xref: /openbmc/linux/drivers/misc/sgi-xp/xpc_uv.c (revision 56a0eccd)
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 <asm/uv/uv_hub.h>
26 #if defined CONFIG_X86_64
27 #include <asm/uv/bios.h>
28 #include <asm/uv/uv_irq.h>
29 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
30 #include <asm/sn/intr.h>
31 #include <asm/sn/sn_sal.h>
32 #endif
33 #include "../sgi-gru/gru.h"
34 #include "../sgi-gru/grukservices.h"
35 #include "xpc.h"
36 
37 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
38 struct uv_IO_APIC_route_entry {
39 	__u64	vector		:  8,
40 		delivery_mode	:  3,
41 		dest_mode	:  1,
42 		delivery_status	:  1,
43 		polarity	:  1,
44 		__reserved_1	:  1,
45 		trigger		:  1,
46 		mask		:  1,
47 		__reserved_2	: 15,
48 		dest		: 32;
49 };
50 #endif
51 
52 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
53 
54 #define XPC_ACTIVATE_MSG_SIZE_UV	(1 * GRU_CACHE_LINE_BYTES)
55 #define XPC_ACTIVATE_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
56 					 XPC_ACTIVATE_MSG_SIZE_UV)
57 #define XPC_ACTIVATE_IRQ_NAME		"xpc_activate"
58 
59 #define XPC_NOTIFY_MSG_SIZE_UV		(2 * GRU_CACHE_LINE_BYTES)
60 #define XPC_NOTIFY_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
61 					 XPC_NOTIFY_MSG_SIZE_UV)
62 #define XPC_NOTIFY_IRQ_NAME		"xpc_notify"
63 
64 static int xpc_mq_node = -1;
65 
66 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
67 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
68 
69 static int
70 xpc_setup_partitions_uv(void)
71 {
72 	short partid;
73 	struct xpc_partition_uv *part_uv;
74 
75 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
76 		part_uv = &xpc_partitions[partid].sn.uv;
77 
78 		mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
79 		spin_lock_init(&part_uv->flags_lock);
80 		part_uv->remote_act_state = XPC_P_AS_INACTIVE;
81 	}
82 	return 0;
83 }
84 
85 static void
86 xpc_teardown_partitions_uv(void)
87 {
88 	short partid;
89 	struct xpc_partition_uv *part_uv;
90 	unsigned long irq_flags;
91 
92 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
93 		part_uv = &xpc_partitions[partid].sn.uv;
94 
95 		if (part_uv->cached_activate_gru_mq_desc != NULL) {
96 			mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
97 			spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
98 			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
99 			spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
100 			kfree(part_uv->cached_activate_gru_mq_desc);
101 			part_uv->cached_activate_gru_mq_desc = NULL;
102 			mutex_unlock(&part_uv->
103 				     cached_activate_gru_mq_desc_mutex);
104 		}
105 	}
106 }
107 
108 static int
109 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
110 {
111 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
112 
113 #if defined CONFIG_X86_64
114 	mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
115 			UV_AFFINITY_CPU);
116 	if (mq->irq < 0)
117 		return mq->irq;
118 
119 	mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
120 
121 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
122 	if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
123 		mq->irq = SGI_XPC_ACTIVATE;
124 	else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
125 		mq->irq = SGI_XPC_NOTIFY;
126 	else
127 		return -EINVAL;
128 
129 	mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
130 	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
131 #else
132 	#error not a supported configuration
133 #endif
134 
135 	return 0;
136 }
137 
138 static void
139 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
140 {
141 #if defined CONFIG_X86_64
142 	uv_teardown_irq(mq->irq);
143 
144 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
145 	int mmr_pnode;
146 	unsigned long mmr_value;
147 
148 	mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
149 	mmr_value = 1UL << 16;
150 
151 	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
152 #else
153 	#error not a supported configuration
154 #endif
155 }
156 
157 static int
158 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
159 {
160 	int ret;
161 
162 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
163 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
164 
165 	ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
166 				    mq->order, &mq->mmr_offset);
167 	if (ret < 0) {
168 		dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
169 			ret);
170 		return -EBUSY;
171 	}
172 #elif defined CONFIG_X86_64
173 	ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
174 					 mq->order, &mq->mmr_offset);
175 	if (ret < 0) {
176 		dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
177 			"ret=%d\n", ret);
178 		return ret;
179 	}
180 #else
181 	#error not a supported configuration
182 #endif
183 
184 	mq->watchlist_num = ret;
185 	return 0;
186 }
187 
188 static void
189 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
190 {
191 	int ret;
192 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
193 
194 #if defined CONFIG_X86_64
195 	ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
196 	BUG_ON(ret != BIOS_STATUS_SUCCESS);
197 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
198 	ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
199 	BUG_ON(ret != SALRET_OK);
200 #else
201 	#error not a supported configuration
202 #endif
203 }
204 
205 static struct xpc_gru_mq_uv *
206 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
207 		     irq_handler_t irq_handler)
208 {
209 	enum xp_retval xp_ret;
210 	int ret;
211 	int nid;
212 	int nasid;
213 	int pg_order;
214 	struct page *page;
215 	struct xpc_gru_mq_uv *mq;
216 	struct uv_IO_APIC_route_entry *mmr_value;
217 
218 	mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
219 	if (mq == NULL) {
220 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
221 			"a xpc_gru_mq_uv structure\n");
222 		ret = -ENOMEM;
223 		goto out_0;
224 	}
225 
226 	mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
227 				  GFP_KERNEL);
228 	if (mq->gru_mq_desc == NULL) {
229 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
230 			"a gru_message_queue_desc structure\n");
231 		ret = -ENOMEM;
232 		goto out_1;
233 	}
234 
235 	pg_order = get_order(mq_size);
