drivers/misc/sgi-xp/xpc_uv.c

/*
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (c) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
 */

/*
 * Cross Partition Communication (XPC) uv-based functions.
 *
 *     Architecture specific implementation of common functions.
 *
 */

#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
#include <linux/device.h>
#include <linux/cpu.h>
#include <linux/module.h>
#include <linux/err.h>
#include <linux/slab.h>
#include <asm/uv/uv_hub.h>
#if defined CONFIG_X86_64
#include <asm/uv/bios.h>
#include <asm/uv/uv_irq.h>
#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
#include <asm/sn/intr.h>
#include <asm/sn/sn_sal.h>
#endif
#include "../sgi-gru/gru.h"
#include "../sgi-gru/grukservices.h"
#include "xpc.h"

#if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
struct uv_IO_APIC_route_entry {
	__u64	vector		:  8,
		delivery_mode	:  3,
		dest_mode	:  1,
		delivery_status	:  1,
		polarity	:  1,
		__reserved_1	:  1,
		trigger		:  1,
		mask		:  1,
		__reserved_2	: 15,
		dest		: 32;
};
#endif

static struct xpc_heartbeat_uv *xpc_heartbeat_uv;

#define XPC_ACTIVATE_MSG_SIZE_UV	(1 * GRU_CACHE_LINE_BYTES)
#define XPC_ACTIVATE_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
					 XPC_ACTIVATE_MSG_SIZE_UV)
#define XPC_ACTIVATE_IRQ_NAME		"xpc_activate"

#define XPC_NOTIFY_MSG_SIZE_UV		(2 * GRU_CACHE_LINE_BYTES)
#define XPC_NOTIFY_MQ_SIZE_UV		(4 * XP_MAX_NPARTITIONS_UV * \
					 XPC_NOTIFY_MSG_SIZE_UV)
#define XPC_NOTIFY_IRQ_NAME		"xpc_notify"

static int xpc_mq_node = -1;

static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
static struct xpc_gru_mq_uv *xpc_notify_mq_uv;

static int
xpc_setup_partitions_uv(void)
{
	short partid;
	struct xpc_partition_uv *part_uv;

	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
		part_uv = &xpc_partitions[partid].sn.uv;

		mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
		spin_lock_init(&part_uv->flags_lock);
		part_uv->remote_act_state = XPC_P_AS_INACTIVE;
	}
	return 0;
}

static void
xpc_teardown_partitions_uv(void)
{
	short partid;
	struct xpc_partition_uv *part_uv;
	unsigned long irq_flags;

	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
		part_uv = &xpc_partitions[partid].sn.uv;

		if (part_uv->cached_activate_gru_mq_desc != NULL) {
			mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
			spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
			spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
			kfree(part_uv->cached_activate_gru_mq_desc);
			part_uv->cached_activate_gru_mq_desc = NULL;
			mutex_unlock(&part_uv->
				     cached_activate_gru_mq_desc_mutex);
		}
	}
}

static int
xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
{
	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);

#if defined CONFIG_X86_64
	mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
			UV_AFFINITY_CPU);
	if (mq->irq < 0)
		return mq->irq;

	mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);

#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
	if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
		mq->irq = SGI_XPC_ACTIVATE;
	else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
		mq->irq = SGI_XPC_NOTIFY;
	else
		return -EINVAL;

	mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
#else
	#error not a supported configuration
#endif

	return 0;
}

static void
xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
{
#if defined CONFIG_X86_64
	uv_teardown_irq(mq->irq);

#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
	int mmr_pnode;
	unsigned long mmr_value;

	mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
	mmr_value = 1UL << 16;

	uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
#else
	#error not a supported configuration
#endif
}

static int
xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
{
	int ret;

#if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);

	ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
				    mq->order, &mq->mmr_offset);
	if (ret < 0) {
		dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
			ret);
		return -EBUSY;
	}
#elif defined CONFIG_X86_64
	ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
					 mq->order, &mq->mmr_offset);
	if (ret < 0) {
		dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
			"ret=%d\n", ret);
		return ret;
	}
#else
	#error not a supported configuration
#endif

	mq->watchlist_num = ret;
	return 0;
}

static void
xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
{
	int ret;
	int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);

#if defined CONFIG_X86_64
	ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
	BUG_ON(ret != BIOS_STATUS_SUCCESS);
#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
	ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
	BUG_ON(ret != SALRET_OK);
#else
	#error not a supported configuration
#endif
}

static struct xpc_gru_mq_uv *
xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
		     irq_handler_t irq_handler)
{
	enum xp_retval xp_ret;
	int ret;
	int nid;
	int nasid;
	int pg_order;
	struct page *page;
	struct xpc_gru_mq_uv *mq;
	struct uv_IO_APIC_route_entry *mmr_value;

	mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
	if (mq == NULL) {
		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
			"a xpc_gru_mq_uv structure\n");
		ret = -ENOMEM;
		goto out_0;
	}

	mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
				  GFP_KERNEL);
	if (mq->gru_mq_desc == NULL) {
		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
			"a gru_message_queue_desc structure\n");
		ret = -ENOMEM;
		goto out_1;
	}

	pg_order = get_order(mq_size);
	mq->order = pg_order + PAGE_SHIFT;
	mq_size = 1UL << mq->order;

	mq->mmr_blade = uv_cpu_to_blade_id(cpu);

	nid = cpu_to_node(cpu);
	page = alloc_pages_exact_node(nid,
				      GFP_KERNEL | __GFP_ZERO | GFP_THISNODE,
				      pg_order);
	if (page == NULL) {
		dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
			"bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
		ret = -ENOMEM;
		goto out_2;
	}
	mq->address = page_address(page);

