xref: /openbmc/linux/drivers/misc/sgi-xp/xpc_uv.c (revision 5a158981)
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 #if defined CONFIG_X86_64
28 #include <asm/uv/bios.h>
29 #include <asm/uv/uv_irq.h>
30 #elif defined CONFIG_IA64_SGI_UV
31 #include <asm/sn/intr.h>
32 #include <asm/sn/sn_sal.h>
33 #endif
34 #include "../sgi-gru/gru.h"
35 #include "../sgi-gru/grukservices.h"
36 #include "xpc.h"
37 
38 #if defined CONFIG_IA64_SGI_UV
39 struct uv_IO_APIC_route_entry {
40 	__u64	vector		:  8,
41 		delivery_mode	:  3,
42 		dest_mode	:  1,
43 		delivery_status	:  1,
44 		polarity	:  1,
45 		__reserved_1	:  1,
46 		trigger		:  1,
47 		mask		:  1,
48 		__reserved_2	: 15,
49 		dest		: 32;
50 };
51 
52 #define sn_partition_id 0
53 #endif
54 
55 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
56 
57 #define XPC_ACTIVATE_MSG_SIZE_UV	(1 * GRU_CACHE_LINE_BYTES)
58 #define XPC_ACTIVATE_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
59 					 XPC_ACTIVATE_MSG_SIZE_UV)
60 #define XPC_ACTIVATE_IRQ_NAME		"xpc_activate"
61 
62 #define XPC_NOTIFY_MSG_SIZE_UV		(2 * GRU_CACHE_LINE_BYTES)
63 #define XPC_NOTIFY_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
64 					 XPC_NOTIFY_MSG_SIZE_UV)
65 #define XPC_NOTIFY_IRQ_NAME		"xpc_notify"
66 
67 static int xpc_mq_node = NUMA_NO_NODE;
68 
69 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
70 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
71 
72 static int
73 xpc_setup_partitions_uv(void)
74 {
75 	short partid;
76 	struct xpc_partition_uv *part_uv;
77 
78 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
79 		part_uv = &xpc_partitions[partid].sn.uv;
80 
81 		mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
82 		spin_lock_init(&part_uv->flags_lock);
83 		part_uv->remote_act_state = XPC_P_AS_INACTIVE;
84 	}
85 	return 0;
86 }
87 
88 static void
89 xpc_teardown_partitions_uv(void)
90 {
91 	short partid;
92 	struct xpc_partition_uv *part_uv;
93 	unsigned long irq_flags;
94 
95 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
96 		part_uv = &xpc_partitions[partid].sn.uv;
97 
98 		if (part_uv->cached_activate_gru_mq_desc != NULL) {
99 			mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
100 			spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
101 			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
102 			spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
103 			kfree(part_uv->cached_activate_gru_mq_desc);
104 			part_uv->cached_activate_gru_mq_desc = NULL;
105 			mutex_unlock(&part_uv->
106 				     cached_activate_gru_mq_desc_mutex);
107 		}
108 	}
109 }
110 
111 static int
112 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
113 {
114 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
115 
116 #if defined CONFIG_X86_64
117 	mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
118 			UV_AFFINITY_CPU);
119 	if (mq->irq < 0)
120 		return mq->irq;
121 
122 	mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
123 
124 #elif defined CONFIG_IA64_SGI_UV
125 	if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
126 		mq->irq = SGI_XPC_ACTIVATE;
127 	else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
128 		mq->irq = SGI_XPC_NOTIFY;
129 	else
130 		return -EINVAL;
131 
132 	mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
133 	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
134 #else
135 	#error not a supported configuration
136 #endif
137 
138 	return 0;
139 }
140 
141 static void
142 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
143 {
144 #if defined CONFIG_X86_64
145 	uv_teardown_irq(mq->irq);
146 
147 #elif defined CONFIG_IA64_SGI_UV
148 	int mmr_pnode;
149 	unsigned long mmr_value;
150 
151 	mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
152 	mmr_value = 1UL << 16;
153 
154 	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
155 #else
156 	#error not a supported configuration
157 #endif
158 }
159 
160 static int
161 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
162 {
163 	int ret;
164 
165 #if defined CONFIG_IA64_SGI_UV
166 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
167 
168 	ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
169 				    mq->order, &mq->mmr_offset);
170 	if (ret < 0) {
171 		dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
172 			ret);
173 		return -EBUSY;
174 	}
175 #elif defined CONFIG_X86_64
176 	ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
177 					 mq->order, &mq->mmr_offset);
178 	if (ret < 0) {
179 		dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
180 			"ret=%d\n", ret);
181 		return ret;
182 	}
183 #else
184 	#error not a supported configuration
185 #endif
186 
187 	mq->watchlist_num = ret;
188 	return 0;
189 }
190 
191 static void
192 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
193 {
194 	int ret;
195 	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
196 
197 #if defined CONFIG_X86_64
198 	ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
199 	BUG_ON(ret != BIOS_STATUS_SUCCESS);
200 #elif defined CONFIG_IA64_SGI_UV
201 	ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
202 	BUG_ON(ret != SALRET_OK);
203 #else
204 	#error not a supported configuration
205 #endif
206 }
207 
208 static struct xpc_gru_mq_uv *
209 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
210 		     irq_handler_t irq_handler)
211 {
212 	enum xp_retval xp_ret;
213 	int ret;
214 	int nid;
215 	int nasid;
216 	int pg_order;
217 	struct page *page;
218 	struct xpc_gru_mq_uv *mq;
219 	struct uv_IO_APIC_route_entry *mmr_value;
220 
221 	mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
222 	if (mq == NULL) {
223 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
224 			"a xpc_gru_mq_uv structure\n");
225 		ret = -ENOMEM;
226 		goto out_0;
227 	}
228 
229 	mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
230 				  GFP_KERNEL);
231 	if (mq->gru_mq_desc == NULL) {
232 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
233 			"a gru_message_queue_desc structure\n");
234 		ret = -ENOMEM;
235 		goto out_1;
236 	}
237 
238 	pg_order = get_order(mq_size);
239 	mq->order = pg_order + PAGE_SHIFT;
