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
6 * Copyright (c) 2008-2009 Silicon Graphics, Inc. All Rights Reserved.
10 * Cross Partition Communication (XPC) sn2-based functions.
12 * Architecture specific implementation of common functions.
16 #include <linux/delay.h>
17 #include <asm/uncached.h>
18 #include <asm/sn/mspec.h>
19 #include <asm/sn/sn_sal.h>
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.
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)
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.
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)
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).
56 static void *xpc_remote_copy_buffer_base_sn2
;
57 static char *xpc_remote_copy_buffer_sn2
;
59 static struct xpc_vars_sn2
*xpc_vars_sn2
;
60 static struct xpc_vars_part_sn2
*xpc_vars_part_sn2
;
63 xpc_setup_partitions_sn_sn2(void)
65 /* nothing needs to be done */
70 xpc_teardown_partitions_sn_sn2(void)
72 /* nothing needs to be done */
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
;
83 * Change protections to allow IPI operations.
86 xpc_allow_IPI_ops_sn2(void)
91 /* !!! The following should get moved into SAL. */
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
));
102 for_each_online_node(node
) {
103 nasid
= cnodeid_to_nasid(node
);
104 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
, SH2_IPI_ACCESS0
),
106 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
, SH2_IPI_ACCESS1
),
108 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
, SH2_IPI_ACCESS2
),
110 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
, SH2_IPI_ACCESS3
),
114 xpc_sh1_IPI_access_sn2
=
115 (u64
)HUB_L((u64
*)LOCAL_MMR_ADDR(SH1_IPI_ACCESS
));
117 for_each_online_node(node
) {
118 nasid
= cnodeid_to_nasid(node
);
119 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
, SH1_IPI_ACCESS
),
126 * Restrict protections to disallow IPI operations.
129 xpc_disallow_IPI_ops_sn2(void)
134 /* !!! The following should get moved into SAL. */
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
);
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
);
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).
164 xpc_receive_IRQ_amo_sn2(struct amo
*amo
)
166 return FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_CLEAR
);
169 static enum xp_retval
170 xpc_send_IRQ_sn2(struct amo
*amo
, u64 flag
, int nasid
, int phys_cpuid
,
174 unsigned long irq_flags
;
176 local_irq_save(irq_flags
);
178 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
, flag
);
179 sn_send_IPI_phys(nasid
, phys_cpuid
, vector
, 0);
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.
187 ret
= xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->variable
),
188 xp_nofault_PIOR_target
));
190 local_irq_restore(irq_flags
);
192 return (ret
== 0) ? xpSuccess
: xpPioReadError
;
196 xpc_init_IRQ_amo_sn2(int index
)
198 struct amo
*amo
= xpc_vars_sn2
->amos_page
+ index
;
200 (void)xpc_receive_IRQ_amo_sn2(amo
); /* clear amo variable */
205 * Functions associated with SGI_XPC_ACTIVATE IRQ.
209 * Notify the heartbeat check thread that an activate IRQ has been received.
212 xpc_handle_activate_IRQ_sn2(int irq
, void *dev_id
)
214 unsigned long irq_flags
;
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
);
220 wake_up_interruptible(&xpc_activate_IRQ_wq
);
225 * Flag the appropriate amo variable and send an IRQ to the specified node.
228 xpc_send_activate_IRQ_sn2(unsigned long amos_page_pa
, int from_nasid
,
229 int to_nasid
, int to_phys_cpuid
)
231 struct amo
*amos
= (struct amo
*)__va(amos_page_pa
+
232 (XPC_ACTIVATE_IRQ_AMOS_SN2
*
233 sizeof(struct amo
)));
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
);
241 xpc_send_local_activate_IRQ_sn2(int from_nasid
)
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
)));
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));
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
);
256 wake_up_interruptible(&xpc_activate_IRQ_wq
);
260 * Functions associated with SGI_XPC_NOTIFY IRQ.
264 * Check to see if any chctl flags were sent from the specified partition.
267 xpc_check_for_sent_chctl_flags_sn2(struct xpc_partition
*part
)
269 union xpc_channel_ctl_flags chctl
;
270 unsigned long irq_flags
;
272 chctl
.all_flags
= xpc_receive_IRQ_amo_sn2(part
->sn
.sn2
.
274 if (chctl
.all_flags
== 0)
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
);
281 dev_dbg(xpc_chan
, "received notify IRQ from partid=%d, chctl.all_flags="
282 "0x%lx\n", XPC_PARTID(part
), chctl
.all_flags
);
284 xpc_wakeup_channel_mgr(part
);
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.
294 * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IRQs sent by XPC
295 * running on other partitions.
297 * Noteworthy Arguments:
299 * irq - Interrupt ReQuest number. NOT USED.
301 * dev_id - partid of IRQ's potential sender.
304 xpc_handle_notify_IRQ_sn2(int irq
, void *dev_id
)
306 short partid
= (short)(u64
)dev_id
;
307 struct xpc_partition
*part
= &xpc_partitions
[partid
];
309 DBUG_ON(partid
< 0 || partid
>= XP_MAX_NPARTITIONS_SN2
);
311 if (xpc_part_ref(part
)) {
312 xpc_check_for_sent_chctl_flags_sn2(part
);
314 xpc_part_deref(part
);
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
325 xpc_check_for_dropped_notify_IRQ_sn2(struct xpc_partition
*part
)
327 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
329 if (xpc_part_ref(part
)) {
330 xpc_check_for_sent_chctl_flags_sn2(part
);
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
);
340 * Send a notify IRQ to the remote partition that is associated with the
344 xpc_send_notify_IRQ_sn2(struct xpc_channel
*ch
, u8 chctl_flag
,
345 char *chctl_flag_string
, unsigned long *irq_flags
)
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 };
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
,
356 part_sn2
->notify_IRQ_nasid
,
357 part_sn2
->notify_IRQ_phys_cpuid
,
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
);
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)
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().
380 xpc_send_local_notify_IRQ_sn2(struct xpc_channel
*ch
, u8 chctl_flag
,
381 char *chctl_flag_string
)
383 struct xpc_partition
*part
= &xpc_partitions
[ch
->partid
];
384 union xpc_channel_ctl_flags chctl
= { 0 };
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
);
393 #define XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(_ch, _ipi_f) \
394 xpc_send_local_notify_IRQ_sn2(_ch, _ipi_f, #_ipi_f)
397 xpc_send_chctl_closerequest_sn2(struct xpc_channel
*ch
,
398 unsigned long *irq_flags
)
400 struct xpc_openclose_args
*args
= ch
->sn
.sn2
.local_openclose_args
;
402 args
->reason
= ch
->reason
;
403 XPC_SEND_NOTIFY_IRQ_SN2(ch
, XPC_CHCTL_CLOSEREQUEST
, irq_flags
);
407 xpc_send_chctl_closereply_sn2(struct xpc_channel
*ch
, unsigned long *irq_flags
)
409 XPC_SEND_NOTIFY_IRQ_SN2(ch
, XPC_CHCTL_CLOSEREPLY
, irq_flags
);
413 xpc_send_chctl_openrequest_sn2(struct xpc_channel
*ch
, unsigned long *irq_flags
)
415 struct xpc_openclose_args
*args
= ch
->sn
.sn2
.local_openclose_args
;
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
);
423 xpc_send_chctl_openreply_sn2(struct xpc_channel
*ch
, unsigned long *irq_flags
)
425 struct xpc_openclose_args
*args
= ch
->sn
.sn2
.local_openclose_args
;
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
);
434 xpc_send_chctl_msgrequest_sn2(struct xpc_channel
*ch
)
436 XPC_SEND_NOTIFY_IRQ_SN2(ch
, XPC_CHCTL_MSGREQUEST
, NULL
);
440 xpc_send_chctl_local_msgrequest_sn2(struct xpc_channel
*ch
)
442 XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(ch
, XPC_CHCTL_MSGREQUEST
);
445 static enum xp_retval
446 xpc_save_remote_msgqueue_pa_sn2(struct xpc_channel
*ch
,
447 unsigned long msgqueue_pa
)
449 ch
->sn
.sn2
.remote_msgqueue_pa
= msgqueue_pa
;
454 * This next set of functions are used to keep track of when a partition is
455 * potentially engaged in accessing memory belonging to another partition.
459 xpc_indicate_partition_engaged_sn2(struct xpc_partition
*part
)
461 unsigned long irq_flags
;
462 struct amo
*amo
= (struct amo
*)__va(part
->sn
.sn2
.remote_amos_page_pa
+
463 (XPC_ENGAGED_PARTITIONS_AMO_SN2
*
464 sizeof(struct amo
)));
466 local_irq_save(irq_flags
);
468 /* set bit corresponding to our partid in remote partition's amo */
469 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
,
470 BIT(sn_partition_id
));
473 * We must always use the nofault function regardless of whether we
474 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
475 * didn't, we'd never know that the other partition is down and would
476 * keep sending IRQs and amos to it until the heartbeat times out.
478 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
480 xp_nofault_PIOR_target
));
482 local_irq_restore(irq_flags
);
486 xpc_indicate_partition_disengaged_sn2(struct xpc_partition
*part
)
488 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
489 unsigned long irq_flags
;
490 struct amo
*amo
= (struct amo
*)__va(part_sn2
->remote_amos_page_pa
+
491 (XPC_ENGAGED_PARTITIONS_AMO_SN2
*
492 sizeof(struct amo
)));
494 local_irq_save(irq_flags
);
496 /* clear bit corresponding to our partid in remote partition's amo */
497 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
498 ~BIT(sn_partition_id
));
501 * We must always use the nofault function regardless of whether we
502 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
503 * didn't, we'd never know that the other partition is down and would
504 * keep sending IRQs and amos to it until the heartbeat times out.
506 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
508 xp_nofault_PIOR_target
));
510 local_irq_restore(irq_flags
);
513 * Send activate IRQ to get other side to see that we've cleared our
514 * bit in their engaged partitions amo.
