Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] minor reformatting to vmlinux.lds.S
  [IA64] CMC/CPE: Reverse the order of fetching log and checking poll threshold
  [IA64] PAL calls need physical mode, stacked
  [IA64] ar.fpsr not set on MCA/INIT kernel entry
  [IA64] printing support for MCA/INIT
  [IA64] trim output of show_mem()
  [IA64] show_mem() printk levels
  [IA64] Make gp value point to Region 5 in mca handler
  Revert "[IA64] Unwire set/get_robust_list"
  [IA64] Implement futex primitives
  [IA64-SGI] Do not request DMA memory for BTE
  [IA64] Move perfmon tables from thread_struct to pfm_context
  [IA64] Add interface so modules can discover whether multithreading is on.
  [IA64] kprobes: fixup the pagefault exception caused by probehandlers
  [IA64] kprobe opcode 16 bytes alignment on IA64
  [IA64] esi-support
  [IA64] Add "model name" to /proc/cpuinfo

19 files changed, 745 insertions, 177 deletions

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index d5ee4fc8fe6..0b7f701d5cf 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -423,6 +423,14 @@ config IA64_PALINFO
 config SGI_SN
 	def_bool y if (IA64_SGI_SN2 || IA64_GENERIC)
 
+config IA64_ESI
+	bool "ESI (Extensible SAL Interface) support"
+	help
+	  If you say Y here, support is built into the kernel to
+	  make ESI calls.  ESI calls are used to support vendor-specific
+	  firmware extensions, such as the ability to inject memory-errors
+	  for test-purposes.  If you're unsure, say N.
+
 source "drivers/sn/Kconfig"
 
 source "drivers/firmware/Kconfig"
diff --git a/arch/ia64/kernel/Makefile b/arch/ia64/kernel/Makefile
index ad8215a3c58..31497496eb4 100644
--- a/arch/ia64/kernel/Makefile
+++ b/arch/ia64/kernel/Makefile
@@ -32,6 +32,11 @@ obj-$(CONFIG_IA64_UNCACHED_ALLOCATOR)	+= uncached.o
 obj-$(CONFIG_AUDIT)		+= audit.o
 mca_recovery-y			+= mca_drv.o mca_drv_asm.o
 
+obj-$(CONFIG_IA64_ESI)		+= esi.o
+ifneq ($(CONFIG_IA64_ESI),)
+obj-y				+= esi_stub.o	# must be in kernel proper
+endif
+
 # The gate DSO image is built using a special linker script.
 targets += gate.so gate-syms.o
 
