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) uv-based functions.
11 *
12 * Architecture specific implementation of common functions.
13 *
14 */
15
16#include <linux/kernel.h>
17#include <linux/mm.h>
18#include <linux/interrupt.h>
19#include <linux/delay.h>
20#include <linux/device.h>
21#include <linux/cpu.h>
22#include <linux/module.h>
23#include <linux/err.h>
24#include <linux/slab.h>
25#include <linux/numa.h>
26#include <asm/uv/uv_hub.h>
27#include <asm/uv/bios.h>
28#include <asm/uv/uv_irq.h>
29#include "../sgi-gru/gru.h"
30#include "../sgi-gru/grukservices.h"
31#include "xpc.h"
32
33static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
34
35#define XPC_ACTIVATE_MSG_SIZE_UV (1 * GRU_CACHE_LINE_BYTES)
36#define XPC_ACTIVATE_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \
37 XPC_ACTIVATE_MSG_SIZE_UV)
38#define XPC_ACTIVATE_IRQ_NAME "xpc_activate"
39
40#define XPC_NOTIFY_MSG_SIZE_UV (2 * GRU_CACHE_LINE_BYTES)
41#define XPC_NOTIFY_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \
42 XPC_NOTIFY_MSG_SIZE_UV)
43#define XPC_NOTIFY_IRQ_NAME "xpc_notify"
44
45static int xpc_mq_node = NUMA_NO_NODE;
46
47static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
48static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
49
50static int
51xpc_setup_partitions_uv(void)
52{
53 short partid;
54 struct xpc_partition_uv *part_uv;
55
56 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
57 part_uv = &xpc_partitions[partid].sn.uv;
58
59 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
60 spin_lock_init(&part_uv->flags_lock);
61 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
62 }
63 return 0;
64}
65
66static void
67xpc_teardown_partitions_uv(void)
68{
69 short partid;
70 struct xpc_partition_uv *part_uv;
71 unsigned long irq_flags;
72
73 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
74 part_uv = &xpc_partitions[partid].sn.uv;
75
76 if (part_uv->cached_activate_gru_mq_desc != NULL) {
77 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
78 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
79 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
80 spin_unlock_irqrestore(lock: &part_uv->flags_lock, flags: irq_flags);
81 kfree(objp: part_uv->cached_activate_gru_mq_desc);
82 part_uv->cached_activate_gru_mq_desc = NULL;
83 mutex_unlock(lock: &part_uv->
84 cached_activate_gru_mq_desc_mutex);
85 }
86 }
87}
88
89static int
90xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
91{
92 int mmr_pnode = uv_blade_to_pnode(bid: mq->mmr_blade);
93
94 mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
95 UV_AFFINITY_CPU);
96 if (mq->irq < 0)
97 return mq->irq;
98
99 mq->mmr_value = uv_read_global_mmr64(pnode: mmr_pnode, offset: mq->mmr_offset);
100
101 return 0;
102}
103
104static void
105xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
106{
107 uv_teardown_irq(mq->irq);
108}
109
110static int
111xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
112{
113 int ret;
114
115 ret = uv_bios_mq_watchlist_alloc(uv_gpa(v: mq->address),
116 mq->order, &mq->mmr_offset);
117 if (ret < 0) {
118 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
119 "ret=%d\n", ret);
120 return ret;
121 }
122
123 mq->watchlist_num = ret;
124 return 0;
125}
126
127static void
128xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
129{
130 int ret;
131 int mmr_pnode = uv_blade_to_pnode(bid: mq->mmr_blade);
132
133 ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
134 BUG_ON(ret != BIOS_STATUS_SUCCESS);
135}
136
137static struct xpc_gru_mq_uv *
138xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
139 irq_handler_t irq_handler)
140{
141 enum xp_retval xp_ret;
142 int ret;
143 int nid;
144 int nasid;
145 int pg_order;
146 struct page *page;
147 struct xpc_gru_mq_uv *mq;
148 struct uv_IO_APIC_route_entry *mmr_value;
149
150 mq = kmalloc(size: sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
151 if (mq == NULL) {
152 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
153 "a xpc_gru_mq_uv structure\n");
154 ret = -ENOMEM;
155 goto out_0;
156 }
157
158 mq->gru_mq_desc = kzalloc(size: sizeof(struct gru_message_queue_desc),
159 GFP_KERNEL);
160 if (mq->gru_mq_desc == NULL) {
161 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
162 "a gru_message_queue_desc structure\n");
163 ret = -ENOMEM;
164 goto out_1;
165 }
166
167 pg_order = get_order(size: mq_size);
168 mq->order = pg_order + PAGE_SHIFT;
169 mq_size = 1UL << mq->order;
170
171 mq->mmr_blade = uv_cpu_to_blade_id(cpu);
172
173 nid = cpu_to_node(cpu);
174 page = __alloc_pages_node(nid,
175 GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
176 order: pg_order);
177 if (page == NULL) {
178 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
179 "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
180 ret = -ENOMEM;
181 goto out_2;
182 }
183 mq->address = page_address(page);
184
185 /* enable generation of irq when GRU mq operation occurs to this mq */
186 ret = xpc_gru_mq_watchlist_alloc_uv(mq);
187 if (ret != 0)
188 goto out_3;
189
190 ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
191 if (ret != 0)
192 goto out_4;
193
194 ret = request_irq(irq: mq->irq, handler: irq_handler, flags: 0, name: irq_name, NULL);
195 if (ret != 0) {
196 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
197 mq->irq, -ret);
198 goto out_5;
199 }
200
201 nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
202
203 mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
204 ret = gru_create_message_queue(mqd: mq->gru_mq_desc, p: mq->address, bytes: mq_size,
205 nasid, vector: mmr_value->vector, apicid: mmr_value->dest);
206 if (ret != 0) {
207 dev_err(xpc_part, "gru_create_message_queue() returned "
208 "error=%d\n", ret);
209 ret = -EINVAL;
210 goto out_6;
211 }
212
213 /* allow other partitions to access this GRU mq */
214 xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
215 if (xp_ret != xpSuccess) {
216 ret = -EACCES;
217 goto out_6;
218 }
219
220 return mq;
221
222 /* something went wrong */
223out_6:
224 free_irq(mq->irq, NULL);
225out_5:
226 xpc_release_gru_mq_irq_uv(mq);
227out_4:
228 xpc_gru_mq_watchlist_free_uv(mq);
229out_3:
230 free_pages(addr: (unsigned long)mq->address, order: pg_order);
231out_2:
232 kfree(objp: mq->gru_mq_desc);
233out_1:
234 kfree(objp: mq);
235out_0:
236 return ERR_PTR(error: ret);
237}
238
239static void
240xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
241{
242 unsigned int mq_size;
243 int pg_order;
244 int ret;
245
246 /* disallow other partitions to access GRU mq */
247 mq_size = 1UL << mq->order;
248 ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
249 BUG_ON(ret != xpSuccess);
250
251 /* unregister irq handler and release mq irq/vector mapping */
252 free_irq(mq->irq, NULL);
253 xpc_release_gru_mq_irq_uv(mq);
254
255 /* disable generation of irq when GRU mq op occurs to this mq */