516 xpc_send_activate_IRQ_sn2(part_sn2
->remote_amos_page_pa
,
518 part_sn2
->activate_IRQ_nasid
,
519 part_sn2
->activate_IRQ_phys_cpuid
);
523 xpc_assume_partition_disengaged_sn2(short partid
)
525 struct amo
*amo
= xpc_vars_sn2
->amos_page
+
526 XPC_ENGAGED_PARTITIONS_AMO_SN2
;
528 /* clear bit(s) based on partid mask in our partition's amo */
529 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
534 xpc_partition_engaged_sn2(short partid
)
536 struct amo
*amo
= xpc_vars_sn2
->amos_page
+
537 XPC_ENGAGED_PARTITIONS_AMO_SN2
;
539 /* our partition's amo variable ANDed with partid mask */
540 return (FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_LOAD
) &
545 xpc_any_partition_engaged_sn2(void)
547 struct amo
*amo
= xpc_vars_sn2
->amos_page
+
548 XPC_ENGAGED_PARTITIONS_AMO_SN2
;
550 /* our partition's amo variable */
551 return FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_LOAD
) != 0;
554 /* original protection values for each node */
555 static u64 xpc_prot_vec_sn2
[MAX_NUMNODES
];
558 * Change protections to allow amo operations on non-Shub 1.1 systems.
560 static enum xp_retval
561 xpc_allow_amo_ops_sn2(struct amo
*amos_page
)
563 enum xp_retval ret
= xpSuccess
;
566 * On SHUB 1.1, we cannot call sn_change_memprotect() since the BIST
567 * collides with memory operations. On those systems we call
568 * xpc_allow_amo_ops_shub_wars_1_1_sn2() instead.
570 if (!enable_shub_wars_1_1())
571 ret
= xp_expand_memprotect(ia64_tpa((u64
)amos_page
), PAGE_SIZE
);
577 * Change protections to allow amo operations on Shub 1.1 systems.
580 xpc_allow_amo_ops_shub_wars_1_1_sn2(void)
585 if (!enable_shub_wars_1_1())
588 for_each_online_node(node
) {
589 nasid
= cnodeid_to_nasid(node
);
590 /* save current protection values */
591 xpc_prot_vec_sn2
[node
] =
592 (u64
)HUB_L((u64
*)GLOBAL_MMR_ADDR(nasid
,
593 SH1_MD_DQLP_MMR_DIR_PRIVEC0
));
594 /* open up everything */
595 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
,
596 SH1_MD_DQLP_MMR_DIR_PRIVEC0
),
598 HUB_S((u64
*)GLOBAL_MMR_ADDR(nasid
,
599 SH1_MD_DQRP_MMR_DIR_PRIVEC0
),
604 static enum xp_retval
605 xpc_get_partition_rsvd_page_pa_sn2(void *buf
, u64
*cookie
, unsigned long *rp_pa
,
611 status
= sn_partition_reserved_page_pa((u64
)buf
, cookie
, rp_pa
, len
);
612 if (status
== SALRET_OK
)
614 else if (status
== SALRET_MORE_PASSES
)
615 ret
= xpNeedMoreInfo
;
624 xpc_setup_rsvd_page_sn_sn2(struct xpc_rsvd_page
*rp
)
626 struct amo
*amos_page
;
630 xpc_vars_sn2
= XPC_RP_VARS(rp
);
632 rp
->sn
.sn2
.vars_pa
= xp_pa(xpc_vars_sn2
);
634 /* vars_part array follows immediately after vars */
635 xpc_vars_part_sn2
= (struct xpc_vars_part_sn2
*)((u8
*)XPC_RP_VARS(rp
) +
639 * Before clearing xpc_vars_sn2, see if a page of amos had been
640 * previously allocated. If not we'll need to allocate one and set
641 * permissions so that cross-partition amos are allowed.
643 * The allocated amo page needs MCA reporting to remain disabled after
644 * XPC has unloaded. To make this work, we keep a copy of the pointer
645 * to this page (i.e., amos_page) in the struct xpc_vars_sn2 structure,
646 * which is pointed to by the reserved page, and re-use that saved copy
647 * on subsequent loads of XPC. This amo page is never freed, and its
648 * memory protections are never restricted.
650 amos_page
= xpc_vars_sn2
->amos_page
;
651 if (amos_page
== NULL
) {
652 amos_page
= (struct amo
*)TO_AMO(uncached_alloc_page(0, 1));
653 if (amos_page
== NULL
) {
654 dev_err(xpc_part
, "can't allocate page of amos\n");
659 * Open up amo-R/W to cpu. This is done on Shub 1.1 systems
660 * when xpc_allow_amo_ops_shub_wars_1_1_sn2() is called.
662 ret
= xpc_allow_amo_ops_sn2(amos_page
);
663 if (ret
!= xpSuccess
) {
664 dev_err(xpc_part
, "can't allow amo operations\n");
665 uncached_free_page(__IA64_UNCACHED_OFFSET
|
666 TO_PHYS((u64
)amos_page
), 1);
671 /* clear xpc_vars_sn2 */
672 memset(xpc_vars_sn2
, 0, sizeof(struct xpc_vars_sn2
));
674 xpc_vars_sn2
->version
= XPC_V_VERSION
;
675 xpc_vars_sn2
->activate_IRQ_nasid
= cpuid_to_nasid(0);
676 xpc_vars_sn2
->activate_IRQ_phys_cpuid
= cpu_physical_id(0);
677 xpc_vars_sn2
->vars_part_pa
= xp_pa(xpc_vars_part_sn2
);
678 xpc_vars_sn2
->amos_page_pa
= ia64_tpa((u64
)amos_page
);
679 xpc_vars_sn2
->amos_page
= amos_page
; /* save for next load of XPC */
681 /* clear xpc_vars_part_sn2 */
682 memset((u64
*)xpc_vars_part_sn2
, 0, sizeof(struct xpc_vars_part_sn2
) *
683 XP_MAX_NPARTITIONS_SN2
);
685 /* initialize the activate IRQ related amo variables */
686 for (i
= 0; i
< xpc_nasid_mask_nlongs
; i
++)
687 (void)xpc_init_IRQ_amo_sn2(XPC_ACTIVATE_IRQ_AMOS_SN2
+ i
);
689 /* initialize the engaged remote partitions related amo variables */
690 (void)xpc_init_IRQ_amo_sn2(XPC_ENGAGED_PARTITIONS_AMO_SN2
);
691 (void)xpc_init_IRQ_amo_sn2(XPC_DEACTIVATE_REQUEST_AMO_SN2
);
697 xpc_hb_allowed_sn2(short partid
, void *heartbeating_to_mask
)
699 return test_bit(partid
, heartbeating_to_mask
);
703 xpc_allow_hb_sn2(short partid
)
705 DBUG_ON(xpc_vars_sn2
== NULL
);
706 set_bit(partid
, xpc_vars_sn2
->heartbeating_to_mask
);
710 xpc_disallow_hb_sn2(short partid
)
712 DBUG_ON(xpc_vars_sn2
== NULL
);
713 clear_bit(partid
, xpc_vars_sn2
->heartbeating_to_mask
);
717 xpc_disallow_all_hbs_sn2(void)
719 DBUG_ON(xpc_vars_sn2
== NULL
);
720 bitmap_zero(xpc_vars_sn2
->heartbeating_to_mask
, xp_max_npartitions
);
724 xpc_increment_heartbeat_sn2(void)
726 xpc_vars_sn2
->heartbeat
++;
730 xpc_offline_heartbeat_sn2(void)
732 xpc_increment_heartbeat_sn2();
733 xpc_vars_sn2
->heartbeat_offline
= 1;
737 xpc_online_heartbeat_sn2(void)
739 xpc_increment_heartbeat_sn2();
740 xpc_vars_sn2
->heartbeat_offline
= 0;
744 xpc_heartbeat_init_sn2(void)
746 DBUG_ON(xpc_vars_sn2
== NULL
);
748 bitmap_zero(xpc_vars_sn2
->heartbeating_to_mask
, XP_MAX_NPARTITIONS_SN2
);
749 xpc_online_heartbeat_sn2();
753 xpc_heartbeat_exit_sn2(void)
755 xpc_offline_heartbeat_sn2();
758 static enum xp_retval
759 xpc_get_remote_heartbeat_sn2(struct xpc_partition
*part
)
761 struct xpc_vars_sn2
*remote_vars
;
764 remote_vars
= (struct xpc_vars_sn2
*)xpc_remote_copy_buffer_sn2
;
766 /* pull the remote vars structure that contains the heartbeat */
767 ret
= xp_remote_memcpy(xp_pa(remote_vars
),
768 part
->sn
.sn2
.remote_vars_pa
,
770 if (ret
!= xpSuccess
)
773 dev_dbg(xpc_part
, "partid=%d, heartbeat=%ld, last_heartbeat=%ld, "
774 "heartbeat_offline=%ld, HB_mask[0]=0x%lx\n", XPC_PARTID(part
),
775 remote_vars
->heartbeat
, part
->last_heartbeat
,
776 remote_vars
->heartbeat_offline
,
777 remote_vars
->heartbeating_to_mask
[0]);
779 if ((remote_vars
->heartbeat
== part
->last_heartbeat
&&
780 !remote_vars
->heartbeat_offline
) ||
781 !xpc_hb_allowed_sn2(sn_partition_id
,
782 remote_vars
->heartbeating_to_mask
)) {
785 part
->last_heartbeat
= remote_vars
->heartbeat
;
792 * Get a copy of the remote partition's XPC variables from the reserved page.
794 * remote_vars points to a buffer that is cacheline aligned for BTE copies and
795 * assumed to be of size XPC_RP_VARS_SIZE.