	/* enable generation of irq when GRU mq operation occurs to this mq */
	ret = xpc_gru_mq_watchlist_alloc_uv(mq);
	if (ret != 0)
		goto out_3;

	ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
	if (ret != 0)
		goto out_4;

	ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
	if (ret != 0) {
		dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
			mq->irq, -ret);
		goto out_5;
	}

	nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));

	mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
	ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
				     nasid, mmr_value->vector, mmr_value->dest);
	if (ret != 0) {
		dev_err(xpc_part, "gru_create_message_queue() returned "
			"error=%d\n", ret);
		ret = -EINVAL;
		goto out_6;
	}

	/* allow other partitions to access this GRU mq */
	xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
	if (xp_ret != xpSuccess) {
		ret = -EACCES;
		goto out_6;
	}

	return mq;

	/* something went wrong */
out_6:
	free_irq(mq->irq, NULL);
out_5:
	xpc_release_gru_mq_irq_uv(mq);
out_4:
	xpc_gru_mq_watchlist_free_uv(mq);
out_3:
	free_pages((unsigned long)mq->address, pg_order);
out_2:
	kfree(mq->gru_mq_desc);
out_1:
	kfree(mq);
out_0:
	return ERR_PTR(ret);
}

static void
xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
{
	unsigned int mq_size;
	int pg_order;
	int ret;

	/* disallow other partitions to access GRU mq */
	mq_size = 1UL << mq->order;
	ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
	BUG_ON(ret != xpSuccess);

	/* unregister irq handler and release mq irq/vector mapping */
	free_irq(mq->irq, NULL);
	xpc_release_gru_mq_irq_uv(mq);

	/* disable generation of irq when GRU mq op occurs to this mq */
	xpc_gru_mq_watchlist_free_uv(mq);

	pg_order = mq->order - PAGE_SHIFT;
	free_pages((unsigned long)mq->address, pg_order);

	kfree(mq);
}

static enum xp_retval
xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
		 size_t msg_size)
{
	enum xp_retval xp_ret;
	int ret;

	while (1) {
		ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
		if (ret == MQE_OK) {
			xp_ret = xpSuccess;
			break;
		}

		if (ret == MQE_QUEUE_FULL) {
			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
				"error=MQE_QUEUE_FULL\n");
			/* !!! handle QLimit reached; delay & try again */
			/* ??? Do we add a limit to the number of retries? */
			(void)msleep_interruptible(10);
		} else if (ret == MQE_CONGESTION) {
			dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
				"error=MQE_CONGESTION\n");
			/* !!! handle LB Overflow; simply try again */
			/* ??? Do we add a limit to the number of retries? */
		} else {
			/* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
			dev_err(xpc_chan, "gru_send_message_gpa() returned "
				"error=%d\n", ret);
			xp_ret = xpGruSendMqError;
			break;
		}
	}
	return xp_ret;
}

static void
xpc_process_activate_IRQ_rcvd_uv(void)
{
	unsigned long irq_flags;
	short partid;
	struct xpc_partition *part;
	u8 act_state_req;

	DBUG_ON(xpc_activate_IRQ_rcvd == 0);

	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
		part = &xpc_partitions[partid];

		if (part->sn.uv.act_state_req == 0)
			continue;

		xpc_activate_IRQ_rcvd--;
		BUG_ON(xpc_activate_IRQ_rcvd < 0);

		act_state_req = part->sn.uv.act_state_req;
		part->sn.uv.act_state_req = 0;
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
			if (part->act_state == XPC_P_AS_INACTIVE)
				xpc_activate_partition(part);
			else if (part->act_state == XPC_P_AS_DEACTIVATING)
				XPC_DEACTIVATE_PARTITION(part, xpReactivating);

		} else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
			if (part->act_state == XPC_P_AS_INACTIVE)
				xpc_activate_partition(part);
			else
				XPC_DEACTIVATE_PARTITION(part, xpReactivating);

		} else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
			XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);

		} else {
			BUG();
		}

		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (xpc_activate_IRQ_rcvd == 0)
			break;
	}
	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

}

static void
xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
			      struct xpc_activate_mq_msghdr_uv *msg_hdr,
			      int part_setup,
			      int *wakeup_hb_checker)
{
	unsigned long irq_flags;
	struct xpc_partition_uv *part_uv = &part->sn.uv;
	struct xpc_openclose_args *args;

	part_uv->remote_act_state = msg_hdr->act_state;

	switch (msg_hdr->type) {
	case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
		/* syncing of remote_act_state was just done above */
		break;

	case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
		struct xpc_activate_mq_msg_activate_req_uv *msg;

		/*
		 * ??? Do we deal here with ts_jiffies being different
		 * ??? if act_state != XPC_P_AS_INACTIVE instead of
		 * ??? below?
		 */
		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_activate_req_uv, hdr);