256 xpc_gru_mq_watchlist_free_uv(mq);
257
258 pg_order = mq->order - PAGE_SHIFT;
259 free_pages(addr: (unsigned long)mq->address, order: pg_order);
260
261 kfree(objp: mq);
262}
263
264static enum xp_retval
265xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
266 size_t msg_size)
267{
268 enum xp_retval xp_ret;
269 int ret;
270
271 while (1) {
272 ret = gru_send_message_gpa(mqd: gru_mq_desc, mesg: msg, bytes: msg_size);
273 if (ret == MQE_OK) {
274 xp_ret = xpSuccess;
275 break;
276 }
277
278 if (ret == MQE_QUEUE_FULL) {
279 dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
280 "error=MQE_QUEUE_FULL\n");
281 /* !!! handle QLimit reached; delay & try again */
282 /* ??? Do we add a limit to the number of retries? */
283 (void)msleep_interruptible(msecs: 10);
284 } else if (ret == MQE_CONGESTION) {
285 dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
286 "error=MQE_CONGESTION\n");
287 /* !!! handle LB Overflow; simply try again */
288 /* ??? Do we add a limit to the number of retries? */
289 } else {
290 /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
291 dev_err(xpc_chan, "gru_send_message_gpa() returned "
292 "error=%d\n", ret);
293 xp_ret = xpGruSendMqError;
294 break;
295 }
296 }
297 return xp_ret;
298}
299
300static void
301xpc_process_activate_IRQ_rcvd_uv(void)
302{
303 unsigned long irq_flags;
304 short partid;
305 struct xpc_partition *part;
306 u8 act_state_req;
307
308 DBUG_ON(xpc_activate_IRQ_rcvd == 0);
309
310 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
311 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
312 part = &xpc_partitions[partid];
313
314 if (part->sn.uv.act_state_req == 0)
315 continue;
316
317 xpc_activate_IRQ_rcvd--;
318 BUG_ON(xpc_activate_IRQ_rcvd < 0);
319
320 act_state_req = part->sn.uv.act_state_req;
321 part->sn.uv.act_state_req = 0;
322 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
323
324 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
325 if (part->act_state == XPC_P_AS_INACTIVE)
326 xpc_activate_partition(part);
327 else if (part->act_state == XPC_P_AS_DEACTIVATING)
328 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
329
330 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
331 if (part->act_state == XPC_P_AS_INACTIVE)
332 xpc_activate_partition(part);
333 else
334 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
335
336 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
337 XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
338
339 } else {
340 BUG();
341 }
342
343 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
344 if (xpc_activate_IRQ_rcvd == 0)
345 break;
346 }
347 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
348
349}
350
351static void
352xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
353 struct xpc_activate_mq_msghdr_uv *msg_hdr,
354 int part_setup,
355 int *wakeup_hb_checker)
356{
357 unsigned long irq_flags;
358 struct xpc_partition_uv *part_uv = &part->sn.uv;
359 struct xpc_openclose_args *args;
360
361 part_uv->remote_act_state = msg_hdr->act_state;
362
363 switch (msg_hdr->type) {
364 case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
365 /* syncing of remote_act_state was just done above */
366 break;
367
368 case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
369 struct xpc_activate_mq_msg_activate_req_uv *msg;
370
371 /*
372 * ??? Do we deal here with ts_jiffies being different
373 * ??? if act_state != XPC_P_AS_INACTIVE instead of
374 * ??? below?
375 */
376 msg = container_of(msg_hdr, struct
377 xpc_activate_mq_msg_activate_req_uv, hdr);
378
379 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
380 if (part_uv->act_state_req == 0)
381 xpc_activate_IRQ_rcvd++;
382 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
383 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
384 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
385 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
386
387 if (msg->activate_gru_mq_desc_gpa !=
388 part_uv->activate_gru_mq_desc_gpa) {
389 spin_lock(lock: &part_uv->flags_lock);
390 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
391 spin_unlock(lock: &part_uv->flags_lock);
392 part_uv->activate_gru_mq_desc_gpa =
393 msg->activate_gru_mq_desc_gpa;
394 }
395 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
396
397 (*wakeup_hb_checker)++;
398 break;
399 }
400 case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
401 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
402
403 msg = container_of(msg_hdr, struct
404 xpc_activate_mq_msg_deactivate_req_uv, hdr);
405
406 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
407 if (part_uv->act_state_req == 0)
408 xpc_activate_IRQ_rcvd++;
409 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
410 part_uv->reason = msg->reason;
411 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
412
413 (*wakeup_hb_checker)++;
414 return;
415 }
416 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
417 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
418
419 if (!part_setup)
420 break;
421
422 msg = container_of(msg_hdr, struct
423 xpc_activate_mq_msg_chctl_closerequest_uv,
424 hdr);
425 args = &part->remote_openclose_args[msg->ch_number];
426 args->reason = msg->reason;
427
428 spin_lock_irqsave(&part->chctl_lock, irq_flags);
429 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
430 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
431
432 xpc_wakeup_channel_mgr(part);
433 break;
434 }
435 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
436 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
437
438 if (!part_setup)
439 break;
440
441 msg = container_of(msg_hdr, struct
442 xpc_activate_mq_msg_chctl_closereply_uv,
443 hdr);
444
445 spin_lock_irqsave(&part->chctl_lock, irq_flags);
446 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
447 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
448
449 xpc_wakeup_channel_mgr(part);
450 break;
451 }
452 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
453 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
454
455 if (!part_setup)
456 break;
457
458 msg = container_of(msg_hdr, struct
459 xpc_activate_mq_msg_chctl_openrequest_uv,
460 hdr);
461 args = &part->remote_openclose_args[msg->ch_number];
462 args->entry_size = msg->entry_size;
463 args->local_nentries = msg->local_nentries;
464
465 spin_lock_irqsave(&part->chctl_lock, irq_flags);
466 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
467 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
468
469 xpc_wakeup_channel_mgr(part);
470 break;
471 }
472 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
473 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
474
475 if (!part_setup)
476 break;
477
478 msg = container_of(msg_hdr, struct
479 xpc_activate_mq_msg_chctl_openreply_uv, hdr);