797 static enum xp_retval
798 xpc_get_remote_vars_sn2(unsigned long remote_vars_pa
,
799 struct xpc_vars_sn2
*remote_vars
)
803 if (remote_vars_pa
== 0)
806 /* pull over the cross partition variables */
807 ret
= xp_remote_memcpy(xp_pa(remote_vars
), remote_vars_pa
,
809 if (ret
!= xpSuccess
)
812 if (XPC_VERSION_MAJOR(remote_vars
->version
) !=
813 XPC_VERSION_MAJOR(XPC_V_VERSION
)) {
821 xpc_request_partition_activation_sn2(struct xpc_rsvd_page
*remote_rp
,
822 unsigned long remote_rp_pa
, int nasid
)
824 xpc_send_local_activate_IRQ_sn2(nasid
);
828 xpc_request_partition_reactivation_sn2(struct xpc_partition
*part
)
830 xpc_send_local_activate_IRQ_sn2(part
->sn
.sn2
.activate_IRQ_nasid
);
834 xpc_request_partition_deactivation_sn2(struct xpc_partition
*part
)
836 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
837 unsigned long irq_flags
;
838 struct amo
*amo
= (struct amo
*)__va(part_sn2
->remote_amos_page_pa
+
839 (XPC_DEACTIVATE_REQUEST_AMO_SN2
*
840 sizeof(struct amo
)));
842 local_irq_save(irq_flags
);
844 /* set bit corresponding to our partid in remote partition's amo */
845 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
,
846 BIT(sn_partition_id
));
849 * We must always use the nofault function regardless of whether we
850 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
851 * didn't, we'd never know that the other partition is down and would
852 * keep sending IRQs and amos to it until the heartbeat times out.
854 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
856 xp_nofault_PIOR_target
));
858 local_irq_restore(irq_flags
);
861 * Send activate IRQ to get other side to see that we've set our
862 * bit in their deactivate request amo.
864 xpc_send_activate_IRQ_sn2(part_sn2
->remote_amos_page_pa
,
866 part_sn2
->activate_IRQ_nasid
,
867 part_sn2
->activate_IRQ_phys_cpuid
);
871 xpc_cancel_partition_deactivation_request_sn2(struct xpc_partition
*part
)
873 unsigned long irq_flags
;
874 struct amo
*amo
= (struct amo
*)__va(part
->sn
.sn2
.remote_amos_page_pa
+
875 (XPC_DEACTIVATE_REQUEST_AMO_SN2
*
876 sizeof(struct amo
)));
878 local_irq_save(irq_flags
);
880 /* clear bit corresponding to our partid in remote partition's amo */
881 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
882 ~BIT(sn_partition_id
));
885 * We must always use the nofault function regardless of whether we
886 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
887 * didn't, we'd never know that the other partition is down and would
888 * keep sending IRQs and amos to it until the heartbeat times out.
890 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
892 xp_nofault_PIOR_target
));
894 local_irq_restore(irq_flags
);
898 xpc_partition_deactivation_requested_sn2(short partid
)
900 struct amo
*amo
= xpc_vars_sn2
->amos_page
+
901 XPC_DEACTIVATE_REQUEST_AMO_SN2
;
903 /* our partition's amo variable ANDed with partid mask */
904 return (FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_LOAD
) &
909 * Update the remote partition's info.
912 xpc_update_partition_info_sn2(struct xpc_partition
*part
, u8 remote_rp_version
,
913 unsigned long *remote_rp_ts_jiffies
,
914 unsigned long remote_rp_pa
,
915 unsigned long remote_vars_pa
,
916 struct xpc_vars_sn2
*remote_vars
)
918 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
920 part
->remote_rp_version
= remote_rp_version
;
921 dev_dbg(xpc_part
, " remote_rp_version = 0x%016x\n",
922 part
->remote_rp_version
);
924 part
->remote_rp_ts_jiffies
= *remote_rp_ts_jiffies
;
925 dev_dbg(xpc_part
, " remote_rp_ts_jiffies = 0x%016lx\n",
926 part
->remote_rp_ts_jiffies
);
928 part
->remote_rp_pa
= remote_rp_pa
;
929 dev_dbg(xpc_part
, " remote_rp_pa = 0x%016lx\n", part
->remote_rp_pa
);
931 part_sn2
->remote_vars_pa
= remote_vars_pa
;
932 dev_dbg(xpc_part
, " remote_vars_pa = 0x%016lx\n",
933 part_sn2
->remote_vars_pa
);
935 part
->last_heartbeat
= remote_vars
->heartbeat
- 1;
936 dev_dbg(xpc_part
, " last_heartbeat = 0x%016lx\n",
937 part
->last_heartbeat
);
939 part_sn2
->remote_vars_part_pa
= remote_vars
->vars_part_pa
;
940 dev_dbg(xpc_part
, " remote_vars_part_pa = 0x%016lx\n",
941 part_sn2
->remote_vars_part_pa
);
943 part_sn2
->activate_IRQ_nasid
= remote_vars
->activate_IRQ_nasid
;
944 dev_dbg(xpc_part
, " activate_IRQ_nasid = 0x%x\n",
945 part_sn2
->activate_IRQ_nasid
);
947 part_sn2
->activate_IRQ_phys_cpuid
=
948 remote_vars
->activate_IRQ_phys_cpuid
;
949 dev_dbg(xpc_part
, " activate_IRQ_phys_cpuid = 0x%x\n",
950 part_sn2
->activate_IRQ_phys_cpuid
);
952 part_sn2
->remote_amos_page_pa
= remote_vars
->amos_page_pa
;
953 dev_dbg(xpc_part
, " remote_amos_page_pa = 0x%lx\n",
954 part_sn2
->remote_amos_page_pa
);
956 part_sn2
->remote_vars_version
= remote_vars
->version
;
957 dev_dbg(xpc_part
, " remote_vars_version = 0x%x\n",
958 part_sn2
->remote_vars_version
);
962 * Prior code has determined the nasid which generated a activate IRQ.
963 * Inspect that nasid to determine if its partition needs to be activated
966 * A partition is considered "awaiting activation" if our partition
967 * flags indicate it is not active and it has a heartbeat. A
968 * partition is considered "awaiting deactivation" if our partition
969 * flags indicate it is active but it has no heartbeat or it is not
970 * sending its heartbeat to us.
972 * To determine the heartbeat, the remote nasid must have a properly
973 * initialized reserved page.
976 xpc_identify_activate_IRQ_req_sn2(int nasid
)
978 struct xpc_rsvd_page
*remote_rp
;
979 struct xpc_vars_sn2
*remote_vars
;
980 unsigned long remote_rp_pa
;
981 unsigned long remote_vars_pa
;
982 int remote_rp_version
;
984 unsigned long remote_rp_ts_jiffies
= 0;
986 struct xpc_partition
*part
;
987 struct xpc_partition_sn2
*part_sn2
;
990 /* pull over the reserved page structure */
992 remote_rp
= (struct xpc_rsvd_page
*)xpc_remote_copy_buffer_sn2
;
994 ret
= xpc_get_remote_rp(nasid
, NULL
, remote_rp
, &remote_rp_pa
);
995 if (ret
!= xpSuccess
) {
996 dev_warn(xpc_part
, "unable to get reserved page from nasid %d, "
997 "which sent interrupt, reason=%d\n", nasid
, ret
);
1001 remote_vars_pa
= remote_rp
->sn
.sn2
.vars_pa
;
1002 remote_rp_version
= remote_rp
->version
;
1003 remote_rp_ts_jiffies
= remote_rp
->ts_jiffies
;
1005 partid
= remote_rp
->SAL_partid
;
1006 part
= &xpc_partitions
[partid
];
1007 part_sn2
= &part
->sn
.sn2
;
1009 /* pull over the cross partition variables */
1011 remote_vars
= (struct xpc_vars_sn2
*)xpc_remote_copy_buffer_sn2
;
1013 ret
= xpc_get_remote_vars_sn2(remote_vars_pa
, remote_vars
);
1014 if (ret
!= xpSuccess
) {
1015 dev_warn(xpc_part
, "unable to get XPC variables from nasid %d, "
1016 "which sent interrupt, reason=%d\n", nasid
, ret
);
1018 XPC_DEACTIVATE_PARTITION(part
, ret
);
1022 part
->activate_IRQ_rcvd
++;
1024 dev_dbg(xpc_part
, "partid for nasid %d is %d; IRQs = %d; HB = "
1025 "%ld:0x%lx\n", (int)nasid
, (int)partid
, part
->activate_IRQ_rcvd
,
1026 remote_vars
->heartbeat
, remote_vars
->heartbeating_to_mask
[0]);
1028 if (xpc_partition_disengaged(part
) &&
1029 part
->act_state
== XPC_P_AS_INACTIVE
) {
1031 xpc_update_partition_info_sn2(part
, remote_rp_version
,
1032 &remote_rp_ts_jiffies
,
1033 remote_rp_pa
, remote_vars_pa
,
1036 if (xpc_partition_deactivation_requested_sn2(partid
)) {
1038 * Other side is waiting on us to deactivate even though
1044 xpc_activate_partition(part
);
1048 DBUG_ON(part
->remote_rp_version
== 0);
1049 DBUG_ON(part_sn2
->remote_vars_version
== 0);
1051 if (remote_rp_ts_jiffies
!= part
->remote_rp_ts_jiffies
) {
1053 /* the other side rebooted */
1055 DBUG_ON(xpc_partition_engaged_sn2(partid
));
1056 DBUG_ON(xpc_partition_deactivation_requested_sn2(partid
));
1058 xpc_update_partition_info_sn2(part
, remote_rp_version
,
1059 &remote_rp_ts_jiffies
,
1060 remote_rp_pa
, remote_vars_pa
,
1065 if (part
->disengage_timeout
> 0 && !xpc_partition_disengaged(part
)) {
1066 /* still waiting on other side to disengage from us */
1071 XPC_DEACTIVATE_PARTITION(part
, xpReactivating
);
1072 else if (xpc_partition_deactivation_requested_sn2(partid
))
1073 XPC_DEACTIVATE_PARTITION(part
, xpOtherGoingDown
);
1077 * Loop through the activation amo variables and process any bits
1078 * which are set. Each bit indicates a nasid sending a partition
1079 * activation or deactivation request.