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index fef06571be9..12701cf32d9 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -1605,8 +1605,8 @@ sys_call_table:
 	data8 sys_ni_syscall			// 1295 reserved for ppoll
 	data8 sys_unshare
 	data8 sys_splice
-	data8 sys_ni_syscall			// reserved for set_robust_list
-	data8 sys_ni_syscall			// reserved for get_robust_list
+	data8 sys_set_robust_list
+	data8 sys_get_robust_list
 	data8 sys_sync_file_range		// 1300
 	data8 sys_tee
 	data8 sys_vmsplice
diff --git a/arch/ia64/kernel/esi.c b/arch/ia64/kernel/esi.c
new file mode 100644
index 00000000000..ebf4e988e78
--- /dev/null
+++ b/arch/ia64/kernel/esi.c
@@ -0,0 +1,205 @@
+/*
+ * Extensible SAL Interface (ESI) support routines.
+ *
+ * Copyright (C) 2006 Hewlett-Packard Co
+ * 	Alex Williamson <alex.williamson@hp.com>
+ */
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/string.h>
+
+#include <asm/esi.h>
+#include <asm/sal.h>
+
+MODULE_AUTHOR("Alex Williamson <alex.williamson@hp.com>");
+MODULE_DESCRIPTION("Extensible SAL Interface (ESI) support");
+MODULE_LICENSE("GPL");
+
+#define MODULE_NAME	"esi"
+
+#define ESI_TABLE_GUID					\
+    EFI_GUID(0x43EA58DC, 0xCF28, 0x4b06, 0xB3,		\
+	     0x91, 0xB7, 0x50, 0x59, 0x34, 0x2B, 0xD4)
+
+enum esi_systab_entry_type {
+	ESI_DESC_ENTRY_POINT = 0
+};
+
+/*
+ * Entry type:	Size:
+ *	0	48
+ */
+#define ESI_DESC_SIZE(type)	"\060"[(unsigned) (type)]
+
+typedef struct ia64_esi_desc_entry_point {
+	u8 type;
+	u8 reserved1[15];
+	u64 esi_proc;
+	u64 gp;
+	efi_guid_t guid;
+} ia64_esi_desc_entry_point_t;
+
+struct pdesc {
+	void *addr;
+	void *gp;
+};
+
+static struct ia64_sal_systab *esi_systab;
+
+static int __init esi_init (void)
+{
+	efi_config_table_t *config_tables;
+	struct ia64_sal_systab *systab;
+	unsigned long esi = 0;
+	char *p;
+	int i;
+
+	config_tables = __va(efi.systab->tables);
+
+	for (i = 0; i < (int) efi.systab->nr_tables; ++i) {
+		if (efi_guidcmp(config_tables[i].guid, ESI_TABLE_GUID) == 0) {
+			esi = config_tables[i].table;
+			break;
+		}
+	}
+
+	if (!esi)
+		return -ENODEV;;
+
+	systab = __va(esi);
+
+	if (strncmp(systab->signature, "ESIT", 4) != 0) {
+		printk(KERN_ERR "bad signature in ESI system table!");
+		return -ENODEV;
+	}
+
+	p = (char *) (systab + 1);
+	for (i = 0; i < systab->entry_count; i++) {
+		/*
+		 * The first byte of each entry type contains the type
+		 * descriptor.
+		 */
+		switch (*p) {
+		      case ESI_DESC_ENTRY_POINT:
+			break;
+		      default:
+			printk(KERN_WARNING "Unkown table type %d found in "
+			       "ESI table, ignoring rest of table\n", *p);
+			return -ENODEV;
+		}
+
+		p += ESI_DESC_SIZE(*p);
+	}
+
+	esi_systab = systab;
+	return 0;
+}
+
+
+int ia64_esi_call (efi_guid_t guid, struct ia64_sal_retval *isrvp,
+		   enum esi_proc_type proc_type, u64 func,
+		   u64 arg1, u64 arg2, u64 arg3, u64 arg4, u64 arg5, u64 arg6,
+		   u64 arg7)
+{
+	struct ia64_fpreg fr[6];
+	unsigned long flags = 0;
+	int i;
+	char *p;
+
+	if (!esi_systab)
+		return -1;
+
+	p = (char *) (esi_systab + 1);
+	for (i = 0; i < esi_systab->entry_count; i++) {
+		if (*p == ESI_DESC_ENTRY_POINT) {
+			ia64_esi_desc_entry_point_t *esi = (void *)p;
+			if (!efi_guidcmp(guid, esi->guid)) {
+				ia64_sal_handler esi_proc;
+				struct pdesc pdesc;
+
+				pdesc.addr = __va(esi->esi_proc);
+				pdesc.gp = __va(esi->gp);
+
+				esi_proc = (ia64_sal_handler) &pdesc;
+
+				ia64_save_scratch_fpregs(fr);
+				if (proc_type == ESI_PROC_SERIALIZED)
+					spin_lock_irqsave(&sal_lock, flags);
+				else if (proc_type == ESI_PROC_MP_SAFE)
+					local_irq_save(flags);
+				else
+					preempt_disable();
+				*isrvp = (*esi_proc)(func, arg1, arg2, arg3,
+						     arg4, arg5, arg6, arg7);
+				if (proc_type == ESI_PROC_SERIALIZED)
+					spin_unlock_irqrestore(&sal_lock,
+							       flags);
+				else if (proc_type == ESI_PROC_MP_SAFE)
+					local_irq_restore(flags);
+				else
+					preempt_enable();
+				ia64_load_scratch_fpregs(fr);
+				return 0;
+			}
+		}
+		p += ESI_DESC_SIZE(*p);
+	}
+	return -1;
+}
+EXPORT_SYMBOL_GPL(ia64_esi_call);
+
+int ia64_esi_call_phys (efi_guid_t guid, struct ia64_sal_retval *isrvp,
+			u64 func, u64 arg1, u64 arg2, u64 arg3, u64 arg4,
+			u64 arg5, u64 arg6, u64 arg7)
+{
+	struct ia64_fpreg fr[6];
+	unsigned long flags;
+	u64 esi_params[8];
+	char *p;
+	int i;
+
+	if (!esi_systab)
+		return -1;
+
+	p = (char *) (esi_systab + 1);
+	for (i = 0; i < esi_systab->entry_count; i++) {
+		if (*p == ESI_DESC_ENTRY_POINT) {
+			ia64_esi_desc_entry_point_t *esi = (void *)p;
+			if (!efi_guidcmp(guid, esi->guid)) {
+				ia64_sal_handler esi_proc;
+				struct pdesc pdesc;
+
+				pdesc.addr = (void *)esi->esi_proc;
+				pdesc.gp = (void *)esi->gp;
+
+				esi_proc = (ia64_sal_handler) &pdesc;
+
+				esi_params[0] = func;
+				esi_params[1] = arg1;
+				esi_params[2] = arg2;
+				esi_params[3] = arg3;
+				esi_params[4] = arg4;
+				esi_params[5] = arg5;
+				esi_params[6] = arg6;
+				esi_params[7] = arg7;
+				ia64_save_scratch_fpregs(fr);
+				spin_lock_irqsave(&sal_lock, flags);
+				*isrvp = esi_call_phys(esi_proc, esi_params);
+				spin_unlock_irqrestore(&sal_lock, flags);
+				ia64_load_scratch_fpregs(fr);
+				return 0;
+			}
+		}
+		p += ESI_DESC_SIZE(*p);
+	}
+	return -1;
+}
+EXPORT_SYMBOL_GPL(ia64_esi_call_phys);
+
+static void __exit esi_exit (void)
+{
+}
+
+module_init(esi_init);
+module_exit(esi_exit);	/* makes module removable... */
diff --git a/arch/ia64/kernel/esi_stub.S b/arch/ia64/kernel/esi_stub.S
new file mode 100644
index 00000000000..6b3d6c1f99b
--- /dev/null
+++ b/arch/ia64/kernel/esi_stub.S
@@ -0,0 +1,96 @@
+/*
+ * ESI call stub.
+ *
+ * Copyright (C) 2005 Hewlett-Packard Co
+ *	Alex Williamson <alex.williamson@hp.com>
+ *
+ * Based on EFI call stub by David Mosberger.  The stub is virtually
+ * identical to the one for EFI phys-mode calls, except that ESI
+ * calls may have up to 8 arguments, so they get passed to this routine
+ * through memory.
+ *
+ * This stub allows us to make ESI calls in physical mode with interrupts
+ * turned off.  ESI calls may not support calling from virtual mode.
+ *
+ * Google for "Extensible SAL specification" for a document describing the
+ * ESI standard.
+ */
+
+/*
+ * PSR settings as per SAL spec (Chapter 8 in the "IA-64 System
+ * Abstraction Layer Specification", revision 2.6e).  Note that
+ * psr.dfl and psr.dfh MUST be cleared, despite what this manual says.
+ * Otherwise, SAL dies whenever it's trying to do an IA-32 BIOS call
+ * (the br.ia instruction fails unless psr.dfl and psr.dfh are
+ * cleared).  Fortunately, SAL promises not to touch the floating
+ * point regs, so at least we don't have to save f2-f127.
+ */
+#define PSR_BITS_TO_CLEAR						\
+	(IA64_PSR_I | IA64_PSR_IT | IA64_PSR_DT | IA64_PSR_RT |		\
+	 IA64_PSR_DD | IA64_PSR_SS | IA64_PSR_RI | IA64_PSR_ED |	\
+	 IA64_PSR_DFL | IA64_PSR_DFH)
+
+#define PSR_BITS_TO_SET							\
+	(IA64_PSR_BN)
+
+#include <asm/processor.h>
+#include <asm/asmmacro.h>
+
+/*
+ * Inputs:
+ *	in0 = address of function descriptor of ESI routine to call
+ *	in1 = address of array of ESI parameters
+ *
+ * Outputs:
+ *	r8 = result returned by called function
+ */
+GLOBAL_ENTRY(esi_call_phys)
+	.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(2)
+	alloc loc1=ar.pfs,2,7,8,0
+	ld8 r2=[in0],8			// load ESI function's entry point
+	mov loc0=rp
+	.body
+	;;
+	ld8 out0=[in1],8		// ESI params loaded from array
+	;;				// passing all as inputs doesn't work
+	ld8 out1=[in1],8
+	;;
+	ld8 out2=[in1],8
+	;;
+	ld8 out3=[in1],8
+	;;
+	ld8 out4=[in1],8
+	;;
+	ld8 out5=[in1],8
+	;;
+	ld8 out6=[in1],8
+	;;
+	ld8 out7=[in1]
+	mov loc2=gp			// save global pointer
+	mov loc4=ar.rsc			// save RSE configuration
+	mov ar.rsc=0			// put RSE in enforced lazy, LE mode
+	;;
+	ld8 gp=[in0]			// load ESI function's global pointer
+	movl r16=PSR_BITS_TO_CLEAR
+	mov loc3=psr			// save processor status word
+	movl r17=PSR_BITS_TO_SET
+	;;
+	or loc3=loc3,r17
+	mov b6=r2
+	;;
+	andcm r16=loc3,r16	// get psr with IT, DT, and RT bits cleared
+	br.call.sptk.many rp=ia64_switch_mode_phys
+.ret0:	mov loc5=r19			// old ar.bsp
+	mov loc6=r20			// old sp
+	br.call.sptk.many rp=b6		// call the ESI function
+.ret1:	mov ar.rsc=0			// put RSE in enforced lazy, LE mode
+	mov r16=loc3			// save virtual mode psr
+	mov r19=loc5			// save virtual mode bspstore
+	mov r20=loc6			// save virtual mode sp
+	br.call.sptk.many rp=ia64_switch_mode_virt // return to virtual mode
+.ret2:	mov ar.rsc=loc4			// restore RSE configuration
+	mov ar.pfs=loc1
+	mov rp=loc0
+	mov gp=loc2
+	br.ret.sptk.many rp
+END(esi_call_phys)
diff --git a/arch/ia64/kernel/ia64_ksyms.c b/arch/ia64/kernel/ia64_ksyms.c
index 3ead20fb6f4..879c1817bd1 100644
--- a/arch/ia64/kernel/ia64_ksyms.c
+++ b/arch/ia64/kernel/ia64_ksyms.c
@@ -105,5 +105,9 @@ EXPORT_SYMBOL(ia64_spinlock_contention);
 # endif
 #endif
 
