[IA64] move XP and XPC to drivers/misc/sgi-xp

Move XPC and XPNET from arch/ia64/sn/kernel to drivers/misc/sgi-xp.

Signed-off-by: Dean Nelson <dcn@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

9 files changed, 7856 insertions, 0 deletions

diff --git a/drivers/misc/sgi-xp/Makefile b/drivers/misc/sgi-xp/Makefile
new file mode 100644
index 000000000000..b6e40a7958ce
--- /dev/null
+++ b/drivers/misc/sgi-xp/Makefile
@@ -0,0 +1,11 @@
+#
+# Makefile for SGI's XP devices.
+#
+
+obj-$(CONFIG_SGI_XP)            += xp.o
+xp-y                            := xp_main.o xp_nofault.o
+
+obj-$(CONFIG_SGI_XP)            += xpc.o
+xpc-y                           := xpc_main.o xpc_channel.o xpc_partition.o
+
+obj-$(CONFIG_SGI_XP)            += xpnet.o
diff --git a/drivers/misc/sgi-xp/xp.h b/drivers/misc/sgi-xp/xp.h
new file mode 100644
index 000000000000..fb65981754c3
--- /dev/null
+++ b/drivers/misc/sgi-xp/xp.h
@@ -0,0 +1,485 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc. All rights reserved.
+ */
+
+
+/*
+ * External Cross Partition (XP) structures and defines.
+ */
+
+
+#ifndef _DRIVERS_MISC_SGIXP_XP_H
+#define _DRIVERS_MISC_SGIXP_XP_H
+
+
+#include <linux/cache.h>
+#include <linux/hardirq.h>
+#include <linux/mutex.h>
+#include <asm/sn/types.h>
+#include <asm/sn/bte.h>
+
+
+#ifdef USE_DBUG_ON
+#define DBUG_ON(condition)      BUG_ON(condition)
+#else
+#define DBUG_ON(condition)
+#endif
+
+
+/*
+ * Define the maximum number of logically defined partitions the system
+ * can support. It is constrained by the maximum number of hardware
+ * partitionable regions. The term 'region' in this context refers to the
+ * minimum number of nodes that can comprise an access protection grouping.
+ * The access protection is in regards to memory, IPI and IOI.
+ *
+ * The maximum number of hardware partitionable regions is equal to the
+ * maximum number of nodes in the entire system divided by the minimum number
+ * of nodes that comprise an access protection grouping.
+ */
+#define XP_MAX_PARTITIONS       64
+
+
+/*
+ * Define the number of u64s required to represent all the C-brick nasids
+ * as a bitmap.  The cross-partition kernel modules deal only with
+ * C-brick nasids, thus the need for bitmaps which don't account for
+ * odd-numbered (non C-brick) nasids.
+ */
+#define XP_MAX_PHYSNODE_ID      (MAX_NUMALINK_NODES / 2)
+#define XP_NASID_MASK_BYTES     ((XP_MAX_PHYSNODE_ID + 7) / 8)
+#define XP_NASID_MASK_WORDS     ((XP_MAX_PHYSNODE_ID + 63) / 64)
+
+
+/*
+ * Wrapper for bte_copy() that should it return a failure status will retry
+ * the bte_copy() once in the hope that the failure was due to a temporary
+ * aberration (i.e., the link going down temporarily).
+ *
+ *      src - physical address of the source of the transfer.
+ *      vdst - virtual address of the destination of the transfer.
+ *      len - number of bytes to transfer from source to destination.
+ *      mode - see bte_copy() for definition.
+ *      notification - see bte_copy() for definition.
+ *
+ * Note: xp_bte_copy() should never be called while holding a spinlock.
+ */
+static inline bte_result_t
+xp_bte_copy(u64 src, u64 vdst, u64 len, u64 mode, void *notification)
+{
+        bte_result_t ret;
+        u64 pdst = ia64_tpa(vdst);
+
+
+        /*
+         * Ensure that the physically mapped memory is contiguous.
+         *
+         * We do this by ensuring that the memory is from region 7 only.
+         * If the need should arise to use memory from one of the other
+         * regions, then modify the BUG_ON() statement to ensure that the
+         * memory from that region is always physically contiguous.
+         */
+        BUG_ON(REGION_NUMBER(vdst) != RGN_KERNEL);
+
+        ret = bte_copy(src, pdst, len, mode, notification);
+        if ((ret != BTE_SUCCESS) && BTE_ERROR_RETRY(ret)) {
+                if (!in_interrupt()) {
+                        cond_resched();
+                }
+                ret = bte_copy(src, pdst, len, mode, notification);
+        }
+
+        return ret;
+}
+
+
+/*
+ * XPC establishes channel connections between the local partition and any
+ * other partition that is currently up. Over these channels, kernel-level
+ * `users' can communicate with their counterparts on the other partitions.
+ *
+ * The maxinum number of channels is limited to eight. For performance reasons,
+ * the internal cross partition structures require sixteen bytes per channel,
+ * and eight allows all of this interface-shared info to fit in one cache line.
+ *
+ * XPC_NCHANNELS reflects the total number of channels currently defined.
+ * If the need for additional channels arises, one can simply increase
+ * XPC_NCHANNELS accordingly. If the day should come where that number
+ * exceeds the MAXIMUM number of channels allowed (eight), then one will need
+ * to make changes to the XPC code to allow for this.
+ */
+#define XPC_MEM_CHANNEL         0       /* memory channel number */
+#define XPC_NET_CHANNEL         1       /* network channel number */
+
+#define XPC_NCHANNELS           2       /* #of defined channels */
+#define XPC_MAX_NCHANNELS       8       /* max #of channels allowed */
+
+#if XPC_NCHANNELS > XPC_MAX_NCHANNELS
+#error  XPC_NCHANNELS exceeds MAXIMUM allowed.
+#endif
+
+
+/*
+ * The format of an XPC message is as follows:
+ *
+ *      +-------+--------------------------------+
+ *      | flags |////////////////////////////////|
+ *      +-------+--------------------------------+
+ *      |             message #                  |
+ *      +----------------------------------------+
+ *      |     payload (user-defined message)     |
+ *      |                                        |
+ *                      :
+ *      |                                        |
+ *      +----------------------------------------+
+ *
+ * The size of the payload is defined by the user via xpc_connect(). A user-
+ * defined message resides in the payload area.
+ *
+ * The user should have no dealings with the message header, but only the
+ * message's payload. When a message entry is allocated (via xpc_allocate())
+ * a pointer to the payload area is returned and not the actual beginning of
+ * the XPC message. The user then constructs a message in the payload area
+ * and passes that pointer as an argument on xpc_send() or xpc_send_notify().
+ *
+ * The size of a message entry (within a message queue) must be a cacheline
+ * sized multiple in order to facilitate the BTE transfer of messages from one
+ * message queue to another. A macro, XPC_MSG_SIZE(), is provided for the user
+ * that wants to fit as many msg entries as possible in a given memory size
+ * (e.g. a memory page).
+ */
+struct xpc_msg {
+        u8 flags;               /* FOR XPC INTERNAL USE ONLY */
+        u8 reserved[7];         /* FOR XPC INTERNAL USE ONLY */
+        s64 number;             /* FOR XPC INTERNAL USE ONLY */
+
+        u64 payload;            /* user defined portion of message */
+};
+
+
+#define XPC_MSG_PAYLOAD_OFFSET  (u64) (&((struct xpc_msg *)0)->payload)
+#define XPC_MSG_SIZE(_payload_size) \
+                L1_CACHE_ALIGN(XPC_MSG_PAYLOAD_OFFSET + (_payload_size))
+
+
+/*
+ * Define the return values and values passed to user's callout functions.
+ * (It is important to add new value codes at the end just preceding
+ * xpcUnknownReason, which must have the highest numerical value.)
+ */
+enum xpc_retval {
+        xpcSuccess = 0,
+
+        xpcNotConnected,        /*  1: channel is not connected */
+        xpcConnected,           /*  2: channel connected (opened) */
+        xpcRETIRED1,            /*  3: (formerly xpcDisconnected) */
+
+        xpcMsgReceived,         /*  4: message received */
+        xpcMsgDelivered,        /*  5: message delivered and acknowledged */
+
+        xpcRETIRED2,            /*  6: (formerly xpcTransferFailed) */
+
+        xpcNoWait,              /*  7: operation would require wait */
+        xpcRetry,               /*  8: retry operation */
+        xpcTimeout,             /*  9: timeout in xpc_allocate_msg_wait() */
+        xpcInterrupted,         /* 10: interrupted wait */
+
+        xpcUnequalMsgSizes,     /* 11: message size disparity between sides */
+        xpcInvalidAddress,      /* 12: invalid address */
+
+        xpcNoMemory,            /* 13: no memory available for XPC structures */
+        xpcLackOfResources,     /* 14: insufficient resources for operation */
+        xpcUnregistered,        /* 15: channel is not registered */
+        xpcAlreadyRegistered,   /* 16: channel is already registered */
+
+        xpcPartitionDown,       /* 17: remote partition is down */
+        xpcNotLoaded,           /* 18: XPC module is not loaded */
+        xpcUnloading,           /* 19: this side is unloading XPC module */
+
+        xpcBadMagic,            /* 20: XPC MAGIC string not found */
+
+        xpcReactivating,        /* 21: remote partition was reactivated */
+
+        xpcUnregistering,       /* 22: this side is unregistering channel */
+        xpcOtherUnregistering,  /* 23: other side is unregistering channel */
+
+        xpcCloneKThread,        /* 24: cloning kernel thread */
+        xpcCloneKThreadFailed,  /* 25: cloning kernel thread failed */
+
+        xpcNoHeartbeat,         /* 26: remote partition has no heartbeat */
+
+        xpcPioReadError,        /* 27: PIO read error */
+        xpcPhysAddrRegFailed,   /* 28: registration of phys addr range failed */
+
+        xpcBteDirectoryError,   /* 29: maps to BTEFAIL_DIR */
+        xpcBtePoisonError,      /* 30: maps to BTEFAIL_POISON */
+        xpcBteWriteError,       /* 31: maps to BTEFAIL_WERR */
+        xpcBteAccessError,      /* 32: maps to BTEFAIL_ACCESS */
+        xpcBtePWriteError,      /* 33: maps to BTEFAIL_PWERR */
+        xpcBtePReadError,       /* 34: maps to BTEFAIL_PRERR */
+        xpcBteTimeOutError,     /* 35: maps to BTEFAIL_TOUT */
+        xpcBteXtalkError,       /* 36: maps to BTEFAIL_XTERR */
+        xpcBteNotAvailable,     /* 37: maps to BTEFAIL_NOTAVAIL */
+        xpcBteUnmappedError,    /* 38: unmapped BTEFAIL_ error */
+
+        xpcBadVersion,          /* 39: bad version number */
+        xpcVarsNotSet,          /* 40: the XPC variables are not set up */
+        xpcNoRsvdPageAddr,      /* 41: unable to get rsvd page's phys addr */
+        xpcInvalidPartid,       /* 42: invalid partition ID */
+        xpcLocalPartid,         /* 43: local partition ID */
+
+        xpcOtherGoingDown,      /* 44: other side going down, reason unknown */
+        xpcSystemGoingDown,     /* 45: system is going down, reason unknown */
+        xpcSystemHalt,          /* 46: system is being halted */
+        xpcSystemReboot,        /* 47: system is being rebooted */
+        xpcSystemPoweroff,      /* 48: system is being powered off */
+
+        xpcDisconnecting,       /* 49: channel disconnecting (closing) */
+
+        xpcOpenCloseError,      /* 50: channel open/close protocol error */
+
+        xpcDisconnected,        /* 51: channel disconnected (closed) */
+
+        xpcBteSh2Start,         /* 52: BTE CRB timeout */
+
+                                /* 53: 0x1 BTE Error Response Short */
+        xpcBteSh2RspShort = xpcBteSh2Start + BTEFAIL_SH2_RESP_SHORT,
+
+                                /* 54: 0x2 BTE Error Response Long */
+        xpcBteSh2RspLong = xpcBteSh2Start + BTEFAIL_SH2_RESP_LONG,
+
+                                /* 56: 0x4 BTE Error Response DSB */
+        xpcBteSh2RspDSB = xpcBteSh2Start + BTEFAIL_SH2_RESP_DSP,
+
+                                /* 60: 0x8 BTE Error Response Access */
+        xpcBteSh2RspAccess = xpcBteSh2Start + BTEFAIL_SH2_RESP_ACCESS,
+
+                                /* 68: 0x10 BTE Error CRB timeout */
+        xpcBteSh2CRBTO = xpcBteSh2Start + BTEFAIL_SH2_CRB_TO,
+
+                                /* 84: 0x20 BTE Error NACK limit */
+        xpcBteSh2NACKLimit = xpcBteSh2Start + BTEFAIL_SH2_NACK_LIMIT,
+
+                                /* 115: BTE end */
+        xpcBteSh2End = xpcBteSh2Start + BTEFAIL_SH2_ALL,
+
+        xpcUnknownReason        /* 116: unknown reason -- must be last in list */
+};
+
+
+/*
+ * Define the callout function types used by XPC to update the user on
+ * connection activity and state changes (via the user function registered by
+ * xpc_connect()) and to notify them of messages received and delivered (via
+ * the user function registered by xpc_send_notify()).
+ *
+ * The two function types are xpc_channel_func and xpc_notify_func and
+ * both share the following arguments, with the exception of "data", which
+ * only xpc_channel_func has.
+ *
+ * Arguments:
+ *
+ *      reason - reason code. (See following table.)
+ *      partid - partition ID associated with condition.
+ *      ch_number - channel # associated with condition.
+ *      data - pointer to optional data. (See following table.)
+ *      key - pointer to optional user-defined value provided as the "key"
+ *            argument to xpc_connect() or xpc_send_notify().
+ *
+ * In the following table the "Optional Data" column applies to callouts made
+ * to functions registered by xpc_connect(). A "NA" in that column indicates
+ * that this reason code can be passed to functions registered by
+ * xpc_send_notify() (i.e. they don't have data arguments).
+ *
+ * Also, the first three reason codes in the following table indicate
+ * success, whereas the others indicate failure. When a failure reason code
+ * is received, one can assume that the channel is not connected.
+ *
+ *
+ * Reason Code          | Cause                          | Optional Data
+ * =====================+================================+=====================
+ * xpcConnected         | connection has been established| max #of entries
+ *                      | to the specified partition on  | allowed in message
+ *                      | the specified channel          | queue
+ * ---------------------+--------------------------------+---------------------
+ * xpcMsgReceived       | an XPC message arrived from    | address of payload
+ *                      | the specified partition on the |
+ *                      | specified channel              | [the user must call
+ *                      |                                | xpc_received() when
+ *                      |                                | finished with the
+ *                      |                                | payload]
+ * ---------------------+--------------------------------+---------------------
+ * xpcMsgDelivered      | notification that the message  | NA
+ *                      | was delivered to the intended  |
+ *                      | recipient and that they have   |
+ *                      | acknowledged its receipt by    |
+ *                      | calling xpc_received()         |
+ * =====================+================================+=====================
+ * xpcUnequalMsgSizes   | can't connect to the specified | NULL
+ *                      | partition on the specified     |
+ *                      | channel because of mismatched  |
+ *                      | message sizes                  |
+ * ---------------------+--------------------------------+---------------------
+ * xpcNoMemory          | insufficient memory avaiable   | NULL
+ *                      | to allocate message queue      |
+ * ---------------------+--------------------------------+---------------------
+ * xpcLackOfResources   | lack of resources to create    | NULL
+ *                      | the necessary kthreads to      |
+ *                      | support the channel            |
+ * ---------------------+--------------------------------+---------------------
+ * xpcUnregistering     | this side's user has           | NULL or NA
+ *                      | unregistered by calling        |
+ *                      | xpc_disconnect()               |
+ * ---------------------+--------------------------------+---------------------
+ * xpcOtherUnregistering| the other side's user has      | NULL or NA
+ *                      | unregistered by calling        |
+ *                      | xpc_disconnect()               |
+ * ---------------------+--------------------------------+---------------------
+ * xpcNoHeartbeat       | the other side's XPC is no     | NULL or NA
+ *                      | longer heartbeating            |
+ *                      |                                |
+ * ---------------------+--------------------------------+---------------------
+ * xpcUnloading         | this side's XPC module is      | NULL or NA
+ *                      | being unloaded                 |
+ *                      |                                |
+ * ---------------------+--------------------------------+---------------------
+ * xpcOtherUnloading    | the other side's XPC module is | NULL or NA
+ *                      | is being unloaded              |
+ *                      |                                |
+ * ---------------------+--------------------------------+---------------------
+ * xpcPioReadError      | xp_nofault_PIOR() returned an  | NULL or NA
+ *                      | error while sending an IPI     |
+ *                      |                                |
+ * ---------------------+--------------------------------+---------------------
+ * xpcInvalidAddress    | the address either received or | NULL or NA
+ *                      | sent by the specified partition|
+ *                      | is invalid                     |
+ * ---------------------+--------------------------------+---------------------
+ * xpcBteNotAvailable   | attempt to pull data from the  | NULL or NA
+ * xpcBtePoisonError    | specified partition over the   |
+ * xpcBteWriteError     | specified channel via a        |
+ * xpcBteAccessError    | bte_copy() failed              |
+ * xpcBteTimeOutError   |                                |
+ * xpcBteXtalkError     |                                |
+ * xpcBteDirectoryError |                                |
+ * xpcBteGenericError   |                                |
+ * xpcBteUnmappedError  |                                |
+ * ---------------------+--------------------------------+---------------------
+ * xpcUnknownReason     | the specified channel to the   | NULL or NA
+ *                      | specified partition was        |
+ *                      | unavailable for unknown reasons|
+ * =====================+================================+=====================
+ */
+
+typedef void (*xpc_channel_func)(enum xpc_retval reason, partid_t partid,
+                int ch_number, void *data, void *key);
+
+typedef void (*xpc_notify_func)(enum xpc_retval reason, partid_t partid,
+                int ch_number, void *key);
+
+
+/*
+ * The following is a registration entry. There is a global array of these,
+ * one per channel. It is used to record the connection registration made
+ * by the users of XPC. As long as a registration entry exists, for any
+ * partition that comes up, XPC will attempt to establish a connection on
+ * that channel. Notification that a connection has been made will occur via
+ * the xpc_channel_func function.
+ *
+ * The 'func' field points to the function to call when aynchronous
+ * notification is required for such events as: a connection established/lost,
+ * or an incoming message received, or an error condition encountered. A
+ * non-NULL 'func' field indicates that there is an active registration for
+ * the channel.
+ */
+struct xpc_registration {
+        struct mutex mutex;
+        xpc_channel_func func;          /* function to call */
+        void *key;                      /* pointer to user's key */
+        u16 nentries;                   /* #of msg entries in local msg queue */
+        u16 msg_size;                   /* message queue's message size */
+        u32 assigned_limit;             /* limit on #of assigned kthreads */
+        u32 idle_limit;                 /* limit on #of idle kthreads */
+} ____cacheline_aligned;
+
+
+#define XPC_CHANNEL_REGISTERED(_c)      (xpc_registrations[_c].func != NULL)
+
+
+/* the following are valid xpc_allocate() flags */
+#define XPC_WAIT        0               /* wait flag */
+#define XPC_NOWAIT      1               /* no wait flag */
+
+
+struct xpc_interface {
+        void (*connect)(int);
+        void (*disconnect)(int);
+        enum xpc_retval (*allocate)(partid_t, int, u32, void **);
+        enum xpc_retval (*send)(partid_t, int, void *);
+        enum xpc_retval (*send_notify)(partid_t, int, void *,
+                                                xpc_notify_func, void *);
+        void (*received)(partid_t, int, void *);
+        enum xpc_retval (*partid_to_nasids)(partid_t, void *);
+};
+
+
+extern struct xpc_interface xpc_interface;
+
+extern void xpc_set_interface(void (*)(int),
+                void (*)(int),
+                enum xpc_retval (*)(partid_t, int, u32, void **),
+                enum xpc_retval (*)(partid_t, int, void *),
+                enum xpc_retval (*)(partid_t, int, void *, xpc_notify_func,
+                                                                void *),
+                void (*)(partid_t, int, void *),
+                enum xpc_retval (*)(partid_t, void *));
+extern void xpc_clear_interface(void);
+
+
+extern enum xpc_retval xpc_connect(int, xpc_channel_func, void *, u16,
+                                                u16, u32, u32);
+extern void xpc_disconnect(int);
+
+static inline enum xpc_retval
+xpc_allocate(partid_t partid, int ch_number, u32 flags, void **payload)
+{
+        return xpc_interface.allocate(partid, ch_number, flags, payload);
+}
+
+static inline enum xpc_retval
+xpc_send(partid_t partid, int ch_number, void *payload)
+{
+        return xpc_interface.send(partid, ch_number, payload);
+}
+
+static inline enum xpc_retval
+xpc_send_notify(partid_t partid, int ch_number, void *payload,
+                        xpc_notify_func func, void *key)
+{
+        return xpc_interface.send_notify(partid, ch_number, payload, func, key);
+}
+
+static inline void
+xpc_received(partid_t partid, int ch_number, void *payload)
+{
+        return xpc_interface.received(partid, ch_number, payload);
+}
+
+static inline enum xpc_retval
+xpc_partid_to_nasids(partid_t partid, void *nasids)
+{
+        return xpc_interface.partid_to_nasids(partid, nasids);
+}
+
+
+extern u64 xp_nofault_PIOR_target;
+extern int xp_nofault_PIOR(void *);
+extern int xp_error_PIOR(void);
+
+
+#endif /* _DRIVERS_MISC_SGIXP_XP_H */
+
diff --git a/drivers/misc/sgi-xp/xp_main.c b/drivers/misc/sgi-xp/xp_main.c
new file mode 100644
index 000000000000..5f9f9c2e9298
--- /dev/null
+++ b/drivers/misc/sgi-xp/xp_main.c
@@ -0,0 +1,290 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * Cross Partition (XP) base.
+ *
+ *      XP provides a base from which its users can interact
+ *      with XPC, yet not be dependent on XPC.
+ *
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <asm/sn/intr.h>
+#include <asm/sn/sn_sal.h>
+#include "xp.h"
+
+
+/*
+ * Target of nofault PIO read.
+ */
+u64 xp_nofault_PIOR_target;
+
+
+/*
+ * xpc_registrations[] keeps track of xpc_connect()'s done by the kernel-level
+ * users of XPC.
+ */
+struct xpc_registration xpc_registrations[XPC_NCHANNELS];
+
+
+/*
+ * Initialize the XPC interface to indicate that XPC isn't loaded.
+ */
+static enum xpc_retval xpc_notloaded(void) { return xpcNotLoaded; }
+
+struct xpc_interface xpc_interface = {
+        (void (*)(int)) xpc_notloaded,
+        (void (*)(int)) xpc_notloaded,
+        (enum xpc_retval (*)(partid_t, int, u32, void **)) xpc_notloaded,
+        (enum xpc_retval (*)(partid_t, int, void *)) xpc_notloaded,
+        (enum xpc_retval (*)(partid_t, int, void *, xpc_notify_func, void *))
+                                                        xpc_notloaded,
+        (void (*)(partid_t, int, void *)) xpc_notloaded,
+        (enum xpc_retval (*)(partid_t, void *)) xpc_notloaded
+};
+
+
+/*
+ * XPC calls this when it (the XPC module) has been loaded.
+ */
+void
+xpc_set_interface(void (*connect)(int),
+                void (*disconnect)(int),
+                enum xpc_retval (*allocate)(partid_t, int, u32, void **),
+                enum xpc_retval (*send)(partid_t, int, void *),
+                enum xpc_retval (*send_notify)(partid_t, int, void *,
+                                                xpc_notify_func, void *),
+                void (*received)(partid_t, int, void *),
+                enum xpc_retval (*partid_to_nasids)(partid_t, void *))
+{
+        xpc_interface.connect = connect;
+        xpc_interface.disconnect = disconnect;
+        xpc_interface.allocate = allocate;
+        xpc_interface.send = send;
+        xpc_interface.send_notify = send_notify;
+        xpc_interface.received = received;
+        xpc_interface.partid_to_nasids = partid_to_nasids;
+}
+
+
+/*
+ * XPC calls this when it (the XPC module) is being unloaded.
+ */
+void
+xpc_clear_interface(void)
+{
+        xpc_interface.connect = (void (*)(int)) xpc_notloaded;
+        xpc_interface.disconnect = (void (*)(int)) xpc_notloaded;
+        xpc_interface.allocate = (enum xpc_retval (*)(partid_t, int, u32,
+                                        void **)) xpc_notloaded;
+        xpc_interface.send = (enum xpc_retval (*)(partid_t, int, void *))
+                                        xpc_notloaded;
+        xpc_interface.send_notify = (enum xpc_retval (*)(partid_t, int, void *,
+                                    xpc_notify_func, void *)) xpc_notloaded;
+        xpc_interface.received = (void (*)(partid_t, int, void *))
+                                        xpc_notloaded;
+        xpc_interface.partid_to_nasids = (enum xpc_retval (*)(partid_t, void *))
+                                        xpc_notloaded;
+}
+
+
+/*
+ * Register for automatic establishment of a channel connection whenever
+ * a partition comes up.
+ *
+ * Arguments:
+ *
+ *      ch_number - channel # to register for connection.
+ *      func - function to call for asynchronous notification of channel
+ *             state changes (i.e., connection, disconnection, error) and
+ *             the arrival of incoming messages.
+ *      key - pointer to optional user-defined value that gets passed back
+ *            to the user on any callouts made to func.
+ *      payload_size - size in bytes of the XPC message's payload area which
+ *                     contains a user-defined message. The user should make
+ *                     this large enough to hold their largest message.
+ *      nentries - max #of XPC message entries a message queue can contain.
+ *                 The actual number, which is determined when a connection
+ *                 is established and may be less then requested, will be
+ *                 passed to the user via the xpcConnected callout.
+ *      assigned_limit - max number of kthreads allowed to be processing
+ *                       messages (per connection) at any given instant.
+ *      idle_limit - max number of kthreads allowed to be idle at any given
+ *                   instant.
+ */
+enum xpc_retval
+xpc_connect(int ch_number, xpc_channel_func func, void *key, u16 payload_size,
+                u16 nentries, u32 assigned_limit, u32 idle_limit)
+{
+        struct xpc_registration *registration;
+
+
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+        DBUG_ON(payload_size == 0 || nentries == 0);
+        DBUG_ON(func == NULL);
+        DBUG_ON(assigned_limit == 0 || idle_limit > assigned_limit);
+
+        registration = &xpc_registrations[ch_number];
+
+        if (mutex_lock_interruptible(&registration->mutex) != 0) {
+                return xpcInterrupted;
+        }
+
+        /* if XPC_CHANNEL_REGISTERED(ch_number) */
+        if (registration->func != NULL) {
+                mutex_unlock(&registration->mutex);
+                return xpcAlreadyRegistered;
+        }
+
+        /* register the channel for connection */
+        registration->msg_size = XPC_MSG_SIZE(payload_size);
+        registration->nentries = nentries;
+        registration->assigned_limit = assigned_limit;
+        registration->idle_limit = idle_limit;
+        registration->key = key;
+        registration->func = func;
+
+        mutex_unlock(&registration->mutex);
+
+        xpc_interface.connect(ch_number);
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Remove the registration for automatic connection of the specified channel
+ * when a partition comes up.
+ *
+ * Before returning this xpc_disconnect() will wait for all connections on the
+ * specified channel have been closed/torndown. So the caller can be assured
+ * that they will not be receiving any more callouts from XPC to their
+ * function registered via xpc_connect().
+ *
+ * Arguments:
+ *
+ *      ch_number - channel # to unregister.
+ */
+void
+xpc_disconnect(int ch_number)
+{
+        struct xpc_registration *registration;
+
+
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+
+        registration = &xpc_registrations[ch_number];
+
+        /*
+         * We've decided not to make this a down_interruptible(), since we
+         * figured XPC's users will just turn around and call xpc_disconnect()
+         * again anyways, so we might as well wait, if need be.
+         */
+        mutex_lock(&registration->mutex);
+
+        /* if !XPC_CHANNEL_REGISTERED(ch_number) */
+        if (registration->func == NULL) {
+                mutex_unlock(&registration->mutex);
+                return;
+        }
+
+        /* remove the connection registration for the specified channel */
+        registration->func = NULL;
+        registration->key = NULL;
+        registration->nentries = 0;
+        registration->msg_size = 0;
+        registration->assigned_limit = 0;
+        registration->idle_limit = 0;
+
+        xpc_interface.disconnect(ch_number);
+
+        mutex_unlock(&registration->mutex);
+
+        return;
+}
+
+
+int __init
+xp_init(void)
+{
+        int ret, ch_number;
+        u64 func_addr = *(u64 *) xp_nofault_PIOR;
+        u64 err_func_addr = *(u64 *) xp_error_PIOR;
+
+
+        if (!ia64_platform_is("sn2")) {
+                return -ENODEV;
+        }
+
+        /*
+         * Register a nofault code region which performs a cross-partition
+         * PIO read. If the PIO read times out, the MCA handler will consume
+         * the error and return to a kernel-provided instruction to indicate
+         * an error. This PIO read exists because it is guaranteed to timeout
+         * if the destination is down (AMO operations do not timeout on at
+         * least some CPUs on Shubs <= v1.2, which unfortunately we have to
+         * work around).
+         */
+        if ((ret = sn_register_nofault_code(func_addr, err_func_addr,
+                                                err_func_addr, 1, 1)) != 0) {
+                printk(KERN_ERR "XP: can't register nofault code, error=%d\n",
+                        ret);
+        }
+        /*
+         * Setup the nofault PIO read target. (There is no special reason why
+         * SH_IPI_ACCESS was selected.)
+         */
+        if (is_shub2()) {
+                xp_nofault_PIOR_target = SH2_IPI_ACCESS0;
+        } else {
+                xp_nofault_PIOR_target = SH1_IPI_ACCESS;
+        }
+
+        /* initialize the connection registration mutex */
+        for (ch_number = 0; ch_number < XPC_NCHANNELS; ch_number++) {
+                mutex_init(&xpc_registrations[ch_number].mutex);
+        }
+
+        return 0;
+}
+module_init(xp_init);
+
+
+void __exit
+xp_exit(void)
+{
+        u64 func_addr = *(u64 *) xp_nofault_PIOR;
+        u64 err_func_addr = *(u64 *) xp_error_PIOR;
+
+
+        /* unregister the PIO read nofault code region */
+        (void) sn_register_nofault_code(func_addr, err_func_addr,
+                                        err_func_addr, 1, 0);
+}
+module_exit(xp_exit);
+
+
+MODULE_AUTHOR("Silicon Graphics, Inc.");
+MODULE_DESCRIPTION("Cross Partition (XP) base");
+MODULE_LICENSE("GPL");
+
+EXPORT_SYMBOL(xp_nofault_PIOR);
+EXPORT_SYMBOL(xp_nofault_PIOR_target);
+EXPORT_SYMBOL(xpc_registrations);
+EXPORT_SYMBOL(xpc_interface);
+EXPORT_SYMBOL(xpc_clear_interface);
+EXPORT_SYMBOL(xpc_set_interface);
+EXPORT_SYMBOL(xpc_connect);
+EXPORT_SYMBOL(xpc_disconnect);
+
diff --git a/drivers/misc/sgi-xp/xp_nofault.S b/drivers/misc/sgi-xp/xp_nofault.S
new file mode 100644
index 000000000000..c13a709c4db5
--- /dev/null
+++ b/drivers/misc/sgi-xp/xp_nofault.S
@@ -0,0 +1,36 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * The xp_nofault_PIOR function takes a pointer to a remote PIO register
+ * and attempts to load and consume a value from it.  This function
+ * will be registered as a nofault code block.  In the event that the
+ * PIO read fails, the MCA handler will force the error to look
+ * corrected and vector to the xp_error_PIOR which will return an error.
+ *
+ * The definition of "consumption" and the time it takes for an MCA
+ * to surface is processor implementation specific.  This code
+ * is sufficient on Itanium through the Montvale processor family.
+ * It may need to be adjusted for future processor implementations.
+ *
+ *      extern int xp_nofault_PIOR(void *remote_register);
+ */
+
+        .global xp_nofault_PIOR
+xp_nofault_PIOR:
+        mov     r8=r0                   // Stage a success return value
+        ld8.acq r9=[r32];;              // PIO Read the specified register
+        adds    r9=1,r9;;               // Add to force consumption
+        srlz.i;;                        // Allow time for MCA to surface
+        br.ret.sptk.many b0;;           // Return success
+
+        .global xp_error_PIOR
+xp_error_PIOR:
+        mov     r8=1                    // Return value of 1
+        br.ret.sptk.many b0;;           // Return failure
diff --git a/drivers/misc/sgi-xp/xpc.h b/drivers/misc/sgi-xp/xpc.h
new file mode 100644
index 000000000000..14e70ee53ebe
--- /dev/null
+++ b/drivers/misc/sgi-xp/xpc.h
@@ -0,0 +1,1267 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * Cross Partition Communication (XPC) structures and macros.