1081 * Return #of IRQs detected.
1084 xpc_identify_activate_IRQ_sender_sn2(void)
1088 unsigned long nasid_mask_long
;
1089 u64 nasid
; /* remote nasid */
1090 int n_IRQs_detected
= 0;
1091 struct amo
*act_amos
;
1093 act_amos
= xpc_vars_sn2
->amos_page
+ XPC_ACTIVATE_IRQ_AMOS_SN2
;
1095 /* scan through activate amo variables looking for non-zero entries */
1096 for (l
= 0; l
< xpc_nasid_mask_nlongs
; l
++) {
1101 nasid_mask_long
= xpc_receive_IRQ_amo_sn2(&act_amos
[l
]);
1103 b
= find_first_bit(&nasid_mask_long
, BITS_PER_LONG
);
1104 if (b
>= BITS_PER_LONG
) {
1105 /* no IRQs from nasids in this amo variable */
1109 dev_dbg(xpc_part
, "amo[%d] gave back 0x%lx\n", l
,
1113 * If this nasid has been added to the machine since
1114 * our partition was reset, this will retain the
1115 * remote nasid in our reserved pages machine mask.
1116 * This is used in the event of module reload.
1118 xpc_mach_nasids
[l
] |= nasid_mask_long
;
1120 /* locate the nasid(s) which sent interrupts */
1124 nasid
= (l
* BITS_PER_LONG
+ b
) * 2;
1125 dev_dbg(xpc_part
, "interrupt from nasid %ld\n", nasid
);
1126 xpc_identify_activate_IRQ_req_sn2(nasid
);
1128 b
= find_next_bit(&nasid_mask_long
, BITS_PER_LONG
,
1130 } while (b
< BITS_PER_LONG
);
1132 return n_IRQs_detected
;
1136 xpc_process_activate_IRQ_rcvd_sn2(void)
1138 unsigned long irq_flags
;
1139 int n_IRQs_expected
;
1140 int n_IRQs_detected
;
1142 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock
, irq_flags
);
1143 n_IRQs_expected
= xpc_activate_IRQ_rcvd
;
1144 xpc_activate_IRQ_rcvd
= 0;
1145 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock
, irq_flags
);
1147 n_IRQs_detected
= xpc_identify_activate_IRQ_sender_sn2();
1148 if (n_IRQs_detected
< n_IRQs_expected
) {
1149 /* retry once to help avoid missing amo */
1150 (void)xpc_identify_activate_IRQ_sender_sn2();
1155 * Setup the channel structures that are sn2 specific.
1157 static enum xp_retval
1158 xpc_setup_ch_structures_sn_sn2(struct xpc_partition
*part
)
1160 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
1161 struct xpc_channel_sn2
*ch_sn2
;
1162 enum xp_retval retval
;
1166 struct timer_list
*timer
;
1167 short partid
= XPC_PARTID(part
);
1169 /* allocate all the required GET/PUT values */
1171 part_sn2
->local_GPs
=
1172 xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE
, GFP_KERNEL
,
1173 &part_sn2
->local_GPs_base
);
1174 if (part_sn2
->local_GPs
== NULL
) {
1175 dev_err(xpc_chan
, "can't get memory for local get/put "
1180 part_sn2
->remote_GPs
=
1181 xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE
, GFP_KERNEL
,
1182 &part_sn2
->remote_GPs_base
);
1183 if (part_sn2
->remote_GPs
== NULL
) {
1184 dev_err(xpc_chan
, "can't get memory for remote get/put "
1186 retval
= xpNoMemory
;
1190 part_sn2
->remote_GPs_pa
= 0;
1192 /* allocate all the required open and close args */
1194 part_sn2
->local_openclose_args
=
1195 xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE
,
1196 GFP_KERNEL
, &part_sn2
->
1197 local_openclose_args_base
);
1198 if (part_sn2
->local_openclose_args
== NULL
) {
1199 dev_err(xpc_chan
, "can't get memory for local connect args\n");
1200 retval
= xpNoMemory
;
1204 part_sn2
->remote_openclose_args_pa
= 0;
1206 part_sn2
->local_chctl_amo_va
= xpc_init_IRQ_amo_sn2(partid
);
1208 part_sn2
->notify_IRQ_nasid
= 0;
1209 part_sn2
->notify_IRQ_phys_cpuid
= 0;
1210 part_sn2
->remote_chctl_amo_va
= NULL
;
1212 sprintf(part_sn2
->notify_IRQ_owner
, "xpc%02d", partid
);
1213 ret
= request_irq(SGI_XPC_NOTIFY
, xpc_handle_notify_IRQ_sn2
,
1214 IRQF_SHARED
, part_sn2
->notify_IRQ_owner
,
1215 (void *)(u64
)partid
);
1217 dev_err(xpc_chan
, "can't register NOTIFY IRQ handler, "
1218 "errno=%d\n", -ret
);
1219 retval
= xpLackOfResources
;
1223 /* Setup a timer to check for dropped notify IRQs */
1224 timer
= &part_sn2
->dropped_notify_IRQ_timer
;
1227 (void (*)(unsigned long))xpc_check_for_dropped_notify_IRQ_sn2
;
1228 timer
->data
= (unsigned long)part
;
1229 timer
->expires
= jiffies
+ XPC_DROPPED_NOTIFY_IRQ_WAIT_INTERVAL
;
1232 for (ch_number
= 0; ch_number
< part
->nchannels
; ch_number
++) {
1233 ch_sn2
= &part
->channels
[ch_number
].sn
.sn2
;
1235 ch_sn2
->local_GP
= &part_sn2
->local_GPs
[ch_number
];
1236 ch_sn2
->local_openclose_args
=
1237 &part_sn2
->local_openclose_args
[ch_number
];
1239 mutex_init(&ch_sn2
->msg_to_pull_mutex
);
1243 * Setup the per partition specific variables required by the
1244 * remote partition to establish channel connections with us.
1246 * The setting of the magic # indicates that these per partition
1247 * specific variables are ready to be used.
1249 xpc_vars_part_sn2
[partid
].GPs_pa
= xp_pa(part_sn2
->local_GPs
);
1250 xpc_vars_part_sn2
[partid
].openclose_args_pa
=
1251 xp_pa(part_sn2
->local_openclose_args
);
1252 xpc_vars_part_sn2
[partid
].chctl_amo_pa
=
1253 xp_pa(part_sn2
->local_chctl_amo_va
);
1254 cpuid
= raw_smp_processor_id(); /* any CPU in this partition will do */
1255 xpc_vars_part_sn2
[partid
].notify_IRQ_nasid
= cpuid_to_nasid(cpuid
);
1256 xpc_vars_part_sn2
[partid
].notify_IRQ_phys_cpuid
=
1257 cpu_physical_id(cpuid
);
1258 xpc_vars_part_sn2
[partid
].nchannels
= part
->nchannels
;
1259 xpc_vars_part_sn2
[partid
].magic
= XPC_VP_MAGIC1_SN2
;
1263 /* setup of ch structures failed */
1265 kfree(part_sn2
->local_openclose_args_base
);
1266 part_sn2
->local_openclose_args
= NULL
;
1268 kfree(part_sn2
->remote_GPs_base
);
1269 part_sn2
->remote_GPs
= NULL
;
1271 kfree(part_sn2
->local_GPs_base
);
1272 part_sn2
->local_GPs
= NULL
;
1277 * Teardown the channel structures that are sn2 specific.
1280 xpc_teardown_ch_structures_sn_sn2(struct xpc_partition
*part
)
1282 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
1283 short partid
= XPC_PARTID(part
);
1286 * Indicate that the variables specific to the remote partition are no
1287 * longer available for its use.
1289 xpc_vars_part_sn2
[partid
].magic
= 0;
1291 /* in case we've still got outstanding timers registered... */
1292 del_timer_sync(&part_sn2
->dropped_notify_IRQ_timer
);
1293 free_irq(SGI_XPC_NOTIFY
, (void *)(u64
)partid
);
1295 kfree(part_sn2
->local_openclose_args_base
);
1296 part_sn2
->local_openclose_args
= NULL
;
1297 kfree(part_sn2
->remote_GPs_base
);
1298 part_sn2
->remote_GPs
= NULL
;
1299 kfree(part_sn2
->local_GPs_base
);
1300 part_sn2
->local_GPs
= NULL
;
1301 part_sn2
->local_chctl_amo_va
= NULL
;
1305 * Create a wrapper that hides the underlying mechanism for pulling a cacheline
1306 * (or multiple cachelines) from a remote partition.
1308 * src_pa must be a cacheline aligned physical address on the remote partition.
1309 * dst must be a cacheline aligned virtual address on this partition.