+#if defined(CONFIG_IA64_ESI) || defined(CONFIG_IA64_ESI_MODULE)
+extern void esi_call_phys (void);
+EXPORT_SYMBOL_GPL(esi_call_phys);
+#endif
 extern char ia64_ivt[];
 EXPORT_SYMBOL(ia64_ivt);
diff --git a/arch/ia64/kernel/kprobes.c b/arch/ia64/kernel/kprobes.c
index 781960f80b6..169ec3a7156 100644
--- a/arch/ia64/kernel/kprobes.c
+++ b/arch/ia64/kernel/kprobes.c
@@ -136,10 +136,8 @@ static void __kprobes update_kprobe_inst_flag(uint template, uint  slot,
 static int __kprobes unsupported_inst(uint template, uint  slot,
 				      uint major_opcode,
 				      unsigned long kprobe_inst,
-				      struct kprobe *p)
+				      unsigned long addr)
 {
-	unsigned long addr = (unsigned long)p->addr;
-
 	if (bundle_encoding[template][slot] == I) {
 		switch (major_opcode) {
 			case 0x0: //I_UNIT_MISC_OPCODE:
@@ -217,7 +215,7 @@ static void __kprobes prepare_break_inst(uint template, uint  slot,
 					 struct kprobe *p)
 {
 	unsigned long break_inst = BREAK_INST;
-	bundle_t *bundle = &p->ainsn.insn.bundle;
+	bundle_t *bundle = &p->opcode.bundle;
 