480 args = &part->remote_openclose_args[msg->ch_number];
481 args->remote_nentries = msg->remote_nentries;
482 args->local_nentries = msg->local_nentries;
483 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
484
485 spin_lock_irqsave(&part->chctl_lock, irq_flags);
486 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
487 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
488
489 xpc_wakeup_channel_mgr(part);
490 break;
491 }
492 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
493 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
494
495 if (!part_setup)
496 break;
497
498 msg = container_of(msg_hdr, struct
499 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
500 spin_lock_irqsave(&part->chctl_lock, irq_flags);
501 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
502 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
503
504 xpc_wakeup_channel_mgr(part);
505 }
506 fallthrough;
507 case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
508 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
509 part_uv->flags |= XPC_P_ENGAGED_UV;
510 spin_unlock_irqrestore(lock: &part_uv->flags_lock, flags: irq_flags);
511 break;
512
513 case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
514 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
515 part_uv->flags &= ~XPC_P_ENGAGED_UV;
516 spin_unlock_irqrestore(lock: &part_uv->flags_lock, flags: irq_flags);
517 break;
518
519 default:
520 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
521 "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
522
523 /* get hb checker to deactivate from the remote partition */
524 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
525 if (part_uv->act_state_req == 0)
526 xpc_activate_IRQ_rcvd++;
527 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
528 part_uv->reason = xpBadMsgType;
529 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
530
531 (*wakeup_hb_checker)++;
532 return;
533 }
534
535 if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
536 part->remote_rp_ts_jiffies != 0) {
537 /*
538 * ??? Does what we do here need to be sensitive to
539 * ??? act_state or remote_act_state?
540 */
541 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
542 if (part_uv->act_state_req == 0)
543 xpc_activate_IRQ_rcvd++;
544 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
545 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
546
547 (*wakeup_hb_checker)++;
548 }
549}
550
551static irqreturn_t
552xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
553{
554 struct xpc_activate_mq_msghdr_uv *msg_hdr;
555 short partid;
556 struct xpc_partition *part;
557 int wakeup_hb_checker = 0;
558 int part_referenced;
559
560 while (1) {
561 msg_hdr = gru_get_next_message(mqd: xpc_activate_mq_uv->gru_mq_desc);
562 if (msg_hdr == NULL)
563 break;
564
565 partid = msg_hdr->partid;
566 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
567 dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
568 "received invalid partid=0x%x in message\n",
569 partid);
570 } else {
571 part = &xpc_partitions[partid];
572
573 part_referenced = xpc_part_ref(part);
574 xpc_handle_activate_mq_msg_uv(part, msg_hdr,
575 part_setup: part_referenced,
576 wakeup_hb_checker: &wakeup_hb_checker);
577 if (part_referenced)
578 xpc_part_deref(part);
579 }
580
581 gru_free_message(mqd: xpc_activate_mq_uv->gru_mq_desc, mesq: msg_hdr);
582 }
583
584 if (wakeup_hb_checker)
585 wake_up_interruptible(&xpc_activate_IRQ_wq);
586
587 return IRQ_HANDLED;
588}
589
590static enum xp_retval
591xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
592 unsigned long gru_mq_desc_gpa)
593{
594 enum xp_retval ret;
595
596 ret = xp_remote_memcpy(uv_gpa(v: gru_mq_desc), gru_mq_desc_gpa,
597 sizeof(struct gru_message_queue_desc));
598 if (ret == xpSuccess)
599 gru_mq_desc->mq = NULL;
600
601 return ret;
602}
603
604static enum xp_retval
605xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
606 int msg_type)
607{
608 struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
609 struct xpc_partition_uv *part_uv = &part->sn.uv;
610 struct gru_message_queue_desc *gru_mq_desc;
611 unsigned long irq_flags;
612 enum xp_retval ret;
613
614 DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
615
616 msg_hdr->type = msg_type;
617 msg_hdr->partid = xp_partition_id;
618 msg_hdr->act_state = part->act_state;
619 msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
620
621 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
622again:
623 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
624 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
625 if (gru_mq_desc == NULL) {
626 gru_mq_desc = kmalloc(size: sizeof(struct
627 gru_message_queue_desc),
628 GFP_ATOMIC);
629 if (gru_mq_desc == NULL) {
630 ret = xpNoMemory;
631 goto done;
632 }
633 part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
634 }
635
636 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
637 gru_mq_desc_gpa: part_uv->
638 activate_gru_mq_desc_gpa);
639 if (ret != xpSuccess)
640 goto done;
641
642 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
643 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
644 spin_unlock_irqrestore(lock: &part_uv->flags_lock, flags: irq_flags);
645 }
646
647 /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
648 ret = xpc_send_gru_msg(gru_mq_desc: part_uv->cached_activate_gru_mq_desc, msg,
649 msg_size);
650 if (ret != xpSuccess) {
651 smp_rmb(); /* ensure a fresh copy of part_uv->flags */
652 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
653 goto again;
654 }
655done:
656 mutex_unlock(lock: &part_uv->cached_activate_gru_mq_desc_mutex);
657 return ret;
658}
659
660static void
661xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
662 size_t msg_size, int msg_type)
663{
664 enum xp_retval ret;
665
666 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
667 if (unlikely(ret != xpSuccess))
668 XPC_DEACTIVATE_PARTITION(part, ret);
669}
670
671static void
672xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
673 void *msg, size_t msg_size, int msg_type)
674{
675 struct xpc_partition *part = &xpc_partitions[ch->partid];
676 enum xp_retval ret;
677
678 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
679 if (unlikely(ret != xpSuccess)) {
680 if (irq_flags != NULL)
681 spin_unlock_irqrestore(lock: &ch->lock, flags: *irq_flags);
682
683 XPC_DEACTIVATE_PARTITION(part, ret);
684
685 if (irq_flags != NULL)
686 spin_lock_irqsave(&ch->lock, *irq_flags);
687 }
688}
689
690static void
691xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
692{
693 unsigned long irq_flags;
694 struct xpc_partition_uv *part_uv = &part->sn.uv;
695
696 /*
697 * !!! Make our side think that the remote partition sent an activate
698 * !!! mq message our way by doing what the activate IRQ handler would
699 * !!! do had one really been sent.