240 	mq_size = 1UL << mq->order;
241 
242 	mq->mmr_blade = uv_cpu_to_blade_id(cpu);
243 
244 	nid = cpu_to_node(cpu);
245 	page = __alloc_pages_node(nid,
246 				      GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
247 				      pg_order);
248 	if (page == NULL) {
249 		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
250 			"bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
251 		ret = -ENOMEM;
252 		goto out_2;
253 	}
254 	mq->address = page_address(page);
255 
256 	/* enable generation of irq when GRU mq operation occurs to this mq */
257 	ret = xpc_gru_mq_watchlist_alloc_uv(mq);
258 	if (ret != 0)
259 		goto out_3;
260 
261 	ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
262 	if (ret != 0)
263 		goto out_4;
264 
265 	ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
266 	if (ret != 0) {
267 		dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
268 			mq->irq, -ret);
269 		goto out_5;
270 	}
271 
272 	nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
273 
274 	mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
275 	ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
276 				     nasid, mmr_value->vector, mmr_value->dest);
277 	if (ret != 0) {
278 		dev_err(xpc_part, "gru_create_message_queue() returned "
279 			"error=%d\n", ret);
280 		ret = -EINVAL;
281 		goto out_6;
282 	}
283 
284 	/* allow other partitions to access this GRU mq */
285 	xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
286 	if (xp_ret != xpSuccess) {
287 		ret = -EACCES;
288 		goto out_6;
289 	}
290 
291 	return mq;
292 
293 	/* something went wrong */
294 out_6:
295 	free_irq(mq->irq, NULL);
296 out_5:
297 	xpc_release_gru_mq_irq_uv(mq);
298 out_4:
299 	xpc_gru_mq_watchlist_free_uv(mq);
300 out_3:
301 	free_pages((unsigned long)mq->address, pg_order);
302 out_2:
303 	kfree(mq->gru_mq_desc);
304 out_1:
305 	kfree(mq);
306 out_0:
307 	return ERR_PTR(ret);
308 }
309 
310 static void
311 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
312 {
313 	unsigned int mq_size;
314 	int pg_order;
315 	int ret;
316 
317 	/* disallow other partitions to access GRU mq */
318 	mq_size = 1UL << mq->order;
319 	ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
320 	BUG_ON(ret != xpSuccess);
321 
322 	/* unregister irq handler and release mq irq/vector mapping */
323 	free_irq(mq->irq, NULL);
324 	xpc_release_gru_mq_irq_uv(mq);
325 
326 	/* disable generation of irq when GRU mq op occurs to this mq */
327 	xpc_gru_mq_watchlist_free_uv(mq);
328 
329 	pg_order = mq->order - PAGE_SHIFT;
330 	free_pages((unsigned long)mq->address, pg_order);
331 
332 	kfree(mq);
333 }
334 
335 static enum xp_retval
336 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
337 		 size_t msg_size)
338 {
339 	enum xp_retval xp_ret;
340 	int ret;
341 
342 	while (1) {
343 		ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
344 		if (ret == MQE_OK) {
345 			xp_ret = xpSuccess;
346 			break;
347 		}
348 
349 		if (ret == MQE_QUEUE_FULL) {
350 			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
351 				"error=MQE_QUEUE_FULL\n");
352 			/* !!! handle QLimit reached; delay & try again */
353 			/* ??? Do we add a limit to the number of retries? */
354 			(void)msleep_interruptible(10);
355 		} else if (ret == MQE_CONGESTION) {
356 			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
357 				"error=MQE_CONGESTION\n");
358 			/* !!! handle LB Overflow; simply try again */
359 			/* ??? Do we add a limit to the number of retries? */
360 		} else {
361 			/* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
362 			dev_err(xpc_chan, "gru_send_message_gpa() returned "
363 				"error=%d\n", ret);
364 			xp_ret = xpGruSendMqError;
365 			break;
366 		}
367 	}
368 	return xp_ret;
369 }
370 
371 static void
372 xpc_process_activate_IRQ_rcvd_uv(void)
373 {
374 	unsigned long irq_flags;
375 	short partid;
376 	struct xpc_partition *part;
377 	u8 act_state_req;
378 
379 	DBUG_ON(xpc_activate_IRQ_rcvd == 0);
380 
381 	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
382 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
383 		part = &xpc_partitions[partid];
384 
385 		if (part->sn.uv.act_state_req == 0)
386 			continue;
387 
388 		xpc_activate_IRQ_rcvd--;
389 		BUG_ON(xpc_activate_IRQ_rcvd < 0);
390 
391 		act_state_req = part->sn.uv.act_state_req;
392 		part->sn.uv.act_state_req = 0;
393 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
394 
395 		if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
396 			if (part->act_state == XPC_P_AS_INACTIVE)
397 				xpc_activate_partition(part);
398 			else if (part->act_state == XPC_P_AS_DEACTIVATING)
399 				XPC_DEACTIVATE_PARTITION(part, xpReactivating);
400 
401 		} else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
402 			if (part->act_state == XPC_P_AS_INACTIVE)
403 				xpc_activate_partition(part);
404 			else
405 				XPC_DEACTIVATE_PARTITION(part, xpReactivating);
406 
407 		} else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
408 			XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
409 
410 		} else {
411 			BUG();
412 		}
413 
414 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
415 		if (xpc_activate_IRQ_rcvd == 0)
416 			break;
417 	}
418 	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
419 
420 }
421 
422 static void
423 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
424 			      struct xpc_activate_mq_msghdr_uv *msg_hdr,
425 			      int part_setup,
426 			      int *wakeup_hb_checker)
427 {
428 	unsigned long irq_flags;
429 	struct xpc_partition_uv *part_uv = &part->sn.uv;
430 	struct xpc_openclose_args *args;
431 
432 	part_uv->remote_act_state = msg_hdr->act_state;
433 
434 	switch (msg_hdr->type) {
435 	case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
436 		/* syncing of remote_act_state was just done above */
437 		break;
438 
439 	case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
440 		struct xpc_activate_mq_msg_activate_req_uv *msg;
441 
442 		/*
443 		 * ??? Do we deal here with ts_jiffies being different
444 		 * ??? if act_state != XPC_P_AS_INACTIVE instead of
445 		 * ??? below?