236 	mq->order = pg_order + PAGE_SHIFT;
237 	mq_size = 1UL << mq->order;
238 
239 	mq->mmr_blade = uv_cpu_to_blade_id(cpu);
240 
241 	nid = cpu_to_node(cpu);
242 	page = __alloc_pages_node(nid,
243 				      GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
244 				      pg_order);
245 	if (page == NULL) {
246 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
247 			"bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
248 		ret = -ENOMEM;
249 		goto out_2;
250 	}
251 	mq->address = page_address(page);
252 
253 	/* enable generation of irq when GRU mq operation occurs to this mq */
254 	ret = xpc_gru_mq_watchlist_alloc_uv(mq);
255 	if (ret != 0)
256 		goto out_3;
257 
258 	ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
259 	if (ret != 0)
260 		goto out_4;
261 
262 	ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
263 	if (ret != 0) {
264 		dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
265 			mq->irq, -ret);
266 		goto out_5;
267 	}
268 
269 	nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
270 
271 	mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
272 	ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
273 				     nasid, mmr_value->vector, mmr_value->dest);
274 	if (ret != 0) {
275 		dev_err(xpc_part, "gru_create_message_queue() returned "
276 			"error=%d\n", ret);
277 		ret = -EINVAL;
278 		goto out_6;
279 	}
280 
281 	/* allow other partitions to access this GRU mq */
282 	xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
283 	if (xp_ret != xpSuccess) {
284 		ret = -EACCES;
285 		goto out_6;
286 	}
287 
288 	return mq;
289 
290 	/* something went wrong */
291 out_6:
292 	free_irq(mq->irq, NULL);
293 out_5:
294 	xpc_release_gru_mq_irq_uv(mq);
295 out_4:
296 	xpc_gru_mq_watchlist_free_uv(mq);
297 out_3:
298 	free_pages((unsigned long)mq->address, pg_order);
299 out_2:
300 	kfree(mq->gru_mq_desc);
301 out_1:
302 	kfree(mq);
303 out_0:
304 	return ERR_PTR(ret);
305 }
306 
307 static void
308 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
309 {
310 	unsigned int mq_size;
311 	int pg_order;
312 	int ret;
313 
314 	/* disallow other partitions to access GRU mq */
315 	mq_size = 1UL << mq->order;
316 	ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
317 	BUG_ON(ret != xpSuccess);
318 
319 	/* unregister irq handler and release mq irq/vector mapping */
320 	free_irq(mq->irq, NULL);
321 	xpc_release_gru_mq_irq_uv(mq);
322 
323 	/* disable generation of irq when GRU mq op occurs to this mq */
324 	xpc_gru_mq_watchlist_free_uv(mq);
325 
326 	pg_order = mq->order - PAGE_SHIFT;
327 	free_pages((unsigned long)mq->address, pg_order);
328 
329 	kfree(mq);
330 }
331 
332 static enum xp_retval
333 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
334 		 size_t msg_size)
335 {
336 	enum xp_retval xp_ret;
337 	int ret;
338 
339 	while (1) {
340 		ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
341 		if (ret == MQE_OK) {
342 			xp_ret = xpSuccess;
343 			break;
344 		}
345 
346 		if (ret == MQE_QUEUE_FULL) {
347 			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
348 				"error=MQE_QUEUE_FULL\n");
349 			/* !!! handle QLimit reached; delay & try again */
350 			/* ??? Do we add a limit to the number of retries? */
351 			(void)msleep_interruptible(10);
352 		} else if (ret == MQE_CONGESTION) {
353 			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
354 				"error=MQE_CONGESTION\n");
355 			/* !!! handle LB Overflow; simply try again */
356 			/* ??? Do we add a limit to the number of retries? */
357 		} else {
358 			/* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
359 			dev_err(xpc_chan, "gru_send_message_gpa() returned "
360 				"error=%d\n", ret);
361 			xp_ret = xpGruSendMqError;
362 			break;
363 		}
364 	}
365 	return xp_ret;
366 }
367 
368 static void
369 xpc_process_activate_IRQ_rcvd_uv(void)
370 {
371 	unsigned long irq_flags;
372 	short partid;
373 	struct xpc_partition *part;
374 	u8 act_state_req;
375 
376 	DBUG_ON(xpc_activate_IRQ_rcvd == 0);
377 
378 	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
379 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
380 		part = &xpc_partitions[partid];
381 
382 		if (part->sn.uv.act_state_req == 0)
383 			continue;
384 
385 		xpc_activate_IRQ_rcvd--;
386 		BUG_ON(xpc_activate_IRQ_rcvd < 0);
387 
388 		act_state_req = part->sn.uv.act_state_req;
389 		part->sn.uv.act_state_req = 0;
390 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
391 
392 		if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
393 			if (part->act_state == XPC_P_AS_INACTIVE)
394 				xpc_activate_partition(part);
395 			else if (part->act_state == XPC_P_AS_DEACTIVATING)
396 				XPC_DEACTIVATE_PARTITION(part, xpReactivating);
397 
398 		} else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
399 			if (part->act_state == XPC_P_AS_INACTIVE)
400 				xpc_activate_partition(part);
401 			else
402 				XPC_DEACTIVATE_PARTITION(part, xpReactivating);
403 
404 		} else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
405 			XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
406 
407 		} else {
408 			BUG();
409 		}
410 
411 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
412 		if (xpc_activate_IRQ_rcvd == 0)
413 			break;
414 	}
415 	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
416 
417 }
418 
419 static void
420 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
421 			      struct xpc_activate_mq_msghdr_uv *msg_hdr,
422 			      int part_setup,
423 			      int *wakeup_hb_checker)
424 {
425 	unsigned long irq_flags;
426 	struct xpc_partition_uv *part_uv = &part->sn.uv;
427 	struct xpc_openclose_args *args;
428 
429 	part_uv->remote_act_state = msg_hdr->act_state;
430 
431 	switch (msg_hdr->type) {
432 	case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
433 		/* syncing of remote_act_state was just done above */
434 		break;
435 
436 	case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
437 		struct xpc_activate_mq_msg_activate_req_uv *msg;
438 
439 		/*
440 		 * ??? Do we deal here with ts_jiffies being different
441 		 * ??? if act_state != XPC_P_AS_INACTIVE instead of
442 		 * ??? below?