+ */
+
+#ifndef _DRIVERS_MISC_SGIXP_XPC_H
+#define _DRIVERS_MISC_SGIXP_XPC_H
+
+
+#include <linux/interrupt.h>
+#include <linux/sysctl.h>
+#include <linux/device.h>
+#include <linux/mutex.h>
+#include <linux/completion.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/sn/bte.h>
+#include <asm/sn/clksupport.h>
+#include <asm/sn/addrs.h>
+#include <asm/sn/mspec.h>
+#include <asm/sn/shub_mmr.h>
+#include "xp.h"
+
+
+/*
+ * XPC Version numbers consist of a major and minor number. XPC can always
+ * talk to versions with same major #, and never talk to versions with a
+ * different major #.
+ */
+#define _XPC_VERSION(_maj, _min)        (((_maj) << 4) | ((_min) & 0xf))
+#define XPC_VERSION_MAJOR(_v)           ((_v) >> 4)
+#define XPC_VERSION_MINOR(_v)           ((_v) & 0xf)
+
+
+/*
+ * The next macros define word or bit representations for given
+ * C-brick nasid in either the SAL provided bit array representing
+ * nasids in the partition/machine or the AMO_t array used for
+ * inter-partition initiation communications.
+ *
+ * For SN2 machines, C-Bricks are alway even numbered NASIDs.  As
+ * such, some space will be saved by insisting that nasid information
+ * passed from SAL always be packed for C-Bricks and the
+ * cross-partition interrupts use the same packing scheme.
+ */
+#define XPC_NASID_W_INDEX(_n)   (((_n) / 64) / 2)
+#define XPC_NASID_B_INDEX(_n)   (((_n) / 2) & (64 - 1))
+#define XPC_NASID_IN_ARRAY(_n, _p) ((_p)[XPC_NASID_W_INDEX(_n)] & \
+                                    (1UL << XPC_NASID_B_INDEX(_n)))
+#define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
+
+#define XPC_HB_DEFAULT_INTERVAL         5       /* incr HB every x secs */
+#define XPC_HB_CHECK_DEFAULT_INTERVAL   20      /* check HB every x secs */
+
+/* define the process name of HB checker and the CPU it is pinned to */
+#define XPC_HB_CHECK_THREAD_NAME        "xpc_hb"
+#define XPC_HB_CHECK_CPU                0
+
+/* define the process name of the discovery thread */
+#define XPC_DISCOVERY_THREAD_NAME       "xpc_discovery"
+
+
+/*
+ * the reserved page
+ *
+ *   SAL reserves one page of memory per partition for XPC. Though a full page
+ *   in length (16384 bytes), its starting address is not page aligned, but it
+ *   is cacheline aligned. The reserved page consists of the following:
+ *
+ *   reserved page header
+ *
+ *     The first cacheline of the reserved page contains the header
+ *     (struct xpc_rsvd_page). Before SAL initialization has completed,
+ *     SAL has set up the following fields of the reserved page header:
+ *     SAL_signature, SAL_version, partid, and nasids_size. The other
+ *     fields are set up by XPC. (xpc_rsvd_page points to the local
+ *     partition's reserved page.)
+ *
+ *   part_nasids mask
+ *   mach_nasids mask
+ *
+ *     SAL also sets up two bitmaps (or masks), one that reflects the actual
+ *     nasids in this partition (part_nasids), and the other that reflects
+ *     the actual nasids in the entire machine (mach_nasids). We're only
+ *     interested in the even numbered nasids (which contain the processors
+ *     and/or memory), so we only need half as many bits to represent the
+ *     nasids. The part_nasids mask is located starting at the first cacheline
+ *     following the reserved page header. The mach_nasids mask follows right
+ *     after the part_nasids mask. The size in bytes of each mask is reflected
+ *     by the reserved page header field 'nasids_size'. (Local partition's
+ *     mask pointers are xpc_part_nasids and xpc_mach_nasids.)
+ *
+ *   vars
+ *   vars part
+ *
+ *     Immediately following the mach_nasids mask are the XPC variables
+ *     required by other partitions. First are those that are generic to all
+ *     partitions (vars), followed on the next available cacheline by those
+ *     which are partition specific (vars part). These are setup by XPC.
+ *     (Local partition's vars pointers are xpc_vars and xpc_vars_part.)
+ *
+ * Note: Until vars_pa is set, the partition XPC code has not been initialized.
+ */
+struct xpc_rsvd_page {
+        u64 SAL_signature;      /* SAL: unique signature */
+        u64 SAL_version;        /* SAL: version */
+        u8 partid;              /* SAL: partition ID */
+        u8 version;
+        u8 pad1[6];             /* align to next u64 in cacheline */
+        volatile u64 vars_pa;
+        struct timespec stamp;  /* time when reserved page was setup by XPC */
+        u64 pad2[9];            /* align to last u64 in cacheline */
+        u64 nasids_size;        /* SAL: size of each nasid mask in bytes */
+};
+
+#define XPC_RP_VERSION _XPC_VERSION(1,1) /* version 1.1 of the reserved page */
+
+#define XPC_SUPPORTS_RP_STAMP(_version) \
+                        (_version >= _XPC_VERSION(1,1))
+
+/*
+ * compare stamps - the return value is:
+ *
+ *      < 0,    if stamp1 < stamp2
+ *      = 0,    if stamp1 == stamp2
+ *      > 0,    if stamp1 > stamp2
+ */
+static inline int
+xpc_compare_stamps(struct timespec *stamp1, struct timespec *stamp2)
+{
+        int ret;
+
+
+        if ((ret = stamp1->tv_sec - stamp2->tv_sec) == 0) {
+                ret = stamp1->tv_nsec - stamp2->tv_nsec;
+        }
+        return ret;
+}
+
+
+/*
+ * Define the structures by which XPC variables can be exported to other
+ * partitions. (There are two: struct xpc_vars and struct xpc_vars_part)
+ */
+
+/*
+ * The following structure describes the partition generic variables
+ * needed by other partitions in order to properly initialize.
+ *
+ * struct xpc_vars version number also applies to struct xpc_vars_part.
+ * Changes to either structure and/or related functionality should be
+ * reflected by incrementing either the major or minor version numbers
+ * of struct xpc_vars.
+ */
+struct xpc_vars {
+        u8 version;
+        u64 heartbeat;
+        u64 heartbeating_to_mask;
+        u64 heartbeat_offline;  /* if 0, heartbeat should be changing */
+        int act_nasid;
+        int act_phys_cpuid;
+        u64 vars_part_pa;
+        u64 amos_page_pa;       /* paddr of page of AMOs from MSPEC driver */
+        AMO_t *amos_page;       /* vaddr of page of AMOs from MSPEC driver */
+};
+
+#define XPC_V_VERSION _XPC_VERSION(3,1) /* version 3.1 of the cross vars */
+
+#define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
+                        (_version >= _XPC_VERSION(3,1))
+
+
+static inline int
+xpc_hb_allowed(partid_t partid, struct xpc_vars *vars)
+{
+        return ((vars->heartbeating_to_mask & (1UL << partid)) != 0);
+}
+
+static inline void
+xpc_allow_hb(partid_t partid, struct xpc_vars *vars)
+{
+        u64 old_mask, new_mask;
+
+        do {
+                old_mask = vars->heartbeating_to_mask;
+                new_mask = (old_mask | (1UL << partid));
+        } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
+                                                        old_mask);
+}
+
+static inline void
+xpc_disallow_hb(partid_t partid, struct xpc_vars *vars)
+{
+        u64 old_mask, new_mask;
+
+        do {
+                old_mask = vars->heartbeating_to_mask;
+                new_mask = (old_mask & ~(1UL << partid));
+        } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
+                                                        old_mask);
+}
+
+
+/*
+ * The AMOs page consists of a number of AMO variables which are divided into
+ * four groups, The first two groups are used to identify an IRQ's sender.
+ * These two groups consist of 64 and 128 AMO variables respectively. The last
+ * two groups, consisting of just one AMO variable each, are used to identify
+ * the remote partitions that are currently engaged (from the viewpoint of
+ * the XPC running on the remote partition).
+ */
+#define XPC_NOTIFY_IRQ_AMOS        0
+#define XPC_ACTIVATE_IRQ_AMOS      (XPC_NOTIFY_IRQ_AMOS + XP_MAX_PARTITIONS)
+#define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
+#define XPC_DISENGAGE_REQUEST_AMO  (XPC_ENGAGED_PARTITIONS_AMO + 1)
+
+
+/*
+ * The following structure describes the per partition specific variables.
+ *
+ * An array of these structures, one per partition, will be defined. As a
+ * partition becomes active XPC will copy the array entry corresponding to
+ * itself from that partition. It is desirable that the size of this
+ * structure evenly divide into a cacheline, such that none of the entries
+ * in this array crosses a cacheline boundary. As it is now, each entry
+ * occupies half a cacheline.
+ */
+struct xpc_vars_part {
+        volatile u64 magic;
+
+        u64 openclose_args_pa;  /* physical address of open and close args */
+        u64 GPs_pa;             /* physical address of Get/Put values */
+
+        u64 IPI_amo_pa;         /* physical address of IPI AMO_t structure */
+        int IPI_nasid;          /* nasid of where to send IPIs */
+        int IPI_phys_cpuid;     /* physical CPU ID of where to send IPIs */
+
+        u8 nchannels;           /* #of defined channels supported */
+
+        u8 reserved[23];        /* pad to a full 64 bytes */
+};
+
+/*
+ * The vars_part MAGIC numbers play a part in the first contact protocol.
+ *
+ * MAGIC1 indicates that the per partition specific variables for a remote
+ * partition have been initialized by this partition.
+ *
+ * MAGIC2 indicates that this partition has pulled the remote partititions
+ * per partition variables that pertain to this partition.
+ */
+#define XPC_VP_MAGIC1   0x0053524156435058L  /* 'XPCVARS\0'L (little endian) */
+#define XPC_VP_MAGIC2   0x0073726176435058L  /* 'XPCvars\0'L (little endian) */
+
+
+/* the reserved page sizes and offsets */
+
+#define XPC_RP_HEADER_SIZE      L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page))
+#define XPC_RP_VARS_SIZE        L1_CACHE_ALIGN(sizeof(struct xpc_vars))
+
+#define XPC_RP_PART_NASIDS(_rp) (u64 *) ((u8 *) _rp + XPC_RP_HEADER_SIZE)
+#define XPC_RP_MACH_NASIDS(_rp) (XPC_RP_PART_NASIDS(_rp) + xp_nasid_mask_words)
+#define XPC_RP_VARS(_rp)        ((struct xpc_vars *) XPC_RP_MACH_NASIDS(_rp) + xp_nasid_mask_words)
+#define XPC_RP_VARS_PART(_rp)   (struct xpc_vars_part *) ((u8 *) XPC_RP_VARS(rp) + XPC_RP_VARS_SIZE)
+
+
+/*
+ * Functions registered by add_timer() or called by kernel_thread() only
+ * allow for a single 64-bit argument. The following macros can be used to
+ * pack and unpack two (32-bit, 16-bit or 8-bit) arguments into or out from
+ * the passed argument.
+ */
+#define XPC_PACK_ARGS(_arg1, _arg2) \
+                        ((((u64) _arg1) & 0xffffffff) | \
+                        ((((u64) _arg2) & 0xffffffff) << 32))
+
+#define XPC_UNPACK_ARG1(_args)  (((u64) _args) & 0xffffffff)
+#define XPC_UNPACK_ARG2(_args)  ((((u64) _args) >> 32) & 0xffffffff)
+
+
+
+/*
+ * Define a Get/Put value pair (pointers) used with a message queue.
+ */
+struct xpc_gp {
+        volatile s64 get;       /* Get value */
+        volatile s64 put;       /* Put value */
+};
+
+#define XPC_GP_SIZE \
+                L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_NCHANNELS)
+
+
+
+/*
+ * Define a structure that contains arguments associated with opening and
+ * closing a channel.
+ */
+struct xpc_openclose_args {
+        u16 reason;             /* reason why channel is closing */
+        u16 msg_size;           /* sizeof each message entry */
+        u16 remote_nentries;    /* #of message entries in remote msg queue */
+        u16 local_nentries;     /* #of message entries in local msg queue */
+        u64 local_msgqueue_pa;  /* physical address of local message queue */
+};
+
+#define XPC_OPENCLOSE_ARGS_SIZE \
+              L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * XPC_NCHANNELS)
+
+
+
+/* struct xpc_msg flags */
+
+#define XPC_M_DONE              0x01    /* msg has been received/consumed */
+#define XPC_M_READY             0x02    /* msg is ready to be sent */
+#define XPC_M_INTERRUPT         0x04    /* send interrupt when msg consumed */
+
+
+#define XPC_MSG_ADDRESS(_payload) \
+                ((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
+
+
+
+/*
+ * Defines notify entry.
+ *
+ * This is used to notify a message's sender that their message was received
+ * and consumed by the intended recipient.
+ */
+struct xpc_notify {
+        volatile u8 type;               /* type of notification */
+
+        /* the following two fields are only used if type == XPC_N_CALL */
+        xpc_notify_func func;           /* user's notify function */
+        void *key;                      /* pointer to user's key */
+};
+
+/* struct xpc_notify type of notification */
+
+#define XPC_N_CALL              0x01    /* notify function provided by user */
+
+
+
+/*
+ * Define the structure that manages all the stuff required by a channel. In
+ * particular, they are used to manage the messages sent across the channel.
+ *
+ * This structure is private to a partition, and is NOT shared across the
+ * partition boundary.
+ *
+ * There is an array of these structures for each remote partition. It is
+ * allocated at the time a partition becomes active. The array contains one
+ * of these structures for each potential channel connection to that partition.
+ *
+ * Each of these structures manages two message queues (circular buffers).
+ * They are allocated at the time a channel connection is made. One of
+ * these message queues (local_msgqueue) holds the locally created messages
+ * that are destined for the remote partition. The other of these message
+ * queues (remote_msgqueue) is a locally cached copy of the remote partition's
+ * own local_msgqueue.
+ *
+ * The following is a description of the Get/Put pointers used to manage these
+ * two message queues. Consider the local_msgqueue to be on one partition
+ * and the remote_msgqueue to be its cached copy on another partition. A
+ * description of what each of the lettered areas contains is included.
+ *
+ *
+ *                     local_msgqueue      remote_msgqueue
+ *
+ *                        |/////////|      |/////////|
+ *    w_remote_GP.get --> +---------+      |/////////|
+ *                        |    F    |      |/////////|
+ *     remote_GP.get  --> +---------+      +---------+ <-- local_GP->get
+ *                        |         |      |         |
+ *                        |         |      |    E    |
+ *                        |         |      |         |
+ *                        |         |      +---------+ <-- w_local_GP.get
+ *                        |    B    |      |/////////|
+ *                        |         |      |////D////|
+ *                        |         |      |/////////|
+ *                        |         |      +---------+ <-- w_remote_GP.put
+ *                        |         |      |////C////|
+ *      local_GP->put --> +---------+      +---------+ <-- remote_GP.put
+ *                        |         |      |/////////|
+ *                        |    A    |      |/////////|
+ *                        |         |      |/////////|
+ *     w_local_GP.put --> +---------+      |/////////|
+ *                        |/////////|      |/////////|
+ *
+ *
+ *          ( remote_GP.[get|put] are cached copies of the remote
+ *            partition's local_GP->[get|put], and thus their values can
+ *            lag behind their counterparts on the remote partition. )
+ *
+ *
+ *  A - Messages that have been allocated, but have not yet been sent to the
+ *      remote partition.
+ *
+ *  B - Messages that have been sent, but have not yet been acknowledged by the
+ *      remote partition as having been received.
+ *
+ *  C - Area that needs to be prepared for the copying of sent messages, by
+ *      the clearing of the message flags of any previously received messages.
+ *
+ *  D - Area into which sent messages are to be copied from the remote
+ *      partition's local_msgqueue and then delivered to their intended
+ *      recipients. [ To allow for a multi-message copy, another pointer
+ *      (next_msg_to_pull) has been added to keep track of the next message
+ *      number needing to be copied (pulled). It chases after w_remote_GP.put.
+ *      Any messages lying between w_local_GP.get and next_msg_to_pull have
+ *      been copied and are ready to be delivered. ]
+ *
+ *  E - Messages that have been copied and delivered, but have not yet been
+ *      acknowledged by the recipient as having been received.
+ *
+ *  F - Messages that have been acknowledged, but XPC has not yet notified the
+ *      sender that the message was received by its intended recipient.
+ *      This is also an area that needs to be prepared for the allocating of
+ *      new messages, by the clearing of the message flags of the acknowledged
+ *      messages.
+ */
+struct xpc_channel {
+        partid_t partid;                /* ID of remote partition connected */
+        spinlock_t lock;                /* lock for updating this structure */
+        u32 flags;                      /* general flags */
+
+        enum xpc_retval reason;         /* reason why channel is disconnect'g */
+        int reason_line;                /* line# disconnect initiated from */
+
+        u16 number;                     /* channel # */
+
+        u16 msg_size;                   /* sizeof each msg entry */
+        u16 local_nentries;             /* #of msg entries in local msg queue */
+        u16 remote_nentries;            /* #of msg entries in remote msg queue*/
+
+        void *local_msgqueue_base;      /* base address of kmalloc'd space */
+        struct xpc_msg *local_msgqueue; /* local message queue */
+        void *remote_msgqueue_base;     /* base address of kmalloc'd space */
+        struct xpc_msg *remote_msgqueue;/* cached copy of remote partition's */
+                                        /* local message queue */
+        u64 remote_msgqueue_pa;         /* phys addr of remote partition's */
+                                        /* local message queue */
+
+        atomic_t references;            /* #of external references to queues */
+
+        atomic_t n_on_msg_allocate_wq;   /* #on msg allocation wait queue */
+        wait_queue_head_t msg_allocate_wq; /* msg allocation wait queue */
+
+        u8 delayed_IPI_flags;           /* IPI flags received, but delayed */
+                                        /* action until channel disconnected */
+
+        /* queue of msg senders who want to be notified when msg received */
+
+        atomic_t n_to_notify;           /* #of msg senders to notify */
+        struct xpc_notify *notify_queue;/* notify queue for messages sent */
+
+        xpc_channel_func func;          /* user's channel function */
+        void *key;                      /* pointer to user's key */
+
+        struct mutex msg_to_pull_mutex; /* next msg to pull serialization */
+        struct completion wdisconnect_wait; /* wait for channel disconnect */
+
+        struct xpc_openclose_args *local_openclose_args; /* args passed on */
+                                        /* opening or closing of channel */
+
+        /* various flavors of local and remote Get/Put values */
+
+        struct xpc_gp *local_GP;        /* local Get/Put values */
+        struct xpc_gp remote_GP;        /* remote Get/Put values */
+        struct xpc_gp w_local_GP;       /* working local Get/Put values */
+        struct xpc_gp w_remote_GP;      /* working remote Get/Put values */
+        s64 next_msg_to_pull;           /* Put value of next msg to pull */
+
+        /* kthread management related fields */
+
+// >>> rethink having kthreads_assigned_limit and kthreads_idle_limit; perhaps
+// >>> allow the assigned limit be unbounded and let the idle limit be dynamic
+// >>> dependent on activity over the last interval of time
+        atomic_t kthreads_assigned;     /* #of kthreads assigned to channel */
+        u32 kthreads_assigned_limit;    /* limit on #of kthreads assigned */
+        atomic_t kthreads_idle;         /* #of kthreads idle waiting for work */
+        u32 kthreads_idle_limit;        /* limit on #of kthreads idle */
+        atomic_t kthreads_active;       /* #of kthreads actively working */
+        // >>> following field is temporary
+        u32 kthreads_created;           /* total #of kthreads created */
+
+        wait_queue_head_t idle_wq;      /* idle kthread wait queue */
+
+} ____cacheline_aligned;
+
+
+/* struct xpc_channel flags */
+
+#define XPC_C_WASCONNECTED      0x00000001 /* channel was connected */
+
+#define XPC_C_ROPENREPLY        0x00000002 /* remote open channel reply */
+#define XPC_C_OPENREPLY         0x00000004 /* local open channel reply */
+#define XPC_C_ROPENREQUEST      0x00000008 /* remote open channel request */
+#define XPC_C_OPENREQUEST       0x00000010 /* local open channel request */
+
+#define XPC_C_SETUP             0x00000020 /* channel's msgqueues are alloc'd */
+#define XPC_C_CONNECTEDCALLOUT  0x00000040 /* connected callout initiated */
+#define XPC_C_CONNECTEDCALLOUT_MADE \
+                                0x00000080 /* connected callout completed */
+#define XPC_C_CONNECTED         0x00000100 /* local channel is connected */
+#define XPC_C_CONNECTING        0x00000200 /* channel is being connected */
+
+#define XPC_C_RCLOSEREPLY       0x00000400 /* remote close channel reply */
+#define XPC_C_CLOSEREPLY        0x00000800 /* local close channel reply */
+#define XPC_C_RCLOSEREQUEST     0x00001000 /* remote close channel request */
+#define XPC_C_CLOSEREQUEST      0x00002000 /* local close channel request */
+
+#define XPC_C_DISCONNECTED      0x00004000 /* channel is disconnected */
+#define XPC_C_DISCONNECTING     0x00008000 /* channel is being disconnected */
+#define XPC_C_DISCONNECTINGCALLOUT \
+                                0x00010000 /* disconnecting callout initiated */
+#define XPC_C_DISCONNECTINGCALLOUT_MADE \
+                                0x00020000 /* disconnecting callout completed */
+#define XPC_C_WDISCONNECT       0x00040000 /* waiting for channel disconnect */
+
+
+
+/*
+ * Manages channels on a partition basis. There is one of these structures
+ * for each partition (a partition will never utilize the structure that
+ * represents itself).
+ */
+struct xpc_partition {
+
+        /* XPC HB infrastructure */
+
+        u8 remote_rp_version;           /* version# of partition's rsvd pg */
+        struct timespec remote_rp_stamp;/* time when rsvd pg was initialized */
+        u64 remote_rp_pa;               /* phys addr of partition's rsvd pg */
+        u64 remote_vars_pa;             /* phys addr of partition's vars */
+        u64 remote_vars_part_pa;        /* phys addr of partition's vars part */
+        u64 last_heartbeat;             /* HB at last read */
+        u64 remote_amos_page_pa;        /* phys addr of partition's amos page */
+        int remote_act_nasid;           /* active part's act/deact nasid */
+        int remote_act_phys_cpuid;      /* active part's act/deact phys cpuid */
+        u32 act_IRQ_rcvd;               /* IRQs since activation */
+        spinlock_t act_lock;            /* protect updating of act_state */
+        u8 act_state;                   /* from XPC HB viewpoint */
+        u8 remote_vars_version;         /* version# of partition's vars */
+        enum xpc_retval reason;         /* reason partition is deactivating */
+        int reason_line;                /* line# deactivation initiated from */
+        int reactivate_nasid;           /* nasid in partition to reactivate */
+
+        unsigned long disengage_request_timeout; /* timeout in jiffies */
+        struct timer_list disengage_request_timer;
+
+
+        /* XPC infrastructure referencing and teardown control */
+
+        volatile u8 setup_state;        /* infrastructure setup state */
+        wait_queue_head_t teardown_wq;  /* kthread waiting to teardown infra */
+        atomic_t references;            /* #of references to infrastructure */
+
+
+        /*
+         * NONE OF THE PRECEDING FIELDS OF THIS STRUCTURE WILL BE CLEARED WHEN
+         * XPC SETS UP THE NECESSARY INFRASTRUCTURE TO SUPPORT CROSS PARTITION
+         * COMMUNICATION. ALL OF THE FOLLOWING FIELDS WILL BE CLEARED. (THE
+         * 'nchannels' FIELD MUST BE THE FIRST OF THE FIELDS TO BE CLEARED.)
+         */
+
+
+        u8 nchannels;              /* #of defined channels supported */
+        atomic_t nchannels_active; /* #of channels that are not DISCONNECTED */
+        atomic_t nchannels_engaged;/* #of channels engaged with remote part */
+        struct xpc_channel *channels;/* array of channel structures */
+
+        void *local_GPs_base;     /* base address of kmalloc'd space */
+        struct xpc_gp *local_GPs; /* local Get/Put values */
+        void *remote_GPs_base;    /* base address of kmalloc'd space */
+        struct xpc_gp *remote_GPs;/* copy of remote partition's local Get/Put */
+                                  /* values */
+        u64 remote_GPs_pa;        /* phys address of remote partition's local */
+                                  /* Get/Put values */
+
+
+        /* fields used to pass args when opening or closing a channel */
+
+        void *local_openclose_args_base;  /* base address of kmalloc'd space */
+        struct xpc_openclose_args *local_openclose_args;  /* local's args */
+        void *remote_openclose_args_base; /* base address of kmalloc'd space */
+        struct xpc_openclose_args *remote_openclose_args; /* copy of remote's */
+                                          /* args */
+        u64 remote_openclose_args_pa;     /* phys addr of remote's args */
+
+
+        /* IPI sending, receiving and handling related fields */
+
+        int remote_IPI_nasid;       /* nasid of where to send IPIs */
+        int remote_IPI_phys_cpuid;  /* phys CPU ID of where to send IPIs */
+        AMO_t *remote_IPI_amo_va;   /* address of remote IPI AMO_t structure */
+
+        AMO_t *local_IPI_amo_va;    /* address of IPI AMO_t structure */
+        u64 local_IPI_amo;          /* IPI amo flags yet to be handled */
+        char IPI_owner[8];          /* IPI owner's name */
+        struct timer_list dropped_IPI_timer; /* dropped IPI timer */
+
+        spinlock_t IPI_lock;        /* IPI handler lock */
+
+
+        /* channel manager related fields */
+
+        atomic_t channel_mgr_requests;  /* #of requests to activate chan mgr */
+        wait_queue_head_t channel_mgr_wq; /* channel mgr's wait queue */
+
+} ____cacheline_aligned;
+
+
+/* struct xpc_partition act_state values (for XPC HB) */
+
+#define XPC_P_INACTIVE          0x00    /* partition is not active */
+#define XPC_P_ACTIVATION_REQ    0x01    /* created thread to activate */
+#define XPC_P_ACTIVATING        0x02    /* activation thread started */
+#define XPC_P_ACTIVE            0x03    /* xpc_partition_up() was called */
+#define XPC_P_DEACTIVATING      0x04    /* partition deactivation initiated */
+
+
+#define XPC_DEACTIVATE_PARTITION(_p, _reason) \
+                        xpc_deactivate_partition(__LINE__, (_p), (_reason))
+
+
+/* struct xpc_partition setup_state values */
+
+#define XPC_P_UNSET             0x00    /* infrastructure was never setup */
+#define XPC_P_SETUP             0x01    /* infrastructure is setup */
+#define XPC_P_WTEARDOWN         0x02    /* waiting to teardown infrastructure */
+#define XPC_P_TORNDOWN          0x03    /* infrastructure is torndown */
+
+
+
+/*
+ * struct xpc_partition IPI_timer #of seconds to wait before checking for
+ * dropped IPIs. These occur whenever an IPI amo write doesn't complete until
+ * after the IPI was received.
+ */
+#define XPC_P_DROPPED_IPI_WAIT  (0.25 * HZ)
+
+
+/* number of seconds to wait for other partitions to disengage */
+#define XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT 90
+
+/* interval in seconds to print 'waiting disengagement' messages */
+#define XPC_DISENGAGE_PRINTMSG_INTERVAL         10
+
+
+#define XPC_PARTID(_p)  ((partid_t) ((_p) - &xpc_partitions[0]))
+
+
+
+/* found in xp_main.c */
+extern struct xpc_registration xpc_registrations[];
+
+
+/* found in xpc_main.c */
+extern struct device *xpc_part;
+extern struct device *xpc_chan;
+extern int xpc_disengage_request_timelimit;
+extern int xpc_disengage_request_timedout;
+extern irqreturn_t xpc_notify_IRQ_handler(int, void *);
+extern void xpc_dropped_IPI_check(struct xpc_partition *);
+extern void xpc_activate_partition(struct xpc_partition *);
+extern void xpc_activate_kthreads(struct xpc_channel *, int);
+extern void xpc_create_kthreads(struct xpc_channel *, int, int);
+extern void xpc_disconnect_wait(int);
+
+
+/* found in xpc_partition.c */
+extern int xpc_exiting;
+extern struct xpc_vars *xpc_vars;
+extern struct xpc_rsvd_page *xpc_rsvd_page;
+extern struct xpc_vars_part *xpc_vars_part;
+extern struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
+extern char *xpc_remote_copy_buffer;
+extern void *xpc_remote_copy_buffer_base;
+extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);
+extern struct xpc_rsvd_page *xpc_rsvd_page_init(void);
+extern void xpc_allow_IPI_ops(void);
+extern void xpc_restrict_IPI_ops(void);
+extern int xpc_identify_act_IRQ_sender(void);
+extern int xpc_partition_disengaged(struct xpc_partition *);
+extern enum xpc_retval xpc_mark_partition_active(struct xpc_partition *);
+extern void xpc_mark_partition_inactive(struct xpc_partition *);
+extern void xpc_discovery(void);
+extern void xpc_check_remote_hb(void);
+extern void xpc_deactivate_partition(const int, struct xpc_partition *,
+                                                enum xpc_retval);
+extern enum xpc_retval xpc_initiate_partid_to_nasids(partid_t, void *);
+
+
+/* found in xpc_channel.c */
+extern void xpc_initiate_connect(int);
+extern void xpc_initiate_disconnect(int);
+extern enum xpc_retval xpc_initiate_allocate(partid_t, int, u32, void **);
+extern enum xpc_retval xpc_initiate_send(partid_t, int, void *);
+extern enum xpc_retval xpc_initiate_send_notify(partid_t, int, void *,
+                                                xpc_notify_func, void *);
+extern void xpc_initiate_received(partid_t, int, void *);
+extern enum xpc_retval xpc_setup_infrastructure(struct xpc_partition *);
+extern enum xpc_retval xpc_pull_remote_vars_part(struct xpc_partition *);
+extern void xpc_process_channel_activity(struct xpc_partition *);
+extern void xpc_connected_callout(struct xpc_channel *);
+extern void xpc_deliver_msg(struct xpc_channel *);
+extern void xpc_disconnect_channel(const int, struct xpc_channel *,
+                                        enum xpc_retval, unsigned long *);
+extern void xpc_disconnect_callout(struct xpc_channel *, enum xpc_retval);
+extern void xpc_partition_going_down(struct xpc_partition *, enum xpc_retval);
+extern void xpc_teardown_infrastructure(struct xpc_partition *);
+
+
+
+static inline void
+xpc_wakeup_channel_mgr(struct xpc_partition *part)
+{
+        if (atomic_inc_return(&part->channel_mgr_requests) == 1) {
+                wake_up(&part->channel_mgr_wq);
+        }
+}
+
+
+
+/*
+ * These next two inlines are used to keep us from tearing down a channel's
+ * msg queues while a thread may be referencing them.
+ */
+static inline void
+xpc_msgqueue_ref(struct xpc_channel *ch)
+{
+        atomic_inc(&ch->references);
+}
+
+static inline void
+xpc_msgqueue_deref(struct xpc_channel *ch)
+{
+        s32 refs = atomic_dec_return(&ch->references);
+
+        DBUG_ON(refs < 0);
+        if (refs == 0) {
+                xpc_wakeup_channel_mgr(&xpc_partitions[ch->partid]);
+        }
+}
+
+
+
+#define XPC_DISCONNECT_CHANNEL(_ch, _reason, _irqflgs) \
+                xpc_disconnect_channel(__LINE__, _ch, _reason, _irqflgs)
+
+
+/*
+ * These two inlines are used to keep us from tearing down a partition's
+ * setup infrastructure while a thread may be referencing it.
+ */
+static inline void
+xpc_part_deref(struct xpc_partition *part)
+{
+        s32 refs = atomic_dec_return(&part->references);
+
+
+        DBUG_ON(refs < 0);
+        if (refs == 0 && part->setup_state == XPC_P_WTEARDOWN) {
+                wake_up(&part->teardown_wq);
+        }
+}
+
+static inline int
+xpc_part_ref(struct xpc_partition *part)
+{
+        int setup;
+
+
+        atomic_inc(&part->references);
+        setup = (part->setup_state == XPC_P_SETUP);
+        if (!setup) {
+                xpc_part_deref(part);
+        }
+        return setup;
+}
+
+
+
+/*
+ * The following macro is to be used for the setting of the reason and
+ * reason_line fields in both the struct xpc_channel and struct xpc_partition
+ * structures.
+ */
+#define XPC_SET_REASON(_p, _reason, _line) \
+        { \
+                (_p)->reason = _reason; \
+                (_p)->reason_line = _line; \
+        }
+
+
+
+/*
+ * This next set of inlines are used to keep track of when a partition is
+ * potentially engaged in accessing memory belonging to another partition.