446 		 */
447 		msg = container_of(msg_hdr, struct
448 				   xpc_activate_mq_msg_activate_req_uv, hdr);
449 
450 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
451 		if (part_uv->act_state_req == 0)
452 			xpc_activate_IRQ_rcvd++;
453 		part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
454 		part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
455 		part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
456 		part_uv->heartbeat_gpa = msg->heartbeat_gpa;
457 
458 		if (msg->activate_gru_mq_desc_gpa !=
459 		    part_uv->activate_gru_mq_desc_gpa) {
460 			spin_lock(&part_uv->flags_lock);
461 			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
462 			spin_unlock(&part_uv->flags_lock);
463 			part_uv->activate_gru_mq_desc_gpa =
464 			    msg->activate_gru_mq_desc_gpa;
465 		}
466 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
467 
468 		(*wakeup_hb_checker)++;
469 		break;
470 	}
471 	case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
472 		struct xpc_activate_mq_msg_deactivate_req_uv *msg;
473 
474 		msg = container_of(msg_hdr, struct
475 				   xpc_activate_mq_msg_deactivate_req_uv, hdr);
476 
477 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
478 		if (part_uv->act_state_req == 0)
479 			xpc_activate_IRQ_rcvd++;
480 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
481 		part_uv->reason = msg->reason;
482 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
483 
484 		(*wakeup_hb_checker)++;
485 		return;
486 	}
487 	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
488 		struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
489 
490 		if (!part_setup)
491 			break;
492 
493 		msg = container_of(msg_hdr, struct
494 				   xpc_activate_mq_msg_chctl_closerequest_uv,
495 				   hdr);
496 		args = &part->remote_openclose_args[msg->ch_number];
497 		args->reason = msg->reason;
498 
499 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
500 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
501 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
502 
503 		xpc_wakeup_channel_mgr(part);
504 		break;
505 	}
506 	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
507 		struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
508 
509 		if (!part_setup)
510 			break;
511 
512 		msg = container_of(msg_hdr, struct
513 				   xpc_activate_mq_msg_chctl_closereply_uv,
514 				   hdr);
515 
516 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
517 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
518 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
519 
520 		xpc_wakeup_channel_mgr(part);
521 		break;
522 	}
523 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
524 		struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
525 
526 		if (!part_setup)
527 			break;
528 
529 		msg = container_of(msg_hdr, struct
530 				   xpc_activate_mq_msg_chctl_openrequest_uv,
531 				   hdr);
532 		args = &part->remote_openclose_args[msg->ch_number];
533 		args->entry_size = msg->entry_size;
534 		args->local_nentries = msg->local_nentries;
535 
536 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
537 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
538 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
539 
540 		xpc_wakeup_channel_mgr(part);
541 		break;
542 	}
543 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
544 		struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
545 
546 		if (!part_setup)
547 			break;
548 
549 		msg = container_of(msg_hdr, struct
550 				   xpc_activate_mq_msg_chctl_openreply_uv, hdr);
551 		args = &part->remote_openclose_args[msg->ch_number];
552 		args->remote_nentries = msg->remote_nentries;
553 		args->local_nentries = msg->local_nentries;
554 		args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
555 
556 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
557 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
558 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
559 
560 		xpc_wakeup_channel_mgr(part);
561 		break;
562 	}
563 	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
564 		struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
565 
566 		if (!part_setup)
567 			break;
568 
569 		msg = container_of(msg_hdr, struct
570 				xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
571 		spin_lock_irqsave(&part->chctl_lock, irq_flags);
572 		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
573 		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
574 
575 		xpc_wakeup_channel_mgr(part);
576 	}
577 		/* fall through */
578 	case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
579 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
580 		part_uv->flags |= XPC_P_ENGAGED_UV;
581 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
582 		break;
583 
584 	case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
585 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
586 		part_uv->flags &= ~XPC_P_ENGAGED_UV;
587 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
588 		break;
589 
590 	default:
591 		dev_err(xpc_part, "received unknown activate_mq msg type=%d "
592 			"from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
593 
594 		/* get hb checker to deactivate from the remote partition */
595 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
596 		if (part_uv->act_state_req == 0)
597 			xpc_activate_IRQ_rcvd++;
598 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
599 		part_uv->reason = xpBadMsgType;
600 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
601 
602 		(*wakeup_hb_checker)++;
603 		return;
604 	}
605 
606 	if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
607 	    part->remote_rp_ts_jiffies != 0) {
608 		/*
609 		 * ??? Does what we do here need to be sensitive to
610 		 * ??? act_state or remote_act_state?
611 		 */
612 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
613 		if (part_uv->act_state_req == 0)
614 			xpc_activate_IRQ_rcvd++;
615 		part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
616 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
617 
618 		(*wakeup_hb_checker)++;
619 	}
620 }
621 
622 static irqreturn_t
623 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
624 {
625 	struct xpc_activate_mq_msghdr_uv *msg_hdr;
626 	short partid;
627 	struct xpc_partition *part;
628 	int wakeup_hb_checker = 0;
629 	int part_referenced;
630 
631 	while (1) {
632 		msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
633 		if (msg_hdr == NULL)
634 			break;
635 
636 		partid = msg_hdr->partid;
637 		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
638 			dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
639 				"received invalid partid=0x%x in message\n",
640 				partid);
641 		} else {
642 			part = &xpc_partitions[partid];
643 
644 			part_referenced = xpc_part_ref(part);
645 			xpc_handle_activate_mq_msg_uv(part, msg_hdr,
646 						      part_referenced,
647 						      &wakeup_hb_checker);
648 			if (part_referenced)
649 				xpc_part_deref(part);
650 		}
651 
652 		gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
653 	}
654 
655 	if (wakeup_hb_checker)
656 		wake_up_interruptible(&xpc_activate_IRQ_wq);
657 
658 	return IRQ_HANDLED;
659 }
660 
661 static enum xp_retval
662 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
663 				unsigned long gru_mq_desc_gpa)
664 {
665 	enum xp_retval ret;
666 
667 	ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
668 			       sizeof(struct gru_message_queue_desc));
669 	if (ret == xpSuccess)
670 		gru_mq_desc->mq = NULL;
671 
672 	return ret;
673 }
674 
675 static enum xp_retval
676 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
677 			 int msg_type)
678 {
679 	struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
680 	struct xpc_partition_uv *part_uv = &part->sn.uv;
681 	struct gru_message_queue_desc *gru_mq_desc;
682 	unsigned long irq_flags;
683 	enum xp_retval ret;
684 
685 	DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
686 
687 	msg_hdr->type = msg_type;
688 	msg_hdr->partid = xp_partition_id;
689 	msg_hdr->act_state = part->act_state;
690 	msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
691 
692 	mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
693 again:
694 	if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
695 		gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
696 		if (gru_mq_desc == NULL) {
697 			gru_mq_desc = kmalloc(sizeof(struct
698 					      gru_message_queue_desc),
699 					      GFP_ATOMIC);
700 			if (gru_mq_desc == NULL) {
701 				ret = xpNoMemory;
702 				goto done;
703 			}
704 			part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
705 		}
706 
707 		ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
708 						      part_uv->
709 						      activate_gru_mq_desc_gpa);
710 		if (ret != xpSuccess)
711 			goto done;
712 
713 		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
714 		part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
715 		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
716 	}
717 
718 	/* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
719 	ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
720 			       msg_size);
721 	if (ret != xpSuccess) {
722 		smp_rmb();	/* ensure a fresh copy of part_uv->flags */
723 		if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
724 			goto again;
725 	}
726 done:
727 	mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
728 	return ret;
729 }
730 
731 static void
732 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
733 			      size_t msg_size, int msg_type)
734 {
735 	enum xp_retval ret;
736 
737 	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
738 	if (unlikely(ret != xpSuccess))
739 		XPC_DEACTIVATE_PARTITION(part, ret);
740 }
741 
742 static void
743 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
744 			 void *msg, size_t msg_size, int msg_type)
745 {
746 	struct xpc_partition *part = &xpc_partitions[ch->partid];
747 	enum xp_retval ret;
748 
749 	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
750 	if (unlikely(ret != xpSuccess)) {
751 		if (irq_flags != NULL)
752 			spin_unlock_irqrestore(&ch->lock, *irq_flags);
753 
754 		XPC_DEACTIVATE_PARTITION(part, ret);
755 
756 		if (irq_flags != NULL)
757 			spin_lock_irqsave(&ch->lock, *irq_flags);
758 	}
759 }
760 
761 static void
762 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
763 {
764 	unsigned long irq_flags;
765 	struct xpc_partition_uv *part_uv = &part->sn.uv;
766 
767 	/*
768 	 * !!! Make our side think that the remote partition sent an activate
769 	 * !!! mq message our way by doing what the activate IRQ handler would
770 	 * !!! do had one really been sent.