700 */
701
702 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
703 if (part_uv->act_state_req == 0)
704 xpc_activate_IRQ_rcvd++;
705 part_uv->act_state_req = act_state_req;
706 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
707
708 wake_up_interruptible(&xpc_activate_IRQ_wq);
709}
710
711static enum xp_retval
712xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
713 size_t *len)
714{
715 s64 status;
716 enum xp_retval ret;
717
718 status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
719 (u64 *)len);
720 if (status == BIOS_STATUS_SUCCESS)
721 ret = xpSuccess;
722 else if (status == BIOS_STATUS_MORE_PASSES)
723 ret = xpNeedMoreInfo;
724 else
725 ret = xpBiosError;
726
727 return ret;
728}
729
730static int
731xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
732{
733 xpc_heartbeat_uv =
734 &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
735 rp->sn.uv.heartbeat_gpa = uv_gpa(v: xpc_heartbeat_uv);
736 rp->sn.uv.activate_gru_mq_desc_gpa =
737 uv_gpa(v: xpc_activate_mq_uv->gru_mq_desc);
738 return 0;
739}
740
741static void
742xpc_allow_hb_uv(short partid)
743{
744}
745
746static void
747xpc_disallow_hb_uv(short partid)
748{
749}
750
751static void
752xpc_disallow_all_hbs_uv(void)
753{
754}
755
756static void
757xpc_increment_heartbeat_uv(void)
758{
759 xpc_heartbeat_uv->value++;
760}
761
762static void
763xpc_offline_heartbeat_uv(void)
764{
765 xpc_increment_heartbeat_uv();
766 xpc_heartbeat_uv->offline = 1;
767}
768
769static void
770xpc_online_heartbeat_uv(void)
771{
772 xpc_increment_heartbeat_uv();
773 xpc_heartbeat_uv->offline = 0;
774}
775
776static void
777xpc_heartbeat_init_uv(void)
778{
779 xpc_heartbeat_uv->value = 1;
780 xpc_heartbeat_uv->offline = 0;
781}
782
783static void
784xpc_heartbeat_exit_uv(void)
785{
786 xpc_offline_heartbeat_uv();
787}
788
789static enum xp_retval
790xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
791{
792 struct xpc_partition_uv *part_uv = &part->sn.uv;
793 enum xp_retval ret;
794
795 ret = xp_remote_memcpy(uv_gpa(v: &part_uv->cached_heartbeat),
796 part_uv->heartbeat_gpa,
797 sizeof(struct xpc_heartbeat_uv));
798 if (ret != xpSuccess)
799 return ret;
800
801 if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
802 !part_uv->cached_heartbeat.offline) {
803
804 ret = xpNoHeartbeat;
805 } else {
806 part->last_heartbeat = part_uv->cached_heartbeat.value;
807 }
808 return ret;
809}
810
811static void
812xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
813 unsigned long remote_rp_gpa, int nasid)
814{
815 short partid = remote_rp->SAL_partid;
816 struct xpc_partition *part = &xpc_partitions[partid];
817 struct xpc_activate_mq_msg_activate_req_uv msg;
818
819 part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
820 part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
821 part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
822 part->sn.uv.activate_gru_mq_desc_gpa =
823 remote_rp->sn.uv.activate_gru_mq_desc_gpa;
824
825 /*
826 * ??? Is it a good idea to make this conditional on what is
827 * ??? potentially stale state information?
828 */
829 if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
830 msg.rp_gpa = uv_gpa(v: xpc_rsvd_page);
831 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
832 msg.activate_gru_mq_desc_gpa =
833 xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
834 xpc_send_activate_IRQ_part_uv(part, msg: &msg, msg_size: sizeof(msg),
835 XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
836 }
837
838 if (part->act_state == XPC_P_AS_INACTIVE)
839 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
840}
841
842static void
843xpc_request_partition_reactivation_uv(struct xpc_partition *part)
844{
845 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
846}
847
848static void
849xpc_request_partition_deactivation_uv(struct xpc_partition *part)
850{
851 struct xpc_activate_mq_msg_deactivate_req_uv msg;
852
853 /*
854 * ??? Is it a good idea to make this conditional on what is
855 * ??? potentially stale state information?
856 */
857 if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
858 part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
859
860 msg.reason = part->reason;
861 xpc_send_activate_IRQ_part_uv(part, msg: &msg, msg_size: sizeof(msg),
862 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
863 }
864}
865
866static void
867xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
868{
869 /* nothing needs to be done */
870 return;
871}
872
873static void
874xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
875{
876 head->first = NULL;
877 head->last = NULL;
878 spin_lock_init(&head->lock);
879 head->n_entries = 0;
880}
881
882static void *
883xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
884{
885 unsigned long irq_flags;
886 struct xpc_fifo_entry_uv *first;
887
888 spin_lock_irqsave(&head->lock, irq_flags);
889 first = head->first;
890 if (head->first != NULL) {
891 head->first = first->next;
892 if (head->first == NULL)
893 head->last = NULL;
894
895 head->n_entries--;
896 BUG_ON(head->n_entries < 0);
897
898 first->next = NULL;
899 }
900 spin_unlock_irqrestore(lock: &head->lock, flags: irq_flags);
901 return first;
902}
903
904static void
905xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
906 struct xpc_fifo_entry_uv *last)
907{
908 unsigned long irq_flags;
909
910 last->next = NULL;
911 spin_lock_irqsave(&head->lock, irq_flags);
912 if (head->last != NULL)
913 head->last->next = last;
914 else
915 head->first = last;
916 head->last = last;
917 head->n_entries++;
918 spin_unlock_irqrestore(lock: &head->lock, flags: irq_flags);
919}
920
921static int
922xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
923{
924 return head->n_entries;
925}
926
927/*
928 * Setup the channel structures that are uv specific.