 	/*
 	 * Copy the original kprobe_inst qualifying predicate(qp)
@@ -423,11 +421,9 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
 	unsigned long *kprobe_addr = (unsigned long *)(addr & ~0xFULL);
 	unsigned long kprobe_inst=0;
 	unsigned int slot = addr & 0xf, template, major_opcode = 0;
-	bundle_t *bundle = &p->ainsn.insn.bundle;
-
-	memcpy(&p->opcode.bundle, kprobe_addr, sizeof(bundle_t));
-	memcpy(&p->ainsn.insn.bundle, kprobe_addr, sizeof(bundle_t));
+	bundle_t *bundle;
 
+	bundle = &((kprobe_opcode_t *)kprobe_addr)->bundle;
  	template = bundle->quad0.template;
 
 	if(valid_kprobe_addr(template, slot, addr))
@@ -440,20 +436,19 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
 	/* Get kprobe_inst and major_opcode from the bundle */
 	get_kprobe_inst(bundle, slot, &kprobe_inst, &major_opcode);
 
-	if (unsupported_inst(template, slot, major_opcode, kprobe_inst, p))
+	if (unsupported_inst(template, slot, major_opcode, kprobe_inst, addr))
 			return -EINVAL;
 
-	prepare_break_inst(template, slot, major_opcode, kprobe_inst, p);
 
-	return 0;
-}
+	p->ainsn.insn = get_insn_slot();
+	if (!p->ainsn.insn)
+		return -ENOMEM;
+	memcpy(&p->opcode, kprobe_addr, sizeof(kprobe_opcode_t));
+	memcpy(p->ainsn.insn, kprobe_addr, sizeof(kprobe_opcode_t));
 
-void __kprobes flush_insn_slot(struct kprobe *p)
-{
-	unsigned long arm_addr;
+	prepare_break_inst(template, slot, major_opcode, kprobe_inst, p);
 