1310 * cnt must be cacheline sized
1312 /* ??? Replace this function by call to xp_remote_memcpy() or bte_copy()? */
1313 static enum xp_retval
1314 xpc_pull_remote_cachelines_sn2(struct xpc_partition
*part
, void *dst
,
1315 const unsigned long src_pa
, size_t cnt
)
1319 DBUG_ON(src_pa
!= L1_CACHE_ALIGN(src_pa
));
1320 DBUG_ON((unsigned long)dst
!= L1_CACHE_ALIGN((unsigned long)dst
));
1321 DBUG_ON(cnt
!= L1_CACHE_ALIGN(cnt
));
1323 if (part
->act_state
== XPC_P_AS_DEACTIVATING
)
1324 return part
->reason
;
1326 ret
= xp_remote_memcpy(xp_pa(dst
), src_pa
, cnt
);
1327 if (ret
!= xpSuccess
) {
1328 dev_dbg(xpc_chan
, "xp_remote_memcpy() from partition %d failed,"
1329 " ret=%d\n", XPC_PARTID(part
), ret
);
1335 * Pull the remote per partition specific variables from the specified
1338 static enum xp_retval
1339 xpc_pull_remote_vars_part_sn2(struct xpc_partition
*part
)
1341 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
1342 u8 buffer
[L1_CACHE_BYTES
* 2];
1343 struct xpc_vars_part_sn2
*pulled_entry_cacheline
=
1344 (struct xpc_vars_part_sn2
*)L1_CACHE_ALIGN((u64
)buffer
);
1345 struct xpc_vars_part_sn2
*pulled_entry
;
1346 unsigned long remote_entry_cacheline_pa
;
1347 unsigned long remote_entry_pa
;
1348 short partid
= XPC_PARTID(part
);
1351 /* pull the cacheline that contains the variables we're interested in */
1353 DBUG_ON(part_sn2
->remote_vars_part_pa
!=
1354 L1_CACHE_ALIGN(part_sn2
->remote_vars_part_pa
));
1355 DBUG_ON(sizeof(struct xpc_vars_part_sn2
) != L1_CACHE_BYTES
/ 2);
1357 remote_entry_pa
= part_sn2
->remote_vars_part_pa
+
1358 sn_partition_id
* sizeof(struct xpc_vars_part_sn2
);
1360 remote_entry_cacheline_pa
= (remote_entry_pa
& ~(L1_CACHE_BYTES
- 1));
1362 pulled_entry
= (struct xpc_vars_part_sn2
*)((u64
)pulled_entry_cacheline
1363 + (remote_entry_pa
&
1364 (L1_CACHE_BYTES
- 1)));
1366 ret
= xpc_pull_remote_cachelines_sn2(part
, pulled_entry_cacheline
,
1367 remote_entry_cacheline_pa
,
1369 if (ret
!= xpSuccess
) {
1370 dev_dbg(xpc_chan
, "failed to pull XPC vars_part from "
1371 "partition %d, ret=%d\n", partid
, ret
);
1375 /* see if they've been set up yet */
1377 if (pulled_entry
->magic
!= XPC_VP_MAGIC1_SN2
&&
1378 pulled_entry
->magic
!= XPC_VP_MAGIC2_SN2
) {
1380 if (pulled_entry
->magic
!= 0) {
1381 dev_dbg(xpc_chan
, "partition %d's XPC vars_part for "
1382 "partition %d has bad magic value (=0x%lx)\n",
1383 partid
, sn_partition_id
, pulled_entry
->magic
);
1387 /* they've not been initialized yet */
1391 if (xpc_vars_part_sn2
[partid
].magic
== XPC_VP_MAGIC1_SN2
) {
1393 /* validate the variables */
1395 if (pulled_entry
->GPs_pa
== 0 ||
1396 pulled_entry
->openclose_args_pa
== 0 ||
1397 pulled_entry
->chctl_amo_pa
== 0) {
1399 dev_err(xpc_chan
, "partition %d's XPC vars_part for "
1400 "partition %d are not valid\n", partid
,
1402 return xpInvalidAddress
;
1405 /* the variables we imported look to be valid */
1407 part_sn2
->remote_GPs_pa
= pulled_entry
->GPs_pa
;
1408 part_sn2
->remote_openclose_args_pa
=
1409 pulled_entry
->openclose_args_pa
;
1410 part_sn2
->remote_chctl_amo_va
=
1411 (struct amo
*)__va(pulled_entry
->chctl_amo_pa
);
1412 part_sn2
->notify_IRQ_nasid
= pulled_entry
->notify_IRQ_nasid
;
1413 part_sn2
->notify_IRQ_phys_cpuid
=
1414 pulled_entry
->notify_IRQ_phys_cpuid
;
1416 if (part
->nchannels
> pulled_entry
->nchannels
)
1417 part
->nchannels
= pulled_entry
->nchannels
;
1419 /* let the other side know that we've pulled their variables */
1421 xpc_vars_part_sn2
[partid
].magic
= XPC_VP_MAGIC2_SN2
;
1424 if (pulled_entry
->magic
== XPC_VP_MAGIC1_SN2
)
1431 * Establish first contact with the remote partititon. This involves pulling
1432 * the XPC per partition variables from the remote partition and waiting for
1433 * the remote partition to pull ours.
1435 static enum xp_retval
1436 xpc_make_first_contact_sn2(struct xpc_partition
*part
)
1438 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
1442 * Register the remote partition's amos with SAL so it can handle
1443 * and cleanup errors within that address range should the remote
1444 * partition go down. We don't unregister this range because it is
1445 * difficult to tell when outstanding writes to the remote partition
1446 * are finished and thus when it is safe to unregister. This should
1447 * not result in wasted space in the SAL xp_addr_region table because
1448 * we should get the same page for remote_amos_page_pa after module
1449 * reloads and system reboots.
1451 if (sn_register_xp_addr_region(part_sn2
->remote_amos_page_pa
,
1452 PAGE_SIZE
, 1) < 0) {
1453 dev_warn(xpc_part
, "xpc_activating(%d) failed to register "
1454 "xp_addr region\n", XPC_PARTID(part
));
1456 ret
= xpPhysAddrRegFailed
;
1457 XPC_DEACTIVATE_PARTITION(part
, ret
);
1462 * Send activate IRQ to get other side to activate if they've not
1463 * already begun to do so.
1465 xpc_send_activate_IRQ_sn2(part_sn2
->remote_amos_page_pa
,
1466 cnodeid_to_nasid(0),
1467 part_sn2
->activate_IRQ_nasid
,
1468 part_sn2
->activate_IRQ_phys_cpuid
);
1470 while ((ret
= xpc_pull_remote_vars_part_sn2(part
)) != xpSuccess
) {
1471 if (ret
!= xpRetry
) {
1472 XPC_DEACTIVATE_PARTITION(part
, ret
);
1476 dev_dbg(xpc_part
, "waiting to make first contact with "
1477 "partition %d\n", XPC_PARTID(part
));
1479 /* wait a 1/4 of a second or so */
1480 (void)msleep_interruptible(250);
1482 if (part
->act_state
== XPC_P_AS_DEACTIVATING
)
1483 return part
->reason
;
1490 * Get the chctl flags and pull the openclose args and/or remote GPs as needed.
1493 xpc_get_chctl_all_flags_sn2(struct xpc_partition
*part
)
1495 struct xpc_partition_sn2
*part_sn2
= &part
->sn
.sn2
;
1496 unsigned long irq_flags
;
1497 union xpc_channel_ctl_flags chctl
;
1501 * See if there are any chctl flags to be handled.
1504 spin_lock_irqsave(&part
->chctl_lock
, irq_flags
);
1505 chctl
= part
->chctl
;
1506 if (chctl
.all_flags
!= 0)
1507 part
->chctl
.all_flags
= 0;
1509 spin_unlock_irqrestore(&part
->chctl_lock
, irq_flags
);
1511 if (xpc_any_openclose_chctl_flags_set(&chctl
)) {
1512 ret
= xpc_pull_remote_cachelines_sn2(part
, part
->
1513 remote_openclose_args
,
1515 remote_openclose_args_pa
,
1516 XPC_OPENCLOSE_ARGS_SIZE
);
1517 if (ret
!= xpSuccess
) {
1518 XPC_DEACTIVATE_PARTITION(part
, ret
);
1520 dev_dbg(xpc_chan
, "failed to pull openclose args from "
1521 "partition %d, ret=%d\n", XPC_PARTID(part
),
1524 /* don't bother processing chctl flags anymore */
1525 chctl
.all_flags
= 0;
1529 if (xpc_any_msg_chctl_flags_set(&chctl
)) {
1530 ret
= xpc_pull_remote_cachelines_sn2(part
, part_sn2
->remote_GPs
,
1531 part_sn2
->remote_GPs_pa
,
1533 if (ret
!= xpSuccess
) {
1534 XPC_DEACTIVATE_PARTITION(part
, ret
);
1536 dev_dbg(xpc_chan
, "failed to pull GPs from partition "
1537 "%d, ret=%d\n", XPC_PARTID(part
), ret
);
1539 /* don't bother processing chctl flags anymore */
1540 chctl
.all_flags
= 0;
1544 return chctl
.all_flags
;
1548 * Allocate the local message queue and the notify queue.
1550 static enum xp_retval
1551 xpc_allocate_local_msgqueue_sn2(struct xpc_channel
*ch
)
1553 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1554 unsigned long irq_flags
;
1558 for (nentries
= ch
->local_nentries
; nentries
> 0; nentries
--) {
1560 nbytes
= nentries
* ch
->entry_size
;
1561 ch_sn2
->local_msgqueue
=
1562 xpc_kzalloc_cacheline_aligned(nbytes
, GFP_KERNEL
,
1563 &ch_sn2
->local_msgqueue_base
);
1564 if (ch_sn2
->local_msgqueue
== NULL
)
1567 nbytes
= nentries
* sizeof(struct xpc_notify_sn2
);
1568 ch_sn2
->notify_queue
= kzalloc(nbytes
, GFP_KERNEL
);
1569 if (ch_sn2
->notify_queue
== NULL
) {
1570 kfree(ch_sn2
->local_msgqueue_base
);
1571 ch_sn2
->local_msgqueue
= NULL
;
1575 spin_lock_irqsave(&ch
->lock
, irq_flags
);
1576 if (nentries
< ch
->local_nentries
) {
1577 dev_dbg(xpc_chan
, "nentries=%d local_nentries=%d, "
1578 "partid=%d, channel=%d\n", nentries
,
1579 ch
->local_nentries
, ch
->partid
, ch
->number
);
1581 ch
->local_nentries
= nentries
;
1583 spin_unlock_irqrestore(&ch
->lock
, irq_flags
);
1587 dev_dbg(xpc_chan
, "can't get memory for local message queue and notify "
1588 "queue, partid=%d, channel=%d\n", ch
->partid
, ch
->number
);
1593 * Allocate the cached remote message queue.
