sgi-xp: support runtime selection of xp_max_npartitions

Support runtime selection of the max number of partitions based on the
hardware being run on.

Signed-off-by: Dean Nelson <dcn@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

10 files changed, 266 insertions, 151 deletions

diff --git a/drivers/misc/sgi-xp/Makefile b/drivers/misc/sgi-xp/Makefile
index b6e40a7958c..b50f2921781 100644
--- a/drivers/misc/sgi-xp/Makefile
+++ b/drivers/misc/sgi-xp/Makefile
@@ -3,7 +3,8 @@
 #
 
 obj-$(CONFIG_SGI_XP)            += xp.o
-xp-y                            := xp_main.o xp_nofault.o
+xp-y                            := xp_main.o xp_uv.o
+xp-$(CONFIG_IA64)               += xp_sn2.o xp_nofault.o
 
 obj-$(CONFIG_SGI_XP)            += xpc.o
 xpc-y                           := xpc_main.o xpc_channel.o xpc_partition.o
diff --git a/drivers/misc/sgi-xp/xp.h b/drivers/misc/sgi-xp/xp.h
index 867fb4863d5..51087e11188 100644
--- a/drivers/misc/sgi-xp/xp.h
+++ b/drivers/misc/sgi-xp/xp.h
@@ -18,6 +18,9 @@
 #include <linux/mutex.h>
 #include <asm/sn/types.h>
 #include <asm/sn/bte.h>
+#ifdef CONFIG_IA64
+#include <asm/sn/arch.h>
+#endif
 
 /* >>> Add this #define to some linux header file some day. */
 #define BYTES_PER_WORD  sizeof(void *)
@@ -45,17 +48,18 @@
 #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.
+ * Define the maximum number of partitions the system can possibly support.
+ * It is based on 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 XP_MAX_NPARTITIONS_SN2  64
+#define XP_MAX_NPARTITIONS_UV   256
 
 /*
  * Define the number of u64s required to represent all the C-brick nasids
@@ -112,24 +116,28 @@ xp_bte_copy(u64 src, u64 vdst, u64 len, u64 mode, void *notification)
  * 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.
+>>> The following described limitation of a max of eight channels possible
+>>> pertains only to ia64-sn2. THIS ISN'T TRUE SINCE I'M PLANNING TO JUST
+>>> TIE INTO THE EXISTING MECHANISM ONCE THE CHANNEL MESSAGES ARE RECEIVED.
+>>> THE 128-BYTE CACHELINE PERFORMANCE ISSUE IS TIED TO IA64-SN2.
  *
- * 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.
+ * XPC_MAX_NCHANNELS accordingly. If the day should come where that number
+ * exceeds the absolute MAXIMUM number of channels possible (eight), then one
+ * will need to make changes to the XPC code to accommodate for this.
+ *
+ * The absolute maximum number of channels possible is currently 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 128-byte cacheline.
  */
 #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 */
+#define XPC_MAX_NCHANNELS       2       /* max #of channels allowed */
 
-#if XPC_NCHANNELS > XPC_MAX_NCHANNELS
-#error  XPC_NCHANNELS exceeds MAXIMUM allowed.
+#if XPC_MAX_NCHANNELS > 8
+#error  XPC_MAX_NCHANNELS exceeds absolute MAXIMUM possible.
 #endif
 
 /*
@@ -254,7 +262,8 @@ enum xp_retval {
         xpBteCopyError,         /* 52: bte_copy() returned error */
         xpSalError,             /* 53: sn SAL error */
 
-        xpUnknownReason         /* 54: unknown reason - must be last in enum */
+        xpUnsupported,          /* 54: unsupported functionality or resource */
+        xpUnknownReason         /* 55: unknown reason - must be last in enum */
 };
 
 /*
@@ -397,8 +406,16 @@ xpc_partid_to_nasids(short partid, void *nasids)
         return xpc_interface.partid_to_nasids(partid, nasids);
 }
 
+extern short xp_max_npartitions;
+
 extern u64 xp_nofault_PIOR_target;
 extern int xp_nofault_PIOR(void *);
 extern int xp_error_PIOR(void);
 