443 		 */
444 		msg = container_of(msg_hdr, struct
445 				   xpc_activate_mq_msg_activate_req_uv, hdr);
446 
447 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
448 		if (part_uv->act_state_req == 0)
449 			xpc_activate_IRQ_rcvd++;
450 		part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
451 		part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
452 		part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
453 		part_uv->heartbeat_gpa = msg->heartbeat_gpa;
454 
455 		if (msg->activate_gru_mq_desc_gpa !=
456 		    part_uv->activate_gru_mq_desc_gpa) {
457 			spin_lock(&part_uv->flags_lock);
458 			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
459 			spin_unlock(&part_uv->flags_lock);
460 			part_uv->activate_gru_mq_desc_gpa =
461 			    msg->activate_gru_mq_desc_gpa;
462 		}
463 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
464 
465 		(*wakeup_hb_checker)++;
466 		break;
467 	}
468 	case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
469 		struct xpc_activate_mq_msg_deactivate_req_uv *msg;
470 
471 		msg = container_of(msg_hdr, struct
472 				   xpc_activate_mq_msg_deactivate_req_uv, hdr);
473 
474 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
475 		if (part_uv->act_state_req == 0)
476 			xpc_activate_IRQ_rcvd++;
477 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
478 		part_uv->reason = msg->reason;
479 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
480 
481 		(*wakeup_hb_checker)++;
482 		return;
483 	}
484 	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
485 		struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
486 
487 		if (!part_setup)
488 			break;
489 
490 		msg = container_of(msg_hdr, struct
491 				   xpc_activate_mq_msg_chctl_closerequest_uv,
492 				   hdr);
493 		args = &part->remote_openclose_args[msg->ch_number];
494 		args->reason = msg->reason;
495 
496 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
497 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
498 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
499 
500 		xpc_wakeup_channel_mgr(part);
501 		break;
502 	}
503 	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
504 		struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
505 
506 		if (!part_setup)
507 			break;
508 
509 		msg = container_of(msg_hdr, struct
510 				   xpc_activate_mq_msg_chctl_closereply_uv,
511 				   hdr);
512 
513 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
514 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
515 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
516 
517 		xpc_wakeup_channel_mgr(part);
518 		break;
519 	}
520 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
521 		struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
522 
523 		if (!part_setup)
524 			break;
525 
526 		msg = container_of(msg_hdr, struct
527 				   xpc_activate_mq_msg_chctl_openrequest_uv,
528 				   hdr);
529 		args = &part->remote_openclose_args[msg->ch_number];
530 		args->entry_size = msg->entry_size;
531 		args->local_nentries = msg->local_nentries;
532 
533 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
534 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
535 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
536 
537 		xpc_wakeup_channel_mgr(part);
538 		break;
539 	}
540 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
541 		struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
542 
543 		if (!part_setup)
544 			break;
545 
546 		msg = container_of(msg_hdr, struct
547 				   xpc_activate_mq_msg_chctl_openreply_uv, hdr);
548 		args = &part->remote_openclose_args[msg->ch_number];
549 		args->remote_nentries = msg->remote_nentries;
550 		args->local_nentries = msg->local_nentries;
551 		args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
552 
553 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
554 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
555 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
556 
557 		xpc_wakeup_channel_mgr(part);
558 		break;
559 	}
560 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
561 		struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
562 
563 		if (!part_setup)
564 			break;
565 
566 		msg = container_of(msg_hdr, struct
567 				xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
568 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
569 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
570 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
571 
572 		xpc_wakeup_channel_mgr(part);
573 	}
574 	case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
575 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
576 		part_uv->flags |= XPC_P_ENGAGED_UV;
577 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
578 		break;
579 
580 	case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
581 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
582 		part_uv->flags &= ~XPC_P_ENGAGED_UV;
583 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
584 		break;
585 
586 	default:
587 		dev_err(xpc_part, "received unknown activate_mq msg type=%d "
588 			"from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
589 
590 		/* get hb checker to deactivate from the remote partition */
591 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
592 		if (part_uv->act_state_req == 0)
593 			xpc_activate_IRQ_rcvd++;
594 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
595 		part_uv->reason = xpBadMsgType;
596 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
597 
598 		(*wakeup_hb_checker)++;
599 		return;
600 	}
601 
602 	if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
603 	    part->remote_rp_ts_jiffies != 0) {
604 		/*
605 		 * ??? Does what we do here need to be sensitive to
606 		 * ??? act_state or remote_act_state?
607 		 */
608 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
609 		if (part_uv->act_state_req == 0)
610 			xpc_activate_IRQ_rcvd++;
611 		part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
612 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
613 
614 		(*wakeup_hb_checker)++;
615 	}
616 }
617 
618 static irqreturn_t
619 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
620 {
621 	struct xpc_activate_mq_msghdr_uv *msg_hdr;
622 	short partid;
623 	struct xpc_partition *part;
624 	int wakeup_hb_checker = 0;
625 	int part_referenced;
626 
627 	while (1) {
628 		msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
629 		if (msg_hdr == NULL)
630 			break;
631 
632 		partid = msg_hdr->partid;
633 		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
634 			dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
635 				"received invalid partid=0x%x in message\n",
636 				partid);
637 		} else {
638 			part = &xpc_partitions[partid];
639 
640 			part_referenced = xpc_part_ref(part);
641 			xpc_handle_activate_mq_msg_uv(part, msg_hdr,
642 						      part_referenced,
643 						      &wakeup_hb_checker);
644 			if (part_referenced)
645 				xpc_part_deref(part);
646 		}
647 
648 		gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
649 	}
650 
651 	if (wakeup_hb_checker)
652 		wake_up_interruptible(&xpc_activate_IRQ_wq);
653 
654 	return IRQ_HANDLED;
655 }
656 
657 static enum xp_retval
658 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
659 				unsigned long gru_mq_desc_gpa)
660 {
661 	enum xp_retval ret;
662 
663 	ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
664 			       sizeof(struct gru_message_queue_desc));
665 	if (ret == xpSuccess)
666 		gru_mq_desc->mq = NULL;
667 
668 	return ret;
669 }
670 
671 static enum xp_retval
672 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
673 			 int msg_type)
674 {
675 	struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
676 	struct xpc_partition_uv *part_uv = &part->sn.uv;
677 	struct gru_message_queue_desc *gru_mq_desc;
678 	unsigned long irq_flags;
679 	enum xp_retval ret;
680 
681 	DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
682 
683 	msg_hdr->type = msg_type;
684 	msg_hdr->partid = xp_partition_id;
685 	msg_hdr->act_state = part->act_state;
686 	msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
687 
688 	mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
689 again:
690 	if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
691 		gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
692 		if (gru_mq_desc == NULL) {
693 			gru_mq_desc = kmalloc(sizeof(struct
694 					      gru_message_queue_desc),
695 					      GFP_KERNEL);
696 			if (gru_mq_desc == NULL) {
697 				ret = xpNoMemory;
698 				goto done;
699 			}
700 			part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
701 		}
702 
703 		ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
704 						      part_uv->
705 						      activate_gru_mq_desc_gpa);
706 		if (ret != xpSuccess)
707 			goto done;
708 
709 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
710 		part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
711 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
712 	}
713 
714 	/* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
715 	ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
716 			       msg_size);
717 	if (ret != xpSuccess) {
718 		smp_rmb();	/* ensure a fresh copy of part_uv->flags */
719 		if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
720 			goto again;
721 	}
722 done:
723 	mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
724 	return ret;
725 }
726 
727 static void
728 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
729 			      size_t msg_size, int msg_type)
730 {
731 	enum xp_retval ret;
732 
733 	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
734 	if (unlikely(ret != xpSuccess))
735 		XPC_DEACTIVATE_PARTITION(part, ret);
736 }
737 
738 static void
739 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
740 			 void *msg, size_t msg_size, int msg_type)
741 {
742 	struct xpc_partition *part = &xpc_partitions[ch->partid];
743 	enum xp_retval ret;
744 
745 	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
746 	if (unlikely(ret != xpSuccess)) {
747 		if (irq_flags != NULL)
748 			spin_unlock_irqrestore(&ch->lock, *irq_flags);
749 
750 		XPC_DEACTIVATE_PARTITION(part, ret);
751 
752 		if (irq_flags != NULL)
753 			spin_lock_irqsave(&ch->lock, *irq_flags);
754 	}
755 }
756 
757 static void
758 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
759 {
760 	unsigned long irq_flags;
761 	struct xpc_partition_uv *part_uv = &part->sn.uv;
762 
763 	/*
764 	 * !!! Make our side think that the remote partition sent an activate
765 	 * !!! mq message our way by doing what the activate IRQ handler would
766 	 * !!! do had one really been sent.