929 */
930static enum xp_retval
931xpc_setup_ch_structures_uv(struct xpc_partition *part)
932{
933 struct xpc_channel_uv *ch_uv;
934 int ch_number;
935
936 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
937 ch_uv = &part->channels[ch_number].sn.uv;
938
939 xpc_init_fifo_uv(head: &ch_uv->msg_slot_free_list);
940 xpc_init_fifo_uv(head: &ch_uv->recv_msg_list);
941 }
942
943 return xpSuccess;
944}
945
946/*
947 * Teardown the channel structures that are uv specific.
948 */
949static void
950xpc_teardown_ch_structures_uv(struct xpc_partition *part)
951{
952 /* nothing needs to be done */
953 return;
954}
955
956static enum xp_retval
957xpc_make_first_contact_uv(struct xpc_partition *part)
958{
959 struct xpc_activate_mq_msg_uv msg;
960
961 /*
962 * We send a sync msg to get the remote partition's remote_act_state
963 * updated to our current act_state which at this point should
964 * be XPC_P_AS_ACTIVATING.
965 */
966 xpc_send_activate_IRQ_part_uv(part, msg: &msg, msg_size: sizeof(msg),
967 XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
968
969 while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
970 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
971
972 dev_dbg(xpc_part, "waiting to make first contact with "
973 "partition %d\n", XPC_PARTID(part));
974
975 /* wait a 1/4 of a second or so */
976 (void)msleep_interruptible(msecs: 250);
977
978 if (part->act_state == XPC_P_AS_DEACTIVATING)
979 return part->reason;
980 }
981
982 return xpSuccess;
983}
984
985static u64
986xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
987{
988 unsigned long irq_flags;
989 union xpc_channel_ctl_flags chctl;
990
991 spin_lock_irqsave(&part->chctl_lock, irq_flags);
992 chctl = part->chctl;
993 if (chctl.all_flags != 0)
994 part->chctl.all_flags = 0;
995
996 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
997 return chctl.all_flags;
998}
999
1000static enum xp_retval
1001xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1002{
1003 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1004 struct xpc_send_msg_slot_uv *msg_slot;
1005 unsigned long irq_flags;
1006 int nentries;
1007 int entry;
1008 size_t nbytes;
1009
1010 for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1011 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1012 ch_uv->send_msg_slots = kzalloc(size: nbytes, GFP_KERNEL);
1013 if (ch_uv->send_msg_slots == NULL)
1014 continue;
1015
1016 for (entry = 0; entry < nentries; entry++) {
1017 msg_slot = &ch_uv->send_msg_slots[entry];
1018
1019 msg_slot->msg_slot_number = entry;
1020 xpc_put_fifo_entry_uv(head: &ch_uv->msg_slot_free_list,
1021 last: &msg_slot->next);
1022 }
1023
1024 spin_lock_irqsave(&ch->lock, irq_flags);
1025 if (nentries < ch->local_nentries)
1026 ch->local_nentries = nentries;
1027 spin_unlock_irqrestore(lock: &ch->lock, flags: irq_flags);
1028 return xpSuccess;
1029 }
1030
1031 return xpNoMemory;
1032}
1033
1034static enum xp_retval
1035xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1036{
1037 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1038 struct xpc_notify_mq_msg_uv *msg_slot;
1039 unsigned long irq_flags;
1040 int nentries;
1041 int entry;
1042 size_t nbytes;
1043
1044 for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1045 nbytes = nentries * ch->entry_size;
1046 ch_uv->recv_msg_slots = kzalloc(size: nbytes, GFP_KERNEL);
1047 if (ch_uv->recv_msg_slots == NULL)
1048 continue;
1049
1050 for (entry = 0; entry < nentries; entry++) {
1051 msg_slot = ch_uv->recv_msg_slots +
1052 entry * ch->entry_size;
1053
1054 msg_slot->hdr.msg_slot_number = entry;
1055 }
1056
1057 spin_lock_irqsave(&ch->lock, irq_flags);
1058 if (nentries < ch->remote_nentries)
1059 ch->remote_nentries = nentries;
1060 spin_unlock_irqrestore(lock: &ch->lock, flags: irq_flags);
1061 return xpSuccess;
1062 }
1063
1064 return xpNoMemory;
1065}
1066
1067/*
1068 * Allocate msg_slots associated with the channel.
1069 */
1070static enum xp_retval
1071xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1072{
1073 static enum xp_retval ret;
1074 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1075
1076 DBUG_ON(ch->flags & XPC_C_SETUP);
1077
1078 ch_uv->cached_notify_gru_mq_desc = kmalloc(size: sizeof(struct
1079 gru_message_queue_desc),
1080 GFP_KERNEL);
1081 if (ch_uv->cached_notify_gru_mq_desc == NULL)
1082 return xpNoMemory;
1083
1084 ret = xpc_allocate_send_msg_slot_uv(ch);
1085 if (ret == xpSuccess) {
1086
1087 ret = xpc_allocate_recv_msg_slot_uv(ch);
1088 if (ret != xpSuccess) {
1089 kfree(objp: ch_uv->send_msg_slots);
1090 xpc_init_fifo_uv(head: &ch_uv->msg_slot_free_list);
1091 }
1092 }
1093 return ret;
1094}
1095
1096/*
1097 * Free up msg_slots and clear other stuff that were setup for the specified
1098 * channel.