+ */
+
+static inline void
+xpc_mark_partition_engaged(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+                                (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
+
+
+        local_irq_save(irq_flags);
+
+        /* set bit corresponding to our partid in remote partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
+                                                (1UL << sn_partition_id));
+        /*
+         * We must always use the nofault function regardless of whether we
+         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+         * didn't, we'd never know that the other partition is down and would
+         * keep sending IPIs and AMOs to it until the heartbeat times out.
+         */
+        (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+                                variable), xp_nofault_PIOR_target));
+
+        local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_mark_partition_disengaged(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+                                (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
+
+
+        local_irq_save(irq_flags);
+
+        /* clear bit corresponding to our partid in remote partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+                                                ~(1UL << sn_partition_id));
+        /*
+         * We must always use the nofault function regardless of whether we
+         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+         * didn't, we'd never know that the other partition is down and would
+         * keep sending IPIs and AMOs to it until the heartbeat times out.
+         */
+        (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+                                variable), xp_nofault_PIOR_target));
+
+        local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_request_partition_disengage(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+                                (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
+
+
+        local_irq_save(irq_flags);
+
+        /* set bit corresponding to our partid in remote partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
+                                                (1UL << sn_partition_id));
+        /*
+         * We must always use the nofault function regardless of whether we
+         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+         * didn't, we'd never know that the other partition is down and would
+         * keep sending IPIs and AMOs to it until the heartbeat times out.
+         */
+        (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+                                variable), xp_nofault_PIOR_target));
+
+        local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_cancel_partition_disengage_request(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+                                (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
+
+
+        local_irq_save(irq_flags);
+
+        /* clear bit corresponding to our partid in remote partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+                                                ~(1UL << sn_partition_id));
+        /*
+         * We must always use the nofault function regardless of whether we
+         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+         * didn't, we'd never know that the other partition is down and would
+         * keep sending IPIs and AMOs to it until the heartbeat times out.
+         */
+        (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+                                variable), xp_nofault_PIOR_target));
+
+        local_irq_restore(irq_flags);
+}
+
+static inline u64
+xpc_partition_engaged(u64 partid_mask)
+{
+        AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
+
+
+        /* return our partition's AMO variable ANDed with partid_mask */
+        return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
+                                                                partid_mask);
+}
+
+static inline u64
+xpc_partition_disengage_requested(u64 partid_mask)
+{
+        AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
+
+
+        /* return our partition's AMO variable ANDed with partid_mask */
+        return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
+                                                                partid_mask);
+}
+
+static inline void
+xpc_clear_partition_engaged(u64 partid_mask)
+{
+        AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
+
+
+        /* clear bit(s) based on partid_mask in our partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+                                                                ~partid_mask);
+}
+
+static inline void
+xpc_clear_partition_disengage_request(u64 partid_mask)
+{
+        AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
+
+
+        /* clear bit(s) based on partid_mask in our partition's AMO */
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+                                                                ~partid_mask);
+}
+
+
+
+/*
+ * The following set of macros and inlines are used for the sending and
+ * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
+ * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
+ * the other that is associated with channel activity (SGI_XPC_NOTIFY).
+ */
+
+static inline u64
+xpc_IPI_receive(AMO_t *amo)
+{
+        return FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_CLEAR);
+}
+
+
+static inline enum xpc_retval
+xpc_IPI_send(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector)
+{
+        int ret = 0;
+        unsigned long irq_flags;
+
+
+        local_irq_save(irq_flags);
+
+        FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR, flag);
+        sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
+
+        /*
+         * We must always use the nofault function regardless of whether we
+         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+         * didn't, we'd never know that the other partition is down and would
+         * keep sending IPIs and AMOs to it until the heartbeat times out.
+         */
+        ret = xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
+                                xp_nofault_PIOR_target));
+
+        local_irq_restore(irq_flags);
+
+        return ((ret == 0) ? xpcSuccess : xpcPioReadError);
+}
+
+
+/*
+ * IPIs associated with SGI_XPC_ACTIVATE IRQ.
+ */
+
+/*
+ * Flag the appropriate AMO variable and send an IPI to the specified node.
+ */
+static inline void
+xpc_activate_IRQ_send(u64 amos_page_pa, int from_nasid, int to_nasid,
+                        int to_phys_cpuid)
+{
+        int w_index = XPC_NASID_W_INDEX(from_nasid);
+        int b_index = XPC_NASID_B_INDEX(from_nasid);
+        AMO_t *amos = (AMO_t *) __va(amos_page_pa +
+                                (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
+
+
+        (void) xpc_IPI_send(&amos[w_index], (1UL << b_index), to_nasid,
+                                to_phys_cpuid, SGI_XPC_ACTIVATE);
+}
+
+static inline void
+xpc_IPI_send_activate(struct xpc_vars *vars)
+{
+        xpc_activate_IRQ_send(vars->amos_page_pa, cnodeid_to_nasid(0),
+                                vars->act_nasid, vars->act_phys_cpuid);
+}
+
+static inline void
+xpc_IPI_send_activated(struct xpc_partition *part)
+{
+        xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
+                        part->remote_act_nasid, part->remote_act_phys_cpuid);
+}
+
+static inline void
+xpc_IPI_send_reactivate(struct xpc_partition *part)
+{
+        xpc_activate_IRQ_send(xpc_vars->amos_page_pa, part->reactivate_nasid,
+                                xpc_vars->act_nasid, xpc_vars->act_phys_cpuid);
+}
+
+static inline void
+xpc_IPI_send_disengage(struct xpc_partition *part)
+{
+        xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
+                        part->remote_act_nasid, part->remote_act_phys_cpuid);
+}
+
+
+/*
+ * IPIs associated with SGI_XPC_NOTIFY IRQ.
+ */
+
+/*
+ * Send an IPI to the remote partition that is associated with the
+ * specified channel.
+ */
+#define XPC_NOTIFY_IRQ_SEND(_ch, _ipi_f, _irq_f) \
+                xpc_notify_IRQ_send(_ch, _ipi_f, #_ipi_f, _irq_f)
+
+static inline void
+xpc_notify_IRQ_send(struct xpc_channel *ch, u8 ipi_flag, char *ipi_flag_string,
+                        unsigned long *irq_flags)
+{
+        struct xpc_partition *part = &xpc_partitions[ch->partid];
+        enum xpc_retval ret;
+
+
+        if (likely(part->act_state != XPC_P_DEACTIVATING)) {
+                ret = xpc_IPI_send(part->remote_IPI_amo_va,
+                                        (u64) ipi_flag << (ch->number * 8),
+                                        part->remote_IPI_nasid,
+                                        part->remote_IPI_phys_cpuid,
+                                        SGI_XPC_NOTIFY);
+                dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
+                        ipi_flag_string, ch->partid, ch->number, ret);
+                if (unlikely(ret != xpcSuccess)) {
+                        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);
+                        }
+                }
+        }
+}
+
+
+/*
+ * Make it look like the remote partition, which is associated with the
+ * specified channel, sent us an IPI. This faked IPI will be handled
+ * by xpc_dropped_IPI_check().
+ */
+#define XPC_NOTIFY_IRQ_SEND_LOCAL(_ch, _ipi_f) \
+                xpc_notify_IRQ_send_local(_ch, _ipi_f, #_ipi_f)
+
+static inline void
+xpc_notify_IRQ_send_local(struct xpc_channel *ch, u8 ipi_flag,
+                                char *ipi_flag_string)
+{
+        struct xpc_partition *part = &xpc_partitions[ch->partid];
+
+
+        FETCHOP_STORE_OP(TO_AMO((u64) &part->local_IPI_amo_va->variable),
+                        FETCHOP_OR, ((u64) ipi_flag << (ch->number * 8)));
+        dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
+                ipi_flag_string, ch->partid, ch->number);
+}
+
+
+/*
+ * The sending and receiving of IPIs includes the setting of an AMO variable
+ * to indicate the reason the IPI was sent. The 64-bit variable is divided
+ * up into eight bytes, ordered from right to left. Byte zero pertains to
+ * channel 0, byte one to channel 1, and so on. Each byte is described by
+ * the following IPI flags.
+ */
+
+#define XPC_IPI_CLOSEREQUEST    0x01
+#define XPC_IPI_CLOSEREPLY      0x02
+#define XPC_IPI_OPENREQUEST     0x04
+#define XPC_IPI_OPENREPLY       0x08
+#define XPC_IPI_MSGREQUEST      0x10
+
+
+/* given an AMO variable and a channel#, get its associated IPI flags */
+#define XPC_GET_IPI_FLAGS(_amo, _c)     ((u8) (((_amo) >> ((_c) * 8)) & 0xff))
+#define XPC_SET_IPI_FLAGS(_amo, _c, _f) (_amo) |= ((u64) (_f) << ((_c) * 8))
+
+#define XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(_amo) ((_amo) & __IA64_UL_CONST(0x0f0f0f0f0f0f0f0f))
+#define XPC_ANY_MSG_IPI_FLAGS_SET(_amo)       ((_amo) & __IA64_UL_CONST(0x1010101010101010))
+
+
+static inline void
+xpc_IPI_send_closerequest(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        struct xpc_openclose_args *args = ch->local_openclose_args;
+
+
+        args->reason = ch->reason;
+
+        XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREQUEST, irq_flags);
+}
+
+static inline void
+xpc_IPI_send_closereply(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREPLY, irq_flags);
+}
+
+static inline void
+xpc_IPI_send_openrequest(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        struct xpc_openclose_args *args = ch->local_openclose_args;
+
+
+        args->msg_size = ch->msg_size;
+        args->local_nentries = ch->local_nentries;
+
+        XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREQUEST, irq_flags);
+}
+
+static inline void
+xpc_IPI_send_openreply(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        struct xpc_openclose_args *args = ch->local_openclose_args;
+
+
+        args->remote_nentries = ch->remote_nentries;
+        args->local_nentries = ch->local_nentries;
+        args->local_msgqueue_pa = __pa(ch->local_msgqueue);
+
+        XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREPLY, irq_flags);
+}
+
+static inline void
+xpc_IPI_send_msgrequest(struct xpc_channel *ch)
+{
+        XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_MSGREQUEST, NULL);
+}
+
+static inline void
+xpc_IPI_send_local_msgrequest(struct xpc_channel *ch)
+{
+        XPC_NOTIFY_IRQ_SEND_LOCAL(ch, XPC_IPI_MSGREQUEST);
+}
+
+
+/*
+ * Memory for XPC's AMO variables is allocated by the MSPEC driver. These
+ * pages are located in the lowest granule. The lowest granule uses 4k pages
+ * for cached references and an alternate TLB handler to never provide a
+ * cacheable mapping for the entire region. This will prevent speculative
+ * reading of cached copies of our lines from being issued which will cause
+ * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
+ * AMO variables (based on XP_MAX_PARTITIONS) for message notification and an
+ * additional 128 AMO variables (based on XP_NASID_MASK_WORDS) for partition
+ * activation and 2 AMO variables for partition deactivation.
+ */
+static inline AMO_t *
+xpc_IPI_init(int index)
+{
+        AMO_t *amo = xpc_vars->amos_page + index;
+
+
+        (void) xpc_IPI_receive(amo);    /* clear AMO variable */
+        return amo;
+}
+
+
+
+static inline enum xpc_retval
+xpc_map_bte_errors(bte_result_t error)
+{
+        if (error == BTE_SUCCESS)
+                return xpcSuccess;
+
+        if (is_shub2()) {
+                if (BTE_VALID_SH2_ERROR(error))
+                        return xpcBteSh2Start + error;
+                return xpcBteUnmappedError;
+        }
+        switch (error) {
+        case BTE_SUCCESS:       return xpcSuccess;
+        case BTEFAIL_DIR:       return xpcBteDirectoryError;
+        case BTEFAIL_POISON:    return xpcBtePoisonError;
+        case BTEFAIL_WERR:      return xpcBteWriteError;
+        case BTEFAIL_ACCESS:    return xpcBteAccessError;
+        case BTEFAIL_PWERR:     return xpcBtePWriteError;
+        case BTEFAIL_PRERR:     return xpcBtePReadError;
+        case BTEFAIL_TOUT:      return xpcBteTimeOutError;
+        case BTEFAIL_XTERR:     return xpcBteXtalkError;
+        case BTEFAIL_NOTAVAIL:  return xpcBteNotAvailable;
+        default:                return xpcBteUnmappedError;
+        }
+}
+
+
+
+/*
+ * Check to see if there is any channel activity to/from the specified
+ * partition.
+ */
+static inline void
+xpc_check_for_channel_activity(struct xpc_partition *part)
+{
+        u64 IPI_amo;
+        unsigned long irq_flags;
+
+
+        IPI_amo = xpc_IPI_receive(part->local_IPI_amo_va);
+        if (IPI_amo == 0) {
+                return;
+        }
+
+        spin_lock_irqsave(&part->IPI_lock, irq_flags);
+        part->local_IPI_amo |= IPI_amo;
+        spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
+
+        dev_dbg(xpc_chan, "received IPI from partid=%d, IPI_amo=0x%lx\n",
+                XPC_PARTID(part), IPI_amo);
+
+        xpc_wakeup_channel_mgr(part);
+}
+
+
+#endif /* _DRIVERS_MISC_SGIXP_XPC_H */
+
diff --git a/drivers/misc/sgi-xp/xpc_channel.c b/drivers/misc/sgi-xp/xpc_channel.c
new file mode 100644
index 000000000000..d7a215eeaaf6
--- /dev/null
+++ b/drivers/misc/sgi-xp/xpc_channel.c
@@ -0,0 +1,2379 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * Cross Partition Communication (XPC) channel support.
+ *
+ *      This is the part of XPC that manages the channels and
+ *      sends/receives messages across them to/from other partitions.
+ *
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/sched.h>
+#include <linux/cache.h>
+#include <linux/interrupt.h>
+#include <linux/mutex.h>
+#include <linux/completion.h>
+#include <asm/sn/bte.h>
+#include <asm/sn/sn_sal.h>
+#include "xpc.h"
+
+
+/*
+ * Guarantee that the kzalloc'd memory is cacheline aligned.
+ */
+static void *
+xpc_kzalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
+{
+        /* see if kzalloc will give us cachline aligned memory by default */
+        *base = kzalloc(size, flags);
+        if (*base == NULL) {
+                return NULL;
+        }
+        if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
+                return *base;
+        }
+        kfree(*base);
+
+        /* nope, we'll have to do it ourselves */
+        *base = kzalloc(size + L1_CACHE_BYTES, flags);
+        if (*base == NULL) {
+                return NULL;
+        }
+        return (void *) L1_CACHE_ALIGN((u64) *base);
+}
+
+
+/*
+ * Set up the initial values for the XPartition Communication channels.
+ */
+static void
+xpc_initialize_channels(struct xpc_partition *part, partid_t partid)
+{
+        int ch_number;
+        struct xpc_channel *ch;
+
+
+        for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
+                ch = &part->channels[ch_number];
+
+                ch->partid = partid;
+                ch->number = ch_number;
+                ch->flags = XPC_C_DISCONNECTED;
+
+                ch->local_GP = &part->local_GPs[ch_number];
+                ch->local_openclose_args =
+                                        &part->local_openclose_args[ch_number];
+
+                atomic_set(&ch->kthreads_assigned, 0);
+                atomic_set(&ch->kthreads_idle, 0);
+                atomic_set(&ch->kthreads_active, 0);
+
+                atomic_set(&ch->references, 0);
+                atomic_set(&ch->n_to_notify, 0);
+
+                spin_lock_init(&ch->lock);
+                mutex_init(&ch->msg_to_pull_mutex);
+                init_completion(&ch->wdisconnect_wait);
+
+                atomic_set(&ch->n_on_msg_allocate_wq, 0);
+                init_waitqueue_head(&ch->msg_allocate_wq);
+                init_waitqueue_head(&ch->idle_wq);
+        }
+}
+
+
+/*
+ * Setup the infrastructure necessary to support XPartition Communication
+ * between the specified remote partition and the local one.
+ */
+enum xpc_retval
+xpc_setup_infrastructure(struct xpc_partition *part)
+{
+        int ret, cpuid;
+        struct timer_list *timer;
+        partid_t partid = XPC_PARTID(part);
+
+
+        /*
+         * Zero out MOST of the entry for this partition. Only the fields
+         * starting with `nchannels' will be zeroed. The preceding fields must
+         * remain `viable' across partition ups and downs, since they may be
+         * referenced during this memset() operation.
+         */
+        memset(&part->nchannels, 0, sizeof(struct xpc_partition) -
+                                offsetof(struct xpc_partition, nchannels));
+
+        /*
+         * Allocate all of the channel structures as a contiguous chunk of
+         * memory.
+         */
+        part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_NCHANNELS,
+                                                                GFP_KERNEL);
+        if (part->channels == NULL) {
+                dev_err(xpc_chan, "can't get memory for channels\n");
+                return xpcNoMemory;
+        }
+
+        part->nchannels = XPC_NCHANNELS;
+
+
+        /* allocate all the required GET/PUT values */
+
+        part->local_GPs = xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE,
+                                        GFP_KERNEL, &part->local_GPs_base);
+        if (part->local_GPs == NULL) {
+                kfree(part->channels);
+                part->channels = NULL;
+                dev_err(xpc_chan, "can't get memory for local get/put "
+                        "values\n");
+                return xpcNoMemory;
+        }
+
+        part->remote_GPs = xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE,
+                                        GFP_KERNEL, &part->remote_GPs_base);
+        if (part->remote_GPs == NULL) {
+                dev_err(xpc_chan, "can't get memory for remote get/put "
+                        "values\n");
+                kfree(part->local_GPs_base);
+                part->local_GPs = NULL;
+                kfree(part->channels);
+                part->channels = NULL;
+                return xpcNoMemory;
+        }
+
+
+        /* allocate all the required open and close args */
+
+        part->local_openclose_args = xpc_kzalloc_cacheline_aligned(
+                                        XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
+                                        &part->local_openclose_args_base);
+        if (part->local_openclose_args == NULL) {
+                dev_err(xpc_chan, "can't get memory for local connect args\n");
+                kfree(part->remote_GPs_base);
+                part->remote_GPs = NULL;
+                kfree(part->local_GPs_base);
+                part->local_GPs = NULL;
+                kfree(part->channels);
+                part->channels = NULL;
+                return xpcNoMemory;
+        }
+
+        part->remote_openclose_args = xpc_kzalloc_cacheline_aligned(
+                                        XPC_OPENCLOSE_ARGS_SIZE, GFP_KERNEL,
+                                        &part->remote_openclose_args_base);
+        if (part->remote_openclose_args == NULL) {
+                dev_err(xpc_chan, "can't get memory for remote connect args\n");
+                kfree(part->local_openclose_args_base);
+                part->local_openclose_args = NULL;
+                kfree(part->remote_GPs_base);
+                part->remote_GPs = NULL;
+                kfree(part->local_GPs_base);
+                part->local_GPs = NULL;
+                kfree(part->channels);
+                part->channels = NULL;
+                return xpcNoMemory;
+        }
+
+
+        xpc_initialize_channels(part, partid);
+
+        atomic_set(&part->nchannels_active, 0);
+        atomic_set(&part->nchannels_engaged, 0);
+
+
+        /* local_IPI_amo were set to 0 by an earlier memset() */
+
+        /* Initialize this partitions AMO_t structure */
+        part->local_IPI_amo_va = xpc_IPI_init(partid);
+
+        spin_lock_init(&part->IPI_lock);
+
+        atomic_set(&part->channel_mgr_requests, 1);
+        init_waitqueue_head(&part->channel_mgr_wq);
+
+        sprintf(part->IPI_owner, "xpc%02d", partid);
+        ret = request_irq(SGI_XPC_NOTIFY, xpc_notify_IRQ_handler, IRQF_SHARED,
+                                part->IPI_owner, (void *) (u64) partid);
+        if (ret != 0) {
+                dev_err(xpc_chan, "can't register NOTIFY IRQ handler, "
+                        "errno=%d\n", -ret);
+                kfree(part->remote_openclose_args_base);
+                part->remote_openclose_args = NULL;
+                kfree(part->local_openclose_args_base);
+                part->local_openclose_args = NULL;
+                kfree(part->remote_GPs_base);
+                part->remote_GPs = NULL;
+                kfree(part->local_GPs_base);
+                part->local_GPs = NULL;
+                kfree(part->channels);
+                part->channels = NULL;
+                return xpcLackOfResources;
+        }
+
+        /* Setup a timer to check for dropped IPIs */
+        timer = &part->dropped_IPI_timer;
+        init_timer(timer);
+        timer->function = (void (*)(unsigned long)) xpc_dropped_IPI_check;
+        timer->data = (unsigned long) part;
+        timer->expires = jiffies + XPC_P_DROPPED_IPI_WAIT;
+        add_timer(timer);
+
+        /*
+         * With the setting of the partition setup_state to XPC_P_SETUP, we're
+         * declaring that this partition is ready to go.
+         */
+        part->setup_state = XPC_P_SETUP;
+
+
+        /*
+         * Setup the per partition specific variables required by the
+         * remote partition to establish channel connections with us.
+         *
+         * The setting of the magic # indicates that these per partition
+         * specific variables are ready to be used.
+         */
+        xpc_vars_part[partid].GPs_pa = __pa(part->local_GPs);
+        xpc_vars_part[partid].openclose_args_pa =
+                                        __pa(part->local_openclose_args);
+        xpc_vars_part[partid].IPI_amo_pa = __pa(part->local_IPI_amo_va);
+        cpuid = raw_smp_processor_id(); /* any CPU in this partition will do */
+        xpc_vars_part[partid].IPI_nasid = cpuid_to_nasid(cpuid);
+        xpc_vars_part[partid].IPI_phys_cpuid = cpu_physical_id(cpuid);
+        xpc_vars_part[partid].nchannels = part->nchannels;
+        xpc_vars_part[partid].magic = XPC_VP_MAGIC1;
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Create a wrapper that hides the underlying mechanism for pulling a cacheline
+ * (or multiple cachelines) from a remote partition.
+ *
+ * src must be a cacheline aligned physical address on the remote partition.
+ * dst must be a cacheline aligned virtual address on this partition.
+ * cnt must be an cacheline sized
+ */
+static enum xpc_retval
+xpc_pull_remote_cachelines(struct xpc_partition *part, void *dst,
+                                const void *src, size_t cnt)
+{
+        bte_result_t bte_ret;
+
+
+        DBUG_ON((u64) src != L1_CACHE_ALIGN((u64) src));
+        DBUG_ON((u64) dst != L1_CACHE_ALIGN((u64) dst));
+        DBUG_ON(cnt != L1_CACHE_ALIGN(cnt));
+
+        if (part->act_state == XPC_P_DEACTIVATING) {
+                return part->reason;
+        }
+
+        bte_ret = xp_bte_copy((u64) src, (u64) dst, (u64) cnt,
+                                        (BTE_NORMAL | BTE_WACQUIRE), NULL);
+        if (bte_ret == BTE_SUCCESS) {
+                return xpcSuccess;
+        }
+
+        dev_dbg(xpc_chan, "xp_bte_copy() from partition %d failed, ret=%d\n",
+                XPC_PARTID(part), bte_ret);
+
+        return xpc_map_bte_errors(bte_ret);
+}
+
+
+/*
+ * Pull the remote per partition specific variables from the specified
+ * partition.
+ */
+enum xpc_retval
+xpc_pull_remote_vars_part(struct xpc_partition *part)
+{
+        u8 buffer[L1_CACHE_BYTES * 2];
+        struct xpc_vars_part *pulled_entry_cacheline =
+                        (struct xpc_vars_part *) L1_CACHE_ALIGN((u64) buffer);
+        struct xpc_vars_part *pulled_entry;
+        u64 remote_entry_cacheline_pa, remote_entry_pa;
+        partid_t partid = XPC_PARTID(part);
+        enum xpc_retval ret;
+
+
+        /* pull the cacheline that contains the variables we're interested in */
+
+        DBUG_ON(part->remote_vars_part_pa !=
+                                L1_CACHE_ALIGN(part->remote_vars_part_pa));
+        DBUG_ON(sizeof(struct xpc_vars_part) != L1_CACHE_BYTES / 2);
+
+        remote_entry_pa = part->remote_vars_part_pa +
+                        sn_partition_id * sizeof(struct xpc_vars_part);
+
+        remote_entry_cacheline_pa = (remote_entry_pa & ~(L1_CACHE_BYTES - 1));
+
+        pulled_entry = (struct xpc_vars_part *) ((u64) pulled_entry_cacheline +
+                                (remote_entry_pa & (L1_CACHE_BYTES - 1)));
+
+        ret = xpc_pull_remote_cachelines(part, pulled_entry_cacheline,
+                                        (void *) remote_entry_cacheline_pa,
+                                        L1_CACHE_BYTES);
+        if (ret != xpcSuccess) {
+                dev_dbg(xpc_chan, "failed to pull XPC vars_part from "
+                        "partition %d, ret=%d\n", partid, ret);
+                return ret;
+        }
+
+
+        /* see if they've been set up yet */
+
+        if (pulled_entry->magic != XPC_VP_MAGIC1 &&
+                                pulled_entry->magic != XPC_VP_MAGIC2) {
+
+                if (pulled_entry->magic != 0) {
+                        dev_dbg(xpc_chan, "partition %d's XPC vars_part for "
+                                "partition %d has bad magic value (=0x%lx)\n",
+                                partid, sn_partition_id, pulled_entry->magic);
+                        return xpcBadMagic;
+                }
+
+                /* they've not been initialized yet */
+                return xpcRetry;
+        }
+
+        if (xpc_vars_part[partid].magic == XPC_VP_MAGIC1) {
+
+                /* validate the variables */
+
+                if (pulled_entry->GPs_pa == 0 ||
+                                pulled_entry->openclose_args_pa == 0 ||
+                                        pulled_entry->IPI_amo_pa == 0) {
+
+                        dev_err(xpc_chan, "partition %d's XPC vars_part for "
+                                "partition %d are not valid\n", partid,
+                                sn_partition_id);
+                        return xpcInvalidAddress;
+                }
+
+                /* the variables we imported look to be valid */
+
+                part->remote_GPs_pa = pulled_entry->GPs_pa;
+                part->remote_openclose_args_pa =
+                                        pulled_entry->openclose_args_pa;
+                part->remote_IPI_amo_va =
+                                      (AMO_t *) __va(pulled_entry->IPI_amo_pa);
+                part->remote_IPI_nasid = pulled_entry->IPI_nasid;
+                part->remote_IPI_phys_cpuid = pulled_entry->IPI_phys_cpuid;
+
+                if (part->nchannels > pulled_entry->nchannels) {
+                        part->nchannels = pulled_entry->nchannels;
+                }
+
+                /* let the other side know that we've pulled their variables */
+
+                xpc_vars_part[partid].magic = XPC_VP_MAGIC2;
+        }
+
+        if (pulled_entry->magic == XPC_VP_MAGIC1) {
+                return xpcRetry;
+        }
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Get the IPI flags and pull the openclose args and/or remote GPs as needed.
+ */
+static u64
+xpc_get_IPI_flags(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        u64 IPI_amo;
+        enum xpc_retval ret;
+
+
+        /*
+         * See if there are any IPI flags to be handled.
+         */
+
+        spin_lock_irqsave(&part->IPI_lock, irq_flags);
+        if ((IPI_amo = part->local_IPI_amo) != 0) {
+                part->local_IPI_amo = 0;
+        }
+        spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
+
+
+        if (XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(IPI_amo)) {
+                ret = xpc_pull_remote_cachelines(part,
+                                        part->remote_openclose_args,
+                                        (void *) part->remote_openclose_args_pa,
+                                        XPC_OPENCLOSE_ARGS_SIZE);
+                if (ret != xpcSuccess) {
+                        XPC_DEACTIVATE_PARTITION(part, ret);
+
+                        dev_dbg(xpc_chan, "failed to pull openclose args from "
+                                "partition %d, ret=%d\n", XPC_PARTID(part),
+                                ret);
+
+                        /* don't bother processing IPIs anymore */
+                        IPI_amo = 0;
+                }
+        }
+
+        if (XPC_ANY_MSG_IPI_FLAGS_SET(IPI_amo)) {
+                ret = xpc_pull_remote_cachelines(part, part->remote_GPs,
+                                                (void *) part->remote_GPs_pa,
+                                                XPC_GP_SIZE);
+                if (ret != xpcSuccess) {
+                        XPC_DEACTIVATE_PARTITION(part, ret);
+
+                        dev_dbg(xpc_chan, "failed to pull GPs from partition "
+                                "%d, ret=%d\n", XPC_PARTID(part), ret);
+
+                        /* don't bother processing IPIs anymore */
+                        IPI_amo = 0;
+                }
+        }
+
+        return IPI_amo;
+}
+
+
+/*
+ * Allocate the local message queue and the notify queue.
+ */
+static enum xpc_retval
+xpc_allocate_local_msgqueue(struct xpc_channel *ch)
+{
+        unsigned long irq_flags;
+        int nentries;
+        size_t nbytes;
+
+
+        // >>> may want to check for ch->flags & XPC_C_DISCONNECTING between
+        // >>> iterations of the for-loop, bail if set?
+
+        // >>> should we impose a minimum #of entries? like 4 or 8?
+        for (nentries = ch->local_nentries; nentries > 0; nentries--) {
+
+                nbytes = nentries * ch->msg_size;
+                ch->local_msgqueue = xpc_kzalloc_cacheline_aligned(nbytes,
+                                                GFP_KERNEL,
+                                                &ch->local_msgqueue_base);
+                if (ch->local_msgqueue == NULL) {
+                        continue;
+                }
+
+                nbytes = nentries * sizeof(struct xpc_notify);
+                ch->notify_queue = kzalloc(nbytes, GFP_KERNEL);
+                if (ch->notify_queue == NULL) {
+                        kfree(ch->local_msgqueue_base);
+                        ch->local_msgqueue = NULL;
+                        continue;
+                }
+
+                spin_lock_irqsave(&ch->lock, irq_flags);
+                if (nentries < ch->local_nentries) {
+                        dev_dbg(xpc_chan, "nentries=%d local_nentries=%d, "
+                                "partid=%d, channel=%d\n", nentries,
+                                ch->local_nentries, ch->partid, ch->number);
+
+                        ch->local_nentries = nentries;
+                }
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                return xpcSuccess;
+        }
+
+        dev_dbg(xpc_chan, "can't get memory for local message queue and notify "
+                "queue, partid=%d, channel=%d\n", ch->partid, ch->number);
+        return xpcNoMemory;
+}
+
+
+/*
+ * Allocate the cached remote message queue.
+ */
+static enum xpc_retval
+xpc_allocate_remote_msgqueue(struct xpc_channel *ch)
+{
+        unsigned long irq_flags;
+        int nentries;
+        size_t nbytes;
+
+
+        DBUG_ON(ch->remote_nentries <= 0);
+
+        // >>> may want to check for ch->flags & XPC_C_DISCONNECTING between
+        // >>> iterations of the for-loop, bail if set?
+
+        // >>> should we impose a minimum #of entries? like 4 or 8?
+        for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
+
+                nbytes = nentries * ch->msg_size;
+                ch->remote_msgqueue = xpc_kzalloc_cacheline_aligned(nbytes,
+                                                GFP_KERNEL,
+                                                &ch->remote_msgqueue_base);
+                if (ch->remote_msgqueue == NULL) {
+                        continue;
+                }
+
+                spin_lock_irqsave(&ch->lock, irq_flags);
+                if (nentries < ch->remote_nentries) {
+                        dev_dbg(xpc_chan, "nentries=%d remote_nentries=%d, "
+                                "partid=%d, channel=%d\n", nentries,
+                                ch->remote_nentries, ch->partid, ch->number);
+
+                        ch->remote_nentries = nentries;
+                }
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                return xpcSuccess;
+        }
+
+        dev_dbg(xpc_chan, "can't get memory for cached remote message queue, "
+                "partid=%d, channel=%d\n", ch->partid, ch->number);
+        return xpcNoMemory;
+}
+
+
+/*
+ * Allocate message queues and other stuff associated with a channel.
+ *
+ * Note: Assumes all of the channel sizes are filled in.
+ */
+static enum xpc_retval
+xpc_allocate_msgqueues(struct xpc_channel *ch)
+{
+        unsigned long irq_flags;
+        enum xpc_retval ret;
+
+
+        DBUG_ON(ch->flags & XPC_C_SETUP);
+
+        if ((ret = xpc_allocate_local_msgqueue(ch)) != xpcSuccess) {
+                return ret;
+        }
+
+        if ((ret = xpc_allocate_remote_msgqueue(ch)) != xpcSuccess) {
+                kfree(ch->local_msgqueue_base);
+                ch->local_msgqueue = NULL;
+                kfree(ch->notify_queue);
+                ch->notify_queue = NULL;
+                return ret;
+        }
+
+        spin_lock_irqsave(&ch->lock, irq_flags);
+        ch->flags |= XPC_C_SETUP;
+        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Process a connect message from a remote partition.
+ *
+ * Note: xpc_process_connect() is expecting to be called with the
+ * spin_lock_irqsave held and will leave it locked upon return.