767 	 */
768 
769 	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
770 	if (part_uv->act_state_req == 0)
771 		xpc_activate_IRQ_rcvd++;
772 	part_uv->act_state_req = act_state_req;
773 	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
774 
775 	wake_up_interruptible(&xpc_activate_IRQ_wq);
776 }
777 
778 static enum xp_retval
779 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
780 				  size_t *len)
781 {
782 	s64 status;
783 	enum xp_retval ret;
784 
785 #if defined CONFIG_X86_64
786 	status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
787 					  (u64 *)len);
788 	if (status == BIOS_STATUS_SUCCESS)
789 		ret = xpSuccess;
790 	else if (status == BIOS_STATUS_MORE_PASSES)
791 		ret = xpNeedMoreInfo;
792 	else
793 		ret = xpBiosError;
794 
795 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
796 	status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
797 	if (status == SALRET_OK)
798 		ret = xpSuccess;
799 	else if (status == SALRET_MORE_PASSES)
800 		ret = xpNeedMoreInfo;
801 	else
802 		ret = xpSalError;
803 
804 #else
805 	#error not a supported configuration
806 #endif
807 
808 	return ret;
809 }
810 
811 static int
812 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
813 {
814 	xpc_heartbeat_uv =
815 	    &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
816 	rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
817 	rp->sn.uv.activate_gru_mq_desc_gpa =
818 	    uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
819 	return 0;
820 }
821 
822 static void
823 xpc_allow_hb_uv(short partid)
824 {
825 }
826 
827 static void
828 xpc_disallow_hb_uv(short partid)
829 {
830 }
831 
832 static void
833 xpc_disallow_all_hbs_uv(void)
834 {
835 }
836 
837 static void
838 xpc_increment_heartbeat_uv(void)
839 {
840 	xpc_heartbeat_uv->value++;
841 }
842 
843 static void
844 xpc_offline_heartbeat_uv(void)
845 {
846 	xpc_increment_heartbeat_uv();
847 	xpc_heartbeat_uv->offline = 1;
848 }
849 
850 static void
851 xpc_online_heartbeat_uv(void)
852 {
853 	xpc_increment_heartbeat_uv();
854 	xpc_heartbeat_uv->offline = 0;
855 }
856 
857 static void
858 xpc_heartbeat_init_uv(void)
859 {
860 	xpc_heartbeat_uv->value = 1;
861 	xpc_heartbeat_uv->offline = 0;
862 }
863 
864 static void
865 xpc_heartbeat_exit_uv(void)
866 {
867 	xpc_offline_heartbeat_uv();
868 }
869 
870 static enum xp_retval
871 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
872 {
873 	struct xpc_partition_uv *part_uv = &part->sn.uv;
874 	enum xp_retval ret;
875 
876 	ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
877 			       part_uv->heartbeat_gpa,
878 			       sizeof(struct xpc_heartbeat_uv));
879 	if (ret != xpSuccess)
880 		return ret;
881 
882 	if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
883 	    !part_uv->cached_heartbeat.offline) {
884 
885 		ret = xpNoHeartbeat;
886 	} else {
887 		part->last_heartbeat = part_uv->cached_heartbeat.value;
888 	}
889 	return ret;
890 }
891 
892 static void
893 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
894 				    unsigned long remote_rp_gpa, int nasid)
895 {
896 	short partid = remote_rp->SAL_partid;
897 	struct xpc_partition *part = &xpc_partitions[partid];
898 	struct xpc_activate_mq_msg_activate_req_uv msg;
899 
900 	part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
901 	part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
902 	part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
903 	part->sn.uv.activate_gru_mq_desc_gpa =
904 	    remote_rp->sn.uv.activate_gru_mq_desc_gpa;
905 
906 	/*
907 	 * ??? Is it a good idea to make this conditional on what is
908 	 * ??? potentially stale state information?
909 	 */
910 	if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
911 		msg.rp_gpa = uv_gpa(xpc_rsvd_page);
912 		msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
913 		msg.activate_gru_mq_desc_gpa =
914 		    xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
915 		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
916 					   XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
917 	}
918 
919 	if (part->act_state == XPC_P_AS_INACTIVE)
920 		xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
921 }
922 
923 static void
924 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
925 {
926 	xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
927 }
928 
929 static void
930 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
931 {
932 	struct xpc_activate_mq_msg_deactivate_req_uv msg;
933 
934 	/*
935 	 * ??? Is it a good idea to make this conditional on what is
936 	 * ??? potentially stale state information?