		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (part_uv->act_state_req == 0)
			xpc_activate_IRQ_rcvd++;
		part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
		part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
		part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
		part_uv->heartbeat_gpa = msg->heartbeat_gpa;

		if (msg->activate_gru_mq_desc_gpa !=
		    part_uv->activate_gru_mq_desc_gpa) {
			spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
			part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
			spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
			part_uv->activate_gru_mq_desc_gpa =
			    msg->activate_gru_mq_desc_gpa;
		}
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		(*wakeup_hb_checker)++;
		break;
	}
	case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
		struct xpc_activate_mq_msg_deactivate_req_uv *msg;

		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_deactivate_req_uv, hdr);

		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (part_uv->act_state_req == 0)
			xpc_activate_IRQ_rcvd++;
		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
		part_uv->reason = msg->reason;
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		(*wakeup_hb_checker)++;
		return;
	}
	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
		struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;

		if (!part_setup)
			break;

		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_chctl_closerequest_uv,
				   hdr);
		args = &part->remote_openclose_args[msg->ch_number];
		args->reason = msg->reason;

		spin_lock_irqsave(&part->chctl_lock, irq_flags);
		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

		xpc_wakeup_channel_mgr(part);
		break;
	}
	case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
		struct xpc_activate_mq_msg_chctl_closereply_uv *msg;

		if (!part_setup)
			break;

		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_chctl_closereply_uv,
				   hdr);

		spin_lock_irqsave(&part->chctl_lock, irq_flags);
		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

		xpc_wakeup_channel_mgr(part);
		break;
	}
	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
		struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;

		if (!part_setup)
			break;

		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_chctl_openrequest_uv,
				   hdr);
		args = &part->remote_openclose_args[msg->ch_number];
		args->entry_size = msg->entry_size;
		args->local_nentries = msg->local_nentries;

		spin_lock_irqsave(&part->chctl_lock, irq_flags);
		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

		xpc_wakeup_channel_mgr(part);
		break;
	}
	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
		struct xpc_activate_mq_msg_chctl_openreply_uv *msg;

		if (!part_setup)
			break;

		msg = container_of(msg_hdr, struct
				   xpc_activate_mq_msg_chctl_openreply_uv, hdr);
		args = &part->remote_openclose_args[msg->ch_number];
		args->remote_nentries = msg->remote_nentries;
		args->local_nentries = msg->local_nentries;
		args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;

		spin_lock_irqsave(&part->chctl_lock, irq_flags);
		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

		xpc_wakeup_channel_mgr(part);
		break;
	}
	case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
		struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;

		if (!part_setup)
			break;

		msg = container_of(msg_hdr, struct
				xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
		spin_lock_irqsave(&part->chctl_lock, irq_flags);
		part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
		spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

		xpc_wakeup_channel_mgr(part);
	}
	case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
		part_uv->flags |= XPC_P_ENGAGED_UV;
		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
		break;

	case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
		part_uv->flags &= ~XPC_P_ENGAGED_UV;
		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
		break;

	default:
		dev_err(xpc_part, "received unknown activate_mq msg type=%d "
			"from partition=%d\n", msg_hdr->type, XPC_PARTID(part));

		/* get hb checker to deactivate from the remote partition */
		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (part_uv->act_state_req == 0)
			xpc_activate_IRQ_rcvd++;
		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
		part_uv->reason = xpBadMsgType;
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		(*wakeup_hb_checker)++;
		return;
	}

	if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
	    part->remote_rp_ts_jiffies != 0) {
		/*
		 * ??? Does what we do here need to be sensitive to
		 * ??? act_state or remote_act_state?
		 */
		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (part_uv->act_state_req == 0)
			xpc_activate_IRQ_rcvd++;
		part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		(*wakeup_hb_checker)++;
	}
}

static irqreturn_t
xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
{
	struct xpc_activate_mq_msghdr_uv *msg_hdr;
	short partid;
	struct xpc_partition *part;
	int wakeup_hb_checker = 0;
	int part_referenced;

	while (1) {
		msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
		if (msg_hdr == NULL)
			break;

		partid = msg_hdr->partid;
		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
			dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
				"received invalid partid=0x%x in message\n",
				partid);
		} else {
			part = &xpc_partitions[partid];

			part_referenced = xpc_part_ref(part);
			xpc_handle_activate_mq_msg_uv(part, msg_hdr,
						      part_referenced,
						      &wakeup_hb_checker);
			if (part_referenced)
				xpc_part_deref(part);
		}

		gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
	}

	if (wakeup_hb_checker)
		wake_up_interruptible(&xpc_activate_IRQ_wq);

	return IRQ_HANDLED;
}

static enum xp_retval
xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
				unsigned long gru_mq_desc_gpa)
{
	enum xp_retval ret;

	ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
			       sizeof(struct gru_message_queue_desc));
	if (ret == xpSuccess)
		gru_mq_desc->mq = NULL;

	return ret;
}

static enum xp_retval
xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
			 int msg_type)
{
	struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
	struct xpc_partition_uv *part_uv = &part->sn.uv;
	struct gru_message_queue_desc *gru_mq_desc;
	unsigned long irq_flags;
	enum xp_retval ret;

	DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);

	msg_hdr->type = msg_type;
	msg_hdr->partid = xp_partition_id;
	msg_hdr->act_state = part->act_state;
	msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;

	mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
again:
	if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
		gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
		if (gru_mq_desc == NULL) {
			gru_mq_desc = kmalloc(sizeof(struct
					      gru_message_queue_desc),
					      GFP_KERNEL);
			if (gru_mq_desc == NULL) {
				ret = xpNoMemory;
				goto done;
			}
			part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
		}

		ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
						      part_uv->
						      activate_gru_mq_desc_gpa);
		if (ret != xpSuccess)
			goto done;

		spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
		part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
		spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
	}

	/* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
	ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
			       msg_size);
	if (ret != xpSuccess) {
		smp_rmb();	/* ensure a fresh copy of part_uv->flags */
		if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
			goto again;
	}
done:
	mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
	return ret;
}

static void
xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
			      size_t msg_size, int msg_type)
{
	enum xp_retval ret;

	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
	if (unlikely(ret != xpSuccess))
		XPC_DEACTIVATE_PARTITION(part, ret);
}

static void
xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
			 void *msg, size_t msg_size, int msg_type)
{
	struct xpc_partition *part = &xpc_partitions[ch->partid];
	enum xp_retval ret;

	ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
	if (unlikely(ret != xpSuccess)) {
		if (irq_flags != NULL)
			spin_unlock_irqrestore(&ch->lock, *irq_flags);

		XPC_DEACTIVATE_PARTITION(part, ret);

		if (irq_flags != NULL)
			spin_lock_irqsave(&ch->lock, *irq_flags);
	}
}

static void
xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
{
	unsigned long irq_flags;
	struct xpc_partition_uv *part_uv = &part->sn.uv;

	/*
	 * !!! Make our side think that the remote partition sent an activate
	 * !!! mq message our way by doing what the activate IRQ handler would
	 * !!! do had one really been sent.
	 */

	spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
	if (part_uv->act_state_req == 0)
		xpc_activate_IRQ_rcvd++;
	part_uv->act_state_req = act_state_req;
	spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

	wake_up_interruptible(&xpc_activate_IRQ_wq);
}

static enum xp_retval
xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
				  size_t *len)
{
	s64 status;
	enum xp_retval ret;

#if defined CONFIG_X86_64
	status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
					  (u64 *)len);
	if (status == BIOS_STATUS_SUCCESS)
		ret = xpSuccess;
	else if (status == BIOS_STATUS_MORE_PASSES)
		ret = xpNeedMoreInfo;
	else
		ret = xpBiosError;

#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
	status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
	if (status == SALRET_OK)
		ret = xpSuccess;
	else if (status == SALRET_MORE_PASSES)
		ret = xpNeedMoreInfo;
	else
		ret = xpSalError;

#else
	#error not a supported configuration
#endif

	return ret;
}

static int
xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
{
	xpc_heartbeat_uv =
	    &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
	rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
	rp->sn.uv.activate_gru_mq_desc_gpa =
	    uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
	return 0;
}

static void
xpc_allow_hb_uv(short partid)
{
}

static void
xpc_disallow_hb_uv(short partid)
{
}

static void
xpc_disallow_all_hbs_uv(void)
{
}

static void
xpc_increment_heartbeat_uv(void)
{
	xpc_heartbeat_uv->value++;
}

static void
xpc_offline_heartbeat_uv(void)
{
	xpc_increment_heartbeat_uv();
	xpc_heartbeat_uv->offline = 1;
}

static void
xpc_online_heartbeat_uv(void)
{
	xpc_increment_heartbeat_uv();
	xpc_heartbeat_uv->offline = 0;
}

static void
xpc_heartbeat_init_uv(void)
{
	xpc_heartbeat_uv->value = 1;
	xpc_heartbeat_uv->offline = 0;
}

static void
xpc_heartbeat_exit_uv(void)
{
	xpc_offline_heartbeat_uv();
}

static enum xp_retval
xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
{
	struct xpc_partition_uv *part_uv = &part->sn.uv;
	enum xp_retval ret;

	ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
			       part_uv->heartbeat_gpa,
			       sizeof(struct xpc_heartbeat_uv));
	if (ret != xpSuccess)
		return ret;

	if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
	    !part_uv->cached_heartbeat.offline) {

		ret = xpNoHeartbeat;
	} else {
		part->last_heartbeat = part_uv->cached_heartbeat.value;
	}
	return ret;
}

static void
xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
				    unsigned long remote_rp_gpa, int nasid)
{
	short partid = remote_rp->SAL_partid;
	struct xpc_partition *part = &xpc_partitions[partid];
	struct xpc_activate_mq_msg_activate_req_uv msg;

	part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
	part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
	part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
	part->sn.uv.activate_gru_mq_desc_gpa =
	    remote_rp->sn.uv.activate_gru_mq_desc_gpa;

	/*
	 * ??? Is it a good idea to make this conditional on what is
	 * ??? potentially stale state information?
	 */
	if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
		msg.rp_gpa = uv_gpa(xpc_rsvd_page);
		msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
		msg.activate_gru_mq_desc_gpa =
		    xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
					   XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
	}

	if (part->act_state == XPC_P_AS_INACTIVE)
		xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
}

static void
xpc_request_partition_reactivation_uv(struct xpc_partition *part)
{
	xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
}

static void
xpc_request_partition_deactivation_uv(struct xpc_partition *part)
{
	struct xpc_activate_mq_msg_deactivate_req_uv msg;