+extern struct device *xp;
+extern enum xp_retval xp_init_sn2(void);
+extern enum xp_retval xp_init_uv(void);
+extern void xp_exit_sn2(void);
+extern void xp_exit_uv(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
index 196480b691a..c5cec606377 100644
--- a/drivers/misc/sgi-xp/xp_main.c
+++ b/drivers/misc/sgi-xp/xp_main.c
@@ -15,28 +15,32 @@
  */
 
 #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 <linux/device.h>
 #include "xp.h"
 
-/*
- * The export of xp_nofault_PIOR needs to happen here since it is defined
- * in drivers/misc/sgi-xp/xp_nofault.S. The target of the nofault read is
- * defined here.
- */
-EXPORT_SYMBOL_GPL(xp_nofault_PIOR);
+/* define the XP debug device structures to be used with dev_dbg() et al */
 
-u64 xp_nofault_PIOR_target;
-EXPORT_SYMBOL_GPL(xp_nofault_PIOR_target);
+struct device_driver xp_dbg_name = {
+        .name = "xp"
+};
+
+struct device xp_dbg_subname = {
+        .bus_id = {0},          /* set to "" */
+        .driver = &xp_dbg_name
+};
+
+struct device *xp = &xp_dbg_subname;
+
+/* max #of partitions possible */
+short xp_max_npartitions;
+EXPORT_SYMBOL_GPL(xp_max_npartitions);
 
 /*
  * xpc_registrations[] keeps track of xpc_connect()'s done by the kernel-level
  * users of XPC.
  */
-struct xpc_registration xpc_registrations[XPC_NCHANNELS];
+struct xpc_registration xpc_registrations[XPC_MAX_NCHANNELS];
 EXPORT_SYMBOL_GPL(xpc_registrations);
 
 /*
@@ -135,7 +139,7 @@ xpc_connect(int ch_number, xpc_channel_func func, void *key, u16 payload_size,
 {
         struct xpc_registration *registration;
 
-        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_MAX_NCHANNELS);
         DBUG_ON(payload_size == 0 || nentries == 0);
         DBUG_ON(func == NULL);
         DBUG_ON(assigned_limit == 0 || idle_limit > assigned_limit);
@@ -185,7 +189,7 @@ xpc_disconnect(int ch_number)
 {
         struct xpc_registration *registration;
 
-        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_MAX_NCHANNELS);
 
         registration = &xpc_registrations[ch_number];
 
@@ -221,39 +225,21 @@ EXPORT_SYMBOL_GPL(xpc_disconnect);
 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;
+        enum xp_retval ret;
+        int ch_number;
 
-        /*
-         * 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).
-         */
-        ret = sn_register_nofault_code(func_addr, err_func_addr, err_func_addr,
-                                       1, 1);
-        if (ret != 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;
+        if (is_shub())
+                ret = xp_init_sn2();
+        else if (is_uv())
+                ret = xp_init_uv();
         else
-                xp_nofault_PIOR_target = SH1_IPI_ACCESS;
+                ret = xpUnsupported;
+
+        if (ret != xpSuccess)
+                return -ENODEV;
 
         /* initialize the connection registration mutex */
-        for (ch_number = 0; ch_number < XPC_NCHANNELS; ch_number++)
+        for (ch_number = 0; ch_number < XPC_MAX_NCHANNELS; ch_number++)
                 mutex_init(&xpc_registrations[ch_number].mutex);
 
         return 0;
@@ -264,12 +250,10 @@ 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);
+        if (is_shub())
+                xp_exit_sn2();
+        else if (is_uv())
+                xp_exit_uv();
 }
 