937 	 */
938 	if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
939 	    part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
940 
941 		msg.reason = part->reason;
942 		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
943 					 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
944 	}
945 }
946 
947 static void
948 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
949 {
950 	/* nothing needs to be done */
951 	return;
952 }
953 
954 static void
955 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
956 {
957 	head->first = NULL;
958 	head->last = NULL;
959 	spin_lock_init(&head->lock);
960 	head->n_entries = 0;
961 }
962 
963 static void *
964 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
965 {
966 	unsigned long irq_flags;
967 	struct xpc_fifo_entry_uv *first;
968 
969 	spin_lock_irqsave(&head->lock, irq_flags);
970 	first = head->first;
971 	if (head->first != NULL) {
972 		head->first = first->next;
973 		if (head->first == NULL)
974 			head->last = NULL;
975 
976 		head->n_entries--;
977 		BUG_ON(head->n_entries < 0);
978 
979 		first->next = NULL;
980 	}
981 	spin_unlock_irqrestore(&head->lock, irq_flags);
982 	return first;
983 }
984 
985 static void
986 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
987 		      struct xpc_fifo_entry_uv *last)
988 {
989 	unsigned long irq_flags;
990 
991 	last->next = NULL;
992 	spin_lock_irqsave(&head->lock, irq_flags);
993 	if (head->last != NULL)
994 		head->last->next = last;
995 	else
996 		head->first = last;
997 	head->last = last;
998 	head->n_entries++;
999 	spin_unlock_irqrestore(&head->lock, irq_flags);
1000 }
1001 
1002 static int
1003 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
1004 {
1005 	return head->n_entries;
1006 }
1007 
1008 /*
1009  * Setup the channel structures that are uv specific.
1010  */
1011 static enum xp_retval
1012 xpc_setup_ch_structures_uv(struct xpc_partition *part)
1013 {
1014 	struct xpc_channel_uv *ch_uv;
1015 	int ch_number;
1016 
1017 	for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1018 		ch_uv = &part->channels[ch_number].sn.uv;
1019 
1020 		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1021 		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1022 	}
1023 
1024 	return xpSuccess;
1025 }
1026 
1027 /*
1028  * Teardown the channel structures that are uv specific.
1029  */
1030 static void
1031 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1032 {
1033 	/* nothing needs to be done */
1034 	return;
1035 }
1036 
1037 static enum xp_retval
1038 xpc_make_first_contact_uv(struct xpc_partition *part)
1039 {
1040 	struct xpc_activate_mq_msg_uv msg;
1041 
1042 	/*
1043 	 * We send a sync msg to get the remote partition's remote_act_state
1044 	 * updated to our current act_state which at this point should
1045 	 * be XPC_P_AS_ACTIVATING.
1046 	 */
1047 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1048 				      XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1049 
1050 	while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1051 		 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1052 
1053 		dev_dbg(xpc_part, "waiting to make first contact with "
1054 			"partition %d\n", XPC_PARTID(part));
1055 
1056 		/* wait a 1/4 of a second or so */
1057 		(void)msleep_interruptible(250);
1058 
1059 		if (part->act_state == XPC_P_AS_DEACTIVATING)
1060 			return part->reason;
1061 	}
1062 
1063 	return xpSuccess;
1064 }
1065 
1066 static u64
1067 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1068 {
1069 	unsigned long irq_flags;
1070 	union xpc_channel_ctl_flags chctl;
1071 
1072 	spin_lock_irqsave(&part->chctl_lock, irq_flags);
1073 	chctl = part->chctl;
1074 	if (chctl.all_flags != 0)
1075 		part->chctl.all_flags = 0;
1076 
1077 	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1078 	return chctl.all_flags;
1079 }
1080 
1081 static enum xp_retval
1082 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1083 {
1084 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1085 	struct xpc_send_msg_slot_uv *msg_slot;
1086 	unsigned long irq_flags;
1087 	int nentries;
1088 	int entry;
1089 	size_t nbytes;
1090 
1091 	for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1092 		nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1093 		ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1094 		if (ch_uv->send_msg_slots == NULL)
1095 			continue;
1096 
1097 		for (entry = 0; entry < nentries; entry++) {
1098 			msg_slot = &ch_uv->send_msg_slots[entry];
1099 
1100 			msg_slot->msg_slot_number = entry;
1101 			xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1102 					      &msg_slot->next);
1103 		}
1104 
1105 		spin_lock_irqsave(&ch->lock, irq_flags);
1106 		if (nentries < ch->local_nentries)
1107 			ch->local_nentries = nentries;
1108 		spin_unlock_irqrestore(&ch->lock, irq_flags);
1109 		return xpSuccess;
1110 	}
1111 
1112 	return xpNoMemory;
1113 }
1114 
1115 static enum xp_retval
1116 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1117 {
1118 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1119 	struct xpc_notify_mq_msg_uv *msg_slot;
1120 	unsigned long irq_flags;
1121 	int nentries;
1122 	int entry;
1123 	size_t nbytes;
1124 
1125 	for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1126 		nbytes = nentries * ch->entry_size;
1127 		ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1128 		if (ch_uv->recv_msg_slots == NULL)
1129 			continue;
1130 
1131 		for (entry = 0; entry < nentries; entry++) {
1132 			msg_slot = ch_uv->recv_msg_slots +
1133 			    entry * ch->entry_size;
1134 
1135 			msg_slot->hdr.msg_slot_number = entry;
1136 		}
1137 
1138 		spin_lock_irqsave(&ch->lock, irq_flags);
1139 		if (nentries < ch->remote_nentries)
1140 			ch->remote_nentries = nentries;
1141 		spin_unlock_irqrestore(&ch->lock, irq_flags);
1142 		return xpSuccess;
1143 	}
1144 
1145 	return xpNoMemory;
1146 }
1147 
1148 /*
1149  * Allocate msg_slots associated with the channel.
1150  */
1151 static enum xp_retval
1152 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1153 {
1154 	static enum xp_retval ret;
1155 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1156 
1157 	DBUG_ON(ch->flags & XPC_C_SETUP);
1158 
1159 	ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1160 						   gru_message_queue_desc),
1161 						   GFP_KERNEL);
1162 	if (ch_uv->cached_notify_gru_mq_desc == NULL)
1163 		return xpNoMemory;
1164 
1165 	ret = xpc_allocate_send_msg_slot_uv(ch);
1166 	if (ret == xpSuccess) {
1167 
1168 		ret = xpc_allocate_recv_msg_slot_uv(ch);
1169 		if (ret != xpSuccess) {
1170 			kfree(ch_uv->send_msg_slots);
1171 			xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1172 		}
1173 	}
1174 	return ret;
1175 }
1176 
1177 /*
1178  * Free up msg_slots and clear other stuff that were setup for the specified
1179  * channel.