+ */
+static void
+xpc_process_connect(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        enum xpc_retval ret;
+
+
+        DBUG_ON(!spin_is_locked(&ch->lock));
+
+        if (!(ch->flags & XPC_C_OPENREQUEST) ||
+                                !(ch->flags & XPC_C_ROPENREQUEST)) {
+                /* nothing more to do for now */
+                return;
+        }
+        DBUG_ON(!(ch->flags & XPC_C_CONNECTING));
+
+        if (!(ch->flags & XPC_C_SETUP)) {
+                spin_unlock_irqrestore(&ch->lock, *irq_flags);
+                ret = xpc_allocate_msgqueues(ch);
+                spin_lock_irqsave(&ch->lock, *irq_flags);
+
+                if (ret != xpcSuccess) {
+                        XPC_DISCONNECT_CHANNEL(ch, ret, irq_flags);
+                }
+                if (ch->flags & (XPC_C_CONNECTED | XPC_C_DISCONNECTING)) {
+                        return;
+                }
+
+                DBUG_ON(!(ch->flags & XPC_C_SETUP));
+                DBUG_ON(ch->local_msgqueue == NULL);
+                DBUG_ON(ch->remote_msgqueue == NULL);
+        }
+
+        if (!(ch->flags & XPC_C_OPENREPLY)) {
+                ch->flags |= XPC_C_OPENREPLY;
+                xpc_IPI_send_openreply(ch, irq_flags);
+        }
+
+        if (!(ch->flags & XPC_C_ROPENREPLY)) {
+                return;
+        }
+
+        DBUG_ON(ch->remote_msgqueue_pa == 0);
+
+        ch->flags = (XPC_C_CONNECTED | XPC_C_SETUP);    /* clear all else */
+
+        dev_info(xpc_chan, "channel %d to partition %d connected\n",
+                ch->number, ch->partid);
+
+        spin_unlock_irqrestore(&ch->lock, *irq_flags);
+        xpc_create_kthreads(ch, 1, 0);
+        spin_lock_irqsave(&ch->lock, *irq_flags);
+}
+
+
+/*
+ * Notify those who wanted to be notified upon delivery of their message.
+ */
+static void
+xpc_notify_senders(struct xpc_channel *ch, enum xpc_retval reason, s64 put)
+{
+        struct xpc_notify *notify;
+        u8 notify_type;
+        s64 get = ch->w_remote_GP.get - 1;
+
+
+        while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
+
+                notify = &ch->notify_queue[get % ch->local_nentries];
+
+                /*
+                 * See if the notify entry indicates it was associated with
+                 * a message who's sender wants to be notified. It is possible
+                 * that it is, but someone else is doing or has done the
+                 * notification.
+                 */
+                notify_type = notify->type;
+                if (notify_type == 0 ||
+                                cmpxchg(&notify->type, notify_type, 0) !=
+                                                                notify_type) {
+                        continue;
+                }
+
+                DBUG_ON(notify_type != XPC_N_CALL);
+
+                atomic_dec(&ch->n_to_notify);
+
+                if (notify->func != NULL) {
+                        dev_dbg(xpc_chan, "notify->func() called, notify=0x%p, "
+                                "msg_number=%ld, partid=%d, channel=%d\n",
+                                (void *) notify, get, ch->partid, ch->number);
+
+                        notify->func(reason, ch->partid, ch->number,
+                                                                notify->key);
+
+                        dev_dbg(xpc_chan, "notify->func() returned, "
+                                "notify=0x%p, msg_number=%ld, partid=%d, "
+                                "channel=%d\n", (void *) notify, get,
+                                ch->partid, ch->number);
+                }
+        }
+}
+
+
+/*
+ * Free up message queues and other stuff that were allocated for the specified
+ * channel.
+ *
+ * Note: ch->reason and ch->reason_line are left set for debugging purposes,
+ * they're cleared when XPC_C_DISCONNECTED is cleared.
+ */
+static void
+xpc_free_msgqueues(struct xpc_channel *ch)
+{
+        DBUG_ON(!spin_is_locked(&ch->lock));
+        DBUG_ON(atomic_read(&ch->n_to_notify) != 0);
+
+        ch->remote_msgqueue_pa = 0;
+        ch->func = NULL;
+        ch->key = NULL;
+        ch->msg_size = 0;
+        ch->local_nentries = 0;
+        ch->remote_nentries = 0;
+        ch->kthreads_assigned_limit = 0;
+        ch->kthreads_idle_limit = 0;
+
+        ch->local_GP->get = 0;
+        ch->local_GP->put = 0;
+        ch->remote_GP.get = 0;
+        ch->remote_GP.put = 0;
+        ch->w_local_GP.get = 0;
+        ch->w_local_GP.put = 0;
+        ch->w_remote_GP.get = 0;
+        ch->w_remote_GP.put = 0;
+        ch->next_msg_to_pull = 0;
+
+        if (ch->flags & XPC_C_SETUP) {
+                ch->flags &= ~XPC_C_SETUP;
+
+                dev_dbg(xpc_chan, "ch->flags=0x%x, partid=%d, channel=%d\n",
+                        ch->flags, ch->partid, ch->number);
+
+                kfree(ch->local_msgqueue_base);
+                ch->local_msgqueue = NULL;
+                kfree(ch->remote_msgqueue_base);
+                ch->remote_msgqueue = NULL;
+                kfree(ch->notify_queue);
+                ch->notify_queue = NULL;
+        }
+}
+
+
+/*
+ * spin_lock_irqsave() is expected to be held on entry.
+ */
+static void
+xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags)
+{
+        struct xpc_partition *part = &xpc_partitions[ch->partid];
+        u32 channel_was_connected = (ch->flags & XPC_C_WASCONNECTED);
+
+
+        DBUG_ON(!spin_is_locked(&ch->lock));
+
+        if (!(ch->flags & XPC_C_DISCONNECTING)) {
+                return;
+        }
+
+        DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
+
+        /* make sure all activity has settled down first */
+
+        if (atomic_read(&ch->kthreads_assigned) > 0 ||
+                                atomic_read(&ch->references) > 0) {
+                return;
+        }
+        DBUG_ON((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
+                        !(ch->flags & XPC_C_DISCONNECTINGCALLOUT_MADE));
+
+        if (part->act_state == XPC_P_DEACTIVATING) {
+                /* can't proceed until the other side disengages from us */
+                if (xpc_partition_engaged(1UL << ch->partid)) {
+                        return;
+                }
+
+        } else {
+
+                /* as long as the other side is up do the full protocol */
+
+                if (!(ch->flags & XPC_C_RCLOSEREQUEST)) {
+                        return;
+                }
+
+                if (!(ch->flags & XPC_C_CLOSEREPLY)) {
+                        ch->flags |= XPC_C_CLOSEREPLY;
+                        xpc_IPI_send_closereply(ch, irq_flags);
+                }
+
+                if (!(ch->flags & XPC_C_RCLOSEREPLY)) {
+                        return;
+                }
+        }
+
+        /* wake those waiting for notify completion */
+        if (atomic_read(&ch->n_to_notify) > 0) {
+                /* >>> we do callout while holding ch->lock */
+                xpc_notify_senders(ch, ch->reason, ch->w_local_GP.put);
+        }
+
+        /* both sides are disconnected now */
+
+        if (ch->flags & XPC_C_DISCONNECTINGCALLOUT_MADE) {
+                spin_unlock_irqrestore(&ch->lock, *irq_flags);
+                xpc_disconnect_callout(ch, xpcDisconnected);
+                spin_lock_irqsave(&ch->lock, *irq_flags);
+        }
+
+        /* it's now safe to free the channel's message queues */
+        xpc_free_msgqueues(ch);
+
+        /* mark disconnected, clear all other flags except XPC_C_WDISCONNECT */
+        ch->flags = (XPC_C_DISCONNECTED | (ch->flags & XPC_C_WDISCONNECT));
+
+        atomic_dec(&part->nchannels_active);
+
+        if (channel_was_connected) {
+                dev_info(xpc_chan, "channel %d to partition %d disconnected, "
+                        "reason=%d\n", ch->number, ch->partid, ch->reason);
+        }
+
+        if (ch->flags & XPC_C_WDISCONNECT) {
+                /* we won't lose the CPU since we're holding ch->lock */
+                complete(&ch->wdisconnect_wait);
+        } else if (ch->delayed_IPI_flags) {
+                if (part->act_state != XPC_P_DEACTIVATING) {
+                        /* time to take action on any delayed IPI flags */
+                        spin_lock(&part->IPI_lock);
+                        XPC_SET_IPI_FLAGS(part->local_IPI_amo, ch->number,
+                                                        ch->delayed_IPI_flags);
+                        spin_unlock(&part->IPI_lock);
+                }
+                ch->delayed_IPI_flags = 0;
+        }
+}
+
+
+/*
+ * Process a change in the channel's remote connection state.
+ */
+static void
+xpc_process_openclose_IPI(struct xpc_partition *part, int ch_number,
+                                u8 IPI_flags)
+{
+        unsigned long irq_flags;
+        struct xpc_openclose_args *args =
+                                &part->remote_openclose_args[ch_number];
+        struct xpc_channel *ch = &part->channels[ch_number];
+        enum xpc_retval reason;
+
+
+
+        spin_lock_irqsave(&ch->lock, irq_flags);
+
+again:
+
+        if ((ch->flags & XPC_C_DISCONNECTED) &&
+                                        (ch->flags & XPC_C_WDISCONNECT)) {
+                /*
+                 * Delay processing IPI flags until thread waiting disconnect
+                 * has had a chance to see that the channel is disconnected.
+                 */
+                ch->delayed_IPI_flags |= IPI_flags;
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                return;
+        }
+
+
+        if (IPI_flags & XPC_IPI_CLOSEREQUEST) {
+
+                dev_dbg(xpc_chan, "XPC_IPI_CLOSEREQUEST (reason=%d) received "
+                        "from partid=%d, channel=%d\n", args->reason,
+                        ch->partid, ch->number);
+
+                /*
+                 * If RCLOSEREQUEST is set, we're probably waiting for
+                 * RCLOSEREPLY. We should find it and a ROPENREQUEST packed
+                 * with this RCLOSEREQUEST in the IPI_flags.
+                 */
+
+                if (ch->flags & XPC_C_RCLOSEREQUEST) {
+                        DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
+                        DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
+                        DBUG_ON(!(ch->flags & XPC_C_CLOSEREPLY));
+                        DBUG_ON(ch->flags & XPC_C_RCLOSEREPLY);
+
+                        DBUG_ON(!(IPI_flags & XPC_IPI_CLOSEREPLY));
+                        IPI_flags &= ~XPC_IPI_CLOSEREPLY;
+                        ch->flags |= XPC_C_RCLOSEREPLY;
+
+                        /* both sides have finished disconnecting */
+                        xpc_process_disconnect(ch, &irq_flags);
+                        DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
+                        goto again;
+                }
+
+                if (ch->flags & XPC_C_DISCONNECTED) {
+                        if (!(IPI_flags & XPC_IPI_OPENREQUEST)) {
+                                if ((XPC_GET_IPI_FLAGS(part->local_IPI_amo,
+                                         ch_number) & XPC_IPI_OPENREQUEST)) {
+
+                                        DBUG_ON(ch->delayed_IPI_flags != 0);
+                                        spin_lock(&part->IPI_lock);
+                                        XPC_SET_IPI_FLAGS(part->local_IPI_amo,
+                                                        ch_number,
+                                                        XPC_IPI_CLOSEREQUEST);
+                                        spin_unlock(&part->IPI_lock);
+                                }
+                                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                                return;
+                        }
+
+                        XPC_SET_REASON(ch, 0, 0);
+                        ch->flags &= ~XPC_C_DISCONNECTED;
+
+                        atomic_inc(&part->nchannels_active);
+                        ch->flags |= (XPC_C_CONNECTING | XPC_C_ROPENREQUEST);
+                }
+
+                IPI_flags &= ~(XPC_IPI_OPENREQUEST | XPC_IPI_OPENREPLY);
+
+                /*
+                 * The meaningful CLOSEREQUEST connection state fields are:
+                 *      reason = reason connection is to be closed
+                 */
+
+                ch->flags |= XPC_C_RCLOSEREQUEST;
+
+                if (!(ch->flags & XPC_C_DISCONNECTING)) {
+                        reason = args->reason;
+                        if (reason <= xpcSuccess || reason > xpcUnknownReason) {
+                                reason = xpcUnknownReason;
+                        } else if (reason == xpcUnregistering) {
+                                reason = xpcOtherUnregistering;
+                        }
+
+                        XPC_DISCONNECT_CHANNEL(ch, reason, &irq_flags);
+
+                        DBUG_ON(IPI_flags & XPC_IPI_CLOSEREPLY);
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                xpc_process_disconnect(ch, &irq_flags);
+        }
+
+
+        if (IPI_flags & XPC_IPI_CLOSEREPLY) {
+
+                dev_dbg(xpc_chan, "XPC_IPI_CLOSEREPLY received from partid=%d,"
+                        " channel=%d\n", ch->partid, ch->number);
+
+                if (ch->flags & XPC_C_DISCONNECTED) {
+                        DBUG_ON(part->act_state != XPC_P_DEACTIVATING);
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                DBUG_ON(!(ch->flags & XPC_C_CLOSEREQUEST));
+
+                if (!(ch->flags & XPC_C_RCLOSEREQUEST)) {
+                        if ((XPC_GET_IPI_FLAGS(part->local_IPI_amo, ch_number)
+                                                & XPC_IPI_CLOSEREQUEST)) {
+
+                                DBUG_ON(ch->delayed_IPI_flags != 0);
+                                spin_lock(&part->IPI_lock);
+                                XPC_SET_IPI_FLAGS(part->local_IPI_amo,
+                                                ch_number, XPC_IPI_CLOSEREPLY);
+                                spin_unlock(&part->IPI_lock);
+                        }
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                ch->flags |= XPC_C_RCLOSEREPLY;
+
+                if (ch->flags & XPC_C_CLOSEREPLY) {
+                        /* both sides have finished disconnecting */
+                        xpc_process_disconnect(ch, &irq_flags);
+                }
+        }
+
+
+        if (IPI_flags & XPC_IPI_OPENREQUEST) {
+
+                dev_dbg(xpc_chan, "XPC_IPI_OPENREQUEST (msg_size=%d, "
+                        "local_nentries=%d) received from partid=%d, "
+                        "channel=%d\n", args->msg_size, args->local_nentries,
+                        ch->partid, ch->number);
+
+                if (part->act_state == XPC_P_DEACTIVATING ||
+                                        (ch->flags & XPC_C_ROPENREQUEST)) {
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_WDISCONNECT)) {
+                        ch->delayed_IPI_flags |= XPC_IPI_OPENREQUEST;
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+                DBUG_ON(!(ch->flags & (XPC_C_DISCONNECTED |
+                                                        XPC_C_OPENREQUEST)));
+                DBUG_ON(ch->flags & (XPC_C_ROPENREQUEST | XPC_C_ROPENREPLY |
+                                        XPC_C_OPENREPLY | XPC_C_CONNECTED));
+
+                /*
+                 * The meaningful OPENREQUEST connection state fields are:
+                 *      msg_size = size of channel's messages in bytes
+                 *      local_nentries = remote partition's local_nentries
+                 */
+                if (args->msg_size == 0 || args->local_nentries == 0) {
+                        /* assume OPENREQUEST was delayed by mistake */
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                ch->flags |= (XPC_C_ROPENREQUEST | XPC_C_CONNECTING);
+                ch->remote_nentries = args->local_nentries;
+
+
+                if (ch->flags & XPC_C_OPENREQUEST) {
+                        if (args->msg_size != ch->msg_size) {
+                                XPC_DISCONNECT_CHANNEL(ch, xpcUnequalMsgSizes,
+                                                                &irq_flags);
+                                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                                return;
+                        }
+                } else {
+                        ch->msg_size = args->msg_size;
+
+                        XPC_SET_REASON(ch, 0, 0);
+                        ch->flags &= ~XPC_C_DISCONNECTED;
+
+                        atomic_inc(&part->nchannels_active);
+                }
+
+                xpc_process_connect(ch, &irq_flags);
+        }
+
+
+        if (IPI_flags & XPC_IPI_OPENREPLY) {
+
+                dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY (local_msgqueue_pa=0x%lx, "
+                        "local_nentries=%d, remote_nentries=%d) received from "
+                        "partid=%d, channel=%d\n", args->local_msgqueue_pa,
+                        args->local_nentries, args->remote_nentries,
+                        ch->partid, ch->number);
+
+                if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_DISCONNECTED)) {
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+                if (!(ch->flags & XPC_C_OPENREQUEST)) {
+                        XPC_DISCONNECT_CHANNEL(ch, xpcOpenCloseError,
+                                                                &irq_flags);
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return;
+                }
+
+                DBUG_ON(!(ch->flags & XPC_C_ROPENREQUEST));
+                DBUG_ON(ch->flags & XPC_C_CONNECTED);
+
+                /*
+                 * The meaningful OPENREPLY connection state fields are:
+                 *      local_msgqueue_pa = physical address of remote
+                 *                          partition's local_msgqueue
+                 *      local_nentries = remote partition's local_nentries
+                 *      remote_nentries = remote partition's remote_nentries
+                 */
+                DBUG_ON(args->local_msgqueue_pa == 0);
+                DBUG_ON(args->local_nentries == 0);
+                DBUG_ON(args->remote_nentries == 0);
+
+                ch->flags |= XPC_C_ROPENREPLY;
+                ch->remote_msgqueue_pa = args->local_msgqueue_pa;
+
+                if (args->local_nentries < ch->remote_nentries) {
+                        dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY: new "
+                                "remote_nentries=%d, old remote_nentries=%d, "
+                                "partid=%d, channel=%d\n",
+                                args->local_nentries, ch->remote_nentries,
+                                ch->partid, ch->number);
+
+                        ch->remote_nentries = args->local_nentries;
+                }
+                if (args->remote_nentries < ch->local_nentries) {
+                        dev_dbg(xpc_chan, "XPC_IPI_OPENREPLY: new "
+                                "local_nentries=%d, old local_nentries=%d, "
+                                "partid=%d, channel=%d\n",
+                                args->remote_nentries, ch->local_nentries,
+                                ch->partid, ch->number);
+
+                        ch->local_nentries = args->remote_nentries;
+                }
+
+                xpc_process_connect(ch, &irq_flags);
+        }
+
+        spin_unlock_irqrestore(&ch->lock, irq_flags);
+}
+
+
+/*
+ * Attempt to establish a channel connection to a remote partition.
+ */
+static enum xpc_retval
+xpc_connect_channel(struct xpc_channel *ch)
+{
+        unsigned long irq_flags;
+        struct xpc_registration *registration = &xpc_registrations[ch->number];
+
+
+        if (mutex_trylock(&registration->mutex) == 0) {
+                return xpcRetry;
+        }
+
+        if (!XPC_CHANNEL_REGISTERED(ch->number)) {
+                mutex_unlock(&registration->mutex);
+                return xpcUnregistered;
+        }
+
+        spin_lock_irqsave(&ch->lock, irq_flags);
+
+        DBUG_ON(ch->flags & XPC_C_CONNECTED);
+        DBUG_ON(ch->flags & XPC_C_OPENREQUEST);
+
+        if (ch->flags & XPC_C_DISCONNECTING) {
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                mutex_unlock(&registration->mutex);
+                return ch->reason;
+        }
+
+
+        /* add info from the channel connect registration to the channel */
+
+        ch->kthreads_assigned_limit = registration->assigned_limit;
+        ch->kthreads_idle_limit = registration->idle_limit;
+        DBUG_ON(atomic_read(&ch->kthreads_assigned) != 0);
+        DBUG_ON(atomic_read(&ch->kthreads_idle) != 0);
+        DBUG_ON(atomic_read(&ch->kthreads_active) != 0);
+
+        ch->func = registration->func;
+        DBUG_ON(registration->func == NULL);
+        ch->key = registration->key;
+
+        ch->local_nentries = registration->nentries;
+
+        if (ch->flags & XPC_C_ROPENREQUEST) {
+                if (registration->msg_size != ch->msg_size) {
+                        /* the local and remote sides aren't the same */
+
+                        /*
+                         * Because XPC_DISCONNECT_CHANNEL() can block we're
+                         * forced to up the registration sema before we unlock
+                         * the channel lock. But that's okay here because we're
+                         * done with the part that required the registration
+                         * sema. XPC_DISCONNECT_CHANNEL() requires that the
+                         * channel lock be locked and will unlock and relock
+                         * the channel lock as needed.
+                         */
+                        mutex_unlock(&registration->mutex);
+                        XPC_DISCONNECT_CHANNEL(ch, xpcUnequalMsgSizes,
+                                                                &irq_flags);
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        return xpcUnequalMsgSizes;
+                }
+        } else {
+                ch->msg_size = registration->msg_size;
+
+                XPC_SET_REASON(ch, 0, 0);
+                ch->flags &= ~XPC_C_DISCONNECTED;
+
+                atomic_inc(&xpc_partitions[ch->partid].nchannels_active);
+        }
+
+        mutex_unlock(&registration->mutex);
+
+
+        /* initiate the connection */
+
+        ch->flags |= (XPC_C_OPENREQUEST | XPC_C_CONNECTING);
+        xpc_IPI_send_openrequest(ch, &irq_flags);
+
+        xpc_process_connect(ch, &irq_flags);
+
+        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Clear some of the msg flags in the local message queue.
+ */
+static inline void
+xpc_clear_local_msgqueue_flags(struct xpc_channel *ch)
+{
+        struct xpc_msg *msg;
+        s64 get;
+
+
+        get = ch->w_remote_GP.get;
+        do {
+                msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
+                                (get % ch->local_nentries) * ch->msg_size);
+                msg->flags = 0;
+        } while (++get < (volatile s64) ch->remote_GP.get);
+}
+
+
+/*
+ * Clear some of the msg flags in the remote message queue.
+ */
+static inline void
+xpc_clear_remote_msgqueue_flags(struct xpc_channel *ch)
+{
+        struct xpc_msg *msg;
+        s64 put;
+
+
+        put = ch->w_remote_GP.put;
+        do {
+                msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
+                                (put % ch->remote_nentries) * ch->msg_size);
+                msg->flags = 0;
+        } while (++put < (volatile s64) ch->remote_GP.put);
+}
+
+
+static void
+xpc_process_msg_IPI(struct xpc_partition *part, int ch_number)
+{
+        struct xpc_channel *ch = &part->channels[ch_number];
+        int nmsgs_sent;
+
+
+        ch->remote_GP = part->remote_GPs[ch_number];
+
+
+        /* See what, if anything, has changed for each connected channel */
+
+        xpc_msgqueue_ref(ch);
+
+        if (ch->w_remote_GP.get == ch->remote_GP.get &&
+                                ch->w_remote_GP.put == ch->remote_GP.put) {
+                /* nothing changed since GPs were last pulled */
+                xpc_msgqueue_deref(ch);
+                return;
+        }
+
+        if (!(ch->flags & XPC_C_CONNECTED)){
+                xpc_msgqueue_deref(ch);
+                return;
+        }
+
+
+        /*
+         * First check to see if messages recently sent by us have been
+         * received by the other side. (The remote GET value will have
+         * changed since we last looked at it.)
+         */
+
+        if (ch->w_remote_GP.get != ch->remote_GP.get) {
+
+                /*
+                 * We need to notify any senders that want to be notified
+                 * that their sent messages have been received by their
+                 * intended recipients. We need to do this before updating
+                 * w_remote_GP.get so that we don't allocate the same message
+                 * queue entries prematurely (see xpc_allocate_msg()).
+                 */
+                if (atomic_read(&ch->n_to_notify) > 0) {
+                        /*
+                         * Notify senders that messages sent have been
+                         * received and delivered by the other side.
+                         */
+                        xpc_notify_senders(ch, xpcMsgDelivered,
+                                                        ch->remote_GP.get);
+                }
+
+                /*
+                 * Clear msg->flags in previously sent messages, so that
+                 * they're ready for xpc_allocate_msg().
+                 */
+                xpc_clear_local_msgqueue_flags(ch);
+
+                ch->w_remote_GP.get = ch->remote_GP.get;
+
+                dev_dbg(xpc_chan, "w_remote_GP.get changed to %ld, partid=%d, "
+                        "channel=%d\n", ch->w_remote_GP.get, ch->partid,
+                        ch->number);
+
+                /*
+                 * If anyone was waiting for message queue entries to become
+                 * available, wake them up.
+                 */
+                if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) {
+                        wake_up(&ch->msg_allocate_wq);
+                }
+        }
+
+
+        /*
+         * Now check for newly sent messages by the other side. (The remote
+         * PUT value will have changed since we last looked at it.)
+         */
+
+        if (ch->w_remote_GP.put != ch->remote_GP.put) {
+                /*
+                 * Clear msg->flags in previously received messages, so that
+                 * they're ready for xpc_get_deliverable_msg().
+                 */
+                xpc_clear_remote_msgqueue_flags(ch);
+
+                ch->w_remote_GP.put = ch->remote_GP.put;
+
+                dev_dbg(xpc_chan, "w_remote_GP.put changed to %ld, partid=%d, "
+                        "channel=%d\n", ch->w_remote_GP.put, ch->partid,
+                        ch->number);
+
+                nmsgs_sent = ch->w_remote_GP.put - ch->w_local_GP.get;
+                if (nmsgs_sent > 0) {
+                        dev_dbg(xpc_chan, "msgs waiting to be copied and "
+                                "delivered=%d, partid=%d, channel=%d\n",
+                                nmsgs_sent, ch->partid, ch->number);
+
+                        if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
+                                xpc_activate_kthreads(ch, nmsgs_sent);
+                        }
+                }
+        }
+
+        xpc_msgqueue_deref(ch);
+}
+
+
+void
+xpc_process_channel_activity(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        u64 IPI_amo, IPI_flags;
+        struct xpc_channel *ch;
+        int ch_number;
+        u32 ch_flags;
+
+
+        IPI_amo = xpc_get_IPI_flags(part);
+
+        /*
+         * Initiate channel connections for registered channels.
+         *
+         * For each connected channel that has pending messages activate idle
+         * kthreads and/or create new kthreads as needed.
+         */
+
+        for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
+                ch = &part->channels[ch_number];
+
+
+                /*
+                 * Process any open or close related IPI flags, and then deal
+                 * with connecting or disconnecting the channel as required.
+                 */
+
+                IPI_flags = XPC_GET_IPI_FLAGS(IPI_amo, ch_number);
+
+                if (XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(IPI_flags)) {
+                        xpc_process_openclose_IPI(part, ch_number, IPI_flags);
+                }
+
+                ch_flags = ch->flags;   /* need an atomic snapshot of flags */
+
+                if (ch_flags & XPC_C_DISCONNECTING) {
+                        spin_lock_irqsave(&ch->lock, irq_flags);
+                        xpc_process_disconnect(ch, &irq_flags);
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        continue;
+                }
+
+                if (part->act_state == XPC_P_DEACTIVATING) {
+                        continue;
+                }
+
+                if (!(ch_flags & XPC_C_CONNECTED)) {
+                        if (!(ch_flags & XPC_C_OPENREQUEST)) {
+                                DBUG_ON(ch_flags & XPC_C_SETUP);
+                                (void) xpc_connect_channel(ch);
+                        } else {
+                                spin_lock_irqsave(&ch->lock, irq_flags);
+                                xpc_process_connect(ch, &irq_flags);
+                                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        }
+                        continue;
+                }
+
+
+                /*
+                 * Process any message related IPI flags, this may involve the
+                 * activation of kthreads to deliver any pending messages sent
+                 * from the other partition.
+                 */
+
+                if (XPC_ANY_MSG_IPI_FLAGS_SET(IPI_flags)) {
+                        xpc_process_msg_IPI(part, ch_number);
+                }
+        }
+}
+
+
+/*
+ * XPC's heartbeat code calls this function to inform XPC that a partition is
+ * going down.  XPC responds by tearing down the XPartition Communication
+ * infrastructure used for the just downed partition.
+ *
+ * XPC's heartbeat code will never call this function and xpc_partition_up()
+ * at the same time. Nor will it ever make multiple calls to either function
+ * at the same time.
+ */
+void
+xpc_partition_going_down(struct xpc_partition *part, enum xpc_retval reason)
+{
+        unsigned long irq_flags;
+        int ch_number;
+        struct xpc_channel *ch;
+
+
+        dev_dbg(xpc_chan, "deactivating partition %d, reason=%d\n",
+                XPC_PARTID(part), reason);
+
+        if (!xpc_part_ref(part)) {
+                /* infrastructure for this partition isn't currently set up */
+                return;
+        }
+
+
+        /* disconnect channels associated with the partition going down */
+
+        for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
+                ch = &part->channels[ch_number];
+
+                xpc_msgqueue_ref(ch);
+                spin_lock_irqsave(&ch->lock, irq_flags);
+
+                XPC_DISCONNECT_CHANNEL(ch, reason, &irq_flags);
+
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                xpc_msgqueue_deref(ch);
+        }
+
+        xpc_wakeup_channel_mgr(part);
+
+        xpc_part_deref(part);
+}
+
+
+/*
+ * Teardown the infrastructure necessary to support XPartition Communication
+ * between the specified remote partition and the local one.
+ */
+void
+xpc_teardown_infrastructure(struct xpc_partition *part)
+{
+        partid_t partid = XPC_PARTID(part);
+
+
+        /*
+         * We start off by making this partition inaccessible to local
+         * processes by marking it as no longer setup. Then we make it
+         * inaccessible to remote processes by clearing the XPC per partition
+         * specific variable's magic # (which indicates that these variables
+         * are no longer valid) and by ignoring all XPC notify IPIs sent to
+         * this partition.
+         */
+
+        DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
+        DBUG_ON(atomic_read(&part->nchannels_active) != 0);
+        DBUG_ON(part->setup_state != XPC_P_SETUP);
+        part->setup_state = XPC_P_WTEARDOWN;
+
+        xpc_vars_part[partid].magic = 0;
+
+
+        free_irq(SGI_XPC_NOTIFY, (void *) (u64) partid);
+
+
+        /*
+         * Before proceeding with the teardown we have to wait until all
+         * existing references cease.
+         */
+        wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
+
+
+        /* now we can begin tearing down the infrastructure */
+
+        part->setup_state = XPC_P_TORNDOWN;
+
+        /* in case we've still got outstanding timers registered... */
+        del_timer_sync(&part->dropped_IPI_timer);
+
+        kfree(part->remote_openclose_args_base);
+        part->remote_openclose_args = NULL;
+        kfree(part->local_openclose_args_base);
+        part->local_openclose_args = NULL;
+        kfree(part->remote_GPs_base);
+        part->remote_GPs = NULL;
+        kfree(part->local_GPs_base);
+        part->local_GPs = NULL;
+        kfree(part->channels);
+        part->channels = NULL;
+        part->local_IPI_amo_va = NULL;
+}
+
+
+/*
+ * Called by XP at the time of channel connection registration to cause
+ * XPC to establish connections to all currently active partitions.
+ */
+void
+xpc_initiate_connect(int ch_number)
+{
+        partid_t partid;
+        struct xpc_partition *part;
+        struct xpc_channel *ch;
+
+
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                part = &xpc_partitions[partid];
+
+                if (xpc_part_ref(part)) {
+                        ch = &part->channels[ch_number];
+
+                        /*
+                         * Initiate the establishment of a connection on the
+                         * newly registered channel to the remote partition.
+                         */
+                        xpc_wakeup_channel_mgr(part);
+                        xpc_part_deref(part);
+                }
+        }
+}
+
+
+void
+xpc_connected_callout(struct xpc_channel *ch)
+{
+        /* let the registerer know that a connection has been established */
+
+        if (ch->func != NULL) {
+                dev_dbg(xpc_chan, "ch->func() called, reason=xpcConnected, "
+                        "partid=%d, channel=%d\n", ch->partid, ch->number);
+
+                ch->func(xpcConnected, ch->partid, ch->number,
+                                (void *) (u64) ch->local_nentries, ch->key);
+
+                dev_dbg(xpc_chan, "ch->func() returned, reason=xpcConnected, "
+                        "partid=%d, channel=%d\n", ch->partid, ch->number);
+        }
+}
+
+
+/*
+ * Called by XP at the time of channel connection unregistration to cause
+ * XPC to teardown all current connections for the specified channel.
+ *
+ * Before returning xpc_initiate_disconnect() will wait until all connections
+ * on the specified channel have been closed/torndown. So the caller can be
+ * assured that they will not be receiving any more callouts from XPC to the
+ * function they registered via xpc_connect().
+ *
+ * Arguments:
+ *
+ *      ch_number - channel # to unregister.
+ */
+void
+xpc_initiate_disconnect(int ch_number)
+{
+        unsigned long irq_flags;
+        partid_t partid;
+        struct xpc_partition *part;
+        struct xpc_channel *ch;
+
+
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+
+        /* initiate the channel disconnect for every active partition */
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                part = &xpc_partitions[partid];
+
+                if (xpc_part_ref(part)) {
+                        ch = &part->channels[ch_number];
+                        xpc_msgqueue_ref(ch);
+
+                        spin_lock_irqsave(&ch->lock, irq_flags);
+
+                        if (!(ch->flags & XPC_C_DISCONNECTED)) {
+                                ch->flags |= XPC_C_WDISCONNECT;
+
+                                XPC_DISCONNECT_CHANNEL(ch, xpcUnregistering,
+                                                                &irq_flags);
+                        }
+
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+                        xpc_msgqueue_deref(ch);
+                        xpc_part_deref(part);
+                }
+        }
+
+        xpc_disconnect_wait(ch_number);
+}
+
+
+/*
+ * To disconnect a channel, and reflect it back to all who may be waiting.