 module_exit(xp_exit);
diff --git a/drivers/misc/sgi-xp/xp_sn2.c b/drivers/misc/sgi-xp/xp_sn2.c
new file mode 100644
index 00000000000..b92579356a6
--- /dev/null
+++ b/drivers/misc/sgi-xp/xp_sn2.c
@@ -0,0 +1,92 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+/*
+ * Cross Partition (XP) sn2-based functions.
+ *
+ *      Architecture specific implementation of common functions.
+ */
+
+#include <linux/device.h>
+#include <asm/sn/sn_sal.h>
+#include "xp.h"
+
+/*
+ * The export of xp_nofault_PIOR needs to happen here since it is defined
+ * in drivers/misc/sgi-xp/xp_nofault.S. The target of the nofault read is
+ * defined here.
+ */
+EXPORT_SYMBOL_GPL(xp_nofault_PIOR);
+
+u64 xp_nofault_PIOR_target;
+EXPORT_SYMBOL_GPL(xp_nofault_PIOR_target);
+
+/*
+ * 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).
+ */
+static enum xp_retval
+xp_register_nofault_code_sn2(void)
+{
+        int ret;
+        u64 func_addr;
+        u64 err_func_addr;
+
+        func_addr = *(u64 *)xp_nofault_PIOR;
+        err_func_addr = *(u64 *)xp_error_PIOR;
+        ret = sn_register_nofault_code(func_addr, err_func_addr, err_func_addr,
+                                       1, 1);
+        if (ret != 0) {
+                dev_err(xp, "can't register nofault code, error=%d\n", ret);
+                return xpSalError;
+        }
+        /*
+         * Setup the nofault PIO read target. (There is no special reason why
+         * SH_IPI_ACCESS was selected.)
+         */
+        if (is_shub1())
+                xp_nofault_PIOR_target = SH1_IPI_ACCESS;
+        else if (is_shub2())
+                xp_nofault_PIOR_target = SH2_IPI_ACCESS0;
+
+        return xpSuccess;
+}
+
+void
+xp_unregister_nofault_code_sn2(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);
+}
+
+enum xp_retval
+xp_init_sn2(void)
+{
+        BUG_ON(!is_shub());
+
+        xp_max_npartitions = XP_MAX_NPARTITIONS_SN2;
+
+        return xp_register_nofault_code_sn2();
+}
+
+void
+xp_exit_sn2(void)
+{
+        BUG_ON(!is_shub());
+
+        xp_unregister_nofault_code_sn2();
+}
+
diff --git a/drivers/misc/sgi-xp/xp_uv.c b/drivers/misc/sgi-xp/xp_uv.c
new file mode 100644
index 00000000000..30888be2cdb
--- /dev/null
+++ b/drivers/misc/sgi-xp/xp_uv.c
@@ -0,0 +1,30 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
+ */
+
+/*
+ * Cross Partition (XP) uv-based functions.
+ *
+ *      Architecture specific implementation of common functions.
+ *
+ */
+
+#include "xp.h"
+
+enum xp_retval
+xp_init_uv(void)
+{
+        BUG_ON(!is_uv());
+
+        xp_max_npartitions = XP_MAX_NPARTITIONS_UV;
+}
+
+void
+xp_exit_uv(void)
+{
+        BUG_ON(!is_uv());
+}
diff --git a/drivers/misc/sgi-xp/xpc.h b/drivers/misc/sgi-xp/xpc.h
index 11ac267ed68..0f2affd01df 100644
--- a/drivers/misc/sgi-xp/xpc.h
+++ b/drivers/misc/sgi-xp/xpc.h
@@ -210,7 +210,7 @@ xpc_disallow_hb(short partid, struct xpc_vars *vars)
  * 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_ACTIVATE_IRQ_AMOS      (XPC_NOTIFY_IRQ_AMOS + XP_MAX_NPARTITIONS_SN2)
 #define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
 #define XPC_DISENGAGE_REQUEST_AMO  (XPC_ENGAGED_PARTITIONS_AMO + 1)
 
@@ -285,7 +285,7 @@ struct xpc_gp {
 };
 
 #define XPC_GP_SIZE \
-                L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_NCHANNELS)
+                L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_MAX_NCHANNELS)
 
 /*
  * Define a structure that contains arguments associated with opening and
@@ -300,7 +300,8 @@ struct xpc_openclose_args {
 };
 