	/*
	 * ??? Is it a good idea to make this conditional on what is
	 * ??? potentially stale state information?
	 */
	if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
	    part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {

		msg.reason = part->reason;
		xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
					 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
	}
}

static void
xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
{
	/* nothing needs to be done */
	return;
}

static void
xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
{
	head->first = NULL;
	head->last = NULL;
	spin_lock_init(&head->lock);
	head->n_entries = 0;
}

static void *
xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
{
	unsigned long irq_flags;
	struct xpc_fifo_entry_uv *first;

	spin_lock_irqsave(&head->lock, irq_flags);
	first = head->first;
	if (head->first != NULL) {
		head->first = first->next;
		if (head->first == NULL)
			head->last = NULL;

		head->n_entries--;
		BUG_ON(head->n_entries < 0);

		first->next = NULL;
	}
	spin_unlock_irqrestore(&head->lock, irq_flags);
	return first;
}

static void
xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
		      struct xpc_fifo_entry_uv *last)
{
	unsigned long irq_flags;

	last->next = NULL;
	spin_lock_irqsave(&head->lock, irq_flags);
	if (head->last != NULL)
		head->last->next = last;
	else
		head->first = last;
	head->last = last;
	head->n_entries++;
	spin_unlock_irqrestore(&head->lock, irq_flags);
}

static int
xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
{
	return head->n_entries;
}

/*
 * Setup the channel structures that are uv specific.
 */
static enum xp_retval
xpc_setup_ch_structures_uv(struct xpc_partition *part)
{
	struct xpc_channel_uv *ch_uv;
	int ch_number;

	for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
		ch_uv = &part->channels[ch_number].sn.uv;

		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
	}

	return xpSuccess;
}

/*
 * Teardown the channel structures that are uv specific.
 */
static void
xpc_teardown_ch_structures_uv(struct xpc_partition *part)
{
	/* nothing needs to be done */
	return;
}

static enum xp_retval
xpc_make_first_contact_uv(struct xpc_partition *part)
{
	struct xpc_activate_mq_msg_uv msg;

	/*
	 * We send a sync msg to get the remote partition's remote_act_state
	 * updated to our current act_state which at this point should
	 * be XPC_P_AS_ACTIVATING.
	 */
	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
				      XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);

	while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
		 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {

		dev_dbg(xpc_part, "waiting to make first contact with "
			"partition %d\n", XPC_PARTID(part));

		/* wait a 1/4 of a second or so */
		(void)msleep_interruptible(250);

		if (part->act_state == XPC_P_AS_DEACTIVATING)
			return part->reason;
	}

	return xpSuccess;
}

static u64
xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
{
	unsigned long irq_flags;
	union xpc_channel_ctl_flags chctl;

	spin_lock_irqsave(&part->chctl_lock, irq_flags);
	chctl = part->chctl;
	if (chctl.all_flags != 0)
		part->chctl.all_flags = 0;

	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
	return chctl.all_flags;
}

static enum xp_retval
xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
{
	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
	struct xpc_send_msg_slot_uv *msg_slot;
	unsigned long irq_flags;
	int nentries;
	int entry;
	size_t nbytes;

	for (nentries = ch->local_nentries; nentries > 0; nentries--) {
		nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
		ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
		if (ch_uv->send_msg_slots == NULL)
			continue;

		for (entry = 0; entry < nentries; entry++) {
			msg_slot = &ch_uv->send_msg_slots[entry];

			msg_slot->msg_slot_number = entry;
			xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
					      &msg_slot->next);
		}

		spin_lock_irqsave(&ch->lock, irq_flags);
		if (nentries < ch->local_nentries)
			ch->local_nentries = nentries;
		spin_unlock_irqrestore(&ch->lock, irq_flags);
		return xpSuccess;
	}

	return xpNoMemory;
}

static enum xp_retval
xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
{
	struct xpc_channel_uv *ch_uv = &ch->sn.uv;
	struct xpc_notify_mq_msg_uv *msg_slot;
	unsigned long irq_flags;
	int nentries;
	int entry;
	size_t nbytes;

	for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
		nbytes = nentries * ch->entry_size;
		ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
		if (ch_uv->recv_msg_slots == NULL)
			continue;

		for (entry = 0; entry < nentries; entry++) {
			msg_slot = ch_uv->recv_msg_slots +
			    entry * ch->entry_size;

			msg_slot->hdr.msg_slot_number = entry;
		}

		spin_lock_irqsave(&ch->lock, irq_flags);
		if (nentries < ch->remote_nentries)
			ch->remote_nentries = nentries;
		spin_unlock_irqrestore(&ch->lock, irq_flags);
		return xpSuccess;
	}

	return xpNoMemory;
}

/*
 * Allocate msg_slots associated with the channel.
 */
static enum xp_retval
xpc_setup_msg_structures_uv(struct xpc_channel *ch)
{
	static enum xp_retval ret;
	struct xpc_channel_uv *ch_uv = &ch->sn.uv;

	DBUG_ON(ch->flags & XPC_C_SETUP);

	ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
						   gru_message_queue_desc),
						   GFP_KERNEL);
	if (ch_uv->cached_notify_gru_mq_desc == NULL)
		return xpNoMemory;

	ret = xpc_allocate_send_msg_slot_uv(ch);
	if (ret == xpSuccess) {

		ret = xpc_allocate_recv_msg_slot_uv(ch);
		if (ret != xpSuccess) {
			kfree(ch_uv->send_msg_slots);
			xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
		}
	}
	return ret;
}