+ *
+ * An OPEN is not allowed until XPC_C_DISCONNECTING is cleared by
+ * xpc_process_disconnect(), and if set, XPC_C_WDISCONNECT is cleared by
+ * xpc_disconnect_wait().
+ *
+ * THE CHANNEL IS TO BE LOCKED BY THE CALLER AND WILL REMAIN LOCKED UPON RETURN.
+ */
+void
+xpc_disconnect_channel(const int line, struct xpc_channel *ch,
+                        enum xpc_retval reason, unsigned long *irq_flags)
+{
+        u32 channel_was_connected = (ch->flags & XPC_C_CONNECTED);
+
+
+        DBUG_ON(!spin_is_locked(&ch->lock));
+
+        if (ch->flags & (XPC_C_DISCONNECTING | XPC_C_DISCONNECTED)) {
+                return;
+        }
+        DBUG_ON(!(ch->flags & (XPC_C_CONNECTING | XPC_C_CONNECTED)));
+
+        dev_dbg(xpc_chan, "reason=%d, line=%d, partid=%d, channel=%d\n",
+                reason, line, ch->partid, ch->number);
+
+        XPC_SET_REASON(ch, reason, line);
+
+        ch->flags |= (XPC_C_CLOSEREQUEST | XPC_C_DISCONNECTING);
+        /* some of these may not have been set */
+        ch->flags &= ~(XPC_C_OPENREQUEST | XPC_C_OPENREPLY |
+                        XPC_C_ROPENREQUEST | XPC_C_ROPENREPLY |
+                        XPC_C_CONNECTING | XPC_C_CONNECTED);
+
+        xpc_IPI_send_closerequest(ch, irq_flags);
+
+        if (channel_was_connected) {
+                ch->flags |= XPC_C_WASCONNECTED;
+        }
+
+        spin_unlock_irqrestore(&ch->lock, *irq_flags);
+
+        /* wake all idle kthreads so they can exit */
+        if (atomic_read(&ch->kthreads_idle) > 0) {
+                wake_up_all(&ch->idle_wq);
+
+        } else if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
+                        !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) {
+                /* start a kthread that will do the xpcDisconnecting callout */
+                xpc_create_kthreads(ch, 1, 1);
+        }
+
+        /* wake those waiting to allocate an entry from the local msg queue */
+        if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) {
+                wake_up(&ch->msg_allocate_wq);
+        }
+
+        spin_lock_irqsave(&ch->lock, *irq_flags);
+}
+
+
+void
+xpc_disconnect_callout(struct xpc_channel *ch, enum xpc_retval reason)
+{
+        /*
+         * Let the channel's registerer know that the channel is being
+         * disconnected. We don't want to do this if the registerer was never
+         * informed of a connection being made.
+         */
+
+        if (ch->func != NULL) {
+                dev_dbg(xpc_chan, "ch->func() called, reason=%d, partid=%d, "
+                        "channel=%d\n", reason, ch->partid, ch->number);
+
+                ch->func(reason, ch->partid, ch->number, NULL, ch->key);
+
+                dev_dbg(xpc_chan, "ch->func() returned, reason=%d, partid=%d, "
+                        "channel=%d\n", reason, ch->partid, ch->number);
+        }
+}
+
+
+/*
+ * Wait for a message entry to become available for the specified channel,
+ * but don't wait any longer than 1 jiffy.
+ */
+static enum xpc_retval
+xpc_allocate_msg_wait(struct xpc_channel *ch)
+{
+        enum xpc_retval ret;
+
+
+        if (ch->flags & XPC_C_DISCONNECTING) {
+                DBUG_ON(ch->reason == xpcInterrupted);  // >>> Is this true?
+                return ch->reason;
+        }
+
+        atomic_inc(&ch->n_on_msg_allocate_wq);
+        ret = interruptible_sleep_on_timeout(&ch->msg_allocate_wq, 1);
+        atomic_dec(&ch->n_on_msg_allocate_wq);
+
+        if (ch->flags & XPC_C_DISCONNECTING) {
+                ret = ch->reason;
+                DBUG_ON(ch->reason == xpcInterrupted);  // >>> Is this true?
+        } else if (ret == 0) {
+                ret = xpcTimeout;
+        } else {
+                ret = xpcInterrupted;
+        }
+
+        return ret;
+}
+
+
+/*
+ * Allocate an entry for a message from the message queue associated with the
+ * specified channel.
+ */
+static enum xpc_retval
+xpc_allocate_msg(struct xpc_channel *ch, u32 flags,
+                        struct xpc_msg **address_of_msg)
+{
+        struct xpc_msg *msg;
+        enum xpc_retval ret;
+        s64 put;
+
+
+        /* this reference will be dropped in xpc_send_msg() */
+        xpc_msgqueue_ref(ch);
+
+        if (ch->flags & XPC_C_DISCONNECTING) {
+                xpc_msgqueue_deref(ch);
+                return ch->reason;
+        }
+        if (!(ch->flags & XPC_C_CONNECTED)) {
+                xpc_msgqueue_deref(ch);
+                return xpcNotConnected;
+        }
+
+
+        /*
+         * Get the next available message entry from the local message queue.
+         * If none are available, we'll make sure that we grab the latest
+         * GP values.
+         */
+        ret = xpcTimeout;
+
+        while (1) {
+
+                put = (volatile s64) ch->w_local_GP.put;
+                if (put - (volatile s64) ch->w_remote_GP.get <
+                                                        ch->local_nentries) {
+
+                        /* There are available message entries. We need to try
+                         * to secure one for ourselves. We'll do this by trying
+                         * to increment w_local_GP.put as long as someone else
+                         * doesn't beat us to it. If they do, we'll have to
+                         * try again.
+                         */
+                        if (cmpxchg(&ch->w_local_GP.put, put, put + 1) ==
+                                                                        put) {
+                                /* we got the entry referenced by put */
+                                break;
+                        }
+                        continue;       /* try again */
+                }
+
+
+                /*
+                 * There aren't any available msg entries at this time.
+                 *
+                 * In waiting for a message entry to become available,
+                 * we set a timeout in case the other side is not
+                 * sending completion IPIs. This lets us fake an IPI
+                 * that will cause the IPI handler to fetch the latest
+                 * GP values as if an IPI was sent by the other side.
+                 */
+                if (ret == xpcTimeout) {
+                        xpc_IPI_send_local_msgrequest(ch);
+                }
+
+                if (flags & XPC_NOWAIT) {
+                        xpc_msgqueue_deref(ch);
+                        return xpcNoWait;
+                }
+
+                ret = xpc_allocate_msg_wait(ch);
+                if (ret != xpcInterrupted && ret != xpcTimeout) {
+                        xpc_msgqueue_deref(ch);
+                        return ret;
+                }
+        }
+
+
+        /* get the message's address and initialize it */
+        msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
+                                (put % ch->local_nentries) * ch->msg_size);
+
+
+        DBUG_ON(msg->flags != 0);
+        msg->number = put;
+
+        dev_dbg(xpc_chan, "w_local_GP.put changed to %ld; msg=0x%p, "
+                "msg_number=%ld, partid=%d, channel=%d\n", put + 1,
+                (void *) msg, msg->number, ch->partid, ch->number);
+
+        *address_of_msg = msg;
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Allocate an entry for a message from the message queue associated with the
+ * specified channel. NOTE that this routine can sleep waiting for a message
+ * entry to become available. To not sleep, pass in the XPC_NOWAIT flag.
+ *
+ * Arguments:
+ *
+ *      partid - ID of partition to which the channel is connected.
+ *      ch_number - channel #.
+ *      flags - see xpc.h for valid flags.
+ *      payload - address of the allocated payload area pointer (filled in on
+ *                return) in which the user-defined message is constructed.
+ */
+enum xpc_retval
+xpc_initiate_allocate(partid_t partid, int ch_number, u32 flags, void **payload)
+{
+        struct xpc_partition *part = &xpc_partitions[partid];
+        enum xpc_retval ret = xpcUnknownReason;
+        struct xpc_msg *msg = NULL;
+
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
+
+        *payload = NULL;
+
+        if (xpc_part_ref(part)) {
+                ret = xpc_allocate_msg(&part->channels[ch_number], flags, &msg);
+                xpc_part_deref(part);
+
+                if (msg != NULL) {
+                        *payload = &msg->payload;
+                }
+        }
+
+        return ret;
+}
+
+
+/*
+ * Now we actually send the messages that are ready to be sent by advancing
+ * the local message queue's Put value and then send an IPI to the recipient
+ * partition.
+ */
+static void
+xpc_send_msgs(struct xpc_channel *ch, s64 initial_put)
+{
+        struct xpc_msg *msg;
+        s64 put = initial_put + 1;
+        int send_IPI = 0;
+
+
+        while (1) {
+
+                while (1) {
+                        if (put == (volatile s64) ch->w_local_GP.put) {
+                                break;
+                        }
+
+                        msg = (struct xpc_msg *) ((u64) ch->local_msgqueue +
+                               (put % ch->local_nentries) * ch->msg_size);
+
+                        if (!(msg->flags & XPC_M_READY)) {
+                                break;
+                        }
+
+                        put++;
+                }
+
+                if (put == initial_put) {
+                        /* nothing's changed */
+                        break;
+                }
+
+                if (cmpxchg_rel(&ch->local_GP->put, initial_put, put) !=
+                                                                initial_put) {
+                        /* someone else beat us to it */
+                        DBUG_ON((volatile s64) ch->local_GP->put < initial_put);
+                        break;
+                }
+
+                /* we just set the new value of local_GP->put */
+
+                dev_dbg(xpc_chan, "local_GP->put changed to %ld, partid=%d, "
+                        "channel=%d\n", put, ch->partid, ch->number);
+
+                send_IPI = 1;
+
+                /*
+                 * We need to ensure that the message referenced by
+                 * local_GP->put is not XPC_M_READY or that local_GP->put
+                 * equals w_local_GP.put, so we'll go have a look.
+                 */
+                initial_put = put;
+        }
+
+        if (send_IPI) {
+                xpc_IPI_send_msgrequest(ch);
+        }
+}
+
+
+/*
+ * Common code that does the actual sending of the message by advancing the
+ * local message queue's Put value and sends an IPI to the partition the
+ * message is being sent to.
+ */
+static enum xpc_retval
+xpc_send_msg(struct xpc_channel *ch, struct xpc_msg *msg, u8 notify_type,
+                        xpc_notify_func func, void *key)
+{
+        enum xpc_retval ret = xpcSuccess;
+        struct xpc_notify *notify = notify;
+        s64 put, msg_number = msg->number;
+
+
+        DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
+        DBUG_ON((((u64) msg - (u64) ch->local_msgqueue) / ch->msg_size) !=
+                                        msg_number % ch->local_nentries);
+        DBUG_ON(msg->flags & XPC_M_READY);
+
+        if (ch->flags & XPC_C_DISCONNECTING) {
+                /* drop the reference grabbed in xpc_allocate_msg() */
+                xpc_msgqueue_deref(ch);
+                return ch->reason;
+        }
+
+        if (notify_type != 0) {
+                /*
+                 * Tell the remote side to send an ACK interrupt when the
+                 * message has been delivered.
+                 */
+                msg->flags |= XPC_M_INTERRUPT;
+
+                atomic_inc(&ch->n_to_notify);
+
+                notify = &ch->notify_queue[msg_number % ch->local_nentries];
+                notify->func = func;
+                notify->key = key;
+                notify->type = notify_type;
+
+                // >>> is a mb() needed here?
+
+                if (ch->flags & XPC_C_DISCONNECTING) {
+                        /*
+                         * An error occurred between our last error check and
+                         * this one. We will try to clear the type field from
+                         * the notify entry. If we succeed then
+                         * xpc_disconnect_channel() didn't already process
+                         * the notify entry.
+                         */
+                        if (cmpxchg(&notify->type, notify_type, 0) ==
+                                                                notify_type) {
+                                atomic_dec(&ch->n_to_notify);
+                                ret = ch->reason;
+                        }
+
+                        /* drop the reference grabbed in xpc_allocate_msg() */
+                        xpc_msgqueue_deref(ch);
+                        return ret;
+                }
+        }
+
+        msg->flags |= XPC_M_READY;
+
+        /*
+         * The preceding store of msg->flags must occur before the following
+         * load of ch->local_GP->put.
+         */
+        mb();
+
+        /* see if the message is next in line to be sent, if so send it */
+
+        put = ch->local_GP->put;
+        if (put == msg_number) {
+                xpc_send_msgs(ch, put);
+        }
+
+        /* drop the reference grabbed in xpc_allocate_msg() */
+        xpc_msgqueue_deref(ch);
+        return ret;
+}
+
+
+/*
+ * Send a message previously allocated using xpc_initiate_allocate() on the
+ * specified channel connected to the specified partition.
+ *
+ * This routine will not wait for the message to be received, nor will
+ * notification be given when it does happen. Once this routine has returned
+ * the message entry allocated via xpc_initiate_allocate() is no longer
+ * accessable to the caller.
+ *
+ * This routine, although called by users, does not call xpc_part_ref() to
+ * ensure that the partition infrastructure is in place. It relies on the
+ * fact that we called xpc_msgqueue_ref() in xpc_allocate_msg().
+ *
+ * Arguments:
+ *
+ *      partid - ID of partition to which the channel is connected.
+ *      ch_number - channel # to send message on.
+ *      payload - pointer to the payload area allocated via
+ *                      xpc_initiate_allocate().
+ */
+enum xpc_retval
+xpc_initiate_send(partid_t partid, int ch_number, void *payload)
+{
+        struct xpc_partition *part = &xpc_partitions[partid];
+        struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
+        enum xpc_retval ret;
+
+
+        dev_dbg(xpc_chan, "msg=0x%p, partid=%d, channel=%d\n", (void *) msg,
+                partid, ch_number);
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
+        DBUG_ON(msg == NULL);
+
+        ret = xpc_send_msg(&part->channels[ch_number], msg, 0, NULL, NULL);
+
+        return ret;
+}
+
+
+/*
+ * Send a message previously allocated using xpc_initiate_allocate on the
+ * specified channel connected to the specified partition.
+ *
+ * This routine will not wait for the message to be sent. Once this routine
+ * has returned the message entry allocated via xpc_initiate_allocate() is no
+ * longer accessable to the caller.
+ *
+ * Once the remote end of the channel has received the message, the function
+ * passed as an argument to xpc_initiate_send_notify() will be called. This
+ * allows the sender to free up or re-use any buffers referenced by the
+ * message, but does NOT mean the message has been processed at the remote
+ * end by a receiver.
+ *
+ * If this routine returns an error, the caller's function will NOT be called.
+ *
+ * This routine, although called by users, does not call xpc_part_ref() to
+ * ensure that the partition infrastructure is in place. It relies on the
+ * fact that we called xpc_msgqueue_ref() in xpc_allocate_msg().
+ *
+ * Arguments:
+ *
+ *      partid - ID of partition to which the channel is connected.
+ *      ch_number - channel # to send message on.
+ *      payload - pointer to the payload area allocated via
+ *                      xpc_initiate_allocate().
+ *      func - function to call with asynchronous notification of message
+ *                receipt. THIS FUNCTION MUST BE NON-BLOCKING.
+ *      key - user-defined key to be passed to the function when it's called.
+ */
+enum xpc_retval
+xpc_initiate_send_notify(partid_t partid, int ch_number, void *payload,
+                                xpc_notify_func func, void *key)
+{
+        struct xpc_partition *part = &xpc_partitions[partid];
+        struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
+        enum xpc_retval ret;
+
+
+        dev_dbg(xpc_chan, "msg=0x%p, partid=%d, channel=%d\n", (void *) msg,
+                partid, ch_number);
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
+        DBUG_ON(msg == NULL);
+        DBUG_ON(func == NULL);
+
+        ret = xpc_send_msg(&part->channels[ch_number], msg, XPC_N_CALL,
+                                                                func, key);
+        return ret;
+}
+
+
+static struct xpc_msg *
+xpc_pull_remote_msg(struct xpc_channel *ch, s64 get)
+{
+        struct xpc_partition *part = &xpc_partitions[ch->partid];
+        struct xpc_msg *remote_msg, *msg;
+        u32 msg_index, nmsgs;
+        u64 msg_offset;
+        enum xpc_retval ret;
+
+
+        if (mutex_lock_interruptible(&ch->msg_to_pull_mutex) != 0) {
+                /* we were interrupted by a signal */
+                return NULL;
+        }
+
+        while (get >= ch->next_msg_to_pull) {
+
+                /* pull as many messages as are ready and able to be pulled */
+
+                msg_index = ch->next_msg_to_pull % ch->remote_nentries;
+
+                DBUG_ON(ch->next_msg_to_pull >=
+                                        (volatile s64) ch->w_remote_GP.put);
+                nmsgs =  (volatile s64) ch->w_remote_GP.put -
+                                                ch->next_msg_to_pull;
+                if (msg_index + nmsgs > ch->remote_nentries) {
+                        /* ignore the ones that wrap the msg queue for now */
+                        nmsgs = ch->remote_nentries - msg_index;
+                }
+
+                msg_offset = msg_index * ch->msg_size;
+                msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
+                                                                msg_offset);
+                remote_msg = (struct xpc_msg *) (ch->remote_msgqueue_pa +
+                                                                msg_offset);
+
+                if ((ret = xpc_pull_remote_cachelines(part, msg, remote_msg,
+                                nmsgs * ch->msg_size)) != xpcSuccess) {
+
+                        dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
+                                " msg %ld from partition %d, channel=%d, "
+                                "ret=%d\n", nmsgs, ch->next_msg_to_pull,
+                                ch->partid, ch->number, ret);
+
+                        XPC_DEACTIVATE_PARTITION(part, ret);
+
+                        mutex_unlock(&ch->msg_to_pull_mutex);
+                        return NULL;
+                }
+
+                mb();   /* >>> this may not be needed, we're not sure */
+
+                ch->next_msg_to_pull += nmsgs;
+        }
+
+        mutex_unlock(&ch->msg_to_pull_mutex);
+
+        /* return the message we were looking for */
+        msg_offset = (get % ch->remote_nentries) * ch->msg_size;
+        msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue + msg_offset);
+
+        return msg;
+}
+
+
+/*
+ * Get a message to be delivered.
+ */
+static struct xpc_msg *
+xpc_get_deliverable_msg(struct xpc_channel *ch)
+{
+        struct xpc_msg *msg = NULL;
+        s64 get;
+
+
+        do {
+                if ((volatile u32) ch->flags & XPC_C_DISCONNECTING) {
+                        break;
+                }
+
+                get = (volatile s64) ch->w_local_GP.get;
+                if (get == (volatile s64) ch->w_remote_GP.put) {
+                        break;
+                }
+
+                /* There are messages waiting to be pulled and delivered.
+                 * We need to try to secure one for ourselves. We'll do this
+                 * by trying to increment w_local_GP.get and hope that no one
+                 * else beats us to it. If they do, we'll we'll simply have
+                 * to try again for the next one.
+                 */
+
+                if (cmpxchg(&ch->w_local_GP.get, get, get + 1) == get) {
+                        /* we got the entry referenced by get */
+
+                        dev_dbg(xpc_chan, "w_local_GP.get changed to %ld, "
+                                "partid=%d, channel=%d\n", get + 1,
+                                ch->partid, ch->number);
+
+                        /* pull the message from the remote partition */
+
+                        msg = xpc_pull_remote_msg(ch, get);
+
+                        DBUG_ON(msg != NULL && msg->number != get);
+                        DBUG_ON(msg != NULL && (msg->flags & XPC_M_DONE));
+                        DBUG_ON(msg != NULL && !(msg->flags & XPC_M_READY));
+
+                        break;
+                }
+
+        } while (1);
+
+        return msg;
+}
+
+
+/*
+ * Deliver a message to its intended recipient.
+ */
+void
+xpc_deliver_msg(struct xpc_channel *ch)
+{
+        struct xpc_msg *msg;
+
+
+        if ((msg = xpc_get_deliverable_msg(ch)) != NULL) {
+
+                /*
+                 * This ref is taken to protect the payload itself from being
+                 * freed before the user is finished with it, which the user
+                 * indicates by calling xpc_initiate_received().
+                 */
+                xpc_msgqueue_ref(ch);
+
+                atomic_inc(&ch->kthreads_active);
+
+                if (ch->func != NULL) {
+                        dev_dbg(xpc_chan, "ch->func() called, msg=0x%p, "
+                                "msg_number=%ld, partid=%d, channel=%d\n",
+                                (void *) msg, msg->number, ch->partid,
+                                ch->number);
+
+                        /* deliver the message to its intended recipient */
+                        ch->func(xpcMsgReceived, ch->partid, ch->number,
+                                        &msg->payload, ch->key);
+
+                        dev_dbg(xpc_chan, "ch->func() returned, msg=0x%p, "
+                                "msg_number=%ld, partid=%d, channel=%d\n",
+                                (void *) msg, msg->number, ch->partid,
+                                ch->number);
+                }
+
+                atomic_dec(&ch->kthreads_active);
+        }
+}
+
+
+/*
+ * Now we actually acknowledge the messages that have been delivered and ack'd
+ * by advancing the cached remote message queue's Get value and if requested
+ * send an IPI to the message sender's partition.
+ */
+static void
+xpc_acknowledge_msgs(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
+{
+        struct xpc_msg *msg;
+        s64 get = initial_get + 1;
+        int send_IPI = 0;
+
+
+        while (1) {
+
+                while (1) {
+                        if (get == (volatile s64) ch->w_local_GP.get) {
+                                break;
+                        }
+
+                        msg = (struct xpc_msg *) ((u64) ch->remote_msgqueue +
+                               (get % ch->remote_nentries) * ch->msg_size);
+
+                        if (!(msg->flags & XPC_M_DONE)) {
+                                break;
+                        }
+
+                        msg_flags |= msg->flags;
+                        get++;
+                }
+
+                if (get == initial_get) {
+                        /* nothing's changed */
+                        break;
+                }
+
+                if (cmpxchg_rel(&ch->local_GP->get, initial_get, get) !=
+                                                                initial_get) {
+                        /* someone else beat us to it */
+                        DBUG_ON((volatile s64) ch->local_GP->get <=
+                                                                initial_get);
+                        break;
+                }
+
+                /* we just set the new value of local_GP->get */
+
+                dev_dbg(xpc_chan, "local_GP->get changed to %ld, partid=%d, "
+                        "channel=%d\n", get, ch->partid, ch->number);
+
+                send_IPI = (msg_flags & XPC_M_INTERRUPT);
+
+                /*
+                 * We need to ensure that the message referenced by
+                 * local_GP->get is not XPC_M_DONE or that local_GP->get
+                 * equals w_local_GP.get, so we'll go have a look.
+                 */
+                initial_get = get;
+        }
+
+        if (send_IPI) {
+                xpc_IPI_send_msgrequest(ch);
+        }
+}
+
+
+/*
+ * Acknowledge receipt of a delivered message.
+ *
+ * If a message has XPC_M_INTERRUPT set, send an interrupt to the partition
+ * that sent the message.
+ *
+ * This function, although called by users, does not call xpc_part_ref() to
+ * ensure that the partition infrastructure is in place. It relies on the
+ * fact that we called xpc_msgqueue_ref() in xpc_deliver_msg().
+ *
+ * Arguments:
+ *
+ *      partid - ID of partition to which the channel is connected.
+ *      ch_number - channel # message received on.
+ *      payload - pointer to the payload area allocated via
+ *                      xpc_initiate_allocate().
+ */
+void
+xpc_initiate_received(partid_t partid, int ch_number, void *payload)
+{
+        struct xpc_partition *part = &xpc_partitions[partid];
+        struct xpc_channel *ch;
+        struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
+        s64 get, msg_number = msg->number;
+
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
+
+        ch = &part->channels[ch_number];
+
+        dev_dbg(xpc_chan, "msg=0x%p, msg_number=%ld, partid=%d, channel=%d\n",
+                (void *) msg, msg_number, ch->partid, ch->number);
+
+        DBUG_ON((((u64) msg - (u64) ch->remote_msgqueue) / ch->msg_size) !=
+                                        msg_number % ch->remote_nentries);
+        DBUG_ON(msg->flags & XPC_M_DONE);
+
+        msg->flags |= XPC_M_DONE;
+
+        /*
+         * The preceding store of msg->flags must occur before the following
+         * load of ch->local_GP->get.
+         */
+        mb();
+
+        /*
+         * See if this message is next in line to be acknowledged as having
+         * been delivered.
+         */
+        get = ch->local_GP->get;
+        if (get == msg_number) {
+                xpc_acknowledge_msgs(ch, get, msg->flags);
+        }
+
+        /* the call to xpc_msgqueue_ref() was done by xpc_deliver_msg()  */
+        xpc_msgqueue_deref(ch);
+}
+
diff --git a/drivers/misc/sgi-xp/xpc_main.c b/drivers/misc/sgi-xp/xpc_main.c
new file mode 100644
index 000000000000..bdb2cf1fcbcc
--- /dev/null
+++ b/drivers/misc/sgi-xp/xpc_main.c
@@ -0,0 +1,1431 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * Cross Partition Communication (XPC) support - standard version.
+ *
+ *      XPC provides a message passing capability that crosses partition
+ *      boundaries. This module is made up of two parts:
+ *
+ *          partition   This part detects the presence/absence of other
+ *                      partitions. It provides a heartbeat and monitors
+ *                      the heartbeats of other partitions.
+ *
+ *          channel     This part manages the channels and sends/receives
+ *                      messages across them to/from other partitions.
+ *
+ *      There are a couple of additional functions residing in XP, which
+ *      provide an interface to XPC for its users.
+ *
+ *
+ *      Caveats:
+ *
+ *        . We currently have no way to determine which nasid an IPI came
+ *          from. Thus, xpc_IPI_send() does a remote AMO write followed by
+ *          an IPI. The AMO indicates where data is to be pulled from, so
+ *          after the IPI arrives, the remote partition checks the AMO word.
+ *          The IPI can actually arrive before the AMO however, so other code
+ *          must periodically check for this case. Also, remote AMO operations
+ *          do not reliably time out. Thus we do a remote PIO read solely to
+ *          know whether the remote partition is down and whether we should
+ *          stop sending IPIs to it. This remote PIO read operation is set up
+ *          in a special nofault region so SAL knows to ignore (and cleanup)
+ *          any errors due to the remote AMO write, PIO read, and/or PIO
+ *          write operations.
+ *
+ *          If/when new hardware solves this IPI problem, we should abandon
+ *          the current approach.
+ *
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/sched.h>
+#include <linux/syscalls.h>
+#include <linux/cache.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+#include <linux/reboot.h>
+#include <linux/completion.h>
+#include <linux/kdebug.h>
+#include <asm/sn/intr.h>
+#include <asm/sn/sn_sal.h>
+#include <asm/uaccess.h>
+#include "xpc.h"
+
+
+/* define two XPC debug device structures to be used with dev_dbg() et al */
+
+struct device_driver xpc_dbg_name = {
+        .name = "xpc"
+};
+
+struct device xpc_part_dbg_subname = {
+        .bus_id = {0},          /* set to "part" at xpc_init() time */
+        .driver = &xpc_dbg_name
+};
+
+struct device xpc_chan_dbg_subname = {
+        .bus_id = {0},          /* set to "chan" at xpc_init() time */
+        .driver = &xpc_dbg_name
+};
+
+struct device *xpc_part = &xpc_part_dbg_subname;
+struct device *xpc_chan = &xpc_chan_dbg_subname;
+
+
+static int xpc_kdebug_ignore;
+
+
+/* systune related variables for /proc/sys directories */
+
+static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
+static int xpc_hb_min_interval = 1;
+static int xpc_hb_max_interval = 10;
+
+static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL;
+static int xpc_hb_check_min_interval = 10;
+static int xpc_hb_check_max_interval = 120;
+
+int xpc_disengage_request_timelimit = XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT;
+static int xpc_disengage_request_min_timelimit = 0;
+static int xpc_disengage_request_max_timelimit = 120;
+
+static ctl_table xpc_sys_xpc_hb_dir[] = {
+        {
+                .ctl_name       = CTL_UNNUMBERED,
+                .procname       = "hb_interval",
+                .data           = &xpc_hb_interval,
+                .maxlen         = sizeof(int),
+                .mode           = 0644,
+                .proc_handler   = &proc_dointvec_minmax,
+                .strategy       = &sysctl_intvec,
+                .extra1         = &xpc_hb_min_interval,
+                .extra2         = &xpc_hb_max_interval
+        },
+        {
+                .ctl_name       = CTL_UNNUMBERED,
+                .procname       = "hb_check_interval",
+                .data           = &xpc_hb_check_interval,
+                .maxlen         = sizeof(int),
+                .mode           = 0644,
+                .proc_handler   = &proc_dointvec_minmax,
+                .strategy       = &sysctl_intvec,
+                .extra1         = &xpc_hb_check_min_interval,
+                .extra2         = &xpc_hb_check_max_interval
+        },
+        {}
+};
+static ctl_table xpc_sys_xpc_dir[] = {
+        {
+                .ctl_name       = CTL_UNNUMBERED,
+                .procname       = "hb",
+                .mode           = 0555,
+                .child          = xpc_sys_xpc_hb_dir
+        },
+        {
+                .ctl_name       = CTL_UNNUMBERED,
+                .procname       = "disengage_request_timelimit",
+                .data           = &xpc_disengage_request_timelimit,
+                .maxlen         = sizeof(int),
+                .mode           = 0644,
+                .proc_handler   = &proc_dointvec_minmax,
+                .strategy       = &sysctl_intvec,
+                .extra1         = &xpc_disengage_request_min_timelimit,
+                .extra2         = &xpc_disengage_request_max_timelimit
+        },
+        {}
+};
+static ctl_table xpc_sys_dir[] = {
+        {
+                .ctl_name       = CTL_UNNUMBERED,
+                .procname       = "xpc",
+                .mode           = 0555,
+                .child          = xpc_sys_xpc_dir
+        },
+        {}
+};
+static struct ctl_table_header *xpc_sysctl;
+
+/* non-zero if any remote partition disengage request was timed out */
+int xpc_disengage_request_timedout;
+
+/* #of IRQs received */
+static atomic_t xpc_act_IRQ_rcvd;
+
+/* IRQ handler notifies this wait queue on receipt of an IRQ */
+static DECLARE_WAIT_QUEUE_HEAD(xpc_act_IRQ_wq);
+
+static unsigned long xpc_hb_check_timeout;
+
+/* notification that the xpc_hb_checker thread has exited */
+static DECLARE_COMPLETION(xpc_hb_checker_exited);
+
+/* notification that the xpc_discovery thread has exited */
+static DECLARE_COMPLETION(xpc_discovery_exited);
+
+
+static struct timer_list xpc_hb_timer;
+
+
+static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *);
+
+
+static int xpc_system_reboot(struct notifier_block *, unsigned long, void *);
+static struct notifier_block xpc_reboot_notifier = {
+        .notifier_call = xpc_system_reboot,
+};
+
+static int xpc_system_die(struct notifier_block *, unsigned long, void *);
+static struct notifier_block xpc_die_notifier = {
+        .notifier_call = xpc_system_die,
+};
+
+
+/*
+ * Timer function to enforce the timelimit on the partition disengage request.
+ */
+static void
+xpc_timeout_partition_disengage_request(unsigned long data)
+{
+        struct xpc_partition *part = (struct xpc_partition *) data;
+
+
+        DBUG_ON(time_before(jiffies, part->disengage_request_timeout));
+
+        (void) xpc_partition_disengaged(part);
+
+        DBUG_ON(part->disengage_request_timeout != 0);
+        DBUG_ON(xpc_partition_engaged(1UL << XPC_PARTID(part)) != 0);
+}
+
+
+/*
+ * Notify the heartbeat check thread that an IRQ has been received.
+ */
+static irqreturn_t
+xpc_act_IRQ_handler(int irq, void *dev_id)
+{
+        atomic_inc(&xpc_act_IRQ_rcvd);
+        wake_up_interruptible(&xpc_act_IRQ_wq);
+        return IRQ_HANDLED;
+}
+
+
+/*
+ * Timer to produce the heartbeat.  The timer structures function is
+ * already set when this is initially called.  A tunable is used to
+ * specify when the next timeout should occur.
+ */
+static void
+xpc_hb_beater(unsigned long dummy)
+{
+        xpc_vars->heartbeat++;
+
+        if (time_after_eq(jiffies, xpc_hb_check_timeout)) {
+                wake_up_interruptible(&xpc_act_IRQ_wq);
+        }
+
+        xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ);
+        add_timer(&xpc_hb_timer);
+}
+
+
+/*
+ * This thread is responsible for nearly all of the partition
+ * activation/deactivation.