 #define XPC_OPENCLOSE_ARGS_SIZE \
-              L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * XPC_NCHANNELS)
+              L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * \
+              XPC_MAX_NCHANNELS)
 
 /* struct xpc_msg flags */
 
@@ -637,7 +638,7 @@ 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 struct xpc_partition *xpc_partitions;
 extern char *xpc_remote_copy_buffer;
 extern void *xpc_remote_copy_buffer_base;
 extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);
@@ -1104,13 +1105,14 @@ xpc_IPI_send_local_msgrequest(struct xpc_channel *ch)
 }
 
 /*
+>>> this block comment needs to be moved and re-written.
  * 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
+ * AMO variables (based on xp_max_npartitions) 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.
  */
diff --git a/drivers/misc/sgi-xp/xpc_channel.c b/drivers/misc/sgi-xp/xpc_channel.c
index 9c90c2d55c0..12d8eb6957a 100644
--- a/drivers/misc/sgi-xp/xpc_channel.c
+++ b/drivers/misc/sgi-xp/xpc_channel.c
@@ -110,14 +110,14 @@ xpc_setup_infrastructure(struct xpc_partition *part)
          * Allocate all of the channel structures as a contiguous chunk of
          * memory.
          */
-        part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_NCHANNELS,
+        part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_MAX_NCHANNELS,
                                  GFP_KERNEL);
         if (part->channels == NULL) {
                 dev_err(xpc_chan, "can't get memory for channels\n");
                 return xpNoMemory;
         }
 
-        part->nchannels = XPC_NCHANNELS;
+        part->nchannels = XPC_MAX_NCHANNELS;
 
         /* allocate all the required GET/PUT values */
 
@@ -1432,9 +1432,9 @@ xpc_initiate_connect(int ch_number)
         struct xpc_partition *part;
         struct xpc_channel *ch;
 
-        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_MAX_NCHANNELS);
 
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
                 part = &xpc_partitions[partid];
 
                 if (xpc_part_ref(part)) {
@@ -1488,10 +1488,10 @@ xpc_initiate_disconnect(int ch_number)
         struct xpc_partition *part;
         struct xpc_channel *ch;
 
-        DBUG_ON(ch_number < 0 || ch_number >= XPC_NCHANNELS);
+        DBUG_ON(ch_number < 0 || ch_number >= XPC_MAX_NCHANNELS);
 
         /* initiate the channel disconnect for every active partition */
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
                 part = &xpc_partitions[partid];
 
                 if (xpc_part_ref(part)) {
@@ -1734,7 +1734,7 @@ xpc_initiate_allocate(short partid, int ch_number, u32 flags, void **payload)
         enum xp_retval ret = xpUnknownReason;
         struct xpc_msg *msg = NULL;
 
-        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
         DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
 
         *payload = NULL;
@@ -1918,7 +1918,7 @@ xpc_initiate_send(short partid, int ch_number, void *payload)
         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(partid < 0 || partid >= xp_max_npartitions);
         DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
         DBUG_ON(msg == NULL);
 
@@ -1968,7 +1968,7 @@ xpc_initiate_send_notify(short partid, int ch_number, void *payload,
         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(partid < 0 || partid >= xp_max_npartitions);
         DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
         DBUG_ON(msg == NULL);
         DBUG_ON(func == NULL);
@@ -2210,7 +2210,7 @@ xpc_initiate_received(short partid, int ch_number, void *payload)
         struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
         s64 get, msg_number = msg->number;
 
-        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
         DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
 
         ch = &part->channels[ch_number];
diff --git a/drivers/misc/sgi-xp/xpc_main.c b/drivers/misc/sgi-xp/xpc_main.c
index c3b4227f48a..a05c7c7da22 100644
--- a/drivers/misc/sgi-xp/xpc_main.c
+++ b/drivers/misc/sgi-xp/xpc_main.c
@@ -433,7 +433,7 @@ xpc_activating(void *__partid)
         struct xpc_partition *part = &xpc_partitions[partid];
         unsigned long irq_flags;
 