771 	 */
772 
773 	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
774 	if (part_uv->act_state_req == 0)
775 		xpc_activate_IRQ_rcvd++;
776 	part_uv->act_state_req = act_state_req;
777 	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
778 
779 	wake_up_interruptible(&xpc_activate_IRQ_wq);
780 }
781 
782 static enum xp_retval
783 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
784 				  size_t *len)
785 {
786 	s64 status;
787 	enum xp_retval ret;
788 
789 #if defined CONFIG_X86_64
790 	status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
791 					  (u64 *)len);
792 	if (status == BIOS_STATUS_SUCCESS)
793 		ret = xpSuccess;
794 	else if (status == BIOS_STATUS_MORE_PASSES)
795 		ret = xpNeedMoreInfo;
796 	else
797 		ret = xpBiosError;
798 
799 #elif defined CONFIG_IA64_SGI_UV
800 	status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
801 	if (status == SALRET_OK)
802 		ret = xpSuccess;
803 	else if (status == SALRET_MORE_PASSES)
804 		ret = xpNeedMoreInfo;
805 	else
806 		ret = xpSalError;
807 
808 #else
809 	#error not a supported configuration
810 #endif
811 
812 	return ret;
813 }
814 
815 static int
816 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
817 {
818 	xpc_heartbeat_uv =
819 	    &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
820 	rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
821 	rp->sn.uv.activate_gru_mq_desc_gpa =
822 	    uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
823 	return 0;
824 }
825 
826 static void
827 xpc_allow_hb_uv(short partid)
828 {
829 }
830 
831 static void
832 xpc_disallow_hb_uv(short partid)
833 {
834 }
835 
836 static void
837 xpc_disallow_all_hbs_uv(void)
838 {
839 }
840 
841 static void
842 xpc_increment_heartbeat_uv(void)
843 {
844 	xpc_heartbeat_uv->value++;
845 }
846 
847 static void
848 xpc_offline_heartbeat_uv(void)
849 {
850 	xpc_increment_heartbeat_uv();
851 	xpc_heartbeat_uv->offline = 1;
852 }
853 
854 static void
855 xpc_online_heartbeat_uv(void)
856 {
857 	xpc_increment_heartbeat_uv();
858 	xpc_heartbeat_uv->offline = 0;
859 }
860 
861 static void
862 xpc_heartbeat_init_uv(void)
863 {
864 	xpc_heartbeat_uv->value = 1;
865 	xpc_heartbeat_uv->offline = 0;
866 }
867 
868 static void
869 xpc_heartbeat_exit_uv(void)
870 {
871 	xpc_offline_heartbeat_uv();
872 }
873 
874 static enum xp_retval
875 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
876 {
877 	struct xpc_partition_uv *part_uv = &part->sn.uv;
878 	enum xp_retval ret;
879 
880 	ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
881 			       part_uv->heartbeat_gpa,
882 			       sizeof(struct xpc_heartbeat_uv));
883 	if (ret != xpSuccess)
884 		return ret;
885 
886 	if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
887 	    !part_uv->cached_heartbeat.offline) {
888 
889 		ret = xpNoHeartbeat;
890 	} else {
891 		part->last_heartbeat = part_uv->cached_heartbeat.value;
892 	}
893 	return ret;
894 }
895 
896 static void
897 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
898 				    unsigned long remote_rp_gpa, int nasid)
899 {
900 	short partid = remote_rp->SAL_partid;
901 	struct xpc_partition *part = &xpc_partitions[partid];
902 	struct xpc_activate_mq_msg_activate_req_uv msg;
903 
904 	part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
905 	part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
906 	part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
907 	part->sn.uv.activate_gru_mq_desc_gpa =
908 	    remote_rp->sn.uv.activate_gru_mq_desc_gpa;
909 
910 	/*
911 	 * ??? Is it a good idea to make this conditional on what is
912 	 * ??? potentially stale state information?
913 	 */
914 	if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
915 		msg.rp_gpa = uv_gpa(xpc_rsvd_page);
916 		msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
917 		msg.activate_gru_mq_desc_gpa =
918 		    xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
919 		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
920 					   XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
921 	}
922 
923 	if (part->act_state == XPC_P_AS_INACTIVE)
924 		xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
925 }
926 
927 static void
928 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
929 {
930 	xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
931 }
932 
933 static void
934 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
935 {
936 	struct xpc_activate_mq_msg_deactivate_req_uv msg;
937 
938 	/*
939 	 * ??? Is it a good idea to make this conditional on what is
940 	 * ??? potentially stale state information?