1180  */
1181 static void
1182 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1183 {
1184 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1185 
1186 	DBUG_ON(!spin_is_locked(&ch->lock));
1187 
1188 	kfree(ch_uv->cached_notify_gru_mq_desc);
1189 	ch_uv->cached_notify_gru_mq_desc = NULL;
1190 
1191 	if (ch->flags & XPC_C_SETUP) {
1192 		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1193 		kfree(ch_uv->send_msg_slots);
1194 		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1195 		kfree(ch_uv->recv_msg_slots);
1196 	}
1197 }
1198 
1199 static void
1200 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1201 {
1202 	struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1203 
1204 	msg.ch_number = ch->number;
1205 	msg.reason = ch->reason;
1206 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1207 				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1208 }
1209 
1210 static void
1211 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1212 {
1213 	struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1214 
1215 	msg.ch_number = ch->number;
1216 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1217 				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1218 }
1219 
1220 static void
1221 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1222 {
1223 	struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1224 
1225 	msg.ch_number = ch->number;
1226 	msg.entry_size = ch->entry_size;
1227 	msg.local_nentries = ch->local_nentries;
1228 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1229 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1230 }
1231 
1232 static void
1233 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1234 {
1235 	struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1236 
1237 	msg.ch_number = ch->number;
1238 	msg.local_nentries = ch->local_nentries;
1239 	msg.remote_nentries = ch->remote_nentries;
1240 	msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1241 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1242 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1243 }
1244 
1245 static void
1246 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1247 {
1248 	struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1249 
1250 	msg.ch_number = ch->number;
1251 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1252 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1253 }
1254 
1255 static void
1256 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1257 {
1258 	unsigned long irq_flags;
1259 
1260 	spin_lock_irqsave(&part->chctl_lock, irq_flags);
1261 	part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1262 	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1263 
1264 	xpc_wakeup_channel_mgr(part);
1265 }
1266 
1267 static enum xp_retval
1268 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1269 			       unsigned long gru_mq_desc_gpa)
1270 {
1271 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1272 
1273 	DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1274 	return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1275 					       gru_mq_desc_gpa);
1276 }
1277 
1278 static void
1279 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1280 {
1281 	struct xpc_activate_mq_msg_uv msg;
1282 
1283 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1284 				      XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1285 }
1286 
1287 static void
1288 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1289 {
1290 	struct xpc_activate_mq_msg_uv msg;
1291 
1292 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1293 				      XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1294 }
1295 
1296 static void
1297 xpc_assume_partition_disengaged_uv(short partid)
1298 {
1299 	struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1300 	unsigned long irq_flags;
1301 
1302 	spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1303 	part_uv->flags &= ~XPC_P_ENGAGED_UV;
1304 	spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1305 }
1306 
1307 static int
1308 xpc_partition_engaged_uv(short partid)
1309 {
1310 	return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1311 }
1312 
1313 static int
1314 xpc_any_partition_engaged_uv(void)
1315 {
1316 	struct xpc_partition_uv *part_uv;
1317 	short partid;
1318 
1319 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1320 		part_uv = &xpc_partitions[partid].sn.uv;
1321 		if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1322 			return 1;
1323 	}
1324 	return 0;
1325 }
1326 
1327 static enum xp_retval
1328 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1329 			 struct xpc_send_msg_slot_uv **address_of_msg_slot)
1330 {
1331 	enum xp_retval ret;
1332 	struct xpc_send_msg_slot_uv *msg_slot;
1333 	struct xpc_fifo_entry_uv *entry;
1334 
1335 	while (1) {
1336 		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1337 		if (entry != NULL)
1338 			break;
1339 
1340 		if (flags & XPC_NOWAIT)
1341 			return xpNoWait;
1342 
1343 		ret = xpc_allocate_msg_wait(ch);
1344 		if (ret != xpInterrupted && ret != xpTimeout)
1345 			return ret;
1346 	}
1347 
1348 	msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1349 	*address_of_msg_slot = msg_slot;
1350 	return xpSuccess;
1351 }
1352 
1353 static void
1354 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1355 		     struct xpc_send_msg_slot_uv *msg_slot)
1356 {
1357 	xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1358 
1359 	/* wakeup anyone waiting for a free msg slot */
1360 	if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1361 		wake_up(&ch->msg_allocate_wq);
1362 }
1363 
1364 static void
1365 xpc_notify_sender_uv(struct xpc_channel *ch,
1366 		     struct xpc_send_msg_slot_uv *msg_slot,
1367 		     enum xp_retval reason)
1368 {
1369 	xpc_notify_func func = msg_slot->func;
1370 
1371 	if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1372 
1373 		atomic_dec(&ch->n_to_notify);
1374 
1375 		dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1376 			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1377 			msg_slot->msg_slot_number, ch->partid, ch->number);
1378 
1379 		func(reason, ch->partid, ch->number, msg_slot->key);
1380 
1381 		dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1382 			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1383 			msg_slot->msg_slot_number, ch->partid, ch->number);
1384 	}
1385 }
1386 
1387 static void
1388 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1389 			    struct xpc_notify_mq_msg_uv *msg)
1390 {
1391 	struct xpc_send_msg_slot_uv *msg_slot;
1392 	int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1393 
1394 	msg_slot = &ch->sn.uv.send_msg_slots[entry];
1395 
1396 	BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1397 	msg_slot->msg_slot_number += ch->local_nentries;
1398 
1399 	if (msg_slot->func != NULL)
1400 		xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1401 
1402 	xpc_free_msg_slot_uv(ch, msg_slot);
1403 }
1404 
1405 static void
1406 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1407 			    struct xpc_notify_mq_msg_uv *msg)
1408 {
1409 	struct xpc_partition_uv *part_uv = &part->sn.uv;
1410 	struct xpc_channel *ch;
1411 	struct xpc_channel_uv *ch_uv;
1412 	struct xpc_notify_mq_msg_uv *msg_slot;
1413 	unsigned long irq_flags;
1414 	int ch_number = msg->hdr.ch_number;
1415 
1416 	if (unlikely(ch_number >= part->nchannels)) {
1417 		dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1418 			"channel number=0x%x in message from partid=%d\n",
1419 			ch_number, XPC_PARTID(part));
1420 
1421 		/* get hb checker to deactivate from the remote partition */
1422 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1423 		if (part_uv->act_state_req == 0)
1424 			xpc_activate_IRQ_rcvd++;
1425 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1426 		part_uv->reason = xpBadChannelNumber;
1427 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1428 
1429 		wake_up_interruptible(&xpc_activate_IRQ_wq);
1430 		return;
1431 	}
1432 
1433 	ch = &part->channels[ch_number];
1434 	xpc_msgqueue_ref(ch);
1435 
1436 	if (!(ch->flags & XPC_C_CONNECTED)) {
1437 		xpc_msgqueue_deref(ch);
1438 		return;
1439 	}
1440 
1441 	/* see if we're really dealing with an ACK for a previously sent msg */
1442 	if (msg->hdr.size == 0) {
1443 		xpc_handle_notify_mq_ack_uv(ch, msg);
1444 		xpc_msgqueue_deref(ch);
1445 		return;
1446 	}
1447 
1448 	/* we're dealing with a normal message sent via the notify_mq */
1449 	ch_uv = &ch->sn.uv;
1450 
1451 	msg_slot = ch_uv->recv_msg_slots +
1452 	    (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1453 
1454 	BUG_ON(msg_slot->hdr.size != 0);
1455 
1456 	memcpy(msg_slot, msg, msg->hdr.size);
1457 
1458 	xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1459 
1460 	if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1461 		/*
1462 		 * If there is an existing idle kthread get it to deliver
1463 		 * the payload, otherwise we'll have to get the channel mgr
1464 		 * for this partition to create a kthread to do the delivery.