/*
 * Free up msg_slots and clear other stuff that were setup for the specified
 * channel.
 */
static void
xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
{
	struct xpc_channel_uv *ch_uv = &ch->sn.uv;

	DBUG_ON(!spin_is_locked(&ch->lock));

	kfree(ch_uv->cached_notify_gru_mq_desc);
	ch_uv->cached_notify_gru_mq_desc = NULL;

	if (ch->flags & XPC_C_SETUP) {
		xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
		kfree(ch_uv->send_msg_slots);
		xpc_init_fifo_uv(&ch_uv->recv_msg_list);
		kfree(ch_uv->recv_msg_slots);
	}
}

static void
xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
{
	struct xpc_activate_mq_msg_chctl_closerequest_uv msg;

	msg.ch_number = ch->number;
	msg.reason = ch->reason;
	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
}

static void
xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
{
	struct xpc_activate_mq_msg_chctl_closereply_uv msg;

	msg.ch_number = ch->number;
	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
				    XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
}

static void
xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
{
	struct xpc_activate_mq_msg_chctl_openrequest_uv msg;

	msg.ch_number = ch->number;
	msg.entry_size = ch->entry_size;
	msg.local_nentries = ch->local_nentries;
	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
}

static void
xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
{
	struct xpc_activate_mq_msg_chctl_openreply_uv msg;

	msg.ch_number = ch->number;
	msg.local_nentries = ch->local_nentries;
	msg.remote_nentries = ch->remote_nentries;
	msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
}

static void
xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
{
	struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;

	msg.ch_number = ch->number;
	xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
				    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
}

static void
xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
{
	unsigned long irq_flags;

	spin_lock_irqsave(&part->chctl_lock, irq_flags);
	part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
	spin_unlock_irqrestore(&part->chctl_lock, irq_flags);

	xpc_wakeup_channel_mgr(part);
}

static enum xp_retval
xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
			       unsigned long gru_mq_desc_gpa)
{
	struct xpc_channel_uv *ch_uv = &ch->sn.uv;

	DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
	return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
					       gru_mq_desc_gpa);
}

static void
xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
{
	struct xpc_activate_mq_msg_uv msg;

	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
				      XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
}

static void
xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
{
	struct xpc_activate_mq_msg_uv msg;

	xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
				      XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
}

static void
xpc_assume_partition_disengaged_uv(short partid)
{
	struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
	unsigned long irq_flags;

	spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
	part_uv->flags &= ~XPC_P_ENGAGED_UV;
	spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
}

static int
xpc_partition_engaged_uv(short partid)
{
	return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
}

static int
xpc_any_partition_engaged_uv(void)
{
	struct xpc_partition_uv *part_uv;
	short partid;

	for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
		part_uv = &xpc_partitions[partid].sn.uv;
		if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
			return 1;
	}
	return 0;
}

static enum xp_retval
xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
			 struct xpc_send_msg_slot_uv **address_of_msg_slot)
{
	enum xp_retval ret;
	struct xpc_send_msg_slot_uv *msg_slot;
	struct xpc_fifo_entry_uv *entry;

	while (1) {
		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
		if (entry != NULL)
			break;

		if (flags & XPC_NOWAIT)
			return xpNoWait;

		ret = xpc_allocate_msg_wait(ch);
		if (ret != xpInterrupted && ret != xpTimeout)
			return ret;
	}

	msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
	*address_of_msg_slot = msg_slot;
	return xpSuccess;
}

static void
xpc_free_msg_slot_uv(struct xpc_channel *ch,
		     struct xpc_send_msg_slot_uv *msg_slot)
{
	xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);

	/* wakeup anyone waiting for a free msg slot */
	if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
		wake_up(&ch->msg_allocate_wq);
}

static void
xpc_notify_sender_uv(struct xpc_channel *ch,
		     struct xpc_send_msg_slot_uv *msg_slot,
		     enum xp_retval reason)
{
	xpc_notify_func func = msg_slot->func;

	if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {

		atomic_dec(&ch->n_to_notify);

		dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
			msg_slot->msg_slot_number, ch->partid, ch->number);

		func(reason, ch->partid, ch->number, msg_slot->key);

		dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
			"msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
			msg_slot->msg_slot_number, ch->partid, ch->number);
	}
}

static void
xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
			    struct xpc_notify_mq_msg_uv *msg)
{
	struct xpc_send_msg_slot_uv *msg_slot;
	int entry = msg->hdr.msg_slot_number % ch->local_nentries;

	msg_slot = &ch->sn.uv.send_msg_slots[entry];

	BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
	msg_slot->msg_slot_number += ch->local_nentries;

	if (msg_slot->func != NULL)
		xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);

	xpc_free_msg_slot_uv(ch, msg_slot);
}

static void
xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
			    struct xpc_notify_mq_msg_uv *msg)
{
	struct xpc_partition_uv *part_uv = &part->sn.uv;
	struct xpc_channel *ch;
	struct xpc_channel_uv *ch_uv;
	struct xpc_notify_mq_msg_uv *msg_slot;
	unsigned long irq_flags;
	int ch_number = msg->hdr.ch_number;

	if (unlikely(ch_number >= part->nchannels)) {
		dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
			"channel number=0x%x in message from partid=%d\n",
			ch_number, XPC_PARTID(part));