+ */
+static int
+xpc_hb_checker(void *ignore)
+{
+        int last_IRQ_count = 0;
+        int new_IRQ_count;
+        int force_IRQ=0;
+
+
+        /* this thread was marked active by xpc_hb_init() */
+
+        daemonize(XPC_HB_CHECK_THREAD_NAME);
+
+        set_cpus_allowed(current, cpumask_of_cpu(XPC_HB_CHECK_CPU));
+
+        /* set our heartbeating to other partitions into motion */
+        xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
+        xpc_hb_beater(0);
+
+        while (!(volatile int) xpc_exiting) {
+
+                dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have "
+                        "been received\n",
+                        (int) (xpc_hb_check_timeout - jiffies),
+                        atomic_read(&xpc_act_IRQ_rcvd) - last_IRQ_count);
+
+
+                /* checking of remote heartbeats is skewed by IRQ handling */
+                if (time_after_eq(jiffies, xpc_hb_check_timeout)) {
+                        dev_dbg(xpc_part, "checking remote heartbeats\n");
+                        xpc_check_remote_hb();
+
+                        /*
+                         * We need to periodically recheck to ensure no
+                         * IPI/AMO pairs have been missed.  That check
+                         * must always reset xpc_hb_check_timeout.
+                         */
+                        force_IRQ = 1;
+                }
+
+
+                /* check for outstanding IRQs */
+                new_IRQ_count = atomic_read(&xpc_act_IRQ_rcvd);
+                if (last_IRQ_count < new_IRQ_count || force_IRQ != 0) {
+                        force_IRQ = 0;
+
+                        dev_dbg(xpc_part, "found an IRQ to process; will be "
+                                "resetting xpc_hb_check_timeout\n");
+
+                        last_IRQ_count += xpc_identify_act_IRQ_sender();
+                        if (last_IRQ_count < new_IRQ_count) {
+                                /* retry once to help avoid missing AMO */
+                                (void) xpc_identify_act_IRQ_sender();
+                        }
+                        last_IRQ_count = new_IRQ_count;
+
+                        xpc_hb_check_timeout = jiffies +
+                                           (xpc_hb_check_interval * HZ);
+                }
+
+                /* wait for IRQ or timeout */
+                (void) wait_event_interruptible(xpc_act_IRQ_wq,
+                            (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) ||
+                                        time_after_eq(jiffies, xpc_hb_check_timeout) ||
+                                                (volatile int) xpc_exiting));
+        }
+
+        dev_dbg(xpc_part, "heartbeat checker is exiting\n");
+
+
+        /* mark this thread as having exited */
+        complete(&xpc_hb_checker_exited);
+        return 0;
+}
+
+
+/*
+ * This thread will attempt to discover other partitions to activate
+ * based on info provided by SAL. This new thread is short lived and
+ * will exit once discovery is complete.
+ */
+static int
+xpc_initiate_discovery(void *ignore)
+{
+        daemonize(XPC_DISCOVERY_THREAD_NAME);
+
+        xpc_discovery();
+
+        dev_dbg(xpc_part, "discovery thread is exiting\n");
+
+        /* mark this thread as having exited */
+        complete(&xpc_discovery_exited);
+        return 0;
+}
+
+
+/*
+ * Establish first contact with the remote partititon. This involves pulling
+ * the XPC per partition variables from the remote partition and waiting for
+ * the remote partition to pull ours.
+ */
+static enum xpc_retval
+xpc_make_first_contact(struct xpc_partition *part)
+{
+        enum xpc_retval ret;
+
+
+        while ((ret = xpc_pull_remote_vars_part(part)) != xpcSuccess) {
+                if (ret != xpcRetry) {
+                        XPC_DEACTIVATE_PARTITION(part, ret);
+                        return ret;
+                }
+
+                dev_dbg(xpc_chan, "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_DEACTIVATING) {
+                        return part->reason;
+                }
+        }
+
+        return xpc_mark_partition_active(part);
+}
+
+
+/*
+ * The first kthread assigned to a newly activated partition is the one
+ * created by XPC HB with which it calls xpc_partition_up(). XPC hangs on to
+ * that kthread until the partition is brought down, at which time that kthread
+ * returns back to XPC HB. (The return of that kthread will signify to XPC HB
+ * that XPC has dismantled all communication infrastructure for the associated
+ * partition.) This kthread becomes the channel manager for that partition.
+ *
+ * Each active partition has a channel manager, who, besides connecting and
+ * disconnecting channels, will ensure that each of the partition's connected
+ * channels has the required number of assigned kthreads to get the work done.
+ */
+static void
+xpc_channel_mgr(struct xpc_partition *part)
+{
+        while (part->act_state != XPC_P_DEACTIVATING ||
+                        atomic_read(&part->nchannels_active) > 0 ||
+                                        !xpc_partition_disengaged(part)) {
+
+                xpc_process_channel_activity(part);
+
+
+                /*
+                 * Wait until we've been requested to activate kthreads or
+                 * all of the channel's message queues have been torn down or
+                 * a signal is pending.
+                 *
+                 * The channel_mgr_requests is set to 1 after being awakened,
+                 * This is done to prevent the channel mgr from making one pass
+                 * through the loop for each request, since he will
+                 * be servicing all the requests in one pass. The reason it's
+                 * set to 1 instead of 0 is so that other kthreads will know
+                 * that the channel mgr is running and won't bother trying to
+                 * wake him up.
+                 */
+                atomic_dec(&part->channel_mgr_requests);
+                (void) wait_event_interruptible(part->channel_mgr_wq,
+                                (atomic_read(&part->channel_mgr_requests) > 0 ||
+                                (volatile u64) part->local_IPI_amo != 0 ||
+                                ((volatile u8) part->act_state ==
+                                                        XPC_P_DEACTIVATING &&
+                                atomic_read(&part->nchannels_active) == 0 &&
+                                xpc_partition_disengaged(part))));
+                atomic_set(&part->channel_mgr_requests, 1);
+
+                // >>> Does it need to wakeup periodically as well? In case we
+                // >>> miscalculated the #of kthreads to wakeup or create?
+        }
+}
+
+
+/*
+ * When XPC HB determines that a partition has come up, it will create a new
+ * kthread and that kthread will call this function to attempt to set up the
+ * basic infrastructure used for Cross Partition Communication with the newly
+ * upped partition.
+ *
+ * The kthread that was created by XPC HB and which setup the XPC
+ * infrastructure will remain assigned to the partition until the partition
+ * goes down. At which time the kthread will teardown the XPC infrastructure
+ * and then exit.
+ *
+ * XPC HB will put the remote partition's XPC per partition specific variables
+ * physical address into xpc_partitions[partid].remote_vars_part_pa prior to
+ * calling xpc_partition_up().
+ */
+static void
+xpc_partition_up(struct xpc_partition *part)
+{
+        DBUG_ON(part->channels != NULL);
+
+        dev_dbg(xpc_chan, "activating partition %d\n", XPC_PARTID(part));
+
+        if (xpc_setup_infrastructure(part) != xpcSuccess) {
+                return;
+        }
+
+        /*
+         * The kthread that XPC HB called us with will become the
+         * channel manager for this partition. It will not return
+         * back to XPC HB until the partition's XPC infrastructure
+         * has been dismantled.
+         */
+
+        (void) xpc_part_ref(part);      /* this will always succeed */
+
+        if (xpc_make_first_contact(part) == xpcSuccess) {
+                xpc_channel_mgr(part);
+        }
+
+        xpc_part_deref(part);
+
+        xpc_teardown_infrastructure(part);
+}
+
+
+static int
+xpc_activating(void *__partid)
+{
+        partid_t partid = (u64) __partid;
+        struct xpc_partition *part = &xpc_partitions[partid];
+        unsigned long irq_flags;
+        struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
+        int ret;
+
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+
+        spin_lock_irqsave(&part->act_lock, irq_flags);
+
+        if (part->act_state == XPC_P_DEACTIVATING) {
+                part->act_state = XPC_P_INACTIVE;
+                spin_unlock_irqrestore(&part->act_lock, irq_flags);
+                part->remote_rp_pa = 0;
+                return 0;
+        }
+
+        /* indicate the thread is activating */
+        DBUG_ON(part->act_state != XPC_P_ACTIVATION_REQ);
+        part->act_state = XPC_P_ACTIVATING;
+
+        XPC_SET_REASON(part, 0, 0);
+        spin_unlock_irqrestore(&part->act_lock, irq_flags);
+
+        dev_dbg(xpc_part, "bringing partition %d up\n", partid);
+
+        daemonize("xpc%02d", partid);
+
+        /*
+         * This thread needs to run at a realtime priority to prevent a
+         * significant performance degradation.
+         */
+        ret = sched_setscheduler(current, SCHED_FIFO, &param);
+        if (ret != 0) {
+                dev_warn(xpc_part, "unable to set pid %d to a realtime "
+                        "priority, ret=%d\n", current->pid, ret);
+        }
+
+        /* allow this thread and its children to run on any CPU */
+        set_cpus_allowed(current, CPU_MASK_ALL);
+
+        /*
+         * Register the remote partition's AMOs with SAL so it can handle
+         * and cleanup errors within that address range should the remote
+         * partition go down. We don't unregister this range because it is
+         * difficult to tell when outstanding writes to the remote partition
+         * are finished and thus when it is safe to unregister. This should
+         * not result in wasted space in the SAL xp_addr_region table because
+         * we should get the same page for remote_amos_page_pa after module
+         * reloads and system reboots.
+         */
+        if (sn_register_xp_addr_region(part->remote_amos_page_pa,
+                                                        PAGE_SIZE, 1) < 0) {
+                dev_warn(xpc_part, "xpc_partition_up(%d) failed to register "
+                        "xp_addr region\n", partid);
+
+                spin_lock_irqsave(&part->act_lock, irq_flags);
+                part->act_state = XPC_P_INACTIVE;
+                XPC_SET_REASON(part, xpcPhysAddrRegFailed, __LINE__);
+                spin_unlock_irqrestore(&part->act_lock, irq_flags);
+                part->remote_rp_pa = 0;
+                return 0;
+        }
+
+        xpc_allow_hb(partid, xpc_vars);
+        xpc_IPI_send_activated(part);
+
+
+        /*
+         * xpc_partition_up() holds this thread and marks this partition as
+         * XPC_P_ACTIVE by calling xpc_hb_mark_active().
+         */
+        (void) xpc_partition_up(part);
+
+        xpc_disallow_hb(partid, xpc_vars);
+        xpc_mark_partition_inactive(part);
+
+        if (part->reason == xpcReactivating) {
+                /* interrupting ourselves results in activating partition */
+                xpc_IPI_send_reactivate(part);
+        }
+
+        return 0;
+}
+
+
+void
+xpc_activate_partition(struct xpc_partition *part)
+{
+        partid_t partid = XPC_PARTID(part);
+        unsigned long irq_flags;
+        pid_t pid;
+
+
+        spin_lock_irqsave(&part->act_lock, irq_flags);
+
+        DBUG_ON(part->act_state != XPC_P_INACTIVE);
+
+        part->act_state = XPC_P_ACTIVATION_REQ;
+        XPC_SET_REASON(part, xpcCloneKThread, __LINE__);
+
+        spin_unlock_irqrestore(&part->act_lock, irq_flags);
+
+        pid = kernel_thread(xpc_activating, (void *) ((u64) partid), 0);
+
+        if (unlikely(pid <= 0)) {
+                spin_lock_irqsave(&part->act_lock, irq_flags);
+                part->act_state = XPC_P_INACTIVE;
+                XPC_SET_REASON(part, xpcCloneKThreadFailed, __LINE__);
+                spin_unlock_irqrestore(&part->act_lock, irq_flags);
+        }
+}
+
+
+/*
+ * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified
+ * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more
+ * than one partition, we use an AMO_t structure per partition to indicate
+ * whether a partition has sent an IPI or not.  >>> If it has, then wake up the
+ * associated kthread to handle it.
+ *
+ * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IPIs sent by XPC
+ * running on other partitions.
+ *
+ * Noteworthy Arguments:
+ *
+ *      irq - Interrupt ReQuest number. NOT USED.
+ *
+ *      dev_id - partid of IPI's potential sender.
+ */
+irqreturn_t
+xpc_notify_IRQ_handler(int irq, void *dev_id)
+{
+        partid_t partid = (partid_t) (u64) dev_id;
+        struct xpc_partition *part = &xpc_partitions[partid];
+
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+
+        if (xpc_part_ref(part)) {
+                xpc_check_for_channel_activity(part);
+
+                xpc_part_deref(part);
+        }
+        return IRQ_HANDLED;
+}
+
+
+/*
+ * Check to see if xpc_notify_IRQ_handler() dropped any IPIs on the floor
+ * because the write to their associated IPI amo completed after the IRQ/IPI
+ * was received.
+ */
+void
+xpc_dropped_IPI_check(struct xpc_partition *part)
+{
+        if (xpc_part_ref(part)) {
+                xpc_check_for_channel_activity(part);
+
+                part->dropped_IPI_timer.expires = jiffies +
+                                                        XPC_P_DROPPED_IPI_WAIT;
+                add_timer(&part->dropped_IPI_timer);
+                xpc_part_deref(part);
+        }
+}
+
+
+void
+xpc_activate_kthreads(struct xpc_channel *ch, int needed)
+{
+        int idle = atomic_read(&ch->kthreads_idle);
+        int assigned = atomic_read(&ch->kthreads_assigned);
+        int wakeup;
+
+
+        DBUG_ON(needed <= 0);
+
+        if (idle > 0) {
+                wakeup = (needed > idle) ? idle : needed;
+                needed -= wakeup;
+
+                dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, "
+                        "channel=%d\n", wakeup, ch->partid, ch->number);
+
+                /* only wakeup the requested number of kthreads */
+                wake_up_nr(&ch->idle_wq, wakeup);
+        }
+
+        if (needed <= 0) {
+                return;
+        }
+
+        if (needed + assigned > ch->kthreads_assigned_limit) {
+                needed = ch->kthreads_assigned_limit - assigned;
+                // >>>should never be less than 0
+                if (needed <= 0) {
+                        return;
+                }
+        }
+
+        dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n",
+                needed, ch->partid, ch->number);
+
+        xpc_create_kthreads(ch, needed, 0);
+}
+
+
+/*
+ * This function is where XPC's kthreads wait for messages to deliver.
+ */
+static void
+xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
+{
+        do {
+                /* deliver messages to their intended recipients */
+
+                while ((volatile s64) ch->w_local_GP.get <
+                                (volatile s64) ch->w_remote_GP.put &&
+                                        !((volatile u32) ch->flags &
+                                                XPC_C_DISCONNECTING)) {
+                        xpc_deliver_msg(ch);
+                }
+
+                if (atomic_inc_return(&ch->kthreads_idle) >
+                                                ch->kthreads_idle_limit) {
+                        /* too many idle kthreads on this channel */
+                        atomic_dec(&ch->kthreads_idle);
+                        break;
+                }
+
+                dev_dbg(xpc_chan, "idle kthread calling "
+                        "wait_event_interruptible_exclusive()\n");
+
+                (void) wait_event_interruptible_exclusive(ch->idle_wq,
+                                ((volatile s64) ch->w_local_GP.get <
+                                        (volatile s64) ch->w_remote_GP.put ||
+                                ((volatile u32) ch->flags &
+                                                XPC_C_DISCONNECTING)));
+
+                atomic_dec(&ch->kthreads_idle);
+
+        } while (!((volatile u32) ch->flags & XPC_C_DISCONNECTING));
+}
+
+
+static int
+xpc_daemonize_kthread(void *args)
+{
+        partid_t partid = XPC_UNPACK_ARG1(args);
+        u16 ch_number = XPC_UNPACK_ARG2(args);
+        struct xpc_partition *part = &xpc_partitions[partid];
+        struct xpc_channel *ch;
+        int n_needed;
+        unsigned long irq_flags;
+
+
+        daemonize("xpc%02dc%d", partid, ch_number);
+
+        dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n",
+                partid, ch_number);
+
+        ch = &part->channels[ch_number];
+
+        if (!(ch->flags & XPC_C_DISCONNECTING)) {
+
+                /* let registerer know that connection has been established */
+
+                spin_lock_irqsave(&ch->lock, irq_flags);
+                if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) {
+                        ch->flags |= XPC_C_CONNECTEDCALLOUT;
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+                        xpc_connected_callout(ch);
+
+                        spin_lock_irqsave(&ch->lock, irq_flags);
+                        ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE;
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+                        /*
+                         * It is possible that while the callout was being
+                         * made that the remote partition sent some messages.
+                         * If that is the case, we may need to activate
+                         * additional kthreads to help deliver them. We only
+                         * need one less than total #of messages to deliver.
+                         */
+                        n_needed = ch->w_remote_GP.put - ch->w_local_GP.get - 1;
+                        if (n_needed > 0 &&
+                                        !(ch->flags & XPC_C_DISCONNECTING)) {
+                                xpc_activate_kthreads(ch, n_needed);
+                        }
+                } else {
+                        spin_unlock_irqrestore(&ch->lock, irq_flags);
+                }
+
+                xpc_kthread_waitmsgs(part, ch);
+        }
+
+        /* let registerer know that connection is disconnecting */
+
+        spin_lock_irqsave(&ch->lock, irq_flags);
+        if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
+                        !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) {
+                ch->flags |= XPC_C_DISCONNECTINGCALLOUT;
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+                xpc_disconnect_callout(ch, xpcDisconnecting);
+
+                spin_lock_irqsave(&ch->lock, irq_flags);
+                ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE;
+        }
+        spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+        if (atomic_dec_return(&ch->kthreads_assigned) == 0) {
+                if (atomic_dec_return(&part->nchannels_engaged) == 0) {
+                        xpc_mark_partition_disengaged(part);
+                        xpc_IPI_send_disengage(part);
+                }
+        }
+
+        xpc_msgqueue_deref(ch);
+
+        dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n",
+                partid, ch_number);
+
+        xpc_part_deref(part);
+        return 0;
+}
+
+
+/*
+ * For each partition that XPC has established communications with, there is
+ * a minimum of one kernel thread assigned to perform any operation that
+ * may potentially sleep or block (basically the callouts to the asynchronous
+ * functions registered via xpc_connect()).
+ *
+ * Additional kthreads are created and destroyed by XPC as the workload
+ * demands.
+ *
+ * A kthread is assigned to one of the active channels that exists for a given
+ * partition.
+ */
+void
+xpc_create_kthreads(struct xpc_channel *ch, int needed,
+                        int ignore_disconnecting)
+{
+        unsigned long irq_flags;
+        pid_t pid;
+        u64 args = XPC_PACK_ARGS(ch->partid, ch->number);
+        struct xpc_partition *part = &xpc_partitions[ch->partid];
+
+
+        while (needed-- > 0) {
+
+                /*
+                 * The following is done on behalf of the newly created
+                 * kthread. That kthread is responsible for doing the
+                 * counterpart to the following before it exits.
+                 */
+                if (ignore_disconnecting) {
+                        if (!atomic_inc_not_zero(&ch->kthreads_assigned)) {
+                                /* kthreads assigned had gone to zero */
+                                BUG_ON(!(ch->flags &
+                                        XPC_C_DISCONNECTINGCALLOUT_MADE));
+                                break;
+                        }
+
+                } else if (ch->flags & XPC_C_DISCONNECTING) {
+                        break;
+
+                } else if (atomic_inc_return(&ch->kthreads_assigned) == 1) {
+                        if (atomic_inc_return(&part->nchannels_engaged) == 1)
+                                xpc_mark_partition_engaged(part);
+                }
+                (void) xpc_part_ref(part);
+                xpc_msgqueue_ref(ch);
+
+                pid = kernel_thread(xpc_daemonize_kthread, (void *) args, 0);
+                if (pid < 0) {
+                        /* the fork failed */
+
+                        /*
+                         * NOTE: if (ignore_disconnecting &&
+                         * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true,
+                         * then we'll deadlock if all other kthreads assigned
+                         * to this channel are blocked in the channel's
+                         * registerer, because the only thing that will unblock
+                         * them is the xpcDisconnecting callout that this
+                         * failed kernel_thread would have made.
+                         */
+
+                        if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
+                            atomic_dec_return(&part->nchannels_engaged) == 0) {
+                                xpc_mark_partition_disengaged(part);
+                                xpc_IPI_send_disengage(part);
+                        }
+                        xpc_msgqueue_deref(ch);
+                        xpc_part_deref(part);
+
+                        if (atomic_read(&ch->kthreads_assigned) <
+                                                ch->kthreads_idle_limit) {
+                                /*
+                                 * Flag this as an error only if we have an
+                                 * insufficient #of kthreads for the channel
+                                 * to function.
+                                 */
+                                spin_lock_irqsave(&ch->lock, irq_flags);
+                                XPC_DISCONNECT_CHANNEL(ch, xpcLackOfResources,
+                                                                &irq_flags);
+                                spin_unlock_irqrestore(&ch->lock, irq_flags);
+                        }
+                        break;
+                }
+
+                ch->kthreads_created++; // >>> temporary debug only!!!
+        }
+}
+
+
+void
+xpc_disconnect_wait(int ch_number)
+{
+        unsigned long irq_flags;
+        partid_t partid;
+        struct xpc_partition *part;
+        struct xpc_channel *ch;
+        int wakeup_channel_mgr;
+
+
+        /* now wait for all callouts to the caller's function to cease */
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                part = &xpc_partitions[partid];
+
+                if (!xpc_part_ref(part)) {
+                        continue;
+                }
+
+                ch = &part->channels[ch_number];
+
+                if (!(ch->flags & XPC_C_WDISCONNECT)) {
+                        xpc_part_deref(part);
+                        continue;
+                }
+
+                wait_for_completion(&ch->wdisconnect_wait);
+
+                spin_lock_irqsave(&ch->lock, irq_flags);
+                DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
+                wakeup_channel_mgr = 0;
+
+                if (ch->delayed_IPI_flags) {
+                        if (part->act_state != XPC_P_DEACTIVATING) {
+                                spin_lock(&part->IPI_lock);
+                                XPC_SET_IPI_FLAGS(part->local_IPI_amo,
+                                        ch->number, ch->delayed_IPI_flags);
+                                spin_unlock(&part->IPI_lock);
+                                wakeup_channel_mgr = 1;
+                        }
+                        ch->delayed_IPI_flags = 0;
+                }
+
+                ch->flags &= ~XPC_C_WDISCONNECT;
+                spin_unlock_irqrestore(&ch->lock, irq_flags);
+
+                if (wakeup_channel_mgr) {
+                        xpc_wakeup_channel_mgr(part);
+                }
+
+                xpc_part_deref(part);
+        }
+}
+
+
+static void
+xpc_do_exit(enum xpc_retval reason)
+{
+        partid_t partid;
+        int active_part_count, printed_waiting_msg = 0;
+        struct xpc_partition *part;
+        unsigned long printmsg_time, disengage_request_timeout = 0;
+
+
+        /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
+        DBUG_ON(xpc_exiting == 1);
+
+        /*
+         * Let the heartbeat checker thread and the discovery thread
+         * (if one is running) know that they should exit. Also wake up
+         * the heartbeat checker thread in case it's sleeping.
+         */
+        xpc_exiting = 1;
+        wake_up_interruptible(&xpc_act_IRQ_wq);
+
+        /* ignore all incoming interrupts */
+        free_irq(SGI_XPC_ACTIVATE, NULL);
+
+        /* wait for the discovery thread to exit */
+        wait_for_completion(&xpc_discovery_exited);
+
+        /* wait for the heartbeat checker thread to exit */
+        wait_for_completion(&xpc_hb_checker_exited);
+
+
+        /* sleep for a 1/3 of a second or so */
+        (void) msleep_interruptible(300);
+
+
+        /* wait for all partitions to become inactive */
+
+        printmsg_time = jiffies + (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ);
+        xpc_disengage_request_timedout = 0;
+
+        do {
+                active_part_count = 0;
+
+                for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                        part = &xpc_partitions[partid];
+
+                        if (xpc_partition_disengaged(part) &&
+                                        part->act_state == XPC_P_INACTIVE) {
+                                continue;
+                        }
+
+                        active_part_count++;
+
+                        XPC_DEACTIVATE_PARTITION(part, reason);
+
+                        if (part->disengage_request_timeout >
+                                                disengage_request_timeout) {
+                                disengage_request_timeout =
+                                                part->disengage_request_timeout;
+                        }
+                }
+
+                if (xpc_partition_engaged(-1UL)) {
+                        if (time_after(jiffies, printmsg_time)) {
+                                dev_info(xpc_part, "waiting for remote "
+                                        "partitions to disengage, timeout in "
+                                        "%ld seconds\n",
+                                        (disengage_request_timeout - jiffies)
+                                                                        / HZ);
+                                printmsg_time = jiffies +
+                                        (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ);
+                                printed_waiting_msg = 1;
+                        }
+
+                } else if (active_part_count > 0) {
+                        if (printed_waiting_msg) {
+                                dev_info(xpc_part, "waiting for local partition"
+                                        " to disengage\n");
+                                printed_waiting_msg = 0;
+                        }
+
+                } else {
+                        if (!xpc_disengage_request_timedout) {
+                                dev_info(xpc_part, "all partitions have "
+                                        "disengaged\n");
+                        }
+                        break;
+                }
+
+                /* sleep for a 1/3 of a second or so */
+                (void) msleep_interruptible(300);
+
+        } while (1);
+
+        DBUG_ON(xpc_partition_engaged(-1UL));
+
+
+        /* indicate to others that our reserved page is uninitialized */
+        xpc_rsvd_page->vars_pa = 0;
+
+        /* now it's time to eliminate our heartbeat */
+        del_timer_sync(&xpc_hb_timer);
+        DBUG_ON(xpc_vars->heartbeating_to_mask != 0);
+
+        if (reason == xpcUnloading) {
+                /* take ourselves off of the reboot_notifier_list */
+                (void) unregister_reboot_notifier(&xpc_reboot_notifier);
+
+                /* take ourselves off of the die_notifier list */
+                (void) unregister_die_notifier(&xpc_die_notifier);
+        }
+
+        /* close down protections for IPI operations */
+        xpc_restrict_IPI_ops();
+
+
+        /* clear the interface to XPC's functions */
+        xpc_clear_interface();
+
+        if (xpc_sysctl) {
+                unregister_sysctl_table(xpc_sysctl);
+        }
+
+        kfree(xpc_remote_copy_buffer_base);
+}
+
+
+/*
+ * This function is called when the system is being rebooted.
+ */
+static int
+xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
+{
+        enum xpc_retval reason;
+
+
+        switch (event) {
+        case SYS_RESTART:
+                reason = xpcSystemReboot;
+                break;
+        case SYS_HALT:
+                reason = xpcSystemHalt;
+                break;
+        case SYS_POWER_OFF:
+                reason = xpcSystemPoweroff;
+                break;
+        default:
+                reason = xpcSystemGoingDown;
+        }
+
+        xpc_do_exit(reason);
+        return NOTIFY_DONE;
+}
+
+
+/*
+ * Notify other partitions to disengage from all references to our memory.
+ */
+static void
+xpc_die_disengage(void)
+{
+        struct xpc_partition *part;
+        partid_t partid;
+        unsigned long engaged;
+        long time, printmsg_time, disengage_request_timeout;
+
+
+        /* keep xpc_hb_checker thread from doing anything (just in case) */
+        xpc_exiting = 1;
+
+        xpc_vars->heartbeating_to_mask = 0;  /* indicate we're deactivated */
+
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                part = &xpc_partitions[partid];
+
+                if (!XPC_SUPPORTS_DISENGAGE_REQUEST(part->
+                                                        remote_vars_version)) {
+
+                        /* just in case it was left set by an earlier XPC */
+                        xpc_clear_partition_engaged(1UL << partid);
+                        continue;
+                }
+
+                if (xpc_partition_engaged(1UL << partid) ||
+                                        part->act_state != XPC_P_INACTIVE) {
+                        xpc_request_partition_disengage(part);
+                        xpc_mark_partition_disengaged(part);
+                        xpc_IPI_send_disengage(part);
+                }
+        }
+
+        time = rtc_time();
+        printmsg_time = time +
+                (XPC_DISENGAGE_PRINTMSG_INTERVAL * sn_rtc_cycles_per_second);
+        disengage_request_timeout = time +
+                (xpc_disengage_request_timelimit * sn_rtc_cycles_per_second);
+
+        /* wait for all other partitions to disengage from us */
+
+        while (1) {
+                engaged = xpc_partition_engaged(-1UL);
+                if (!engaged) {
+                        dev_info(xpc_part, "all partitions have disengaged\n");
+                        break;
+                }
+
+                time = rtc_time();
+                if (time >= disengage_request_timeout) {
+                        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                                if (engaged & (1UL << partid)) {
+                                        dev_info(xpc_part, "disengage from "
+                                                "remote partition %d timed "
+                                                "out\n", partid);
+                                }
+                        }
+                        break;
+                }
+
+                if (time >= printmsg_time) {
+                        dev_info(xpc_part, "waiting for remote partitions to "
+                                "disengage, timeout in %ld seconds\n",
+                                (disengage_request_timeout - time) /
+                                                sn_rtc_cycles_per_second);
+                        printmsg_time = time +
+                                        (XPC_DISENGAGE_PRINTMSG_INTERVAL *
+                                                sn_rtc_cycles_per_second);
+                }
+        }
+}
+
+
+/*
+ * This function is called when the system is being restarted or halted due
+ * to some sort of system failure. If this is the case we need to notify the
+ * other partitions to disengage from all references to our memory.
+ * This function can also be called when our heartbeater could be offlined
+ * for a time. In this case we need to notify other partitions to not worry
+ * about the lack of a heartbeat.
+ */
+static int
+xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused)
+{
+        switch (event) {
+        case DIE_MACHINE_RESTART:
+        case DIE_MACHINE_HALT:
+                xpc_die_disengage();
+                break;
+
+        case DIE_KDEBUG_ENTER:
+                /* Should lack of heartbeat be ignored by other partitions? */
+                if (!xpc_kdebug_ignore) {
+                        break;
+                }
+                /* fall through */
+        case DIE_MCA_MONARCH_ENTER:
+        case DIE_INIT_MONARCH_ENTER:
+                xpc_vars->heartbeat++;
+                xpc_vars->heartbeat_offline = 1;
+                break;
+
+        case DIE_KDEBUG_LEAVE:
+                /* Is lack of heartbeat being ignored by other partitions? */
+                if (!xpc_kdebug_ignore) {
+                        break;
+                }
+                /* fall through */
+        case DIE_MCA_MONARCH_LEAVE:
+        case DIE_INIT_MONARCH_LEAVE:
+                xpc_vars->heartbeat++;
+                xpc_vars->heartbeat_offline = 0;
+                break;
+        }
+
+        return NOTIFY_DONE;
+}
+
+
+int __init
+xpc_init(void)
+{
+        int ret;
+        partid_t partid;
+        struct xpc_partition *part;
+        pid_t pid;
+        size_t buf_size;
+
+
+        if (!ia64_platform_is("sn2")) {
+                return -ENODEV;
+        }
+
+
+        buf_size = max(XPC_RP_VARS_SIZE,
+                                XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES);
+        xpc_remote_copy_buffer = xpc_kmalloc_cacheline_aligned(buf_size,
+                                     GFP_KERNEL, &xpc_remote_copy_buffer_base);
+        if (xpc_remote_copy_buffer == NULL)
+                return -ENOMEM;
+
+        snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part");
+        snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan");
+
+        xpc_sysctl = register_sysctl_table(xpc_sys_dir);
+
+        /*
+         * The first few fields of each entry of xpc_partitions[] need to
+         * be initialized now so that calls to xpc_connect() and
+         * xpc_disconnect() can be made prior to the activation of any remote
+         * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE
+         * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING
+         * PARTITION HAS BEEN ACTIVATED.
+         */
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                part = &xpc_partitions[partid];
+
+                DBUG_ON((u64) part != L1_CACHE_ALIGN((u64) part));
+
+                part->act_IRQ_rcvd = 0;
+                spin_lock_init(&part->act_lock);
+                part->act_state = XPC_P_INACTIVE;
+                XPC_SET_REASON(part, 0, 0);
+
+                init_timer(&part->disengage_request_timer);
+                part->disengage_request_timer.function =
+                                xpc_timeout_partition_disengage_request;
+                part->disengage_request_timer.data = (unsigned long) part;
+
+                part->setup_state = XPC_P_UNSET;
+                init_waitqueue_head(&part->teardown_wq);
+                atomic_set(&part->references, 0);
+        }
+
+        /*
+         * Open up protections for IPI operations (and AMO operations on
+         * Shub 1.1 systems).
+         */
+        xpc_allow_IPI_ops();
+
+        /*
+         * Interrupts being processed will increment this atomic variable and
+         * awaken the heartbeat thread which will process the interrupts.
+         */
+        atomic_set(&xpc_act_IRQ_rcvd, 0);
+
+        /*
+         * This is safe to do before the xpc_hb_checker thread has started
+         * because the handler releases a wait queue.  If an interrupt is
+         * received before the thread is waiting, it will not go to sleep,
+         * but rather immediately process the interrupt.