1465 		 */
1466 		if (atomic_read(&ch->kthreads_idle) > 0)
1467 			wake_up_nr(&ch->idle_wq, 1);
1468 		else
1469 			xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1470 	}
1471 	xpc_msgqueue_deref(ch);
1472 }
1473 
1474 static irqreturn_t
1475 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1476 {
1477 	struct xpc_notify_mq_msg_uv *msg;
1478 	short partid;
1479 	struct xpc_partition *part;
1480 
1481 	while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1482 	       NULL) {
1483 
1484 		partid = msg->hdr.partid;
1485 		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1486 			dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1487 				"invalid partid=0x%x in message\n", partid);
1488 		} else {
1489 			part = &xpc_partitions[partid];
1490 
1491 			if (xpc_part_ref(part)) {
1492 				xpc_handle_notify_mq_msg_uv(part, msg);
1493 				xpc_part_deref(part);
1494 			}
1495 		}
1496 
1497 		gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1498 	}
1499 
1500 	return IRQ_HANDLED;
1501 }
1502 
1503 static int
1504 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1505 {
1506 	return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1507 }
1508 
1509 static void
1510 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1511 {
1512 	struct xpc_channel *ch = &part->channels[ch_number];
1513 	int ndeliverable_payloads;
1514 
1515 	xpc_msgqueue_ref(ch);
1516 
1517 	ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1518 
1519 	if (ndeliverable_payloads > 0 &&
1520 	    (ch->flags & XPC_C_CONNECTED) &&
1521 	    (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1522 
1523 		xpc_activate_kthreads(ch, ndeliverable_payloads);
1524 	}
1525 
1526 	xpc_msgqueue_deref(ch);
1527 }
1528 
1529 static enum xp_retval
1530 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1531 		    u16 payload_size, u8 notify_type, xpc_notify_func func,
1532 		    void *key)
1533 {
1534 	enum xp_retval ret = xpSuccess;
1535 	struct xpc_send_msg_slot_uv *msg_slot = NULL;
1536 	struct xpc_notify_mq_msg_uv *msg;
1537 	u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1538 	size_t msg_size;
1539 
1540 	DBUG_ON(notify_type != XPC_N_CALL);
1541 
1542 	msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1543 	if (msg_size > ch->entry_size)
1544 		return xpPayloadTooBig;
1545 
1546 	xpc_msgqueue_ref(ch);
1547 
1548 	if (ch->flags & XPC_C_DISCONNECTING) {
1549 		ret = ch->reason;
1550 		goto out_1;
1551 	}
1552 	if (!(ch->flags & XPC_C_CONNECTED)) {
1553 		ret = xpNotConnected;
1554 		goto out_1;
1555 	}
1556 
1557 	ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1558 	if (ret != xpSuccess)
1559 		goto out_1;
1560 
1561 	if (func != NULL) {
1562 		atomic_inc(&ch->n_to_notify);
1563 
1564 		msg_slot->key = key;
1565 		smp_wmb(); /* a non-NULL func must hit memory after the key */
1566 		msg_slot->func = func;
1567 
1568 		if (ch->flags & XPC_C_DISCONNECTING) {
1569 			ret = ch->reason;
1570 			goto out_2;
1571 		}
1572 	}
1573 
1574 	msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1575 	msg->hdr.partid = xp_partition_id;
1576 	msg->hdr.ch_number = ch->number;
1577 	msg->hdr.size = msg_size;
1578 	msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1579 	memcpy(&msg->payload, payload, payload_size);
1580 
1581 	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1582 			       msg_size);
1583 	if (ret == xpSuccess)
1584 		goto out_1;
1585 
1586 	XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1587 out_2:
1588 	if (func != NULL) {
1589 		/*
1590 		 * Try to NULL the msg_slot's func field. If we fail, then
1591 		 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1592 		 * case we need to pretend we succeeded to send the message
1593 		 * since the user will get a callout for the disconnect error
1594 		 * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1595 		 * error returned here will confuse them. Additionally, since
1596 		 * in this case the channel is being disconnected we don't need
1597 		 * to put the the msg_slot back on the free list.
1598 		 */
1599 		if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1600 			ret = xpSuccess;
1601 			goto out_1;
1602 		}
1603 
1604 		msg_slot->key = NULL;
1605 		atomic_dec(&ch->n_to_notify);
1606 	}
1607 	xpc_free_msg_slot_uv(ch, msg_slot);
1608 out_1:
1609 	xpc_msgqueue_deref(ch);
1610 	return ret;
1611 }
1612 
1613 /*
1614  * Tell the callers of xpc_send_notify() that the status of their payloads
1615  * is unknown because the channel is now disconnecting.
1616  *
1617  * We don't worry about putting these msg_slots on the free list since the
1618  * msg_slots themselves are about to be kfree'd.