941 	 */
942 	if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
943 	    part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
944 
945 		msg.reason = part->reason;
946 		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
947 					 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
948 	}
949 }
950 
951 static void
952 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
953 {
954 	/* nothing needs to be done */
955 	return;
956 }
957 
958 static void
959 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
960 {
961 	head->first = NULL;
962 	head->last = NULL;
963 	spin_lock_init(&head->lock);
964 	head->n_entries = 0;
965 }
966 
967 static void *
968 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
969 {
970 	unsigned long irq_flags;
971 	struct xpc_fifo_entry_uv *first;
972 
973 	spin_lock_irqsave(&head->lock, irq_flags);
974 	first = head->first;
975 	if (head->first != NULL) {
976 		head->first = first->next;
977 		if (head->first == NULL)
978 			head->last = NULL;
979 
980 		head->n_entries--;
981 		BUG_ON(head->n_entries < 0);
982 
983 		first->next = NULL;
984 	}
985 	spin_unlock_irqrestore(&head->lock, irq_flags);
986 	return first;
987 }
988 
989 static void
990 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
991 		      struct xpc_fifo_entry_uv *last)
992 {
993 	unsigned long irq_flags;
994 
995 	last->next = NULL;
996 	spin_lock_irqsave(&head->lock, irq_flags);
997 	if (head->last != NULL)
998 		head->last->next = last;
999 	else
1000 		head->first = last;
1001 	head->last = last;
1002 	head->n_entries++;
1003 	spin_unlock_irqrestore(&head->lock, irq_flags);
1004 }
1005 
1006 static int
1007 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
1008 {
1009 	return head->n_entries;
1010 }
1011 
1012 /*
1013  * Setup the channel structures that are uv specific.
1014  */
1015 static enum xp_retval
1016 xpc_setup_ch_structures_uv(struct xpc_partition *part)
1017 {
1018 	struct xpc_channel_uv *ch_uv;
1019 	int ch_number;
1020 
1021 	for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1022 		ch_uv = &part->channels[ch_number].sn.uv;
1023 
1024 		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1025 		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1026 	}
1027 
1028 	return xpSuccess;
1029 }
1030 
1031 /*
1032  * Teardown the channel structures that are uv specific.
1033  */
1034 static void
1035 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1036 {
1037 	/* nothing needs to be done */
1038 	return;
1039 }
1040 
1041 static enum xp_retval
1042 xpc_make_first_contact_uv(struct xpc_partition *part)
1043 {
1044 	struct xpc_activate_mq_msg_uv msg;
1045 
1046 	/*
1047 	 * We send a sync msg to get the remote partition's remote_act_state
1048 	 * updated to our current act_state which at this point should
1049 	 * be XPC_P_AS_ACTIVATING.
1050 	 */
1051 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1052 				      XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1053 
1054 	while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1055 		 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1056 
1057 		dev_dbg(xpc_part, "waiting to make first contact with "
1058 			"partition %d\n", XPC_PARTID(part));
1059 
1060 		/* wait a 1/4 of a second or so */
1061 		(void)msleep_interruptible(250);
1062 
1063 		if (part->act_state == XPC_P_AS_DEACTIVATING)
1064 			return part->reason;
1065 	}
1066 
1067 	return xpSuccess;
1068 }
1069 
1070 static u64
1071 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1072 {
1073 	unsigned long irq_flags;
1074 	union xpc_channel_ctl_flags chctl;
1075 
1076 	spin_lock_irqsave(&part->chctl_lock, irq_flags);
1077 	chctl = part->chctl;
1078 	if (chctl.all_flags != 0)
1079 		part->chctl.all_flags = 0;
1080 
1081 	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1082 	return chctl.all_flags;
1083 }
1084 
1085 static enum xp_retval
1086 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1087 {
1088 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1089 	struct xpc_send_msg_slot_uv *msg_slot;
1090 	unsigned long irq_flags;
1091 	int nentries;
1092 	int entry;
1093 	size_t nbytes;
1094 
1095 	for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1096 		nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1097 		ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1098 		if (ch_uv->send_msg_slots == NULL)
1099 			continue;
1100 
1101 		for (entry = 0; entry < nentries; entry++) {
1102 			msg_slot = &ch_uv->send_msg_slots[entry];
1103 
1104 			msg_slot->msg_slot_number = entry;
1105 			xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1106 					      &msg_slot->next);
1107 		}
1108 
1109 		spin_lock_irqsave(&ch->lock, irq_flags);
1110 		if (nentries < ch->local_nentries)
1111 			ch->local_nentries = nentries;
1112 		spin_unlock_irqrestore(&ch->lock, irq_flags);
1113 		return xpSuccess;
1114 	}
1115 
1116 	return xpNoMemory;
1117 }
1118 
1119 static enum xp_retval
1120 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1121 {
1122 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1123 	struct xpc_notify_mq_msg_uv *msg_slot;
1124 	unsigned long irq_flags;
1125 	int nentries;
1126 	int entry;
1127 	size_t nbytes;
1128 
1129 	for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1130 		nbytes = nentries * ch->entry_size;
1131 		ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1132 		if (ch_uv->recv_msg_slots == NULL)
1133 			continue;
1134 
1135 		for (entry = 0; entry < nentries; entry++) {
1136 			msg_slot = ch_uv->recv_msg_slots +
1137 			    entry * ch->entry_size;
1138 
1139 			msg_slot->hdr.msg_slot_number = entry;
1140 		}
1141 
1142 		spin_lock_irqsave(&ch->lock, irq_flags);
1143 		if (nentries < ch->remote_nentries)
1144 			ch->remote_nentries = nentries;
1145 		spin_unlock_irqrestore(&ch->lock, irq_flags);
1146 		return xpSuccess;
1147 	}
1148 
1149 	return xpNoMemory;
1150 }
1151 
1152 /*
1153  * Allocate msg_slots associated with the channel.
1154  */
1155 static enum xp_retval
1156 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1157 {
1158 	static enum xp_retval ret;
1159 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1160 
1161 	DBUG_ON(ch->flags & XPC_C_SETUP);
1162 
1163 	ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1164 						   gru_message_queue_desc),
1165 						   GFP_KERNEL);
1166 	if (ch_uv->cached_notify_gru_mq_desc == NULL)
1167 		return xpNoMemory;
1168 
1169 	ret = xpc_allocate_send_msg_slot_uv(ch);
1170 	if (ret == xpSuccess) {
1171 
1172 		ret = xpc_allocate_recv_msg_slot_uv(ch);
1173 		if (ret != xpSuccess) {
1174 			kfree(ch_uv->send_msg_slots);
1175 			xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1176 		}
1177 	}
1178 	return ret;
1179 }
1180 
1181 /*
1182  * Free up msg_slots and clear other stuff that were setup for the specified
1183  * channel.