+         */
+        ret = request_irq(SGI_XPC_ACTIVATE, xpc_act_IRQ_handler, 0,
+                                                        "xpc hb", NULL);
+        if (ret != 0) {
+                dev_err(xpc_part, "can't register ACTIVATE IRQ handler, "
+                        "errno=%d\n", -ret);
+
+                xpc_restrict_IPI_ops();
+
+                if (xpc_sysctl) {
+                        unregister_sysctl_table(xpc_sysctl);
+                }
+
+                kfree(xpc_remote_copy_buffer_base);
+                return -EBUSY;
+        }
+
+        /*
+         * Fill the partition reserved page with the information needed by
+         * other partitions to discover we are alive and establish initial
+         * communications.
+         */
+        xpc_rsvd_page = xpc_rsvd_page_init();
+        if (xpc_rsvd_page == NULL) {
+                dev_err(xpc_part, "could not setup our reserved page\n");
+
+                free_irq(SGI_XPC_ACTIVATE, NULL);
+                xpc_restrict_IPI_ops();
+
+                if (xpc_sysctl) {
+                        unregister_sysctl_table(xpc_sysctl);
+                }
+
+                kfree(xpc_remote_copy_buffer_base);
+                return -EBUSY;
+        }
+
+
+        /* add ourselves to the reboot_notifier_list */
+        ret = register_reboot_notifier(&xpc_reboot_notifier);
+        if (ret != 0) {
+                dev_warn(xpc_part, "can't register reboot notifier\n");
+        }
+
+        /* add ourselves to the die_notifier list */
+        ret = register_die_notifier(&xpc_die_notifier);
+        if (ret != 0) {
+                dev_warn(xpc_part, "can't register die notifier\n");
+        }
+
+        init_timer(&xpc_hb_timer);
+        xpc_hb_timer.function = xpc_hb_beater;
+
+        /*
+         * The real work-horse behind xpc.  This processes incoming
+         * interrupts and monitors remote heartbeats.
+         */
+        pid = kernel_thread(xpc_hb_checker, NULL, 0);
+        if (pid < 0) {
+                dev_err(xpc_part, "failed while forking hb check thread\n");
+
+                /* indicate to others that our reserved page is uninitialized */
+                xpc_rsvd_page->vars_pa = 0;
+
+                /* take ourselves off of the reboot_notifier_list */
+                (void) unregister_reboot_notifier(&xpc_reboot_notifier);
+
+                /* take ourselves off of the die_notifier list */
+                (void) unregister_die_notifier(&xpc_die_notifier);
+
+                del_timer_sync(&xpc_hb_timer);
+                free_irq(SGI_XPC_ACTIVATE, NULL);
+                xpc_restrict_IPI_ops();
+
+                if (xpc_sysctl) {
+                        unregister_sysctl_table(xpc_sysctl);
+                }
+
+                kfree(xpc_remote_copy_buffer_base);
+                return -EBUSY;
+        }
+
+
+        /*
+         * Startup a thread that will attempt to discover other partitions to
+         * activate based on info provided by SAL. This new thread is short
+         * lived and will exit once discovery is complete.
+         */
+        pid = kernel_thread(xpc_initiate_discovery, NULL, 0);
+        if (pid < 0) {
+                dev_err(xpc_part, "failed while forking discovery thread\n");
+
+                /* mark this new thread as a non-starter */
+                complete(&xpc_discovery_exited);
+
+                xpc_do_exit(xpcUnloading);
+                return -EBUSY;
+        }
+
+
+        /* set the interface to point at XPC's functions */
+        xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect,
+                          xpc_initiate_allocate, xpc_initiate_send,
+                          xpc_initiate_send_notify, xpc_initiate_received,
+                          xpc_initiate_partid_to_nasids);
+
+        return 0;
+}
+module_init(xpc_init);
+
+
+void __exit
+xpc_exit(void)
+{
+        xpc_do_exit(xpcUnloading);
+}
+module_exit(xpc_exit);
+
+
+MODULE_AUTHOR("Silicon Graphics, Inc.");
+MODULE_DESCRIPTION("Cross Partition Communication (XPC) support");
+MODULE_LICENSE("GPL");
+
+module_param(xpc_hb_interval, int, 0);
+MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between "
+                "heartbeat increments.");
+
+module_param(xpc_hb_check_interval, int, 0);
+MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between "
+                "heartbeat checks.");
+
+module_param(xpc_disengage_request_timelimit, int, 0);
+MODULE_PARM_DESC(xpc_disengage_request_timelimit, "Number of seconds to wait "
+                "for disengage request to complete.");
+
+module_param(xpc_kdebug_ignore, int, 0);
+MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by "
+                "other partitions when dropping into kdebug.");
+
diff --git a/drivers/misc/sgi-xp/xpc_partition.c b/drivers/misc/sgi-xp/xpc_partition.c
new file mode 100644
index 000000000000..7412dc7351cd
--- /dev/null
+++ b/drivers/misc/sgi-xp/xpc_partition.c
@@ -0,0 +1,1239 @@
+/*
+ * 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) 2004-2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+
+/*
+ * Cross Partition Communication (XPC) partition support.
+ *
+ *      This is the part of XPC that detects the presence/absence of
+ *      other partitions. It provides a heartbeat and monitors the
+ *      heartbeats of other partitions.
+ *
+ */
+
+
+#include <linux/kernel.h>
+#include <linux/sysctl.h>
+#include <linux/cache.h>
+#include <linux/mmzone.h>
+#include <linux/nodemask.h>
+#include <asm/uncached.h>
+#include <asm/sn/bte.h>
+#include <asm/sn/intr.h>
+#include <asm/sn/sn_sal.h>
+#include <asm/sn/nodepda.h>
+#include <asm/sn/addrs.h>
+#include "xpc.h"
+
+
+/* XPC is exiting flag */
+int xpc_exiting;
+
+
+/* SH_IPI_ACCESS shub register value on startup */
+static u64 xpc_sh1_IPI_access;
+static u64 xpc_sh2_IPI_access0;
+static u64 xpc_sh2_IPI_access1;
+static u64 xpc_sh2_IPI_access2;
+static u64 xpc_sh2_IPI_access3;
+
+
+/* original protection values for each node */
+u64 xpc_prot_vec[MAX_NUMNODES];
+
+
+/* this partition's reserved page pointers */
+struct xpc_rsvd_page *xpc_rsvd_page;
+static u64 *xpc_part_nasids;
+static u64 *xpc_mach_nasids;
+struct xpc_vars *xpc_vars;
+struct xpc_vars_part *xpc_vars_part;
+
+static int xp_nasid_mask_bytes; /* actual size in bytes of nasid mask */
+static int xp_nasid_mask_words; /* actual size in words of nasid mask */
+
+
+/*
+ * For performance reasons, each entry of xpc_partitions[] is cacheline
+ * aligned. And xpc_partitions[] is padded with an additional entry at the
+ * end so that the last legitimate entry doesn't share its cacheline with
+ * another variable.
+ */
+struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
+
+
+/*
+ * Generic buffer used to store a local copy of portions of a remote
+ * partition's reserved page (either its header and part_nasids mask,
+ * or its vars).
+ */
+char *xpc_remote_copy_buffer;
+void *xpc_remote_copy_buffer_base;
+
+
+/*
+ * Guarantee that the kmalloc'd memory is cacheline aligned.
+ */
+void *
+xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
+{
+        /* see if kmalloc will give us cachline aligned memory by default */
+        *base = kmalloc(size, flags);
+        if (*base == NULL) {
+                return NULL;
+        }
+        if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
+                return *base;
+        }
+        kfree(*base);
+
+        /* nope, we'll have to do it ourselves */
+        *base = kmalloc(size + L1_CACHE_BYTES, flags);
+        if (*base == NULL) {
+                return NULL;
+        }
+        return (void *) L1_CACHE_ALIGN((u64) *base);
+}
+
+
+/*
+ * Given a nasid, get the physical address of the  partition's reserved page
+ * for that nasid. This function returns 0 on any error.
+ */
+static u64
+xpc_get_rsvd_page_pa(int nasid)
+{
+        bte_result_t bte_res;
+        s64 status;
+        u64 cookie = 0;
+        u64 rp_pa = nasid;      /* seed with nasid */
+        u64 len = 0;
+        u64 buf = buf;
+        u64 buf_len = 0;
+        void *buf_base = NULL;
+
+
+        while (1) {
+
+                status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
+                                                                &len);
+
+                dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
+                        "0x%016lx, address=0x%016lx, len=0x%016lx\n",
+                        status, cookie, rp_pa, len);
+
+                if (status != SALRET_MORE_PASSES) {
+                        break;
+                }
+
+                if (L1_CACHE_ALIGN(len) > buf_len) {
+                        kfree(buf_base);
+                        buf_len = L1_CACHE_ALIGN(len);
+                        buf = (u64) xpc_kmalloc_cacheline_aligned(buf_len,
+                                                        GFP_KERNEL, &buf_base);
+                        if (buf_base == NULL) {
+                                dev_err(xpc_part, "unable to kmalloc "
+                                        "len=0x%016lx\n", buf_len);
+                                status = SALRET_ERROR;
+                                break;
+                        }
+                }
+
+                bte_res = xp_bte_copy(rp_pa, buf, buf_len,
+                                        (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+                if (bte_res != BTE_SUCCESS) {
+                        dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
+                        status = SALRET_ERROR;
+                        break;
+                }
+        }
+
+        kfree(buf_base);
+
+        if (status != SALRET_OK) {
+                rp_pa = 0;
+        }
+        dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
+        return rp_pa;
+}
+
+
+/*
+ * Fill the partition reserved page with the information needed by
+ * other partitions to discover we are alive and establish initial
+ * communications.
+ */
+struct xpc_rsvd_page *
+xpc_rsvd_page_init(void)
+{
+        struct xpc_rsvd_page *rp;
+        AMO_t *amos_page;
+        u64 rp_pa, nasid_array = 0;
+        int i, ret;
+
+
+        /* get the local reserved page's address */
+
+        preempt_disable();
+        rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
+        preempt_enable();
+        if (rp_pa == 0) {
+                dev_err(xpc_part, "SAL failed to locate the reserved page\n");
+                return NULL;
+        }
+        rp = (struct xpc_rsvd_page *) __va(rp_pa);
+
+        if (rp->partid != sn_partition_id) {
+                dev_err(xpc_part, "the reserved page's partid of %d should be "
+                        "%d\n", rp->partid, sn_partition_id);
+                return NULL;
+        }
+
+        rp->version = XPC_RP_VERSION;
+
+        /* establish the actual sizes of the nasid masks */
+        if (rp->SAL_version == 1) {
+                /* SAL_version 1 didn't set the nasids_size field */
+                rp->nasids_size = 128;
+        }
+        xp_nasid_mask_bytes = rp->nasids_size;
+        xp_nasid_mask_words = xp_nasid_mask_bytes / 8;
+
+        /* setup the pointers to the various items in the reserved page */
+        xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
+        xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
+        xpc_vars = XPC_RP_VARS(rp);
+        xpc_vars_part = XPC_RP_VARS_PART(rp);
+
+        /*
+         * Before clearing xpc_vars, see if a page of AMOs had been previously
+         * allocated. If not we'll need to allocate one and set permissions
+         * so that cross-partition AMOs are allowed.
+         *
+         * The allocated AMO page needs MCA reporting to remain disabled after
+         * XPC has unloaded.  To make this work, we keep a copy of the pointer
+         * to this page (i.e., amos_page) in the struct xpc_vars structure,
+         * which is pointed to by the reserved page, and re-use that saved copy
+         * on subsequent loads of XPC. This AMO page is never freed, and its
+         * memory protections are never restricted.
+         */
+        if ((amos_page = xpc_vars->amos_page) == NULL) {
+                amos_page = (AMO_t *) TO_AMO(uncached_alloc_page(0));
+                if (amos_page == NULL) {
+                        dev_err(xpc_part, "can't allocate page of AMOs\n");
+                        return NULL;
+                }
+
+                /*
+                 * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
+                 * when xpc_allow_IPI_ops() is called via xpc_hb_init().
+                 */
+                if (!enable_shub_wars_1_1()) {
+                        ret = sn_change_memprotect(ia64_tpa((u64) amos_page),
+                                        PAGE_SIZE, SN_MEMPROT_ACCESS_CLASS_1,
+                                        &nasid_array);
+                        if (ret != 0) {
+                                dev_err(xpc_part, "can't change memory "
+                                        "protections\n");
+                                uncached_free_page(__IA64_UNCACHED_OFFSET |
+                                                   TO_PHYS((u64) amos_page));
+                                return NULL;
+                        }
+                }
+        } else if (!IS_AMO_ADDRESS((u64) amos_page)) {
+                /*
+                 * EFI's XPBOOT can also set amos_page in the reserved page,
+                 * but it happens to leave it as an uncached physical address
+                 * and we need it to be an uncached virtual, so we'll have to
+                 * convert it.
+                 */
+                if (!IS_AMO_PHYS_ADDRESS((u64) amos_page)) {
+                        dev_err(xpc_part, "previously used amos_page address "
+                                "is bad = 0x%p\n", (void *) amos_page);
+                        return NULL;
+                }
+                amos_page = (AMO_t *) TO_AMO((u64) amos_page);
+        }
+
+        /* clear xpc_vars */
+        memset(xpc_vars, 0, sizeof(struct xpc_vars));
+
+        xpc_vars->version = XPC_V_VERSION;
+        xpc_vars->act_nasid = cpuid_to_nasid(0);
+        xpc_vars->act_phys_cpuid = cpu_physical_id(0);
+        xpc_vars->vars_part_pa = __pa(xpc_vars_part);
+        xpc_vars->amos_page_pa = ia64_tpa((u64) amos_page);
+        xpc_vars->amos_page = amos_page;  /* save for next load of XPC */
+
+
+        /* clear xpc_vars_part */
+        memset((u64 *) xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
+                                                        XP_MAX_PARTITIONS);
+
+        /* initialize the activate IRQ related AMO variables */
+        for (i = 0; i < xp_nasid_mask_words; i++) {
+                (void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
+        }
+
+        /* initialize the engaged remote partitions related AMO variables */
+        (void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
+        (void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
+
+        /* timestamp of when reserved page was setup by XPC */
+        rp->stamp = CURRENT_TIME;
+
+        /*
+         * This signifies to the remote partition that our reserved
+         * page is initialized.
+         */
+        rp->vars_pa = __pa(xpc_vars);
+
+        return rp;
+}
+
+
+/*
+ * Change protections to allow IPI operations (and AMO operations on
+ * Shub 1.1 systems).
+ */
+void
+xpc_allow_IPI_ops(void)
+{
+        int node;
+        int nasid;
+
+
+        // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
+
+        if (is_shub2()) {
+                xpc_sh2_IPI_access0 =
+                        (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
+                xpc_sh2_IPI_access1 =
+                        (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
+                xpc_sh2_IPI_access2 =
+                        (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
+                xpc_sh2_IPI_access3 =
+                        (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
+
+                for_each_online_node(node) {
+                        nasid = cnodeid_to_nasid(node);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
+                                                                -1UL);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
+                                                                -1UL);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
+                                                                -1UL);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
+                                                                -1UL);
+                }
+
+        } else {
+                xpc_sh1_IPI_access =
+                        (u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
+
+                for_each_online_node(node) {
+                        nasid = cnodeid_to_nasid(node);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
+                                                                -1UL);
+
+                        /*
+                         * Since the BIST collides with memory operations on
+                         * SHUB 1.1 sn_change_memprotect() cannot be used.
+                         */
+                        if (enable_shub_wars_1_1()) {
+                                /* open up everything */
+                                xpc_prot_vec[node] = (u64) HUB_L((u64 *)
+                                                GLOBAL_MMR_ADDR(nasid,
+                                                SH1_MD_DQLP_MMR_DIR_PRIVEC0));
+                                HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
+                                                SH1_MD_DQLP_MMR_DIR_PRIVEC0),
+                                                                -1UL);
+                                HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
+                                                SH1_MD_DQRP_MMR_DIR_PRIVEC0),
+                                                                -1UL);
+                        }
+                }
+        }
+}
+
+
+/*
+ * Restrict protections to disallow IPI operations (and AMO operations on
+ * Shub 1.1 systems).
+ */
+void
+xpc_restrict_IPI_ops(void)
+{
+        int node;
+        int nasid;
+
+
+        // >>> Change SH_IPI_ACCESS code to use SAL call once it is available.
+
+        if (is_shub2()) {
+
+                for_each_online_node(node) {
+                        nasid = cnodeid_to_nasid(node);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
+                                                        xpc_sh2_IPI_access0);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
+                                                        xpc_sh2_IPI_access1);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
+                                                        xpc_sh2_IPI_access2);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
+                                                        xpc_sh2_IPI_access3);
+                }
+
+        } else {
+
+                for_each_online_node(node) {
+                        nasid = cnodeid_to_nasid(node);
+                        HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
+                                                        xpc_sh1_IPI_access);
+
+                        if (enable_shub_wars_1_1()) {
+                                HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
+                                                SH1_MD_DQLP_MMR_DIR_PRIVEC0),
+                                                        xpc_prot_vec[node]);
+                                HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
+                                                SH1_MD_DQRP_MMR_DIR_PRIVEC0),
+                                                        xpc_prot_vec[node]);
+                        }
+                }
+        }
+}
+
+
+/*
+ * At periodic intervals, scan through all active partitions and ensure
+ * their heartbeat is still active.  If not, the partition is deactivated.
+ */
+void
+xpc_check_remote_hb(void)
+{
+        struct xpc_vars *remote_vars;
+        struct xpc_partition *part;
+        partid_t partid;
+        bte_result_t bres;
+
+
+        remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
+
+        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+
+                if (xpc_exiting) {
+                        break;
+                }
+
+                if (partid == sn_partition_id) {
+                        continue;
+                }
+
+                part = &xpc_partitions[partid];
+
+                if (part->act_state == XPC_P_INACTIVE ||
+                                part->act_state == XPC_P_DEACTIVATING) {
+                        continue;
+                }
+
+                /* pull the remote_hb cache line */
+                bres = xp_bte_copy(part->remote_vars_pa,
+                                        (u64) remote_vars,
+                                        XPC_RP_VARS_SIZE,
+                                        (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+                if (bres != BTE_SUCCESS) {
+                        XPC_DEACTIVATE_PARTITION(part,
+                                                xpc_map_bte_errors(bres));
+                        continue;
+                }
+
+                dev_dbg(xpc_part, "partid = %d, heartbeat = %ld, last_heartbeat"
+                        " = %ld, heartbeat_offline = %ld, HB_mask = 0x%lx\n",
+                        partid, remote_vars->heartbeat, part->last_heartbeat,
+                        remote_vars->heartbeat_offline,
+                        remote_vars->heartbeating_to_mask);
+
+                if (((remote_vars->heartbeat == part->last_heartbeat) &&
+                        (remote_vars->heartbeat_offline == 0)) ||
+                             !xpc_hb_allowed(sn_partition_id, remote_vars)) {
+
+                        XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
+                        continue;
+                }
+
+                part->last_heartbeat = remote_vars->heartbeat;
+        }
+}
+
+
+/*
+ * Get a copy of a portion of the remote partition's rsvd page.
+ *
+ * remote_rp points to a buffer that is cacheline aligned for BTE copies and
+ * is large enough to contain a copy of their reserved page header and
+ * part_nasids mask.
+ */
+static enum xpc_retval
+xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
+                struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
+{
+        int bres, i;
+
+
+        /* get the reserved page's physical address */
+
+        *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
+        if (*remote_rp_pa == 0) {
+                return xpcNoRsvdPageAddr;
+        }
+
+
+        /* pull over the reserved page header and part_nasids mask */
+        bres = xp_bte_copy(*remote_rp_pa, (u64) remote_rp,
+                                XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
+                                (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+        if (bres != BTE_SUCCESS) {
+                return xpc_map_bte_errors(bres);
+        }
+
+
+        if (discovered_nasids != NULL) {
+                u64 *remote_part_nasids = XPC_RP_PART_NASIDS(remote_rp);
+
+
+                for (i = 0; i < xp_nasid_mask_words; i++) {
+                        discovered_nasids[i] |= remote_part_nasids[i];
+                }
+        }
+
+
+        /* check that the partid is for another partition */
+
+        if (remote_rp->partid < 1 ||
+                                remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
+                return xpcInvalidPartid;
+        }
+
+        if (remote_rp->partid == sn_partition_id) {
+                return xpcLocalPartid;
+        }
+
+
+        if (XPC_VERSION_MAJOR(remote_rp->version) !=
+                                        XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
+                return xpcBadVersion;
+        }
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Get a copy of the remote partition's XPC variables from the reserved page.
+ *
+ * remote_vars points to a buffer that is cacheline aligned for BTE copies and
+ * assumed to be of size XPC_RP_VARS_SIZE.
+ */
+static enum xpc_retval
+xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
+{
+        int bres;
+
+
+        if (remote_vars_pa == 0) {
+                return xpcVarsNotSet;
+        }
+
+        /* pull over the cross partition variables */
+        bres = xp_bte_copy(remote_vars_pa, (u64) remote_vars, XPC_RP_VARS_SIZE,
+                                (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+        if (bres != BTE_SUCCESS) {
+                return xpc_map_bte_errors(bres);
+        }
+
+        if (XPC_VERSION_MAJOR(remote_vars->version) !=
+                                        XPC_VERSION_MAJOR(XPC_V_VERSION)) {
+                return xpcBadVersion;
+        }
+
+        return xpcSuccess;
+}
+
+
+/*
+ * Update the remote partition's info.
+ */
+static void
+xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
+                struct timespec *remote_rp_stamp, u64 remote_rp_pa,
+                u64 remote_vars_pa, struct xpc_vars *remote_vars)
+{
+        part->remote_rp_version = remote_rp_version;
+        dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
+                part->remote_rp_version);
+
+        part->remote_rp_stamp = *remote_rp_stamp;
+        dev_dbg(xpc_part, "  remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
+                part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
+
+        part->remote_rp_pa = remote_rp_pa;
+        dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
+
+        part->remote_vars_pa = remote_vars_pa;
+        dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
+                part->remote_vars_pa);
+
+        part->last_heartbeat = remote_vars->heartbeat;
+        dev_dbg(xpc_part, "  last_heartbeat = 0x%016lx\n",
+                part->last_heartbeat);
+
+        part->remote_vars_part_pa = remote_vars->vars_part_pa;
+        dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
+                part->remote_vars_part_pa);
+
+        part->remote_act_nasid = remote_vars->act_nasid;
+        dev_dbg(xpc_part, "  remote_act_nasid = 0x%x\n",
+                part->remote_act_nasid);
+
+        part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
+        dev_dbg(xpc_part, "  remote_act_phys_cpuid = 0x%x\n",
+                part->remote_act_phys_cpuid);
+
+        part->remote_amos_page_pa = remote_vars->amos_page_pa;
+        dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
+                part->remote_amos_page_pa);
+
+        part->remote_vars_version = remote_vars->version;
+        dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
+                part->remote_vars_version);
+}
+
+
+/*
+ * Prior code has determined the nasid which generated an IPI.  Inspect
+ * that nasid to determine if its partition needs to be activated or
+ * deactivated.
+ *
+ * A partition is consider "awaiting activation" if our partition
+ * flags indicate it is not active and it has a heartbeat.  A
+ * partition is considered "awaiting deactivation" if our partition
+ * flags indicate it is active but it has no heartbeat or it is not
+ * sending its heartbeat to us.
+ *
+ * To determine the heartbeat, the remote nasid must have a properly
+ * initialized reserved page.
+ */
+static void
+xpc_identify_act_IRQ_req(int nasid)
+{
+        struct xpc_rsvd_page *remote_rp;
+        struct xpc_vars *remote_vars;
+        u64 remote_rp_pa;
+        u64 remote_vars_pa;
+        int remote_rp_version;
+        int reactivate = 0;
+        int stamp_diff;
+        struct timespec remote_rp_stamp = { 0, 0 };
+        partid_t partid;
+        struct xpc_partition *part;
+        enum xpc_retval ret;
+
+
+        /* pull over the reserved page structure */
+
+        remote_rp = (struct xpc_rsvd_page *) xpc_remote_copy_buffer;
+
+        ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
+        if (ret != xpcSuccess) {
+                dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
+                        "which sent interrupt, reason=%d\n", nasid, ret);
+                return;
+        }
+
+        remote_vars_pa = remote_rp->vars_pa;
+        remote_rp_version = remote_rp->version;
+        if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+                remote_rp_stamp = remote_rp->stamp;
+        }
+        partid = remote_rp->partid;
+        part = &xpc_partitions[partid];
+
+
+        /* pull over the cross partition variables */
+
+        remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
+
+        ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
+        if (ret != xpcSuccess) {
+
+                dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
+                        "which sent interrupt, reason=%d\n", nasid, ret);
+
+                XPC_DEACTIVATE_PARTITION(part, ret);
+                return;
+        }
+
+
+        part->act_IRQ_rcvd++;
+
+        dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
+                "%ld:0x%lx\n", (int) nasid, (int) partid, part->act_IRQ_rcvd,
+                remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
+
+        if (xpc_partition_disengaged(part) &&
+                                        part->act_state == XPC_P_INACTIVE) {
+
+                xpc_update_partition_info(part, remote_rp_version,
+                                        &remote_rp_stamp, remote_rp_pa,
+                                        remote_vars_pa, remote_vars);
+
+                if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+                        if (xpc_partition_disengage_requested(1UL << partid)) {
+                                /*
+                                 * Other side is waiting on us to disengage,
+                                 * even though we already have.
+                                 */
+                                return;
+                        }
+                } else {
+                        /* other side doesn't support disengage requests */
+                        xpc_clear_partition_disengage_request(1UL << partid);
+                }
+
+                xpc_activate_partition(part);
+                return;
+        }
+
+        DBUG_ON(part->remote_rp_version == 0);
+        DBUG_ON(part->remote_vars_version == 0);
+
+        if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
+                DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
+                                                        remote_vars_version));
+
+                if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+                        DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
+                                                                version));
+                        /* see if the other side rebooted */
+                        if (part->remote_amos_page_pa ==
+                                remote_vars->amos_page_pa &&
+                                        xpc_hb_allowed(sn_partition_id,
+                                                                remote_vars)) {
+                                /* doesn't look that way, so ignore the IPI */
+                                return;
+                        }
+                }
+
+                /*
+                 * Other side rebooted and previous XPC didn't support the
+                 * disengage request, so we don't need to do anything special.
+                 */
+
+                xpc_update_partition_info(part, remote_rp_version,
+                                                &remote_rp_stamp, remote_rp_pa,
+                                                remote_vars_pa, remote_vars);
+                part->reactivate_nasid = nasid;
+                XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
+                return;
+        }
+
+        DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
+
+        if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+                DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
+
+                /*
+                 * Other side rebooted and previous XPC did support the
+                 * disengage request, but the new one doesn't.
+                 */
+
+                xpc_clear_partition_engaged(1UL << partid);
+                xpc_clear_partition_disengage_request(1UL << partid);
+
+                xpc_update_partition_info(part, remote_rp_version,
+                                                &remote_rp_stamp, remote_rp_pa,
+                                                remote_vars_pa, remote_vars);
+                reactivate = 1;
+
+        } else {
+                DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
+
+                stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
+                                                        &remote_rp_stamp);
+                if (stamp_diff != 0) {
+                        DBUG_ON(stamp_diff >= 0);
+
+                        /*
+                         * Other side rebooted and the previous XPC did support
+                         * the disengage request, as does the new one.
+                         */
+
+                        DBUG_ON(xpc_partition_engaged(1UL << partid));
+                        DBUG_ON(xpc_partition_disengage_requested(1UL <<
+                                                                partid));
+
+                        xpc_update_partition_info(part, remote_rp_version,
+                                                &remote_rp_stamp, remote_rp_pa,
+                                                remote_vars_pa, remote_vars);
+                        reactivate = 1;
+                }
+        }
+
+        if (part->disengage_request_timeout > 0 &&
+                                        !xpc_partition_disengaged(part)) {
+                /* still waiting on other side to disengage from us */
+                return;
+        }
+
+        if (reactivate) {
+                part->reactivate_nasid = nasid;
+                XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
+
+        } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
+                        xpc_partition_disengage_requested(1UL << partid)) {
+                XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
+        }
+}
+
+
+/*
+ * Loop through the activation AMO variables and process any bits
+ * which are set.  Each bit indicates a nasid sending a partition
+ * activation or deactivation request.
+ *
+ * Return #of IRQs detected.
+ */
+int
+xpc_identify_act_IRQ_sender(void)
+{
+        int word, bit;
+        u64 nasid_mask;
+        u64 nasid;                      /* remote nasid */
+        int n_IRQs_detected = 0;
+        AMO_t *act_amos;
+
+
+        act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
+
+
+        /* scan through act AMO variable looking for non-zero entries */
+        for (word = 0; word < xp_nasid_mask_words; word++) {
+
+                if (xpc_exiting) {
+                        break;
+                }
+
+                nasid_mask = xpc_IPI_receive(&act_amos[word]);
+                if (nasid_mask == 0) {
+                        /* no IRQs from nasids in this variable */
+                        continue;
+                }
+
+                dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
+                        nasid_mask);
+
+
+                /*
+                 * If this nasid has been added to the machine since
+                 * our partition was reset, this will retain the
+                 * remote nasid in our reserved pages machine mask.
+                 * This is used in the event of module reload.
+                 */
+                xpc_mach_nasids[word] |= nasid_mask;
+
+
+                /* locate the nasid(s) which sent interrupts */
+
+                for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
+                        if (nasid_mask & (1UL << bit)) {
+                                n_IRQs_detected++;
+                                nasid = XPC_NASID_FROM_W_B(word, bit);
+                                dev_dbg(xpc_part, "interrupt from nasid %ld\n",
+                                        nasid);
+                                xpc_identify_act_IRQ_req(nasid);
+                        }
+                }
+        }
+        return n_IRQs_detected;
+}
+
+
+/*
+ * See if the other side has responded to a partition disengage request
+ * from us.
+ */
+int
+xpc_partition_disengaged(struct xpc_partition *part)
+{
+        partid_t partid = XPC_PARTID(part);
+        int disengaged;
+
+
+        disengaged = (xpc_partition_engaged(1UL << partid) == 0);
+        if (part->disengage_request_timeout) {
+                if (!disengaged) {
+                        if (time_before(jiffies, part->disengage_request_timeout)) {
+                                /* timelimit hasn't been reached yet */
+                                return 0;
+                        }
+
+                        /*
+                         * Other side hasn't responded to our disengage
+                         * request in a timely fashion, so assume it's dead.
+                         */
+
+                        dev_info(xpc_part, "disengage from remote partition %d "
+                                "timed out\n", partid);
+                        xpc_disengage_request_timedout = 1;
+                        xpc_clear_partition_engaged(1UL << partid);
+                        disengaged = 1;
+                }
+                part->disengage_request_timeout = 0;
+
+                /* cancel the timer function, provided it's not us */
+                if (!in_interrupt()) {
+                        del_singleshot_timer_sync(&part->
+                                                      disengage_request_timer);
+                }
+
+                DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
+                                        part->act_state != XPC_P_INACTIVE);
+                if (part->act_state != XPC_P_INACTIVE) {
+                        xpc_wakeup_channel_mgr(part);
+                }
+
+                if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+                        xpc_cancel_partition_disengage_request(part);
+                }
+        }
+        return disengaged;
+}
+
+
+/*
+ * Mark specified partition as active.
+ */
+enum xpc_retval
+xpc_mark_partition_active(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+        enum xpc_retval ret;
+
+
+        dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
+
+        spin_lock_irqsave(&part->act_lock, irq_flags);
+        if (part->act_state == XPC_P_ACTIVATING) {
+                part->act_state = XPC_P_ACTIVE;
+                ret = xpcSuccess;
+        } else {
+                DBUG_ON(part->reason == xpcSuccess);
+                ret = part->reason;
+        }
+        spin_unlock_irqrestore(&part->act_lock, irq_flags);
+
+        return ret;
+}
+
+
+/*
+ * Notify XPC that the partition is down.
+ */
+void
+xpc_deactivate_partition(const int line, struct xpc_partition *part,
+                                enum xpc_retval reason)
+{
+        unsigned long irq_flags;
+
+
+        spin_lock_irqsave(&part->act_lock, irq_flags);
+
+        if (part->act_state == XPC_P_INACTIVE) {
+                XPC_SET_REASON(part, reason, line);
+                spin_unlock_irqrestore(&part->act_lock, irq_flags);
+                if (reason == xpcReactivating) {
+                        /* we interrupt ourselves to reactivate partition */
+                        xpc_IPI_send_reactivate(part);
+                }
+                return;
+        }
+        if (part->act_state == XPC_P_DEACTIVATING) {
+                if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
+                                        reason == xpcReactivating) {
+                        XPC_SET_REASON(part, reason, line);
+                }
+                spin_unlock_irqrestore(&part->act_lock, irq_flags);
+                return;
+        }
+
+        part->act_state = XPC_P_DEACTIVATING;
+        XPC_SET_REASON(part, reason, line);
+
+        spin_unlock_irqrestore(&part->act_lock, irq_flags);
+
+        if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+                xpc_request_partition_disengage(part);
+                xpc_IPI_send_disengage(part);
+
+                /* set a timelimit on the disengage request */
+                part->disengage_request_timeout = jiffies +
+                                        (xpc_disengage_request_timelimit * HZ);
+                part->disengage_request_timer.expires =
+                                        part->disengage_request_timeout;
+                add_timer(&part->disengage_request_timer);
+        }
+
+        dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
+                XPC_PARTID(part), reason);
+
+        xpc_partition_going_down(part, reason);
+}
+
+
+/*
+ * Mark specified partition as inactive.