1619  */
1620 static void
1621 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1622 {
1623 	struct xpc_send_msg_slot_uv *msg_slot;
1624 	int entry;
1625 
1626 	DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1627 
1628 	for (entry = 0; entry < ch->local_nentries; entry++) {
1629 
1630 		if (atomic_read(&ch->n_to_notify) == 0)
1631 			break;
1632 
1633 		msg_slot = &ch->sn.uv.send_msg_slots[entry];
1634 		if (msg_slot->func != NULL)
1635 			xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1636 	}
1637 }
1638 
1639 /*
1640  * Get the next deliverable message's payload.
1641  */
1642 static void *
1643 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1644 {
1645 	struct xpc_fifo_entry_uv *entry;
1646 	struct xpc_notify_mq_msg_uv *msg;
1647 	void *payload = NULL;
1648 
1649 	if (!(ch->flags & XPC_C_DISCONNECTING)) {
1650 		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1651 		if (entry != NULL) {
1652 			msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1653 					   hdr.u.next);
1654 			payload = &msg->payload;
1655 		}
1656 	}
1657 	return payload;
1658 }
1659 
1660 static void
1661 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1662 {
1663 	struct xpc_notify_mq_msg_uv *msg;
1664 	enum xp_retval ret;
1665 
1666 	msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1667 
1668 	/* return an ACK to the sender of this message */
1669 
1670 	msg->hdr.partid = xp_partition_id;
1671 	msg->hdr.size = 0;	/* size of zero indicates this is an ACK */
1672 
1673 	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1674 			       sizeof(struct xpc_notify_mq_msghdr_uv));
1675 	if (ret != xpSuccess)
1676 		XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1677 }
1678 
1679 static struct xpc_arch_operations xpc_arch_ops_uv = {
1680 	.setup_partitions = xpc_setup_partitions_uv,
1681 	.teardown_partitions = xpc_teardown_partitions_uv,
1682 	.process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1683 	.get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1684 	.setup_rsvd_page = xpc_setup_rsvd_page_uv,
1685 
1686 	.allow_hb = xpc_allow_hb_uv,
1687 	.disallow_hb = xpc_disallow_hb_uv,
1688 	.disallow_all_hbs = xpc_disallow_all_hbs_uv,
1689 	.increment_heartbeat = xpc_increment_heartbeat_uv,
1690 	.offline_heartbeat = xpc_offline_heartbeat_uv,
1691 	.online_heartbeat = xpc_online_heartbeat_uv,
1692 	.heartbeat_init = xpc_heartbeat_init_uv,
1693 	.heartbeat_exit = xpc_heartbeat_exit_uv,
1694 	.get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1695 
1696 	.request_partition_activation =
1697 		xpc_request_partition_activation_uv,
1698 	.request_partition_reactivation =
1699 		xpc_request_partition_reactivation_uv,
1700 	.request_partition_deactivation =
1701 		xpc_request_partition_deactivation_uv,
1702 	.cancel_partition_deactivation_request =
1703 		xpc_cancel_partition_deactivation_request_uv,
1704 
1705 	.setup_ch_structures = xpc_setup_ch_structures_uv,
1706 	.teardown_ch_structures = xpc_teardown_ch_structures_uv,
1707 
1708 	.make_first_contact = xpc_make_first_contact_uv,
1709 
1710 	.get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1711 	.send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1712 	.send_chctl_closereply = xpc_send_chctl_closereply_uv,
1713 	.send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1714 	.send_chctl_openreply = xpc_send_chctl_openreply_uv,
1715 	.send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1716 	.process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1717 
1718 	.save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1719 
1720 	.setup_msg_structures = xpc_setup_msg_structures_uv,
1721 	.teardown_msg_structures = xpc_teardown_msg_structures_uv,
1722 
1723 	.indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1724 	.indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1725 	.assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1726 	.partition_engaged = xpc_partition_engaged_uv,
1727 	.any_partition_engaged = xpc_any_partition_engaged_uv,
1728 
1729 	.n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1730 	.send_payload = xpc_send_payload_uv,
1731 	.get_deliverable_payload = xpc_get_deliverable_payload_uv,
1732 	.received_payload = xpc_received_payload_uv,
1733 	.notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1734 };
1735 
1736 static int
1737 xpc_init_mq_node(int nid)
1738 {
1739 	int cpu;
1740 
1741 	get_online_cpus();
1742 
1743 	for_each_cpu(cpu, cpumask_of_node(nid)) {
1744 		xpc_activate_mq_uv =
1745 			xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
1746 					     XPC_ACTIVATE_IRQ_NAME,
1747 					     xpc_handle_activate_IRQ_uv);
1748 		if (!IS_ERR(xpc_activate_mq_uv))
1749 			break;
1750 	}
1751 	if (IS_ERR(xpc_activate_mq_uv)) {
1752 		put_online_cpus();
1753 		return PTR_ERR(xpc_activate_mq_uv);
1754 	}
1755 
1756 	for_each_cpu(cpu, cpumask_of_node(nid)) {
1757 		xpc_notify_mq_uv =
1758 			xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
1759 					     XPC_NOTIFY_IRQ_NAME,
1760 					     xpc_handle_notify_IRQ_uv);
1761 		if (!IS_ERR(xpc_notify_mq_uv))
1762 			break;
1763 	}
1764 	if (IS_ERR(xpc_notify_mq_uv)) {
1765 		xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1766 		put_online_cpus();
1767 		return PTR_ERR(xpc_notify_mq_uv);
1768 	}
1769 
1770 	put_online_cpus();
1771 	return 0;
1772 }
1773 
1774 int
1775 xpc_init_uv(void)
1776 {
1777 	int nid;
1778 	int ret = 0;
1779 
1780 	xpc_arch_ops = xpc_arch_ops_uv;
1781 
1782 	if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1783 		dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1784 			XPC_MSG_HDR_MAX_SIZE);
1785 		return -E2BIG;
1786 	}
1787 
1788 	if (xpc_mq_node < 0)
1789 		for_each_online_node(nid) {
1790 			ret = xpc_init_mq_node(nid);
1791 
1792 			if (!ret)
1793 				break;
1794 		}
1795 	else
1796 		ret = xpc_init_mq_node(xpc_mq_node);
1797 
1798 	if (ret < 0)
1799 		dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1800 			-ret);
1801 
1802 	return ret;
1803 }
1804 
1805 void
1806 xpc_exit_uv(void)
1807 {
1808 	xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1809 	xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1810 }
1811 
1812 module_param(xpc_mq_node, int, 0);
1813 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");
1814