 msg_number;
2323         s64 get;
2324
2325         msg = container_of(payload, struct xpc_msg_sn2, payload);
2326         msg_number = msg->number;
2327
2328         dev_dbg(xpc_chan, "msg=0x%p, msg_number=%lld, partid=%d, channel=%d\n",
2329                 (void *)msg, msg_number, ch->partid, ch->number);
2330
2331         DBUG_ON((((u64)msg - (u64)ch->sn.sn2.remote_msgqueue) / ch->entry_size) !=
2332                 msg_number % ch->remote_nentries);
2333         DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
2334         DBUG_ON(msg->flags & XPC_M_SN2_DONE);
2335
2336         msg->flags |= XPC_M_SN2_DONE;
2337
2338         /*
2339          * The preceding store of msg->flags must occur before the following
2340          * load of local_GP->get.
2341          */
2342         smp_mb();
2343
2344         /*
2345          * See if this message is next in line to be acknowledged as having
2346          * been delivered.
2347          */
2348         get = ch->sn.sn2.local_GP->get;
2349         if (get == msg_number)
2350                 xpc_acknowledge_msgs_sn2(ch, get, msg->flags);
2351 }
2352
2353 static struct xpc_arch_operations xpc_arch_ops_sn2 = {
2354         .setup_partitions = xpc_setup_partitions_sn2,
2355         .teardown_partitions = xpc_teardown_partitions_sn2,
2356         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_sn2,
2357         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_sn2,
2358         .setup_rsvd_page = xpc_setup_rsvd_page_sn2,
2359
2360         .allow_hb = xpc_allow_hb_sn2,
2361         .disallow_hb = xpc_disallow_hb_sn2,
2362         .disallow_all_hbs = xpc_disallow_all_hbs_sn2,
2363         .increment_heartbeat = xpc_increment_heartbeat_sn2,
2364         .offline_heartbeat = xpc_offline_heartbeat_sn2,
2365         .online_heartbeat = xpc_online_heartbeat_sn2,
2366         .heartbeat_init = xpc_heartbeat_init_sn2,
2367         .heartbeat_exit = xpc_heartbeat_exit_sn2,
2368         .get_remote_heartbeat = xpc_get_remote_heartbeat_sn2,
2369
2370         .request_partition_activation =
2371                 xpc_request_partition_activation_sn2,
2372         .request_partition_reactivation =
2373                 xpc_request_partition_reactivation_sn2,
2374         .request_partition_deactivation =
2375                 xpc_request_partition_deactivation_sn2,
2376         .cancel_partition_deactivation_request =
2377                 xpc_cancel_partition_deactivation_request_sn2,
2378
2379         .setup_ch_structures = xpc_setup_ch_structures_sn2,
2380         .teardown_ch_structures = xpc_teardown_ch_structures_sn2,
2381
2382         .make_first_contact = xpc_make_first_contact_sn2,
2383
2384         .get_chctl_all_flags = xpc_get_chctl_all_flags_sn2,
2385         .send_chctl_closerequest = xpc_send_chctl_closerequest_sn2,
2386         .send_chctl_closereply = xpc_send_chctl_closereply_sn2,
2387         .send_chctl_openrequest = xpc_send_chctl_openrequest_sn2,
2388         .send_chctl_openreply = xpc_send_chctl_openreply_sn2,
2389         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_sn2,
2390         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_sn2,
2391
2392         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_sn2,
2393
2394         .setup_msg_structures = xpc_setup_msg_structures_sn2,
2395         .teardown_msg_structures = xpc_teardown_msg_structures_sn2,
2396
2397         .indicate_partition_engaged = xpc_indicate_partition_engaged_sn2,
2398         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_sn2,
2399         .partition_engaged = xpc_partition_engaged_sn2,
2400         .any_partition_engaged = xpc_any_partition_engaged_sn2,
2401         .assume_partition_disengaged = xpc_assume_partition_disengaged_sn2,
2402
2403         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_sn2,
2404         .send_payload = xpc_send_payload_sn2,
2405         .get_deliverable_payload = xpc_get_deliverable_payload_sn2,
2406         .received_payload = xpc_received_payload_sn2,
2407         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_sn2,
2408 };
2409
2410 int
2411 xpc_init_sn2(void)
2412 {
2413         int ret;
2414         size_t buf_size;
2415
2416         xpc_arch_ops = xpc_arch_ops_sn2;
2417
2418         if (offsetof(struct xpc_msg_sn2, payload) > XPC_MSG_HDR_MAX_SIZE) {
2419                 dev_err(xpc_part, "header portion of struct xpc_msg_sn2 is "
2420                         "larger than %d\n", XPC_MSG_HDR_MAX_SIZE);
2421                 return -E2BIG;
2422         }
2423
2424         buf_size = max(XPC_RP_VARS_SIZE,
2425                        XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES_SN2);
2426         xpc_remote_copy_buffer_sn2 = xpc_kmalloc_cacheline_aligned(buf_size,
2427                                                                    GFP_KERNEL,
2428                                               &xpc_remote_copy_buffer_base_sn2);
2429         if (xpc_remote_copy_buffer_sn2 == NULL) {
2430                 dev_err(xpc_part, "can't get memory for remote copy buffer\n");
2431                 return -ENOMEM;
2432         }
2433
2434         /* open up protections for IPI and [potentially] amo operations */
2435         xpc_allow_IPI_ops_sn2();
2436         xpc_allow_amo_ops_shub_wars_1_1_sn2();
2437
2438         /*
2439          * This is safe to do before the xpc_hb_checker thread has started
2440          * because the handler releases a wait queue.  If an interrupt is
2441          * received before the thread is waiting, it will not go to sleep,
2442          * but rather immediately process the interrupt.
2443          */
2444         ret = request_irq(SGI_XPC_ACTIVATE, xpc_handle_activate_IRQ_sn2, 0,
2445                           "xpc hb", NULL);
2446         if (ret != 0) {
2447                 dev_err(xpc_part, "can't register ACTIVATE IRQ handler, "
2448                         "errno=%d\n", -ret);
2449                 xpc_disallow_IPI_ops_sn2();
2450                 kfree(xpc_remote_copy_buffer_base_sn2);
2451         }
2452         return ret;
2453 }
2454
2455 void
2456 xpc_exit_sn2(void)
2457 {
2458         free_irq(SGI_XPC_ACTIVATE, NULL);
2459         xpc_disallow_IPI_ops_sn2();
2460         kfree(xpc_remote_copy_buffer_base_sn2);
2461 }