1099 */
1100static void
1101xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1102{
1103 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1104
1105 lockdep_assert_held(&ch->lock);
1106
1107 kfree(objp: ch_uv->cached_notify_gru_mq_desc);
1108 ch_uv->cached_notify_gru_mq_desc = NULL;
1109
1110 if (ch->flags & XPC_C_SETUP) {
1111 xpc_init_fifo_uv(head: &ch_uv->msg_slot_free_list);
1112 kfree(objp: ch_uv->send_msg_slots);
1113 xpc_init_fifo_uv(head: &ch_uv->recv_msg_list);
1114 kfree(objp: ch_uv->recv_msg_slots);
1115 }
1116}
1117
1118static void
1119xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1120{
1121 struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1122
1123 msg.ch_number = ch->number;
1124 msg.reason = ch->reason;
1125 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, msg: &msg, msg_size: sizeof(msg),
1126 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1127}
1128
1129static void
1130xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1131{
1132 struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1133
1134 msg.ch_number = ch->number;
1135 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, msg: &msg, msg_size: sizeof(msg),
1136 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1137}
1138
1139static void
1140xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1141{
1142 struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1143
1144 msg.ch_number = ch->number;
1145 msg.entry_size = ch->entry_size;
1146 msg.local_nentries = ch->local_nentries;
1147 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, msg: &msg, msg_size: sizeof(msg),
1148 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1149}
1150
1151static void
1152xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1153{
1154 struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1155
1156 msg.ch_number = ch->number;
1157 msg.local_nentries = ch->local_nentries;
1158 msg.remote_nentries = ch->remote_nentries;
1159 msg.notify_gru_mq_desc_gpa = uv_gpa(v: xpc_notify_mq_uv->gru_mq_desc);
1160 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, msg: &msg, msg_size: sizeof(msg),
1161 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1162}
1163
1164static void
1165xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1166{
1167 struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1168
1169 msg.ch_number = ch->number;
1170 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, msg: &msg, msg_size: sizeof(msg),
1171 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1172}
1173
1174static void
1175xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1176{
1177 unsigned long irq_flags;
1178
1179 spin_lock_irqsave(&part->chctl_lock, irq_flags);
1180 part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1181 spin_unlock_irqrestore(lock: &part->chctl_lock, flags: irq_flags);
1182
1183 xpc_wakeup_channel_mgr(part);
1184}
1185
1186static enum xp_retval
1187xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1188 unsigned long gru_mq_desc_gpa)
1189{
1190 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1191
1192 DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1193 return xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc: ch_uv->cached_notify_gru_mq_desc,
1194 gru_mq_desc_gpa);
1195}
1196
1197static void
1198xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1199{
1200 struct xpc_activate_mq_msg_uv msg;
1201
1202 xpc_send_activate_IRQ_part_uv(part, msg: &msg, msg_size: sizeof(msg),
1203 XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1204}
1205
1206static void
1207xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1208{
1209 struct xpc_activate_mq_msg_uv msg;
1210
1211 xpc_send_activate_IRQ_part_uv(part, msg: &msg, msg_size: sizeof(msg),
1212 XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1213}
1214
1215static void
1216xpc_assume_partition_disengaged_uv(short partid)
1217{
1218 struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1219 unsigned long irq_flags;
1220
1221 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1222 part_uv->flags &= ~XPC_P_ENGAGED_UV;
1223 spin_unlock_irqrestore(lock: &part_uv->flags_lock, flags: irq_flags);
1224}
1225
1226static int
1227xpc_partition_engaged_uv(short partid)
1228{
1229 return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1230}
1231
1232static int
1233xpc_any_partition_engaged_uv(void)
1234{
1235 struct xpc_partition_uv *part_uv;
1236 short partid;
1237
1238 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1239 part_uv = &xpc_partitions[partid].sn.uv;
1240 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1241 return 1;
1242 }
1243 return 0;
1244}
1245
1246static enum xp_retval
1247xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1248 struct xpc_send_msg_slot_uv **address_of_msg_slot)
1249{
1250 enum xp_retval ret;
1251 struct xpc_send_msg_slot_uv *msg_slot;
1252 struct xpc_fifo_entry_uv *entry;
1253
1254 while (1) {
1255 entry = xpc_get_fifo_entry_uv(head: &ch->sn.uv.msg_slot_free_list);
1256 if (entry != NULL)
1257 break;
1258
1259 if (flags & XPC_NOWAIT)
1260 return xpNoWait;
1261
1262 ret = xpc_allocate_msg_wait(ch);
1263 if (ret != xpInterrupted && ret != xpTimeout)
1264 return ret;
1265 }
1266
1267 msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1268 *address_of_msg_slot = msg_slot;
1269 return xpSuccess;
1270}
1271
1272static void
1273xpc_free_msg_slot_uv(struct xpc_channel *ch,
1274 struct xpc_send_msg_slot_uv *msg_slot)
1275{
1276 xpc_put_fifo_entry_uv(head: &ch->sn.uv.msg_slot_free_list, last: &msg_slot->next);
1277
1278 /* wakeup anyone waiting for a free msg slot */
1279 if (atomic_read(v: &ch->n_on_msg_allocate_wq) > 0)
1280 wake_up(&ch->msg_allocate_wq);
1281}
1282
1283static void
1284xpc_notify_sender_uv(struct xpc_channel *ch,
1285 struct xpc_send_msg_slot_uv *msg_slot,
1286 enum xp_retval reason)
1287{
1288 xpc_notify_func func = msg_slot->func;
1289
1290 if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1291
1292 atomic_dec(v: &ch->n_to_notify);
1293
1294 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1295 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1296 msg_slot->msg_slot_number, ch->partid, ch->number);
1297
1298 func(reason, ch->partid, ch->number, msg_slot->key);
1299
1300 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1301 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1302 msg_slot->msg_slot_number, ch->partid, ch->number);
1303 }
1304}
1305
1306static void
1307xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1308 struct xpc_notify_mq_msg_uv *msg)
1309{
1310 struct xpc_send_msg_slot_uv *msg_slot;
1311 int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1312
1313 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1314
1315 BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1316 msg_slot->msg_slot_number += ch->local_nentries;
1317
1318 if (msg_slot->func != NULL)
1319 xpc_notify_sender_uv(ch, msg_slot, reason: xpMsgDelivered);
1320
1321 xpc_free_msg_slot_uv(ch, msg_slot);
1322}
1323
1324static void
1325xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1326 struct xpc_notify_mq_msg_uv *msg)
1327{
1328 struct xpc_partition_uv *part_uv = &part->sn.uv;
1329 struct xpc_channel *ch;
1330 struct xpc_channel_uv *ch_uv;
1331 struct xpc_notify_mq_msg_uv *msg_slot;
1332 unsigned long irq_flags;
1333 int ch_number = msg->hdr.ch_number;
1334
1335 if (unlikely(ch_number >= part->nchannels)) {
1336 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1337 "channel number=0x%x in message from partid=%d\n",
1338 ch_number, XPC_PARTID(part));
1339
1340 /* get hb checker to deactivate from the remote partition */
1341 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1342 if (part_uv->act_state_req == 0)
1343 xpc_activate_IRQ_rcvd++;
1344 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1345 part_uv->reason = xpBadChannelNumber;
1346 spin_unlock_irqrestore(lock: &xpc_activate_IRQ_rcvd_lock, flags: irq_flags);
1347
1348 wake_up_interruptible(&xpc_activate_IRQ_wq);
1349 return;
1350 }
1351
1352 ch = &part->channels[ch_number];
1353 xpc_msgqueue_ref(ch);
1354
1355 if (!(ch->flags & XPC_C_CONNECTED)) {
1356 xpc_msgqueue_deref(ch);
1357 return;
1358 }
1359
1360 /* see if we're really dealing with an ACK for a previously sent msg */
1361 if (msg->hdr.size == 0) {
1362 xpc_handle_notify_mq_ack_uv(ch, msg);
1363 xpc_msgqueue_deref(ch);
1364 return;
1365 }
1366
1367 /* we're dealing with a normal message sent via the notify_mq */
1368 ch_uv = &ch->sn.uv;
1369
1370 msg_slot = ch_uv->recv_msg_slots +
1371 (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1372
1373 BUG_ON(msg_slot->hdr.size != 0);
1374
1375 memcpy(msg_slot, msg, msg->hdr.size);
1376
1377 xpc_put_fifo_entry_uv(head: &ch_uv->recv_msg_list, last: &msg_slot->hdr.u.next);
1378
1379 if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1380 /*
1381 * If there is an existing idle kthread get it to deliver
1382 * the payload, otherwise we'll have to get the channel mgr
1383 * for this partition to create a kthread to do the delivery.