1184  */
1185 static void
1186 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1187 {
1188 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1189 
1190 	lockdep_assert_held(&ch->lock);
1191 
1192 	kfree(ch_uv->cached_notify_gru_mq_desc);
1193 	ch_uv->cached_notify_gru_mq_desc = NULL;
1194 
1195 	if (ch->flags & XPC_C_SETUP) {
1196 		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1197 		kfree(ch_uv->send_msg_slots);
1198 		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1199 		kfree(ch_uv->recv_msg_slots);
1200 	}
1201 }
1202 
1203 static void
1204 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1205 {
1206 	struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1207 
1208 	msg.ch_number = ch->number;
1209 	msg.reason = ch->reason;
1210 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1211 				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1212 }
1213 
1214 static void
1215 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1216 {
1217 	struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1218 
1219 	msg.ch_number = ch->number;
1220 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1221 				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1222 }
1223 
1224 static void
1225 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1226 {
1227 	struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1228 
1229 	msg.ch_number = ch->number;
1230 	msg.entry_size = ch->entry_size;
1231 	msg.local_nentries = ch->local_nentries;
1232 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1233 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1234 }
1235 
1236 static void
1237 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1238 {
1239 	struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1240 
1241 	msg.ch_number = ch->number;
1242 	msg.local_nentries = ch->local_nentries;
1243 	msg.remote_nentries = ch->remote_nentries;
1244 	msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1245 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1246 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1247 }
1248 
1249 static void
1250 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1251 {
1252 	struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1253 
1254 	msg.ch_number = ch->number;
1255 	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1256 				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1257 }
1258 
1259 static void
1260 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1261 {
1262 	unsigned long irq_flags;
1263 
1264 	spin_lock_irqsave(&part->chctl_lock, irq_flags);
1265 	part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1266 	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1267 
1268 	xpc_wakeup_channel_mgr(part);
1269 }
1270 
1271 static enum xp_retval
1272 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1273 			       unsigned long gru_mq_desc_gpa)
1274 {
1275 	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1276 
1277 	DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1278 	return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1279 					       gru_mq_desc_gpa);
1280 }
1281 
1282 static void
1283 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1284 {
1285 	struct xpc_activate_mq_msg_uv msg;
1286 
1287 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1288 				      XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1289 }
1290 
1291 static void
1292 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1293 {
1294 	struct xpc_activate_mq_msg_uv msg;
1295 
1296 	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1297 				      XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1298 }
1299 
1300 static void
1301 xpc_assume_partition_disengaged_uv(short partid)
1302 {
1303 	struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1304 	unsigned long irq_flags;
1305 
1306 	spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1307 	part_uv->flags &= ~XPC_P_ENGAGED_UV;
1308 	spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1309 }
1310 
1311 static int
1312 xpc_partition_engaged_uv(short partid)
1313 {
1314 	return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1315 }
1316 
1317 static int
1318 xpc_any_partition_engaged_uv(void)
1319 {
1320 	struct xpc_partition_uv *part_uv;
1321 	short partid;
1322 
1323 	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1324 		part_uv = &xpc_partitions[partid].sn.uv;
1325 		if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1326 			return 1;
1327 	}
1328 	return 0;
1329 }
1330 
1331 static enum xp_retval
1332 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1333 			 struct xpc_send_msg_slot_uv **address_of_msg_slot)
1334 {
1335 	enum xp_retval ret;
1336 	struct xpc_send_msg_slot_uv *msg_slot;
1337 	struct xpc_fifo_entry_uv *entry;
1338 
1339 	while (1) {
1340 		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1341 		if (entry != NULL)
1342 			break;
1343 
1344 		if (flags & XPC_NOWAIT)
1345 			return xpNoWait;
1346 
1347 		ret = xpc_allocate_msg_wait(ch);
1348 		if (ret != xpInterrupted && ret != xpTimeout)
1349 			return ret;
1350 	}
1351 
1352 	msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1353 	*address_of_msg_slot = msg_slot;
1354 	return xpSuccess;
1355 }
1356 
1357 static void
1358 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1359 		     struct xpc_send_msg_slot_uv *msg_slot)
1360 {
1361 	xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1362 
1363 	/* wakeup anyone waiting for a free msg slot */
1364 	if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1365 		wake_up(&ch->msg_allocate_wq);
1366 }
1367 
1368 static void
1369 xpc_notify_sender_uv(struct xpc_channel *ch,
1370 		     struct xpc_send_msg_slot_uv *msg_slot,
1371 		     enum xp_retval reason)
1372 {
1373 	xpc_notify_func func = msg_slot->func;
1374 
1375 	if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1376 
1377 		atomic_dec(&ch->n_to_notify);
1378 
1379 		dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1380 			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1381 			msg_slot->msg_slot_number, ch->partid, ch->number);
1382 
1383 		func(reason, ch->partid, ch->number, msg_slot->key);
1384 
1385 		dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1386 			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1387 			msg_slot->msg_slot_number, ch->partid, ch->number);
1388 	}
1389 }
1390 
1391 static void
1392 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1393 			    struct xpc_notify_mq_msg_uv *msg)
1394 {
1395 	struct xpc_send_msg_slot_uv *msg_slot;
1396 	int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1397 
1398 	msg_slot = &ch->sn.uv.send_msg_slots[entry];
1399 
1400 	BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1401 	msg_slot->msg_slot_number += ch->local_nentries;
1402 
1403 	if (msg_slot->func != NULL)
1404 		xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1405 
1406 	xpc_free_msg_slot_uv(ch, msg_slot);
1407 }
1408 
1409 static void
1410 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1411 			    struct xpc_notify_mq_msg_uv *msg)
1412 {
1413 	struct xpc_partition_uv *part_uv = &part->sn.uv;
1414 	struct xpc_channel *ch;
1415 	struct xpc_channel_uv *ch_uv;
1416 	struct xpc_notify_mq_msg_uv *msg_slot;
1417 	unsigned long irq_flags;
1418 	int ch_number = msg->hdr.ch_number;
1419 
1420 	if (unlikely(ch_number >= part->nchannels)) {
1421 		dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1422 			"channel number=0x%x in message from partid=%d\n",
1423 			ch_number, XPC_PARTID(part));
1424 
1425 		/* get hb checker to deactivate from the remote partition */
1426 		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1427 		if (part_uv->act_state_req == 0)
1428 			xpc_activate_IRQ_rcvd++;
1429 		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1430 		part_uv->reason = xpBadChannelNumber;
1431 		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1432 
1433 		wake_up_interruptible(&xpc_activate_IRQ_wq);
1434 		return;
1435 	}
1436 
1437 	ch = &part->channels[ch_number];
1438 	xpc_msgqueue_ref(ch);
1439 
1440 	if (!(ch->flags & XPC_C_CONNECTED)) {
1441 		xpc_msgqueue_deref(ch);
1442 		return;
1443 	}
1444 
1445 	/* see if we're really dealing with an ACK for a previously sent msg */
1446 	if (msg->hdr.size == 0) {
1447 		xpc_handle_notify_mq_ack_uv(ch, msg);
1448 		xpc_msgqueue_deref(ch);
1449 		return;
1450 	}
1451 
1452 	/* we're dealing with a normal message sent via the notify_mq */
1453 	ch_uv = &ch->sn.uv;
1454 
1455 	msg_slot = ch_uv->recv_msg_slots +
1456 	    (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1457 
1458 	BUG_ON(msg_slot->hdr.size != 0);
1459 
1460 	memcpy(msg_slot, msg, msg->hdr.size);
1461 
1462 	xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1463 
1464 	if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1465 		/*
1466 		 * If there is an existing idle kthread get it to deliver
1467 		 * the payload, otherwise we'll have to get the channel mgr
1468 		 * for this partition to create a kthread to do the delivery.