+ */
+void
+xpc_mark_partition_inactive(struct xpc_partition *part)
+{
+        unsigned long irq_flags;
+
+
+        dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
+                XPC_PARTID(part));
+
+        spin_lock_irqsave(&part->act_lock, irq_flags);
+        part->act_state = XPC_P_INACTIVE;
+        spin_unlock_irqrestore(&part->act_lock, irq_flags);
+        part->remote_rp_pa = 0;
+}
+
+
+/*
+ * SAL has provided a partition and machine mask.  The partition mask
+ * contains a bit for each even nasid in our partition.  The machine
+ * mask contains a bit for each even nasid in the entire machine.
+ *
+ * Using those two bit arrays, we can determine which nasids are
+ * known in the machine.  Each should also have a reserved page
+ * initialized if they are available for partitioning.
+ */
+void
+xpc_discovery(void)
+{
+        void *remote_rp_base;
+        struct xpc_rsvd_page *remote_rp;
+        struct xpc_vars *remote_vars;
+        u64 remote_rp_pa;
+        u64 remote_vars_pa;
+        int region;
+        int region_size;
+        int max_regions;
+        int nasid;
+        struct xpc_rsvd_page *rp;
+        partid_t partid;
+        struct xpc_partition *part;
+        u64 *discovered_nasids;
+        enum xpc_retval ret;
+
+
+        remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
+                                                xp_nasid_mask_bytes,
+                                                GFP_KERNEL, &remote_rp_base);
+        if (remote_rp == NULL) {
+                return;
+        }
+        remote_vars = (struct xpc_vars *) remote_rp;
+
+
+        discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
+                                                        GFP_KERNEL);
+        if (discovered_nasids == NULL) {
+                kfree(remote_rp_base);
+                return;
+        }
+
+        rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
+
+        /*
+         * The term 'region' in this context refers to the minimum number of
+         * nodes that can comprise an access protection grouping. The access
+         * protection is in regards to memory, IOI and IPI.
+         */
+        max_regions = 64;
+        region_size = sn_region_size;
+
+        switch (region_size) {
+        case 128:
+                max_regions *= 2;
+        case 64:
+                max_regions *= 2;
+        case 32:
+                max_regions *= 2;
+                region_size = 16;
+                DBUG_ON(!is_shub2());
+        }
+
+        for (region = 0; region < max_regions; region++) {
+
+                if ((volatile int) xpc_exiting) {
+                        break;
+                }
+
+                dev_dbg(xpc_part, "searching region %d\n", region);
+
+                for (nasid = (region * region_size * 2);
+                     nasid < ((region + 1) * region_size * 2);
+                     nasid += 2) {
+
+                        if ((volatile int) xpc_exiting) {
+                                break;
+                        }
+
+                        dev_dbg(xpc_part, "checking nasid %d\n", nasid);
+
+
+                        if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
+                                dev_dbg(xpc_part, "PROM indicates Nasid %d is "
+                                        "part of the local partition; skipping "
+                                        "region\n", nasid);
+                                break;
+                        }
+
+                        if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
+                                dev_dbg(xpc_part, "PROM indicates Nasid %d was "
+                                        "not on Numa-Link network at reset\n",
+                                        nasid);
+                                continue;
+                        }
+
+                        if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
+                                dev_dbg(xpc_part, "Nasid %d is part of a "
+                                        "partition which was previously "
+                                        "discovered\n", nasid);
+                                continue;
+                        }
+
+
+                        /* pull over the reserved page structure */
+
+                        ret = xpc_get_remote_rp(nasid, discovered_nasids,
+                                              remote_rp, &remote_rp_pa);
+                        if (ret != xpcSuccess) {
+                                dev_dbg(xpc_part, "unable to get reserved page "
+                                        "from nasid %d, reason=%d\n", nasid,
+                                        ret);
+
+                                if (ret == xpcLocalPartid) {
+                                        break;
+                                }
+                                continue;
+                        }
+
+                        remote_vars_pa = remote_rp->vars_pa;
+
+                        partid = remote_rp->partid;
+                        part = &xpc_partitions[partid];
+
+
+                        /* pull over the cross partition variables */
+
+                        ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
+                        if (ret != xpcSuccess) {
+                                dev_dbg(xpc_part, "unable to get XPC variables "
+                                        "from nasid %d, reason=%d\n", nasid,
+                                        ret);
+
+                                XPC_DEACTIVATE_PARTITION(part, ret);
+                                continue;
+                        }
+
+                        if (part->act_state != XPC_P_INACTIVE) {
+                                dev_dbg(xpc_part, "partition %d on nasid %d is "
+                                        "already activating\n", partid, nasid);
+                                break;
+                        }
+
+                        /*
+                         * Register the remote partition's AMOs with SAL so it
+                         * can handle and cleanup errors within that address
+                         * range should the remote partition go down. We don't
+                         * unregister this range because it is difficult to
+                         * tell when outstanding writes to the remote partition
+                         * are finished and thus when it is thus safe to
+                         * unregister. This should not result in wasted space
+                         * in the SAL xp_addr_region table because we should
+                         * get the same page for remote_act_amos_pa after
+                         * module reloads and system reboots.
+                         */
+                        if (sn_register_xp_addr_region(
+                                            remote_vars->amos_page_pa,
+                                                        PAGE_SIZE, 1) < 0) {
+                                dev_dbg(xpc_part, "partition %d failed to "
+                                        "register xp_addr region 0x%016lx\n",
+                                        partid, remote_vars->amos_page_pa);
+
+                                XPC_SET_REASON(part, xpcPhysAddrRegFailed,
+                                                __LINE__);
+                                break;
+                        }
+
+                        /*
+                         * The remote nasid is valid and available.
+                         * Send an interrupt to that nasid to notify
+                         * it that we are ready to begin activation.
+                         */
+                        dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
+                                "nasid %d, phys_cpuid 0x%x\n",
+                                remote_vars->amos_page_pa,
+                                remote_vars->act_nasid,
+                                remote_vars->act_phys_cpuid);
+
+                        if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
+                                                                version)) {
+                                part->remote_amos_page_pa =
+                                                remote_vars->amos_page_pa;
+                                xpc_mark_partition_disengaged(part);
+                                xpc_cancel_partition_disengage_request(part);
+                        }
+                        xpc_IPI_send_activate(remote_vars);
+                }
+        }
+
+        kfree(discovered_nasids);
+        kfree(remote_rp_base);
+}
+
+
+/*
+ * Given a partid, get the nasids owned by that partition from the
+ * remote partition's reserved page.
+ */
+enum xpc_retval
+xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
+{
+        struct xpc_partition *part;
+        u64 part_nasid_pa;
+        int bte_res;
+
+
+        part = &xpc_partitions[partid];
+        if (part->remote_rp_pa == 0) {
+                return xpcPartitionDown;
+        }
+
+        memset(nasid_mask, 0, XP_NASID_MASK_BYTES);
+
+        part_nasid_pa = (u64) XPC_RP_PART_NASIDS(part->remote_rp_pa);
+
+        bte_res = xp_bte_copy(part_nasid_pa, (u64) nasid_mask,
+                        xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+
+        return xpc_map_bte_errors(bte_res);
+}
+
diff --git a/drivers/misc/sgi-xp/xpnet.c b/drivers/misc/sgi-xp/xpnet.c
new file mode 100644
index 000000000000..38552f37e53d
--- /dev/null
+++ b/drivers/misc/sgi-xp/xpnet.c
@@ -0,0 +1,718 @@
+/*
+ * 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) 1999-2008 Silicon Graphics, Inc. All rights reserved.
+ */
+
+
+/*
+ * Cross Partition Network Interface (XPNET) support
+ *
+ *      XPNET provides a virtual network layered on top of the Cross
+ *      Partition communication layer.
+ *
+ *      XPNET provides direct point-to-point and broadcast-like support
+ *      for an ethernet-like device.  The ethernet broadcast medium is
+ *      replaced with a point-to-point message structure which passes
+ *      pointers to a DMA-capable block that a remote partition should
+ *      retrieve and pass to the upper level networking layer.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/ioport.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/delay.h>
+#include <linux/ethtool.h>
+#include <linux/mii.h>
+#include <linux/smp.h>
+#include <linux/string.h>
+#include <asm/sn/bte.h>
+#include <asm/sn/io.h>
+#include <asm/sn/sn_sal.h>
+#include <asm/types.h>
+#include <asm/atomic.h>
+#include "xp.h"
+
+
+/*
+ * The message payload transferred by XPC.
+ *
+ * buf_pa is the physical address where the DMA should pull from.
+ *
+ * NOTE: for performance reasons, buf_pa should _ALWAYS_ begin on a
+ * cacheline boundary.  To accomplish this, we record the number of
+ * bytes from the beginning of the first cacheline to the first useful
+ * byte of the skb (leadin_ignore) and the number of bytes from the
+ * last useful byte of the skb to the end of the last cacheline
+ * (tailout_ignore).
+ *
+ * size is the number of bytes to transfer which includes the skb->len
+ * (useful bytes of the senders skb) plus the leadin and tailout
+ */
+struct xpnet_message {
+        u16 version;            /* Version for this message */
+        u16 embedded_bytes;     /* #of bytes embedded in XPC message */
+        u32 magic;              /* Special number indicating this is xpnet */
+        u64 buf_pa;             /* phys address of buffer to retrieve */
+        u32 size;               /* #of bytes in buffer */
+        u8 leadin_ignore;       /* #of bytes to ignore at the beginning */
+        u8 tailout_ignore;      /* #of bytes to ignore at the end */
+        unsigned char data;     /* body of small packets */
+};
+
+/*
+ * Determine the size of our message, the cacheline aligned size,
+ * and then the number of message will request from XPC.
+ *
+ * XPC expects each message to exist in an individual cacheline.
+ */
+#define XPNET_MSG_SIZE          (L1_CACHE_BYTES - XPC_MSG_PAYLOAD_OFFSET)
+#define XPNET_MSG_DATA_MAX      \
+                (XPNET_MSG_SIZE - (u64)(&((struct xpnet_message *)0)->data))
+#define XPNET_MSG_ALIGNED_SIZE  (L1_CACHE_ALIGN(XPNET_MSG_SIZE))
+#define XPNET_MSG_NENTRIES      (PAGE_SIZE / XPNET_MSG_ALIGNED_SIZE)
+
+
+#define XPNET_MAX_KTHREADS      (XPNET_MSG_NENTRIES + 1)
+#define XPNET_MAX_IDLE_KTHREADS (XPNET_MSG_NENTRIES + 1)
+
+/*
+ * Version number of XPNET implementation. XPNET can always talk to versions
+ * with same major #, and never talk to versions with a different version.
+ */
+#define _XPNET_VERSION(_major, _minor)  (((_major) << 4) | (_minor))
+#define XPNET_VERSION_MAJOR(_v)         ((_v) >> 4)
+#define XPNET_VERSION_MINOR(_v)         ((_v) & 0xf)
+
+#define XPNET_VERSION _XPNET_VERSION(1,0)               /* version 1.0 */
+#define XPNET_VERSION_EMBED _XPNET_VERSION(1,1)         /* version 1.1 */
+#define XPNET_MAGIC     0x88786984 /* "XNET" */
+
+#define XPNET_VALID_MSG(_m)                                                  \
+   ((XPNET_VERSION_MAJOR(_m->version) == XPNET_VERSION_MAJOR(XPNET_VERSION)) \
+    && (msg->magic == XPNET_MAGIC))
+
+#define XPNET_DEVICE_NAME               "xp0"
+
+
+/*
+ * When messages are queued with xpc_send_notify, a kmalloc'd buffer
+ * of the following type is passed as a notification cookie.  When the
+ * notification function is called, we use the cookie to decide
+ * whether all outstanding message sends have completed.  The skb can
+ * then be released.
+ */
+struct xpnet_pending_msg {
+        struct list_head free_list;
+        struct sk_buff *skb;
+        atomic_t use_count;
+};
+
+/* driver specific structure pointed to by the device structure */
+struct xpnet_dev_private {
+        struct net_device_stats stats;
+};
+
+struct net_device *xpnet_device;
+
+/*
+ * When we are notified of other partitions activating, we add them to
+ * our bitmask of partitions to which we broadcast.
+ */
+static u64 xpnet_broadcast_partitions;
+/* protect above */
+static DEFINE_SPINLOCK(xpnet_broadcast_lock);
+
+/*
+ * Since the Block Transfer Engine (BTE) is being used for the transfer
+ * and it relies upon cache-line size transfers, we need to reserve at
+ * least one cache-line for head and tail alignment.  The BTE is
+ * limited to 8MB transfers.
+ *
+ * Testing has shown that changing MTU to greater than 64KB has no effect
+ * on TCP as the two sides negotiate a Max Segment Size that is limited
+ * to 64K.  Other protocols May use packets greater than this, but for
+ * now, the default is 64KB.
+ */
+#define XPNET_MAX_MTU (0x800000UL - L1_CACHE_BYTES)
+/* 32KB has been determined to be the ideal */
+#define XPNET_DEF_MTU (0x8000UL)
+
+
+/*
+ * The partition id is encapsulated in the MAC address.  The following
+ * define locates the octet the partid is in.
+ */
+#define XPNET_PARTID_OCTET      1
+#define XPNET_LICENSE_OCTET     2
+
+
+/*
+ * Define the XPNET debug device structure that is to be used with dev_dbg(),
+ * dev_err(), dev_warn(), and dev_info().
+ */
+struct device_driver xpnet_dbg_name = {
+        .name = "xpnet"
+};
+
+struct device xpnet_dbg_subname = {
+        .bus_id = {0},                  /* set to "" */
+        .driver = &xpnet_dbg_name
+};
+
+struct device *xpnet = &xpnet_dbg_subname;
+
+/*
+ * Packet was recevied by XPC and forwarded to us.
+ */
+static void
+xpnet_receive(partid_t partid, int channel, struct xpnet_message *msg)
+{
+        struct sk_buff *skb;
+        bte_result_t bret;
+        struct xpnet_dev_private *priv =
+                (struct xpnet_dev_private *) xpnet_device->priv;
+
+
+        if (!XPNET_VALID_MSG(msg)) {
+                /*
+                 * Packet with a different XPC version.  Ignore.
+                 */
+                xpc_received(partid, channel, (void *) msg);
+
+                priv->stats.rx_errors++;
+
+                return;
+        }
+        dev_dbg(xpnet, "received 0x%lx, %d, %d, %d\n", msg->buf_pa, msg->size,
+                msg->leadin_ignore, msg->tailout_ignore);
+
+
+        /* reserve an extra cache line */
+        skb = dev_alloc_skb(msg->size + L1_CACHE_BYTES);
+        if (!skb) {
+                dev_err(xpnet, "failed on dev_alloc_skb(%d)\n",
+                        msg->size + L1_CACHE_BYTES);
+
+                xpc_received(partid, channel, (void *) msg);
+
+                priv->stats.rx_errors++;
+
+                return;
+        }
+
+        /*
+         * The allocated skb has some reserved space.
+         * In order to use bte_copy, we need to get the
+         * skb->data pointer moved forward.
+         */
+        skb_reserve(skb, (L1_CACHE_BYTES - ((u64)skb->data &
+                                            (L1_CACHE_BYTES - 1)) +
+                          msg->leadin_ignore));
+
+        /*
+         * Update the tail pointer to indicate data actually
+         * transferred.
+         */
+        skb_put(skb, (msg->size - msg->leadin_ignore - msg->tailout_ignore));
+
+        /*
+         * Move the data over from the other side.
+         */
+        if ((XPNET_VERSION_MINOR(msg->version) == 1) &&
+                                                (msg->embedded_bytes != 0)) {
+                dev_dbg(xpnet, "copying embedded message. memcpy(0x%p, 0x%p, "
+                        "%lu)\n", skb->data, &msg->data,
+                        (size_t) msg->embedded_bytes);
+
+                skb_copy_to_linear_data(skb, &msg->data, (size_t)msg->embedded_bytes);
+        } else {
+                dev_dbg(xpnet, "transferring buffer to the skb->data area;\n\t"
+                        "bte_copy(0x%p, 0x%p, %hu)\n", (void *)msg->buf_pa,
+                        (void *)__pa((u64)skb->data & ~(L1_CACHE_BYTES - 1)),
+                        msg->size);
+
+                bret = bte_copy(msg->buf_pa,
+                                __pa((u64)skb->data & ~(L1_CACHE_BYTES - 1)),
+                                msg->size, (BTE_NOTIFY | BTE_WACQUIRE), NULL);
+
+                if (bret != BTE_SUCCESS) {
+                        // >>> Need better way of cleaning skb.  Currently skb
+                        // >>> appears in_use and we can't just call
+                        // >>> dev_kfree_skb.
+                        dev_err(xpnet, "bte_copy(0x%p, 0x%p, 0x%hx) returned "
+                                "error=0x%x\n", (void *)msg->buf_pa,
+                                (void *)__pa((u64)skb->data &
+                                                        ~(L1_CACHE_BYTES - 1)),
+                                msg->size, bret);
+
+                        xpc_received(partid, channel, (void *) msg);
+
+                        priv->stats.rx_errors++;
+
+                        return;
+                }
+        }
+
+        dev_dbg(xpnet, "<skb->head=0x%p skb->data=0x%p skb->tail=0x%p "
+                "skb->end=0x%p skb->len=%d\n", (void *) skb->head,
+                (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb),
+                skb->len);
+
+        skb->protocol = eth_type_trans(skb, xpnet_device);
+        skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+        dev_dbg(xpnet, "passing skb to network layer\n"
+                KERN_DEBUG "\tskb->head=0x%p skb->data=0x%p skb->tail=0x%p "
+                "skb->end=0x%p skb->len=%d\n",
+                (void *)skb->head, (void *)skb->data, skb_tail_pointer(skb),
+                skb_end_pointer(skb), skb->len);
+
+
+        xpnet_device->last_rx = jiffies;
+        priv->stats.rx_packets++;
+        priv->stats.rx_bytes += skb->len + ETH_HLEN;
+
+        netif_rx_ni(skb);
+        xpc_received(partid, channel, (void *) msg);
+}
+
+
+/*
+ * This is the handler which XPC calls during any sort of change in
+ * state or message reception on a connection.
+ */
+static void
+xpnet_connection_activity(enum xpc_retval reason, partid_t partid, int channel,
+                          void *data, void *key)
+{
+        long bp;
+
+
+        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(channel != XPC_NET_CHANNEL);
+
+        switch(reason) {
+        case xpcMsgReceived:    /* message received */
+                DBUG_ON(data == NULL);
+
+                xpnet_receive(partid, channel, (struct xpnet_message *) data);
+                break;
+
+        case xpcConnected:      /* connection completed to a partition */
+                spin_lock_bh(&xpnet_broadcast_lock);
+                xpnet_broadcast_partitions |= 1UL << (partid -1 );
+                bp = xpnet_broadcast_partitions;
+                spin_unlock_bh(&xpnet_broadcast_lock);
+
+                netif_carrier_on(xpnet_device);
+
+                dev_dbg(xpnet, "%s connection created to partition %d; "
+                        "xpnet_broadcast_partitions=0x%lx\n",
+                        xpnet_device->name, partid, bp);
+                break;
+
+        default:
+                spin_lock_bh(&xpnet_broadcast_lock);
+                xpnet_broadcast_partitions &= ~(1UL << (partid -1 ));
+                bp = xpnet_broadcast_partitions;
+                spin_unlock_bh(&xpnet_broadcast_lock);
+
+                if (bp == 0) {
+                        netif_carrier_off(xpnet_device);
+                }
+
+                dev_dbg(xpnet, "%s disconnected from partition %d; "
+                        "xpnet_broadcast_partitions=0x%lx\n",
+                        xpnet_device->name, partid, bp);
+                break;
+
+        }
+}
+
+
+static int
+xpnet_dev_open(struct net_device *dev)
+{
+        enum xpc_retval ret;
+
+
+        dev_dbg(xpnet, "calling xpc_connect(%d, 0x%p, NULL, %ld, %ld, %ld, "
+                "%ld)\n", XPC_NET_CHANNEL, xpnet_connection_activity,
+                XPNET_MSG_SIZE, XPNET_MSG_NENTRIES, XPNET_MAX_KTHREADS,
+                XPNET_MAX_IDLE_KTHREADS);
+
+        ret = xpc_connect(XPC_NET_CHANNEL, xpnet_connection_activity, NULL,
+                          XPNET_MSG_SIZE, XPNET_MSG_NENTRIES,
+                          XPNET_MAX_KTHREADS, XPNET_MAX_IDLE_KTHREADS);
+        if (ret != xpcSuccess) {
+                dev_err(xpnet, "ifconfig up of %s failed on XPC connect, "
+                        "ret=%d\n", dev->name, ret);
+
+                return -ENOMEM;
+        }
+
+        dev_dbg(xpnet, "ifconfig up of %s; XPC connected\n", dev->name);
+
+        return 0;
+}
+
+
+static int
+xpnet_dev_stop(struct net_device *dev)
+{
+        xpc_disconnect(XPC_NET_CHANNEL);
+
+        dev_dbg(xpnet, "ifconfig down of %s; XPC disconnected\n", dev->name);
+
+        return 0;
+}
+
+
+static int
+xpnet_dev_change_mtu(struct net_device *dev, int new_mtu)
+{
+        /* 68 comes from min TCP+IP+MAC header */
+        if ((new_mtu < 68) || (new_mtu > XPNET_MAX_MTU)) {
+                dev_err(xpnet, "ifconfig %s mtu %d failed; value must be "
+                        "between 68 and %ld\n", dev->name, new_mtu,
+                        XPNET_MAX_MTU);
+                return -EINVAL;
+        }
+
+        dev->mtu = new_mtu;
+        dev_dbg(xpnet, "ifconfig %s mtu set to %d\n", dev->name, new_mtu);
+        return 0;
+}
+
+
+/*
+ * Required for the net_device structure.
+ */
+static int
+xpnet_dev_set_config(struct net_device *dev, struct ifmap *new_map)
+{
+        return 0;
+}
+
+
+/*
+ * Return statistics to the caller.
+ */
+static struct net_device_stats *
+xpnet_dev_get_stats(struct net_device *dev)
+{
+        struct xpnet_dev_private *priv;
+
+
+        priv = (struct xpnet_dev_private *) dev->priv;
+
+        return &priv->stats;
+}
+
+
+/*
+ * Notification that the other end has received the message and
+ * DMA'd the skb information.  At this point, they are done with
+ * our side.  When all recipients are done processing, we
+ * release the skb and then release our pending message structure.
+ */
+static void
+xpnet_send_completed(enum xpc_retval reason, partid_t partid, int channel,
+                        void *__qm)
+{
+        struct xpnet_pending_msg *queued_msg =
+                (struct xpnet_pending_msg *) __qm;
+
+
+        DBUG_ON(queued_msg == NULL);
+
+        dev_dbg(xpnet, "message to %d notified with reason %d\n",
+                partid, reason);
+
+        if (atomic_dec_return(&queued_msg->use_count) == 0) {
+                dev_dbg(xpnet, "all acks for skb->head=-x%p\n",
+                        (void *) queued_msg->skb->head);
+
+                dev_kfree_skb_any(queued_msg->skb);
+                kfree(queued_msg);
+        }
+}
+
+
+/*
+ * Network layer has formatted a packet (skb) and is ready to place it
+ * "on the wire".  Prepare and send an xpnet_message to all partitions
+ * which have connected with us and are targets of this packet.
+ *
+ * MAC-NOTE:  For the XPNET driver, the MAC address contains the
+ * destination partition_id.  If the destination partition id word
+ * is 0xff, this packet is to broadcast to all partitions.
+ */
+static int
+xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+        struct xpnet_pending_msg *queued_msg;
+        enum xpc_retval ret;
+        struct xpnet_message *msg;
+        u64 start_addr, end_addr;
+        long dp;
+        u8 second_mac_octet;
+        partid_t dest_partid;
+        struct xpnet_dev_private *priv;
+        u16 embedded_bytes;
+
+
+        priv = (struct xpnet_dev_private *) dev->priv;
+
+
+        dev_dbg(xpnet, ">skb->head=0x%p skb->data=0x%p skb->tail=0x%p "
+                "skb->end=0x%p skb->len=%d\n", (void *) skb->head,
+                (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb),
+                skb->len);
+
+
+        /*
+         * The xpnet_pending_msg tracks how many outstanding
+         * xpc_send_notifies are relying on this skb.  When none
+         * remain, release the skb.
+         */
+        queued_msg = kmalloc(sizeof(struct xpnet_pending_msg), GFP_ATOMIC);
+        if (queued_msg == NULL) {
+                dev_warn(xpnet, "failed to kmalloc %ld bytes; dropping "
+                        "packet\n", sizeof(struct xpnet_pending_msg));
+
+                priv->stats.tx_errors++;
+
+                return -ENOMEM;
+        }
+
+
+        /* get the beginning of the first cacheline and end of last */
+        start_addr = ((u64) skb->data & ~(L1_CACHE_BYTES - 1));
+        end_addr = L1_CACHE_ALIGN((u64)skb_tail_pointer(skb));
+
+        /* calculate how many bytes to embed in the XPC message */
+        embedded_bytes = 0;
+        if (unlikely(skb->len <= XPNET_MSG_DATA_MAX)) {
+                /* skb->data does fit so embed */
+                embedded_bytes = skb->len;
+        }
+
+
+        /*
+         * Since the send occurs asynchronously, we set the count to one
+         * and begin sending.  Any sends that happen to complete before
+         * we are done sending will not free the skb.  We will be left
+         * with that task during exit.  This also handles the case of
+         * a packet destined for a partition which is no longer up.
+         */
+        atomic_set(&queued_msg->use_count, 1);
+        queued_msg->skb = skb;
+
+
+        second_mac_octet = skb->data[XPNET_PARTID_OCTET];
+        if (second_mac_octet == 0xff) {
+                /* we are being asked to broadcast to all partitions */
+                dp = xpnet_broadcast_partitions;
+        } else if (second_mac_octet != 0) {
+                dp = xpnet_broadcast_partitions &
+                                        (1UL << (second_mac_octet - 1));
+        } else {
+                /* 0 is an invalid partid.  Ignore */
+                dp = 0;
+        }
+        dev_dbg(xpnet, "destination Partitions mask (dp) = 0x%lx\n", dp);
+
+        /*
+         * If we wanted to allow promiscuous mode to work like an
+         * unswitched network, this would be a good point to OR in a
+         * mask of partitions which should be receiving all packets.
+         */
+
+        /*
+         * Main send loop.
+         */
+        for (dest_partid = 1; dp && dest_partid < XP_MAX_PARTITIONS;
+             dest_partid++) {
+
+
+                if (!(dp & (1UL << (dest_partid - 1)))) {
+                        /* not destined for this partition */
+                        continue;
+                }
+
+                /* remove this partition from the destinations mask */
+                dp &= ~(1UL << (dest_partid - 1));
+
+
+                /* found a partition to send to */
+
+                ret = xpc_allocate(dest_partid, XPC_NET_CHANNEL,
+                                   XPC_NOWAIT, (void **)&msg);
+                if (unlikely(ret != xpcSuccess)) {
+                        continue;
+                }
+
+                msg->embedded_bytes = embedded_bytes;
+                if (unlikely(embedded_bytes != 0)) {
+                        msg->version = XPNET_VERSION_EMBED;
+                        dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n",
+                                &msg->data, skb->data, (size_t) embedded_bytes);
+                        skb_copy_from_linear_data(skb, &msg->data,
+                                                  (size_t)embedded_bytes);
+                } else {
+                        msg->version = XPNET_VERSION;
+                }
+                msg->magic = XPNET_MAGIC;
+                msg->size = end_addr - start_addr;
+                msg->leadin_ignore = (u64) skb->data - start_addr;
+                msg->tailout_ignore = end_addr - (u64)skb_tail_pointer(skb);
+                msg->buf_pa = __pa(start_addr);
+
+                dev_dbg(xpnet, "sending XPC message to %d:%d\n"
+                        KERN_DEBUG "msg->buf_pa=0x%lx, msg->size=%u, "
+                        "msg->leadin_ignore=%u, msg->tailout_ignore=%u\n",
+                        dest_partid, XPC_NET_CHANNEL, msg->buf_pa, msg->size,
+                        msg->leadin_ignore, msg->tailout_ignore);
+
+
+                atomic_inc(&queued_msg->use_count);
+
+                ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, msg,
+                                      xpnet_send_completed, queued_msg);
+                if (unlikely(ret != xpcSuccess)) {
+                        atomic_dec(&queued_msg->use_count);
+                        continue;
+                }
+
+        }
+
+        if (atomic_dec_return(&queued_msg->use_count) == 0) {
+                dev_dbg(xpnet, "no partitions to receive packet destined for "
+                        "%d\n", dest_partid);
+
+
+                dev_kfree_skb(skb);
+                kfree(queued_msg);
+        }
+
+        priv->stats.tx_packets++;
+        priv->stats.tx_bytes += skb->len;
+
+        return 0;
+}
+
+
+/*
+ * Deal with transmit timeouts coming from the network layer.
+ */
+static void
+xpnet_dev_tx_timeout (struct net_device *dev)
+{
+        struct xpnet_dev_private *priv;
+
+
+        priv = (struct xpnet_dev_private *) dev->priv;
+
+        priv->stats.tx_errors++;
+        return;
+}
+
+
+static int __init
+xpnet_init(void)
+{
+        int i;
+        u32 license_num;
+        int result = -ENOMEM;
+
+
+        if (!ia64_platform_is("sn2")) {
+                return -ENODEV;
+        }
+
+        dev_info(xpnet, "registering network device %s\n", XPNET_DEVICE_NAME);
+
+        /*
+         * use ether_setup() to init the majority of our device
+         * structure and then override the necessary pieces.
+         */
+        xpnet_device = alloc_netdev(sizeof(struct xpnet_dev_private),
+                                    XPNET_DEVICE_NAME, ether_setup);
+        if (xpnet_device == NULL) {
+                return -ENOMEM;
+        }
+
+        netif_carrier_off(xpnet_device);
+
+        xpnet_device->mtu = XPNET_DEF_MTU;
+        xpnet_device->change_mtu = xpnet_dev_change_mtu;
+        xpnet_device->open = xpnet_dev_open;
+        xpnet_device->get_stats = xpnet_dev_get_stats;
+        xpnet_device->stop = xpnet_dev_stop;
+        xpnet_device->hard_start_xmit = xpnet_dev_hard_start_xmit;
+        xpnet_device->tx_timeout = xpnet_dev_tx_timeout;
+        xpnet_device->set_config = xpnet_dev_set_config;
+
+        /*
+         * Multicast assumes the LSB of the first octet is set for multicast
+         * MAC addresses.  We chose the first octet of the MAC to be unlikely
+         * to collide with any vendor's officially issued MAC.
+         */
+        xpnet_device->dev_addr[0] = 0xfe;
+        xpnet_device->dev_addr[XPNET_PARTID_OCTET] = sn_partition_id;
+        license_num = sn_partition_serial_number_val();
+        for (i = 3; i >= 0; i--) {
+                xpnet_device->dev_addr[XPNET_LICENSE_OCTET + i] =
+                                                        license_num & 0xff;
+                license_num = license_num >> 8;
+        }
+
+        /*
+         * ether_setup() sets this to a multicast device.  We are
+         * really not supporting multicast at this time.
+         */
+        xpnet_device->flags &= ~IFF_MULTICAST;
+
+        /*
+         * No need to checksum as it is a DMA transfer.  The BTE will
+         * report an error if the data is not retrievable and the
+         * packet will be dropped.
+         */
+        xpnet_device->features = NETIF_F_NO_CSUM;
+
+        result = register_netdev(xpnet_device);
+        if (result != 0) {
+                free_netdev(xpnet_device);
+        }
+
+        return result;
+}
+module_init(xpnet_init);
+
+
+static void __exit
+xpnet_exit(void)
+{
+        dev_info(xpnet, "unregistering network device %s\n",
+                xpnet_device[0].name);
+
+        unregister_netdev(xpnet_device);
+
+        free_netdev(xpnet_device);
+}
+module_exit(xpnet_exit);
+
+
+MODULE_AUTHOR("Silicon Graphics, Inc.");
+MODULE_DESCRIPTION("Cross Partition Network adapter (XPNET)");
+MODULE_LICENSE("GPL");
+