1384 */
1385 if (atomic_read(v: &ch->kthreads_idle) > 0)
1386 wake_up_nr(&ch->idle_wq, 1);
1387 else
1388 xpc_send_chctl_local_msgrequest_uv(part, ch_number: ch->number);
1389 }
1390 xpc_msgqueue_deref(ch);
1391}
1392
1393static irqreturn_t
1394xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1395{
1396 struct xpc_notify_mq_msg_uv *msg;
1397 short partid;
1398 struct xpc_partition *part;
1399
1400 while ((msg = gru_get_next_message(mqd: xpc_notify_mq_uv->gru_mq_desc)) !=
1401 NULL) {
1402
1403 partid = msg->hdr.partid;
1404 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1405 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1406 "invalid partid=0x%x in message\n", partid);
1407 } else {
1408 part = &xpc_partitions[partid];
1409
1410 if (xpc_part_ref(part)) {
1411 xpc_handle_notify_mq_msg_uv(part, msg);
1412 xpc_part_deref(part);
1413 }
1414 }
1415
1416 gru_free_message(mqd: xpc_notify_mq_uv->gru_mq_desc, mesq: msg);
1417 }
1418
1419 return IRQ_HANDLED;
1420}
1421
1422static int
1423xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1424{
1425 return xpc_n_of_fifo_entries_uv(head: &ch->sn.uv.recv_msg_list);
1426}
1427
1428static void
1429xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1430{
1431 struct xpc_channel *ch = &part->channels[ch_number];
1432 int ndeliverable_payloads;
1433
1434 xpc_msgqueue_ref(ch);
1435
1436 ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1437
1438 if (ndeliverable_payloads > 0 &&
1439 (ch->flags & XPC_C_CONNECTED) &&
1440 (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1441
1442 xpc_activate_kthreads(ch, ndeliverable_payloads);
1443 }
1444
1445 xpc_msgqueue_deref(ch);
1446}
1447
1448static enum xp_retval
1449xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1450 u16 payload_size, u8 notify_type, xpc_notify_func func,
1451 void *key)
1452{
1453 enum xp_retval ret = xpSuccess;
1454 struct xpc_send_msg_slot_uv *msg_slot = NULL;
1455 struct xpc_notify_mq_msg_uv *msg;
1456 u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1457 size_t msg_size;
1458
1459 DBUG_ON(notify_type != XPC_N_CALL);
1460
1461 msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1462 if (msg_size > ch->entry_size)
1463 return xpPayloadTooBig;
1464
1465 xpc_msgqueue_ref(ch);
1466
1467 if (ch->flags & XPC_C_DISCONNECTING) {
1468 ret = ch->reason;
1469 goto out_1;
1470 }
1471 if (!(ch->flags & XPC_C_CONNECTED)) {
1472 ret = xpNotConnected;
1473 goto out_1;
1474 }
1475
1476 ret = xpc_allocate_msg_slot_uv(ch, flags, address_of_msg_slot: &msg_slot);
1477 if (ret != xpSuccess)
1478 goto out_1;
1479
1480 if (func != NULL) {
1481 atomic_inc(v: &ch->n_to_notify);
1482
1483 msg_slot->key = key;
1484 smp_wmb(); /* a non-NULL func must hit memory after the key */
1485 msg_slot->func = func;
1486
1487 if (ch->flags & XPC_C_DISCONNECTING) {
1488 ret = ch->reason;
1489 goto out_2;
1490 }
1491 }
1492
1493 msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1494 msg->hdr.partid = xp_partition_id;
1495 msg->hdr.ch_number = ch->number;
1496 msg->hdr.size = msg_size;
1497 msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1498 memcpy(&msg->payload, payload, payload_size);
1499
1500 ret = xpc_send_gru_msg(gru_mq_desc: ch->sn.uv.cached_notify_gru_mq_desc, msg,
1501 msg_size);
1502 if (ret == xpSuccess)
1503 goto out_1;
1504
1505 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1506out_2:
1507 if (func != NULL) {
1508 /*
1509 * Try to NULL the msg_slot's func field. If we fail, then
1510 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1511 * case we need to pretend we succeeded to send the message
1512 * since the user will get a callout for the disconnect error
1513 * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1514 * error returned here will confuse them. Additionally, since
1515 * in this case the channel is being disconnected we don't need
1516 * to put the msg_slot back on the free list.
1517 */
1518 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1519 ret = xpSuccess;
1520 goto out_1;
1521 }
1522
1523 msg_slot->key = NULL;
1524 atomic_dec(v: &ch->n_to_notify);
1525 }
1526 xpc_free_msg_slot_uv(ch, msg_slot);
1527out_1:
1528 xpc_msgqueue_deref(ch);
1529 return ret;
1530}
1531
1532/*
1533 * Tell the callers of xpc_send_notify() that the status of their payloads
1534 * is unknown because the channel is now disconnecting.
1535 *
1536 * We don't worry about putting these msg_slots on the free list since the
1537 * msg_slots themselves are about to be kfree'd.