-        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
 
         spin_lock_irqsave(&part->act_lock, irq_flags);
 
@@ -544,7 +544,7 @@ xpc_notify_IRQ_handler(int irq, void *dev_id)
         short partid = (short)(u64)dev_id;
         struct xpc_partition *part = &xpc_partitions[partid];
 
-        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
 
         if (xpc_part_ref(part)) {
                 xpc_check_for_channel_activity(part);
@@ -815,7 +815,7 @@ xpc_disconnect_wait(int ch_number)
         int wakeup_channel_mgr;
 
         /* now wait for all callouts to the caller's function to cease */
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
                 part = &xpc_partitions[partid];
 
                 if (!xpc_part_ref(part))
@@ -895,7 +895,7 @@ xpc_do_exit(enum xp_retval reason)
         do {
                 active_part_count = 0;
 
-                for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                for (partid = 0; partid < xp_max_npartitions; partid++) {
                         part = &xpc_partitions[partid];
 
                         if (xpc_partition_disengaged(part) &&
@@ -956,11 +956,8 @@ xpc_do_exit(enum xp_retval reason)
         DBUG_ON(xpc_vars->heartbeating_to_mask != 0);
 
         if (reason == xpUnloading) {
-                /* 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);
+                (void)unregister_reboot_notifier(&xpc_reboot_notifier);
         }
 
         /* close down protections for IPI operations */
@@ -972,6 +969,7 @@ xpc_do_exit(enum xp_retval reason)
         if (xpc_sysctl)
                 unregister_sysctl_table(xpc_sysctl);
 
+        kfree(xpc_partitions);
         kfree(xpc_remote_copy_buffer_base);
 }
 
@@ -1017,7 +1015,7 @@ xpc_die_disengage(void)
 
         xpc_vars->heartbeating_to_mask = 0;     /* indicate we're deactivated */
 
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
                 part = &xpc_partitions[partid];
 
                 if (!XPC_SUPPORTS_DISENGAGE_REQUEST(part->
@@ -1053,7 +1051,8 @@ xpc_die_disengage(void)
 
                 time = rtc_time();
                 if (time >= disengage_request_timeout) {
-                        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+                        for (partid = 0; partid < xp_max_npartitions;
+                             partid++) {
                                 if (engaged & (1UL << partid)) {
                                         dev_info(xpc_part, "disengage from "
                                                  "remote partition %d timed "
@@ -1132,18 +1131,26 @@ xpc_init(void)
         if (!ia64_platform_is("sn2"))
                 return -ENODEV;
 
+        snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part");
+        snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan");
+
         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)
+        if (xpc_remote_copy_buffer == NULL) {
+                dev_err(xpc_part, "can't get memory for remote copy buffer\n");
                 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);
+        xpc_partitions = kzalloc(sizeof(struct xpc_partition) *
+                                 xp_max_npartitions, GFP_KERNEL);
+        if (xpc_partitions == NULL) {
+                dev_err(xpc_part, "can't get memory for partition structure\n");
+                ret = -ENOMEM;
+                goto out_1;
+        }
 
         /*
          * The first few fields of each entry of xpc_partitions[] need to
@@ -1153,7 +1160,7 @@ xpc_init(void)
          * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING
          * PARTITION HAS BEEN ACTIVATED.
          */
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
                 part = &xpc_partitions[partid];
 
                 DBUG_ON((u64)part != L1_CACHE_ALIGN((u64)part));
@@ -1173,6 +1180,8 @@ xpc_init(void)
                 atomic_set(&part->references, 0);
         }
 
+        xpc_sysctl = register_sysctl_table(xpc_sys_dir);
+
         /*
          * Open up protections for IPI operations (and AMO operations on
          * Shub 1.1 systems).
@@ -1196,14 +1205,8 @@ xpc_init(void)
         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;
+                ret = -EBUSY;
+                goto out_2;
         }
 
         /*
@@ -1213,16 +1216,9 @@ xpc_init(void)
          */
         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;
+                dev_err(xpc_part, "can't setup our reserved page\n");
+                ret = -EBUSY;
+                goto out_3;
         }
 