		/* get hb checker to deactivate from the remote partition */
		spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
		if (part_uv->act_state_req == 0)
			xpc_activate_IRQ_rcvd++;
		part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
		part_uv->reason = xpBadChannelNumber;
		spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);

		wake_up_interruptible(&xpc_activate_IRQ_wq);
		return;
	}

	ch = &part->channels[ch_number];
	xpc_msgqueue_ref(ch);

	if (!(ch->flags & XPC_C_CONNECTED)) {
		xpc_msgqueue_deref(ch);
		return;
	}

	/* see if we're really dealing with an ACK for a previously sent msg */
	if (msg->hdr.size == 0) {
		xpc_handle_notify_mq_ack_uv(ch, msg);
		xpc_msgqueue_deref(ch);
		return;
	}

	/* we're dealing with a normal message sent via the notify_mq */
	ch_uv = &ch->sn.uv;

	msg_slot = ch_uv->recv_msg_slots +
	    (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;

	BUG_ON(msg_slot->hdr.size != 0);

	memcpy(msg_slot, msg, msg->hdr.size);

	xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);

	if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
		/*
		 * If there is an existing idle kthread get it to deliver
		 * the payload, otherwise we'll have to get the channel mgr
		 * for this partition to create a kthread to do the delivery.
		 */
		if (atomic_read(&ch->kthreads_idle) > 0)
			wake_up_nr(&ch->idle_wq, 1);
		else
			xpc_send_chctl_local_msgrequest_uv(part, ch->number);
	}
	xpc_msgqueue_deref(ch);
}

static irqreturn_t
xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
{
	struct xpc_notify_mq_msg_uv *msg;
	short partid;
	struct xpc_partition *part;

	while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
	       NULL) {

		partid = msg->hdr.partid;
		if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
			dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
				"invalid partid=0x%x in message\n", partid);
		} else {
			part = &xpc_partitions[partid];

			if (xpc_part_ref(part)) {
				xpc_handle_notify_mq_msg_uv(part, msg);
				xpc_part_deref(part);
			}
		}

		gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
	}

	return IRQ_HANDLED;
}

static int
xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
{
	return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
}

static void
xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
{
	struct xpc_channel *ch = &part->channels[ch_number];
	int ndeliverable_payloads;

	xpc_msgqueue_ref(ch);

	ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);

	if (ndeliverable_payloads > 0 &&
	    (ch->flags & XPC_C_CONNECTED) &&
	    (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {

		xpc_activate_kthreads(ch, ndeliverable_payloads);
	}

	xpc_msgqueue_deref(ch);
}

static enum xp_retval
xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
		    u16 payload_size, u8 notify_type, xpc_notify_func func,
		    void *key)
{
	enum xp_retval ret = xpSuccess;
	struct xpc_send_msg_slot_uv *msg_slot = NULL;
	struct xpc_notify_mq_msg_uv *msg;
	u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
	size_t msg_size;

	DBUG_ON(notify_type != XPC_N_CALL);

	msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
	if (msg_size > ch->entry_size)
		return xpPayloadTooBig;

	xpc_msgqueue_ref(ch);

	if (ch->flags & XPC_C_DISCONNECTING) {
		ret = ch->reason;
		goto out_1;
	}
	if (!(ch->flags & XPC_C_CONNECTED)) {
		ret = xpNotConnected;
		goto out_1;
	}

	ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
	if (ret != xpSuccess)
		goto out_1;

	if (func != NULL) {
		atomic_inc(&ch->n_to_notify);

		msg_slot->key = key;
		smp_wmb(); /* a non-NULL func must hit memory after the key */
		msg_slot->func = func;

		if (ch->flags & XPC_C_DISCONNECTING) {
			ret = ch->reason;
			goto out_2;
		}
	}

	msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
	msg->hdr.partid = xp_partition_id;
	msg->hdr.ch_number = ch->number;
	msg->hdr.size = msg_size;
	msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
	memcpy(&msg->payload, payload, payload_size);

	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
			       msg_size);
	if (ret == xpSuccess)
		goto out_1;

	XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
out_2:
	if (func != NULL) {
		/*
		 * Try to NULL the msg_slot's func field. If we fail, then
		 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
		 * case we need to pretend we succeeded to send the message
		 * since the user will get a callout for the disconnect error
		 * by xpc_notify_senders_of_disconnect_uv(), and to also get an
		 * error returned here will confuse them. Additionally, since
		 * in this case the channel is being disconnected we don't need
		 * to put the the msg_slot back on the free list.
		 */
		if (cmpxchg(&msg_slot->func, func, NULL) != func) {
			ret = xpSuccess;
			goto out_1;
		}

		msg_slot->key = NULL;
		atomic_dec(&ch->n_to_notify);
	}
	xpc_free_msg_slot_uv(ch, msg_slot);
out_1:
	xpc_msgqueue_deref(ch);
	return ret;
}

/*
 * Tell the callers of xpc_send_notify() that the status of their payloads
 * is unknown because the channel is now disconnecting.
 *
 * We don't worry about putting these msg_slots on the free list since the
 * msg_slots themselves are about to be kfree'd.
 */
static void
xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
{
	struct xpc_send_msg_slot_uv *msg_slot;
	int entry;

	DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));

	for (entry = 0; entry < ch->local_nentries; entry++) {

		if (atomic_read(&ch->n_to_notify) == 0)
			break;

		msg_slot = &ch->sn.uv.send_msg_slots[entry];
		if (msg_slot->func != NULL)
			xpc_notify_sender_uv(ch, msg_slot, ch->reason);
	}
}

/*
 * Get the next deliverable message's payload.
 */
static void *
xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
{
	struct xpc_fifo_entry_uv *entry;
	struct xpc_notify_mq_msg_uv *msg;
	void *payload = NULL;

	if (!(ch->flags & XPC_C_DISCONNECTING)) {
		entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
		if (entry != NULL) {
			msg = container_of(entry, struct xpc_notify_mq_msg_uv,
					   hdr.u.next);
			payload = &msg->payload;
		}
	}
	return payload;
}

static void
xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
{
	struct xpc_notify_mq_msg_uv *msg;
	enum xp_retval ret;

	msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);

	/* return an ACK to the sender of this message */

	msg->hdr.partid = xp_partition_id;
	msg->hdr.size = 0;	/* size of zero indicates this is an ACK */

	ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
			       sizeof(struct xpc_notify_mq_msghdr_uv));
	if (ret != xpSuccess)
		XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
}

static struct xpc_arch_operations xpc_arch_ops_uv = {
	.setup_partitions = xpc_setup_partitions_uv,
	.teardown_partitions = xpc_teardown_partitions_uv,
	.process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
	.get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
	.setup_rsvd_page = xpc_setup_rsvd_page_uv,

	.allow_hb = xpc_allow_hb_uv,
	.disallow_hb = xpc_disallow_hb_uv,
	.disallow_all_hbs = xpc_disallow_all_hbs_uv,
	.increment_heartbeat = xpc_increment_heartbeat_uv,
	.offline_heartbeat = xpc_offline_heartbeat_uv,
	.online_heartbeat = xpc_online_heartbeat_uv,
	.heartbeat_init = xpc_heartbeat_init_uv,
	.heartbeat_exit = xpc_heartbeat_exit_uv,
	.get_remote_heartbeat = xpc_get_remote_heartbeat_uv,

	.request_partition_activation =
		xpc_request_partition_activation_uv,
	.request_partition_reactivation =
		xpc_request_partition_reactivation_uv,
	.request_partition_deactivation =
		xpc_request_partition_deactivation_uv,
	.cancel_partition_deactivation_request =
		xpc_cancel_partition_deactivation_request_uv,

	.setup_ch_structures = xpc_setup_ch_structures_uv,
	.teardown_ch_structures = xpc_teardown_ch_structures_uv,

	.make_first_contact = xpc_make_first_contact_uv,

	.get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
	.send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
	.send_chctl_closereply = xpc_send_chctl_closereply_uv,
	.send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
	.send_chctl_openreply = xpc_send_chctl_openreply_uv,
	.send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
	.process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,

	.save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,

	.setup_msg_structures = xpc_setup_msg_structures_uv,
	.teardown_msg_structures = xpc_teardown_msg_structures_uv,

	.indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
	.indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
	.assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
	.partition_engaged = xpc_partition_engaged_uv,
	.any_partition_engaged = xpc_any_partition_engaged_uv,

	.n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
	.send_payload = xpc_send_payload_uv,
	.get_deliverable_payload = xpc_get_deliverable_payload_uv,
	.received_payload = xpc_received_payload_uv,
	.notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
};

static int
xpc_init_mq_node(int nid)
{
	int cpu;

	get_online_cpus();

	for_each_cpu(cpu, cpumask_of_node(nid)) {
		xpc_activate_mq_uv =
			xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
					     XPC_ACTIVATE_IRQ_NAME,
					     xpc_handle_activate_IRQ_uv);
		if (!IS_ERR(xpc_activate_mq_uv))
			break;
	}
	if (IS_ERR(xpc_activate_mq_uv)) {
		put_online_cpus();
		return PTR_ERR(xpc_activate_mq_uv);
	}

	for_each_cpu(cpu, cpumask_of_node(nid)) {
		xpc_notify_mq_uv =
			xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
					     XPC_NOTIFY_IRQ_NAME,
					     xpc_handle_notify_IRQ_uv);
		if (!IS_ERR(xpc_notify_mq_uv))
			break;
	}
	if (IS_ERR(xpc_notify_mq_uv)) {
		xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
		put_online_cpus();
		return PTR_ERR(xpc_notify_mq_uv);
	}

	put_online_cpus();
	return 0;
}

int
xpc_init_uv(void)
{
	int nid;
	int ret = 0;

	xpc_arch_ops = xpc_arch_ops_uv;

	if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
		dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
			XPC_MSG_HDR_MAX_SIZE);
		return -E2BIG;
	}

	if (xpc_mq_node < 0)
		for_each_online_node(nid) {
			ret = xpc_init_mq_node(nid);

			if (!ret)
				break;
		}
	else
		ret = xpc_init_mq_node(xpc_mq_node);

	if (ret < 0)
		dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
			-ret);

	return ret;
}

void
xpc_exit_uv(void)
{
	xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
	xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
}

module_param(xpc_mq_node, int, 0);
MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");