1469 		 */
1470 		if (atomic_read(&ch->kthreads_idle) > 0)
1471 			wake_up_nr(&ch->idle_wq, 1);
1472 		else
1473 			xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1474 	}
1475 	xpc_msgqueue_deref(ch);
1476 }
1477 
1478 static irqreturn_t
1479 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1480 {
1481 	struct xpc_notify_mq_msg_uv *msg;
1482 	short partid;
1483 	struct xpc_partition *part;
1484 
1485 	while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1486 	       NULL) {
1487 
1488 		partid = msg->hdr.partid;
1489 		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1490 			dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1491 				"invalid partid=0x%x in message\n", partid);
1492 		} else {
1493 			part = &xpc_partitions[partid];
1494 
1495 			if (xpc_part_ref(part)) {
1496 				xpc_handle_notify_mq_msg_uv(part, msg);
1497 				xpc_part_deref(part);
1498 			}
1499 		}
1500 
1501 		gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1502 	}
1503 
1504 	return IRQ_HANDLED;
1505 }
1506 
1507 static int
1508 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1509 {
1510 	return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1511 }
1512 
1513 static void
1514 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1515 {
1516 	struct xpc_channel *ch = &part->channels[ch_number];
1517 	int ndeliverable_payloads;
1518 
1519 	xpc_msgqueue_ref(ch);
1520 
1521 	ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1522 
1523 	if (ndeliverable_payloads > 0 &&
1524 	    (ch->flags & XPC_C_CONNECTED) &&
1525 	    (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1526 
1527 		xpc_activate_kthreads(ch, ndeliverable_payloads);
1528 	}
1529 
1530 	xpc_msgqueue_deref(ch);
1531 }
1532 
1533 static enum xp_retval
1534 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1535 		    u16 payload_size, u8 notify_type, xpc_notify_func func,
1536 		    void *key)
1537 {
1538 	enum xp_retval ret = xpSuccess;
1539 	struct xpc_send_msg_slot_uv *msg_slot = NULL;
1540 	struct xpc_notify_mq_msg_uv *msg;
1541 	u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1542 	size_t msg_size;
1543 
1544 	DBUG_ON(notify_type != XPC_N_CALL);
1545 
1546 	msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1547 	if (msg_size > ch->entry_size)
1548 		return xpPayloadTooBig;
1549 
1550 	xpc_msgqueue_ref(ch);
1551 
1552 	if (ch->flags & XPC_C_DISCONNECTING) {
1553 		ret = ch->reason;
1554 		goto out_1;
1555 	}
1556 	if (!(ch->flags & XPC_C_CONNECTED)) {
1557 		ret = xpNotConnected;
1558 		goto out_1;
1559 	}
1560 
1561 	ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1562 	if (ret != xpSuccess)
1563 		goto out_1;
1564 
1565 	if (func != NULL) {
1566 		atomic_inc(&ch->n_to_notify);
1567 
1568 		msg_slot->key = key;
1569 		smp_wmb(); /* a non-NULL func must hit memory after the key */
1570 		msg_slot->func = func;
1571 
1572 		if (ch->flags & XPC_C_DISCONNECTING) {
1573 			ret = ch->reason;
1574 			goto out_2;
1575 		}
1576 	}
1577 
1578 	msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1579 	msg->hdr.partid = xp_partition_id;
1580 	msg->hdr.ch_number = ch->number;
1581 	msg->hdr.size = msg_size;
1582 	msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1583 	memcpy(&msg->payload, payload, payload_size);
1584 
1585 	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1586 			       msg_size);
1587 	if (ret == xpSuccess)
1588 		goto out_1;
1589 
1590 	XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1591 out_2:
1592 	if (func != NULL) {
1593 		/*
1594 		 * Try to NULL the msg_slot's func field. If we fail, then
1595 		 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1596 		 * case we need to pretend we succeeded to send the message
1597 		 * since the user will get a callout for the disconnect error
1598 		 * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1599 		 * error returned here will confuse them. Additionally, since
1600 		 * in this case the channel is being disconnected we don't need
1601 		 * to put the the msg_slot back on the free list.
1602 		 */
1603 		if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1604 			ret = xpSuccess;
1605 			goto out_1;
1606 		}
1607 
1608 		msg_slot->key = NULL;
1609 		atomic_dec(&ch->n_to_notify);
1610 	}
1611 	xpc_free_msg_slot_uv(ch, msg_slot);
1612 out_1:
1613 	xpc_msgqueue_deref(ch);
1614 	return ret;
1615 }
1616 
1617 /*
1618  * Tell the callers of xpc_send_notify() that the status of their payloads
1619  * is unknown because the channel is now disconnecting.
1620  *
1621  * We don't worry about putting these msg_slots on the free list since the
1622  * msg_slots themselves are about to be kfree'd.