         /* add ourselves to the reboot_notifier_list */
@@ -1245,25 +1241,8 @@ xpc_init(void)
         kthread = kthread_run(xpc_hb_checker, NULL, XPC_HB_CHECK_THREAD_NAME);
         if (IS_ERR(kthread)) {
                 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;
+                ret = -EBUSY;
+                goto out_4;
         }
 
         /*
@@ -1290,6 +1269,24 @@ xpc_init(void)
                           xpc_initiate_partid_to_nasids);
 
         return 0;
+
+        /* initialization was not successful */
+out_4:
+        /* indicate to others that our reserved page is uninitialized */
+        xpc_rsvd_page->vars_pa = 0;
+        del_timer_sync(&xpc_hb_timer);
+        (void)unregister_die_notifier(&xpc_die_notifier);
+        (void)unregister_reboot_notifier(&xpc_reboot_notifier);
+out_3:
+        free_irq(SGI_XPC_ACTIVATE, NULL);
+out_2:
+        xpc_restrict_IPI_ops();
+        if (xpc_sysctl)
+                unregister_sysctl_table(xpc_sysctl);
+        kfree(xpc_partitions);
+out_1:
+        kfree(xpc_remote_copy_buffer_base);
+        return ret;
 }
 
 module_init(xpc_init);
diff --git a/drivers/misc/sgi-xp/xpc_partition.c b/drivers/misc/sgi-xp/xpc_partition.c
index 7dd4b5812c4..02a858eddd8 100644
--- a/drivers/misc/sgi-xp/xpc_partition.c
+++ b/drivers/misc/sgi-xp/xpc_partition.c
@@ -51,13 +51,7 @@ 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];
+struct xpc_partition *xpc_partitions;
 
 /*
  * Generic buffer used to store a local copy of portions of a remote
@@ -261,7 +255,7 @@ xpc_rsvd_page_init(void)
 
         /* clear xpc_vars_part */
         memset((u64 *)xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
-               XP_MAX_PARTITIONS);
+               xp_max_npartitions);
 
         /* initialize the activate IRQ related AMO variables */
         for (i = 0; i < xp_nasid_mask_words; i++)
@@ -408,7 +402,7 @@ xpc_check_remote_hb(void)
 
         remote_vars = (struct xpc_vars *)xpc_remote_copy_buffer;
 
-        for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+        for (partid = 0; partid < xp_max_npartitions; partid++) {
 
                 if (xpc_exiting)
                         break;
@@ -487,10 +481,8 @@ xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
 
         /* check that the partid is for another partition */
 
-        if (remote_rp->partid < 1 ||
-            remote_rp->partid > (XP_MAX_PARTITIONS - 1)) {
+        if (remote_rp->partid < 0 || remote_rp->partid >= xp_max_npartitions)
                 return xpInvalidPartid;
-        }
 
         if (remote_rp->partid == sn_partition_id)
                 return xpLocalPartid;
diff --git a/drivers/misc/sgi-xp/xpnet.c b/drivers/misc/sgi-xp/xpnet.c
index 822dc8e8d7f..cc252f201b2 100644
--- a/drivers/misc/sgi-xp/xpnet.c
+++ b/drivers/misc/sgi-xp/xpnet.c
@@ -287,7 +287,7 @@ xpnet_connection_activity(enum xp_retval reason, short partid, int channel,
 {
         long bp;
 
-        DBUG_ON(partid <= 0 || partid >= XP_MAX_PARTITIONS);
+        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
         DBUG_ON(channel != XPC_NET_CHANNEL);
 
         switch (reason) {
@@ -513,7 +513,7 @@ xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
         /*
          * Main send loop.
          */
-        for (dest_partid = 1; dp && dest_partid < XP_MAX_PARTITIONS;
+        for (dest_partid = 0; dp && dest_partid < xp_max_npartitions;
              dest_partid++) {
 
                 if (!(dp & (1UL << (dest_partid - 1)))) {