1538 */
1539static void
1540xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1541{
1542 struct xpc_send_msg_slot_uv *msg_slot;
1543 int entry;
1544
1545 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1546
1547 for (entry = 0; entry < ch->local_nentries; entry++) {
1548
1549 if (atomic_read(v: &ch->n_to_notify) == 0)
1550 break;
1551
1552 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1553 if (msg_slot->func != NULL)
1554 xpc_notify_sender_uv(ch, msg_slot, reason: ch->reason);
1555 }
1556}
1557
1558/*
1559 * Get the next deliverable message's payload.
1560 */
1561static void *
1562xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1563{
1564 struct xpc_fifo_entry_uv *entry;
1565 struct xpc_notify_mq_msg_uv *msg;
1566 void *payload = NULL;
1567
1568 if (!(ch->flags & XPC_C_DISCONNECTING)) {
1569 entry = xpc_get_fifo_entry_uv(head: &ch->sn.uv.recv_msg_list);
1570 if (entry != NULL) {
1571 msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1572 hdr.u.next);
1573 payload = &msg->payload;
1574 }
1575 }
1576 return payload;
1577}
1578
1579static void
1580xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1581{
1582 struct xpc_notify_mq_msg_uv *msg;
1583 enum xp_retval ret;
1584
1585 msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1586
1587 /* return an ACK to the sender of this message */
1588
1589 msg->hdr.partid = xp_partition_id;
1590 msg->hdr.size = 0; /* size of zero indicates this is an ACK */
1591
1592 ret = xpc_send_gru_msg(gru_mq_desc: ch->sn.uv.cached_notify_gru_mq_desc, msg,
1593 msg_size: sizeof(struct xpc_notify_mq_msghdr_uv));
1594 if (ret != xpSuccess)
1595 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1596}
1597
1598static const struct xpc_arch_operations xpc_arch_ops_uv = {
1599 .setup_partitions = xpc_setup_partitions_uv,
1600 .teardown_partitions = xpc_teardown_partitions_uv,
1601 .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1602 .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1603 .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1604
1605 .allow_hb = xpc_allow_hb_uv,
1606 .disallow_hb = xpc_disallow_hb_uv,
1607 .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1608 .increment_heartbeat = xpc_increment_heartbeat_uv,
1609 .offline_heartbeat = xpc_offline_heartbeat_uv,
1610 .online_heartbeat = xpc_online_heartbeat_uv,
1611 .heartbeat_init = xpc_heartbeat_init_uv,
1612 .heartbeat_exit = xpc_heartbeat_exit_uv,
1613 .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1614
1615 .request_partition_activation =
1616 xpc_request_partition_activation_uv,
1617 .request_partition_reactivation =
1618 xpc_request_partition_reactivation_uv,
1619 .request_partition_deactivation =
1620 xpc_request_partition_deactivation_uv,
1621 .cancel_partition_deactivation_request =
1622 xpc_cancel_partition_deactivation_request_uv,
1623
1624 .setup_ch_structures = xpc_setup_ch_structures_uv,
1625 .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1626
1627 .make_first_contact = xpc_make_first_contact_uv,
1628
1629 .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1630 .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1631 .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1632 .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1633 .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1634 .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1635 .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1636
1637 .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1638
1639 .setup_msg_structures = xpc_setup_msg_structures_uv,
1640 .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1641
1642 .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1643 .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1644 .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1645 .partition_engaged = xpc_partition_engaged_uv,
1646 .any_partition_engaged = xpc_any_partition_engaged_uv,
1647
1648 .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1649 .send_payload = xpc_send_payload_uv,
1650 .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1651 .received_payload = xpc_received_payload_uv,
1652 .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1653};
1654
1655static int
1656xpc_init_mq_node(int nid)
1657{
1658 int cpu;
1659
1660 cpus_read_lock();
1661
1662 for_each_cpu(cpu, cpumask_of_node(nid)) {
1663 xpc_activate_mq_uv =
1664 xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, cpu: nid,
1665 XPC_ACTIVATE_IRQ_NAME,
1666 irq_handler: xpc_handle_activate_IRQ_uv);
1667 if (!IS_ERR(ptr: xpc_activate_mq_uv))
1668 break;
1669 }
1670 if (IS_ERR(ptr: xpc_activate_mq_uv)) {
1671 cpus_read_unlock();
1672 return PTR_ERR(ptr: xpc_activate_mq_uv);
1673 }
1674
1675 for_each_cpu(cpu, cpumask_of_node(nid)) {
1676 xpc_notify_mq_uv =
1677 xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, cpu: nid,
1678 XPC_NOTIFY_IRQ_NAME,
1679 irq_handler: xpc_handle_notify_IRQ_uv);
1680 if (!IS_ERR(ptr: xpc_notify_mq_uv))
1681 break;
1682 }
1683 if (IS_ERR(ptr: xpc_notify_mq_uv)) {
1684 xpc_destroy_gru_mq_uv(mq: xpc_activate_mq_uv);
1685 cpus_read_unlock();
1686 return PTR_ERR(ptr: xpc_notify_mq_uv);
1687 }
1688
1689 cpus_read_unlock();
1690 return 0;
1691}
1692
1693int
1694xpc_init_uv(void)
1695{
1696 int nid;
1697 int ret = 0;
1698
1699 xpc_arch_ops = xpc_arch_ops_uv;
1700
1701 if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1702 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1703 XPC_MSG_HDR_MAX_SIZE);
1704 return -E2BIG;
1705 }
1706
1707 if (xpc_mq_node < 0)
1708 for_each_online_node(nid) {
1709 ret = xpc_init_mq_node(nid);
1710
1711 if (!ret)
1712 break;
1713 }
1714 else
1715 ret = xpc_init_mq_node(nid: xpc_mq_node);
1716
1717 if (ret < 0)
1718 dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1719 -ret);
1720
1721 return ret;
1722}
1723
1724void
1725xpc_exit_uv(void)
1726{
1727 xpc_destroy_gru_mq_uv(mq: xpc_notify_mq_uv);
1728 xpc_destroy_gru_mq_uv(mq: xpc_activate_mq_uv);
1729}
1730
1731module_param(xpc_mq_node, int, 0);
1732MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");
1733

source code of linux/drivers/misc/sgi-xp/xpc_uv.c