1623  */
1624 static void
1625 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1626 {
1627 	struct xpc_send_msg_slot_uv *msg_slot;
1628 	int entry;
1629 
1630 	DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1631 
1632 	for (entry = 0; entry < ch->local_nentries; entry++) {
1633 
1634 		if (atomic_read(&ch->n_to_notify) == 0)
1635 			break;
1636 
1637 		msg_slot = &ch->sn.uv.send_msg_slots[entry];
1638 		if (msg_slot->func != NULL)
1639 			xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1640 	}
1641 }
1642 
1643 /*
1644  * Get the next deliverable message's payload.
1645  */
1646 static void *
1647 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1648 {
1649 	struct xpc_fifo_entry_uv *entry;
1650 	struct xpc_notify_mq_msg_uv *msg;
1651 	void *payload = NULL;
1652 
1653 	if (!(ch->flags & XPC_C_DISCONNECTING)) {
1654 		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1655 		if (entry != NULL) {
1656 			msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1657 					   hdr.u.next);
1658 			payload = &msg->payload;
1659 		}
1660 	}
1661 	return payload;
1662 }
1663 
1664 static void
1665 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1666 {
1667 	struct xpc_notify_mq_msg_uv *msg;
1668 	enum xp_retval ret;
1669 
1670 	msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1671 
1672 	/* return an ACK to the sender of this message */
1673 
1674 	msg->hdr.partid = xp_partition_id;
1675 	msg->hdr.size = 0;	/* size of zero indicates this is an ACK */
1676 
1677 	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1678 			       sizeof(struct xpc_notify_mq_msghdr_uv));
1679 	if (ret != xpSuccess)
1680 		XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1681 }
1682 
1683 static const struct xpc_arch_operations xpc_arch_ops_uv = {
1684 	.setup_partitions = xpc_setup_partitions_uv,
1685 	.teardown_partitions = xpc_teardown_partitions_uv,
1686 	.process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1687 	.get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1688 	.setup_rsvd_page = xpc_setup_rsvd_page_uv,
1689 
1690 	.allow_hb = xpc_allow_hb_uv,
1691 	.disallow_hb = xpc_disallow_hb_uv,
1692 	.disallow_all_hbs = xpc_disallow_all_hbs_uv,
1693 	.increment_heartbeat = xpc_increment_heartbeat_uv,
1694 	.offline_heartbeat = xpc_offline_heartbeat_uv,
1695 	.online_heartbeat = xpc_online_heartbeat_uv,
1696 	.heartbeat_init = xpc_heartbeat_init_uv,
1697 	.heartbeat_exit = xpc_heartbeat_exit_uv,
1698 	.get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1699 
1700 	.request_partition_activation =
1701 		xpc_request_partition_activation_uv,
1702 	.request_partition_reactivation =
1703 		xpc_request_partition_reactivation_uv,
1704 	.request_partition_deactivation =
1705 		xpc_request_partition_deactivation_uv,
1706 	.cancel_partition_deactivation_request =
1707 		xpc_cancel_partition_deactivation_request_uv,
1708 
1709 	.setup_ch_structures = xpc_setup_ch_structures_uv,
1710 	.teardown_ch_structures = xpc_teardown_ch_structures_uv,
1711 
1712 	.make_first_contact = xpc_make_first_contact_uv,
1713 
1714 	.get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1715 	.send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1716 	.send_chctl_closereply = xpc_send_chctl_closereply_uv,
1717 	.send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1718 	.send_chctl_openreply = xpc_send_chctl_openreply_uv,
1719 	.send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1720 	.process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1721 
1722 	.save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1723 
1724 	.setup_msg_structures = xpc_setup_msg_structures_uv,
1725 	.teardown_msg_structures = xpc_teardown_msg_structures_uv,
1726 
1727 	.indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1728 	.indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1729 	.assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1730 	.partition_engaged = xpc_partition_engaged_uv,
1731 	.any_partition_engaged = xpc_any_partition_engaged_uv,
1732 
1733 	.n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1734 	.send_payload = xpc_send_payload_uv,
1735 	.get_deliverable_payload = xpc_get_deliverable_payload_uv,
1736 	.received_payload = xpc_received_payload_uv,
1737 	.notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1738 };
1739 
1740 static int
1741 xpc_init_mq_node(int nid)
1742 {
1743 	int cpu;
1744 
1745 	get_online_cpus();
1746 
1747 	for_each_cpu(cpu, cpumask_of_node(nid)) {
1748 		xpc_activate_mq_uv =
1749 			xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
1750 					     XPC_ACTIVATE_IRQ_NAME,
1751 					     xpc_handle_activate_IRQ_uv);
1752 		if (!IS_ERR(xpc_activate_mq_uv))
1753 			break;
1754 	}
1755 	if (IS_ERR(xpc_activate_mq_uv)) {
1756 		put_online_cpus();
1757 		return PTR_ERR(xpc_activate_mq_uv);
1758 	}
1759 
1760 	for_each_cpu(cpu, cpumask_of_node(nid)) {
1761 		xpc_notify_mq_uv =
1762 			xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
1763 					     XPC_NOTIFY_IRQ_NAME,
1764 					     xpc_handle_notify_IRQ_uv);
1765 		if (!IS_ERR(xpc_notify_mq_uv))
1766 			break;
1767 	}
1768 	if (IS_ERR(xpc_notify_mq_uv)) {
1769 		xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1770 		put_online_cpus();
1771 		return PTR_ERR(xpc_notify_mq_uv);
1772 	}
1773 
1774 	put_online_cpus();
1775 	return 0;
1776 }
1777 
1778 int
1779 xpc_init_uv(void)
1780 {
1781 	int nid;
1782 	int ret = 0;
1783 
1784 	xpc_arch_ops = xpc_arch_ops_uv;
1785 
1786 	if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1787 		dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1788 			XPC_MSG_HDR_MAX_SIZE);
1789 		return -E2BIG;
1790 	}
1791 
1792 	if (xpc_mq_node < 0)
1793 		for_each_online_node(nid) {
1794 			ret = xpc_init_mq_node(nid);
1795 
1796 			if (!ret)
1797 				break;
1798 		}
1799 	else
1800 		ret = xpc_init_mq_node(xpc_mq_node);
1801 
1802 	if (ret < 0)
1803 		dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1804 			-ret);
1805 
1806 	return ret;
1807 }
1808 
1809 void
1810 xpc_exit_uv(void)
1811 {
1812 	xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1813 	xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1814 }
1815 
1816 module_param(xpc_mq_node, int, 0);
1817 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");
1818