1595 static enum xp_retval
1596 xpc_allocate_remote_msgqueue_sn2(struct xpc_channel
*ch
)
1598 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1599 unsigned long irq_flags
;
1603 DBUG_ON(ch
->remote_nentries
<= 0);
1605 for (nentries
= ch
->remote_nentries
; nentries
> 0; nentries
--) {
1607 nbytes
= nentries
* ch
->entry_size
;
1608 ch_sn2
->remote_msgqueue
=
1609 xpc_kzalloc_cacheline_aligned(nbytes
, GFP_KERNEL
, &ch_sn2
->
1610 remote_msgqueue_base
);
1611 if (ch_sn2
->remote_msgqueue
== NULL
)
1614 spin_lock_irqsave(&ch
->lock
, irq_flags
);
1615 if (nentries
< ch
->remote_nentries
) {
1616 dev_dbg(xpc_chan
, "nentries=%d remote_nentries=%d, "
1617 "partid=%d, channel=%d\n", nentries
,
1618 ch
->remote_nentries
, ch
->partid
, ch
->number
);
1620 ch
->remote_nentries
= nentries
;
1622 spin_unlock_irqrestore(&ch
->lock
, irq_flags
);
1626 dev_dbg(xpc_chan
, "can't get memory for cached remote message queue, "
1627 "partid=%d, channel=%d\n", ch
->partid
, ch
->number
);
1632 * Allocate message queues and other stuff associated with a channel.
1634 * Note: Assumes all of the channel sizes are filled in.
1636 static enum xp_retval
1637 xpc_setup_msg_structures_sn2(struct xpc_channel
*ch
)
1639 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1642 DBUG_ON(ch
->flags
& XPC_C_SETUP
);
1644 ret
= xpc_allocate_local_msgqueue_sn2(ch
);
1645 if (ret
== xpSuccess
) {
1647 ret
= xpc_allocate_remote_msgqueue_sn2(ch
);
1648 if (ret
!= xpSuccess
) {
1649 kfree(ch_sn2
->local_msgqueue_base
);
1650 ch_sn2
->local_msgqueue
= NULL
;
1651 kfree(ch_sn2
->notify_queue
);
1652 ch_sn2
->notify_queue
= NULL
;
1659 * Free up message queues and other stuff that were allocated for the specified
1663 xpc_teardown_msg_structures_sn2(struct xpc_channel
*ch
)
1665 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1667 DBUG_ON(!spin_is_locked(&ch
->lock
));
1669 ch_sn2
->remote_msgqueue_pa
= 0;
1671 ch_sn2
->local_GP
->get
= 0;
1672 ch_sn2
->local_GP
->put
= 0;
1673 ch_sn2
->remote_GP
.get
= 0;
1674 ch_sn2
->remote_GP
.put
= 0;
1675 ch_sn2
->w_local_GP
.get
= 0;
1676 ch_sn2
->w_local_GP
.put
= 0;
1677 ch_sn2
->w_remote_GP
.get
= 0;
1678 ch_sn2
->w_remote_GP
.put
= 0;
1679 ch_sn2
->next_msg_to_pull
= 0;
1681 if (ch
->flags
& XPC_C_SETUP
) {
1682 dev_dbg(xpc_chan
, "ch->flags=0x%x, partid=%d, channel=%d\n",
1683 ch
->flags
, ch
->partid
, ch
->number
);
1685 kfree(ch_sn2
->local_msgqueue_base
);
1686 ch_sn2
->local_msgqueue
= NULL
;
1687 kfree(ch_sn2
->remote_msgqueue_base
);
1688 ch_sn2
->remote_msgqueue
= NULL
;
1689 kfree(ch_sn2
->notify_queue
);
1690 ch_sn2
->notify_queue
= NULL
;
1695 * Notify those who wanted to be notified upon delivery of their message.
1698 xpc_notify_senders_sn2(struct xpc_channel
*ch
, enum xp_retval reason
, s64 put
)
1700 struct xpc_notify_sn2
*notify
;
1702 s64 get
= ch
->sn
.sn2
.w_remote_GP
.get
- 1;
1704 while (++get
< put
&& atomic_read(&ch
->n_to_notify
) > 0) {
1706 notify
= &ch
->sn
.sn2
.notify_queue
[get
% ch
->local_nentries
];
1709 * See if the notify entry indicates it was associated with
1710 * a message who's sender wants to be notified. It is possible
1711 * that it is, but someone else is doing or has done the
1714 notify_type
= notify
->type
;
1715 if (notify_type
== 0 ||
1716 cmpxchg(¬ify
->type
, notify_type
, 0) != notify_type
) {
1720 DBUG_ON(notify_type
!= XPC_N_CALL
);
1722 atomic_dec(&ch
->n_to_notify
);
1724 if (notify
->func
!= NULL
) {
1725 dev_dbg(xpc_chan
, "notify->func() called, notify=0x%p "
1726 "msg_number=%ld partid=%d channel=%d\n",
1727 (void *)notify
, get
, ch
->partid
, ch
->number
);
1729 notify
->func(reason
, ch
->partid
, ch
->number
,
1732 dev_dbg(xpc_chan
, "notify->func() returned, notify=0x%p"
1733 " msg_number=%ld partid=%d channel=%d\n",
1734 (void *)notify
, get
, ch
->partid
, ch
->number
);
1740 xpc_notify_senders_of_disconnect_sn2(struct xpc_channel
*ch
)
1742 xpc_notify_senders_sn2(ch
, ch
->reason
, ch
->sn
.sn2
.w_local_GP
.put
);
1746 * Clear some of the msg flags in the local message queue.
1749 xpc_clear_local_msgqueue_flags_sn2(struct xpc_channel
*ch
)
1751 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1752 struct xpc_msg_sn2
*msg
;
1755 get
= ch_sn2
->w_remote_GP
.get
;
1757 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->local_msgqueue
+
1758 (get
% ch
->local_nentries
) *
1760 DBUG_ON(!(msg
->flags
& XPC_M_SN2_READY
));
1762 } while (++get
< ch_sn2
->remote_GP
.get
);
1766 * Clear some of the msg flags in the remote message queue.
1769 xpc_clear_remote_msgqueue_flags_sn2(struct xpc_channel
*ch
)
1771 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1772 struct xpc_msg_sn2
*msg
;
1773 s64 put
, remote_nentries
= ch
->remote_nentries
;
1775 /* flags are zeroed when the buffer is allocated */
1776 if (ch_sn2
->remote_GP
.put
< remote_nentries
)
1779 put
= max(ch_sn2
->w_remote_GP
.put
, remote_nentries
);
1781 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->remote_msgqueue
+
1782 (put
% remote_nentries
) *
1784 DBUG_ON(!(msg
->flags
& XPC_M_SN2_READY
));
1785 DBUG_ON(!(msg
->flags
& XPC_M_SN2_DONE
));
1786 DBUG_ON(msg
->number
!= put
- remote_nentries
);
1788 } while (++put
< ch_sn2
->remote_GP
.put
);
1792 xpc_n_of_deliverable_payloads_sn2(struct xpc_channel
*ch
)
1794 return ch
->sn
.sn2
.w_remote_GP
.put
- ch
->sn
.sn2
.w_local_GP
.get
;
1798 xpc_process_msg_chctl_flags_sn2(struct xpc_partition
*part
, int ch_number
)
1800 struct xpc_channel
*ch
= &part
->channels
[ch_number
];
1801 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1804 ch_sn2
->remote_GP
= part
->sn
.sn2
.remote_GPs
[ch_number
];
1806 /* See what, if anything, has changed for each connected channel */
1808 xpc_msgqueue_ref(ch
);
1810 if (ch_sn2
->w_remote_GP
.get
== ch_sn2
->remote_GP
.get
&&
1811 ch_sn2
->w_remote_GP
.put
== ch_sn2
->remote_GP
.put
) {
1812 /* nothing changed since GPs were last pulled */
1813 xpc_msgqueue_deref(ch
);
1817 if (!(ch
->flags
& XPC_C_CONNECTED
)) {
1818 xpc_msgqueue_deref(ch
);
1823 * First check to see if messages recently sent by us have been
1824 * received by the other side. (The remote GET value will have
1825 * changed since we last looked at it.)
1828 if (ch_sn2
->w_remote_GP
.get
!= ch_sn2
->remote_GP
.get
) {
1831 * We need to notify any senders that want to be notified
1832 * that their sent messages have been received by their
1833 * intended recipients. We need to do this before updating
1834 * w_remote_GP.get so that we don't allocate the same message
1835 * queue entries prematurely (see xpc_allocate_msg()).
1837 if (atomic_read(&ch
->n_to_notify
) > 0) {
1839 * Notify senders that messages sent have been
1840 * received and delivered by the other side.
1842 xpc_notify_senders_sn2(ch
, xpMsgDelivered
,
1843 ch_sn2
->remote_GP
.get
);
1847 * Clear msg->flags in previously sent messages, so that
1848 * they're ready for xpc_allocate_msg().
1850 xpc_clear_local_msgqueue_flags_sn2(ch
);
1852 ch_sn2
->w_remote_GP
.get
= ch_sn2
->remote_GP
.get
;
1854 dev_dbg(xpc_chan
, "w_remote_GP.get changed to %ld, partid=%d, "
1855 "channel=%d\n", ch_sn2
->w_remote_GP
.get
, ch
->partid
,
1859 * If anyone was waiting for message queue entries to become
1860 * available, wake them up.
1862 if (atomic_read(&ch
->n_on_msg_allocate_wq
) > 0)
1863 wake_up(&ch
->msg_allocate_wq
);
1867 * Now check for newly sent messages by the other side. (The remote
1868 * PUT value will have changed since we last looked at it.)
1871 if (ch_sn2
->w_remote_GP
.put
!= ch_sn2
->remote_GP
.put
) {
1873 * Clear msg->flags in previously received messages, so that
1874 * they're ready for xpc_get_deliverable_payload_sn2().
1876 xpc_clear_remote_msgqueue_flags_sn2(ch
);
1878 smp_wmb(); /* ensure flags have been cleared before bte_copy */
1879 ch_sn2
->w_remote_GP
.put
= ch_sn2
->remote_GP
.put
;
1881 dev_dbg(xpc_chan
, "w_remote_GP.put changed to %ld, partid=%d, "
1882 "channel=%d\n", ch_sn2
->w_remote_GP
.put
, ch
->partid
,
1885 npayloads_sent
= xpc_n_of_deliverable_payloads_sn2(ch
);
1886 if (npayloads_sent
> 0) {
1887 dev_dbg(xpc_chan
, "msgs waiting to be copied and "
1888 "delivered=%d, partid=%d, channel=%d\n",
1889 npayloads_sent
, ch
->partid
, ch
->number
);
1891 if (ch
->flags
& XPC_C_CONNECTEDCALLOUT_MADE
)
1892 xpc_activate_kthreads(ch
, npayloads_sent
);
1896 xpc_msgqueue_deref(ch
);
1899 static struct xpc_msg_sn2
*
1900 xpc_pull_remote_msg_sn2(struct xpc_channel
*ch
, s64 get
)
1902 struct xpc_partition
*part
= &xpc_partitions
[ch
->partid
];
1903 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1904 unsigned long remote_msg_pa
;
1905 struct xpc_msg_sn2
*msg
;
1911 if (mutex_lock_interruptible(&ch_sn2
->msg_to_pull_mutex
) != 0) {
1912 /* we were interrupted by a signal */
1916 while (get
>= ch_sn2
->next_msg_to_pull
) {
1918 /* pull as many messages as are ready and able to be pulled */
1920 msg_index
= ch_sn2
->next_msg_to_pull
% ch
->remote_nentries
;
1922 DBUG_ON(ch_sn2
->next_msg_to_pull
>= ch_sn2
->w_remote_GP
.put
);
1923 nmsgs
= ch_sn2
->w_remote_GP
.put
- ch_sn2
->next_msg_to_pull
;
1924 if (msg_index
+ nmsgs
> ch
->remote_nentries
) {
1925 /* ignore the ones that wrap the msg queue for now */
1926 nmsgs
= ch
->remote_nentries
- msg_index
;
1929 msg_offset
= msg_index
* ch
->entry_size
;
1930 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->remote_msgqueue
+
1932 remote_msg_pa
= ch_sn2
->remote_msgqueue_pa
+ msg_offset
;
1934 ret
= xpc_pull_remote_cachelines_sn2(part
, msg
, remote_msg_pa
,
1935 nmsgs
* ch
->entry_size
);
1936 if (ret
!= xpSuccess
) {
1938 dev_dbg(xpc_chan
, "failed to pull %d msgs starting with"
1939 " msg %ld from partition %d, channel=%d, "
1940 "ret=%d\n", nmsgs
, ch_sn2
->next_msg_to_pull
,
1941 ch
->partid
, ch
->number
, ret
);
1943 XPC_DEACTIVATE_PARTITION(part
, ret
);
1945 mutex_unlock(&ch_sn2
->msg_to_pull_mutex
);
1949 ch_sn2
->next_msg_to_pull
+= nmsgs
;
1952 mutex_unlock(&ch_sn2
->msg_to_pull_mutex
);
1954 /* return the message we were looking for */
1955 msg_offset
= (get
% ch
->remote_nentries
) * ch
->entry_size
;
1956 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->remote_msgqueue
+ msg_offset
);
1962 * Get the next deliverable message's payload.
1965 xpc_get_deliverable_payload_sn2(struct xpc_channel
*ch
)
1967 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
1968 struct xpc_msg_sn2
*msg
;
1969 void *payload
= NULL
;
1973 if (ch
->flags
& XPC_C_DISCONNECTING
)
1976 get
= ch_sn2
->w_local_GP
.get
;
1977 smp_rmb(); /* guarantee that .get loads before .put */
1978 if (get
== ch_sn2
->w_remote_GP
.put
)
1981 /* There are messages waiting to be pulled and delivered.
1982 * We need to try to secure one for ourselves. We'll do this
1983 * by trying to increment w_local_GP.get and hope that no one
1984 * else beats us to it. If they do, we'll we'll simply have
1985 * to try again for the next one.
1988 if (cmpxchg(&ch_sn2
->w_local_GP
.get
, get
, get
+ 1) == get
) {
1989 /* we got the entry referenced by get */
1991 dev_dbg(xpc_chan
, "w_local_GP.get changed to %ld, "
1992 "partid=%d, channel=%d\n", get
+ 1,
1993 ch
->partid
, ch
->number
);
1995 /* pull the message from the remote partition */
1997 msg
= xpc_pull_remote_msg_sn2(ch
, get
);
2000 DBUG_ON(msg
->number
!= get
);
2001 DBUG_ON(msg
->flags
& XPC_M_SN2_DONE
);
2002 DBUG_ON(!(msg
->flags
& XPC_M_SN2_READY
));
2004 payload
= &msg
->payload
;
2015 * Now we actually send the messages that are ready to be sent by advancing
2016 * the local message queue's Put value and then send a chctl msgrequest to the
2017 * recipient partition.
2020 xpc_send_msgs_sn2(struct xpc_channel
*ch
, s64 initial_put
)
2022 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
2023 struct xpc_msg_sn2
*msg
;
2024 s64 put
= initial_put
+ 1;
2025 int send_msgrequest
= 0;
2030 if (put
== ch_sn2
->w_local_GP
.put
)
2033 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->
2034 local_msgqueue
+ (put
%
2035 ch
->local_nentries
) *
2038 if (!(msg
->flags
& XPC_M_SN2_READY
))
2044 if (put
== initial_put
) {
2045 /* nothing's changed */
2049 if (cmpxchg_rel(&ch_sn2
->local_GP
->put
, initial_put
, put
) !=
2051 /* someone else beat us to it */
2052 DBUG_ON(ch_sn2
->local_GP
->put
< initial_put
);
2056 /* we just set the new value of local_GP->put */
2058 dev_dbg(xpc_chan
, "local_GP->put changed to %ld, partid=%d, "
2059 "channel=%d\n", put
, ch
->partid
, ch
->number
);
2061 send_msgrequest
= 1;
2064 * We need to ensure that the message referenced by
2065 * local_GP->put is not XPC_M_SN2_READY or that local_GP->put
2066 * equals w_local_GP.put, so we'll go have a look.
2071 if (send_msgrequest
)
2072 xpc_send_chctl_msgrequest_sn2(ch
);
2076 * Allocate an entry for a message from the message queue associated with the
2077 * specified channel.
2079 static enum xp_retval
2080 xpc_allocate_msg_sn2(struct xpc_channel
*ch
, u32 flags
,
2081 struct xpc_msg_sn2
**address_of_msg
)
2083 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
2084 struct xpc_msg_sn2
*msg
;
2089 * Get the next available message entry from the local message queue.
2090 * If none are available, we'll make sure that we grab the latest
2097 put
= ch_sn2
->w_local_GP
.put
;
2098 smp_rmb(); /* guarantee that .put loads before .get */
2099 if (put
- ch_sn2
->w_remote_GP
.get
< ch
->local_nentries
) {
2101 /* There are available message entries. We need to try
2102 * to secure one for ourselves. We'll do this by trying
2103 * to increment w_local_GP.put as long as someone else
2104 * doesn't beat us to it. If they do, we'll have to
2107 if (cmpxchg(&ch_sn2
->w_local_GP
.put
, put
, put
+ 1) ==
2109 /* we got the entry referenced by put */
2112 continue; /* try again */
2116 * There aren't any available msg entries at this time.
2118 * In waiting for a message entry to become available,
2119 * we set a timeout in case the other side is not sending
2120 * completion interrupts. This lets us fake a notify IRQ
2121 * that will cause the notify IRQ handler to fetch the latest
2122 * GP values as if an interrupt was sent by the other side.
2124 if (ret
== xpTimeout
)
2125 xpc_send_chctl_local_msgrequest_sn2(ch
);
2127 if (flags
& XPC_NOWAIT
)
2130 ret
= xpc_allocate_msg_wait(ch
);
2131 if (ret
!= xpInterrupted
&& ret
!= xpTimeout
)
2135 /* get the message's address and initialize it */
2136 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->local_msgqueue
+
2137 (put
% ch
->local_nentries
) *
2140 DBUG_ON(msg
->flags
!= 0);
2143 dev_dbg(xpc_chan
, "w_local_GP.put changed to %ld; msg=0x%p, "
2144 "msg_number=%ld, partid=%d, channel=%d\n", put
+ 1,
2145 (void *)msg
, msg
->number
, ch
->partid
, ch
->number
);
2147 *address_of_msg
= msg
;
2152 * Common code that does the actual sending of the message by advancing the
2153 * local message queue's Put value and sends a chctl msgrequest to the
2154 * partition the message is being sent to.
2156 static enum xp_retval
2157 xpc_send_payload_sn2(struct xpc_channel
*ch
, u32 flags
, void *payload
,
2158 u16 payload_size
, u8 notify_type
, xpc_notify_func func
,
2161 enum xp_retval ret
= xpSuccess
;
2162 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
2163 struct xpc_msg_sn2
*msg
= msg
;
2164 struct xpc_notify_sn2
*notify
= notify
;
2168 DBUG_ON(notify_type
== XPC_N_CALL
&& func
== NULL
);
2170 if (XPC_MSG_SIZE(payload_size
) > ch
->entry_size
)
2171 return xpPayloadTooBig
;
2173 xpc_msgqueue_ref(ch
);
2175 if (ch
->flags
& XPC_C_DISCONNECTING
) {
2179 if (!(ch
->flags
& XPC_C_CONNECTED
)) {
2180 ret
= xpNotConnected
;
2184 ret
= xpc_allocate_msg_sn2(ch
, flags
, &msg
);
2185 if (ret
!= xpSuccess
)
2188 msg_number
= msg
->number
;
2190 if (notify_type
!= 0) {
2192 * Tell the remote side to send an ACK interrupt when the
2193 * message has been delivered.
2195 msg
->flags
|= XPC_M_SN2_INTERRUPT
;
2197 atomic_inc(&ch
->n_to_notify
);
2199 notify
= &ch_sn2
->notify_queue
[msg_number
% ch
->local_nentries
];
2200 notify
->func
= func
;
2202 notify
->type
= notify_type
;
2204 /* ??? Is a mb() needed here? */
2206 if (ch
->flags
& XPC_C_DISCONNECTING
) {
2208 * An error occurred between our last error check and
2209 * this one. We will try to clear the type field from
2210 * the notify entry. If we succeed then
2211 * xpc_disconnect_channel() didn't already process
2214 if (cmpxchg(¬ify
->type
, notify_type
, 0) ==
2216 atomic_dec(&ch
->n_to_notify
);
2223 memcpy(&msg
->payload
, payload
, payload_size
);
2225 msg
->flags
|= XPC_M_SN2_READY
;
2228 * The preceding store of msg->flags must occur before the following
2229 * load of local_GP->put.
2233 /* see if the message is next in line to be sent, if so send it */
2235 put
= ch_sn2
->local_GP
->put
;
2236 if (put
== msg_number
)
2237 xpc_send_msgs_sn2(ch
, put
);
2240 xpc_msgqueue_deref(ch
);
2245 * Now we actually acknowledge the messages that have been delivered and ack'd
2246 * by advancing the cached remote message queue's Get value and if requested
2247 * send a chctl msgrequest to the message sender's partition.
2249 * If a message has XPC_M_SN2_INTERRUPT set, send an interrupt to the partition
2250 * that sent the message.
2253 xpc_acknowledge_msgs_sn2(struct xpc_channel
*ch
, s64 initial_get
, u8 msg_flags
)
2255 struct xpc_channel_sn2
*ch_sn2
= &ch
->sn
.sn2
;
2256 struct xpc_msg_sn2
*msg
;
2257 s64 get
= initial_get
+ 1;
2258 int send_msgrequest
= 0;
2263 if (get
== ch_sn2
->w_local_GP
.get
)
2266 msg
= (struct xpc_msg_sn2
*)((u64
)ch_sn2
->
2267 remote_msgqueue
+ (get
%
2268 ch
->remote_nentries
) *
2271 if (!(msg
->flags
& XPC_M_SN2_DONE
))
2274 msg_flags
|= msg
->flags
;
2278 if (get
== initial_get
) {
2279 /* nothing's changed */
2283 if (cmpxchg_rel(&ch_sn2
->local_GP
->get
, initial_get
, get
) !=
2285 /* someone else beat us to it */
2286 DBUG_ON(ch_sn2
->local_GP
->get
<= initial_get
);
2290 /* we just set the new value of local_GP->get */
2292 dev_dbg(xpc_chan
, "local_GP->get changed to %ld, partid=%d, "
2293 "channel=%d\n", get
, ch
->partid
, ch
->number
);
2295 send_msgrequest
= (msg_flags
& XPC_M_SN2_INTERRUPT
);
2298 * We need to ensure that the message referenced by
2299 * local_GP->get is not XPC_M_SN2_DONE or that local_GP->get
2300 * equals w_local_GP.get, so we'll go have a look.
2305 if (send_msgrequest
)
2306 xpc_send_chctl_msgrequest_sn2(ch
);
2310 xpc_received_payload_sn2(struct xpc_channel
*ch
, void *payload
)
2312 struct xpc_msg_sn2
*msg
;
2316 msg
= container_of(payload
, struct xpc_msg_sn2
, payload
);
2317 msg_number
= msg
->number
;
2319 dev_dbg(xpc_chan
, "msg=0x%p, msg_number=%ld, partid=%d, channel=%d\n",
2320 (void *)msg
, msg_number
, ch
->partid
, ch
->number
);
2322 DBUG_ON((((u64
)msg
- (u64
)ch
->sn
.sn2
.remote_msgqueue
) / ch
->entry_size
) !=
2323 msg_number
% ch
->remote_nentries
);
2324 DBUG_ON(!(msg
->flags
& XPC_M_SN2_READY
));
2325 DBUG_ON(msg
->flags
& XPC_M_SN2_DONE
);
2327 msg
->flags
|= XPC_M_SN2_DONE
;
2330 * The preceding store of msg->flags must occur before the following
2331 * load of local_GP->get.
2336 * See if this message is next in line to be acknowledged as having
2339 get
= ch
->sn
.sn2
.local_GP
->get
;
2340 if (get
== msg_number
)
2341 xpc_acknowledge_msgs_sn2(ch
, get
, msg
->flags
);
2350 xpc_setup_partitions_sn
= xpc_setup_partitions_sn_sn2
;
2351 xpc_teardown_partitions_sn
= xpc_teardown_partitions_sn_sn2
;
2352 xpc_get_partition_rsvd_page_pa
= xpc_get_partition_rsvd_page_pa_sn2
;
2353 xpc_setup_rsvd_page_sn
= xpc_setup_rsvd_page_sn_sn2
;
2355 xpc_allow_hb
= xpc_allow_hb_sn2
;
2356 xpc_disallow_hb
= xpc_disallow_hb_sn2
;
2357 xpc_disallow_all_hbs
= xpc_disallow_all_hbs_sn2
;
2358 xpc_increment_heartbeat
= xpc_increment_heartbeat_sn2
;
2359 xpc_offline_heartbeat
= xpc_offline_heartbeat_sn2
;
2360 xpc_online_heartbeat
= xpc_online_heartbeat_sn2
;
2361 xpc_heartbeat_init
= xpc_heartbeat_init_sn2
;
2362 xpc_heartbeat_exit
= xpc_heartbeat_exit_sn2
;
2363 xpc_get_remote_heartbeat
= xpc_get_remote_heartbeat_sn2
;
2365 xpc_request_partition_activation
= xpc_request_partition_activation_sn2
;
2366 xpc_request_partition_reactivation
=
2367 xpc_request_partition_reactivation_sn2
;
2368 xpc_request_partition_deactivation
=
2369 xpc_request_partition_deactivation_sn2
;
2370 xpc_cancel_partition_deactivation_request
=
2371 xpc_cancel_partition_deactivation_request_sn2
;
2373 xpc_process_activate_IRQ_rcvd
= xpc_process_activate_IRQ_rcvd_sn2
;
2374 xpc_setup_ch_structures_sn
= xpc_setup_ch_structures_sn_sn2
;
2375 xpc_teardown_ch_structures_sn
= xpc_teardown_ch_structures_sn_sn2
;
2376 xpc_make_first_contact
= xpc_make_first_contact_sn2
;
2378 xpc_get_chctl_all_flags
= xpc_get_chctl_all_flags_sn2
;
2379 xpc_send_chctl_closerequest
= xpc_send_chctl_closerequest_sn2
;
2380 xpc_send_chctl_closereply
= xpc_send_chctl_closereply_sn2
;
2381 xpc_send_chctl_openrequest
= xpc_send_chctl_openrequest_sn2
;
2382 xpc_send_chctl_openreply
= xpc_send_chctl_openreply_sn2
;
2384 xpc_save_remote_msgqueue_pa
= xpc_save_remote_msgqueue_pa_sn2
;
2386 xpc_setup_msg_structures
= xpc_setup_msg_structures_sn2
;
2387 xpc_teardown_msg_structures
= xpc_teardown_msg_structures_sn2
;
2389 xpc_notify_senders_of_disconnect
= xpc_notify_senders_of_disconnect_sn2
;
2390 xpc_process_msg_chctl_flags
= xpc_process_msg_chctl_flags_sn2
;
2391 xpc_n_of_deliverable_payloads
= xpc_n_of_deliverable_payloads_sn2
;
2392 xpc_get_deliverable_payload
= xpc_get_deliverable_payload_sn2
;
2394 xpc_indicate_partition_engaged
= xpc_indicate_partition_engaged_sn2
;
2395 xpc_indicate_partition_disengaged
=
2396 xpc_indicate_partition_disengaged_sn2
;
2397 xpc_partition_engaged
= xpc_partition_engaged_sn2
;
2398 xpc_any_partition_engaged
= xpc_any_partition_engaged_sn2
;
2399 xpc_assume_partition_disengaged
= xpc_assume_partition_disengaged_sn2
;
2401 xpc_send_payload
= xpc_send_payload_sn2
;
2402 xpc_received_payload
= xpc_received_payload_sn2
;
2404 if (offsetof(struct xpc_msg_sn2
, payload
) > XPC_MSG_HDR_MAX_SIZE
) {
2405 dev_err(xpc_part
, "header portion of struct xpc_msg_sn2 is "
2406 "larger than %d\n", XPC_MSG_HDR_MAX_SIZE
);
2410 buf_size
= max(XPC_RP_VARS_SIZE
,
2411 XPC_RP_HEADER_SIZE
+ XP_NASID_MASK_BYTES_SN2
);
2412 xpc_remote_copy_buffer_sn2
= xpc_kmalloc_cacheline_aligned(buf_size
,
2414 &xpc_remote_copy_buffer_base_sn2
);
2415 if (xpc_remote_copy_buffer_sn2
== NULL
) {
2416 dev_err(xpc_part
, "can't get memory for remote copy buffer\n");
2420 /* open up protections for IPI and [potentially] amo operations */
2421 xpc_allow_IPI_ops_sn2();
2422 xpc_allow_amo_ops_shub_wars_1_1_sn2();
2425 * This is safe to do before the xpc_hb_checker thread has started
2426 * because the handler releases a wait queue. If an interrupt is
2427 * received before the thread is waiting, it will not go to sleep,
2428 * but rather immediately process the interrupt.
2430 ret
= request_irq(SGI_XPC_ACTIVATE
, xpc_handle_activate_IRQ_sn2
, 0,
2433 dev_err(xpc_part
, "can't register ACTIVATE IRQ handler, "
2434 "errno=%d\n", -ret
);
2435 xpc_disallow_IPI_ops_sn2();
2436 kfree(xpc_remote_copy_buffer_base_sn2
);
2444 free_irq(SGI_XPC_ACTIVATE
, NULL
);
2445 xpc_disallow_IPI_ops_sn2();
2446 kfree(xpc_remote_copy_buffer_base_sn2
);