-	arm_addr = ((unsigned long)&p->opcode.bundle) & ~0xFULL;
-	flush_icache_range(arm_addr, arm_addr + sizeof(bundle_t));
+	return 0;
 }
 
 void __kprobes arch_arm_kprobe(struct kprobe *p)
@@ -461,9 +456,10 @@ void __kprobes arch_arm_kprobe(struct kprobe *p)
 	unsigned long addr = (unsigned long)p->addr;
 	unsigned long arm_addr = addr & ~0xFULL;
 
-	flush_insn_slot(p);
-	memcpy((char *)arm_addr, &p->ainsn.insn.bundle, sizeof(bundle_t));
-	flush_icache_range(arm_addr, arm_addr + sizeof(bundle_t));
+	flush_icache_range((unsigned long)p->ainsn.insn,
+			(unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t));
+	memcpy((char *)arm_addr, &p->opcode, sizeof(kprobe_opcode_t));
+	flush_icache_range(arm_addr, arm_addr + sizeof(kprobe_opcode_t));
 }
 
 void __kprobes arch_disarm_kprobe(struct kprobe *p)
@@ -471,11 +467,18 @@ void __kprobes arch_disarm_kprobe(struct kprobe *p)
 	unsigned long addr = (unsigned long)p->addr;
 	unsigned long arm_addr = addr & ~0xFULL;
 
-	/* p->opcode contains the original unaltered bundle */
-	memcpy((char *) arm_addr, (char *) &p->opcode.bundle, sizeof(bundle_t));
-	flush_icache_range(arm_addr, arm_addr + sizeof(bundle_t));
+	/* p->ainsn.insn contains the original unaltered kprobe_opcode_t */
+	memcpy((char *) arm_addr, (char *) p->ainsn.insn,
+					 sizeof(kprobe_opcode_t));
+	flush_icache_range(arm_addr, arm_addr + sizeof(kprobe_opcode_t));
 }
 
+void __kprobes arch_remove_kprobe(struct kprobe *p)
+{
+	mutex_lock(&kprobe_mutex);
+	free_insn_slot(p->ainsn.insn);
+	mutex_unlock(&kprobe_mutex);
+}
 /*
  * We are resuming execution after a single step fault, so the pt_regs
  * structure reflects the register state after we executed the instruction
@@ -486,12 +489,12 @@ void __kprobes arch_disarm_kprobe(struct kprobe *p)
  */
 static void __kprobes resume_execution(struct kprobe *p, struct pt_regs *regs)
 {
-  	unsigned long bundle_addr = ((unsigned long) (&p->opcode.bundle)) & ~0xFULL;
+  	unsigned long bundle_addr = (unsigned long) (&p->ainsn.insn->bundle);
   	unsigned long resume_addr = (unsigned long)p->addr & ~0xFULL;
  	unsigned long template;
  	int slot = ((unsigned long)p->addr & 0xf);
 
-	template = p->opcode.bundle.quad0.template;
+	template = p->ainsn.insn->bundle.quad0.template;
 
  	if (slot == 1 && bundle_encoding[template][1] == L)
  		slot = 2;
@@ -553,7 +556,7 @@ turn_ss_off:
 
 static void __kprobes prepare_ss(struct kprobe *p, struct pt_regs *regs)
 {
-	unsigned long bundle_addr = (unsigned long) &p->opcode.bundle;
+	unsigned long bundle_addr = (unsigned long) &p->ainsn.insn->bundle;
 	unsigned long slot = (unsigned long)p->addr & 0xf;
 
 	/* single step inline if break instruction */
@@ -768,6 +771,12 @@ static int __kprobes kprobes_fault_handler(struct pt_regs *regs, int trapnr)
 		 */
 		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
 			return 1;
+		/*
+		 * In case the user-specified fault handler returned
+		 * zero, try to fix up.
+		 */
+		if (ia64_done_with_exception(regs))
+			return 1;
 
 		/*
 		 * Let ia64_do_page_fault() fix it.
diff --git a/arch/ia64/kernel/mca.c b/arch/ia64/kernel/mca.c
index 2fbe4536fe1..bfbd8986153 100644
--- a/arch/ia64/kernel/mca.c
+++ b/arch/ia64/kernel/mca.c
@@ -54,6 +54,9 @@
  *
  * 2005-10-07 Keith Owens <kaos@sgi.com>
  *	      Add notify_die() hooks.
+ *
+ * 2006-09-15 Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
+ * 	      Add printing support for MCA/INIT.
  */
 #include <linux/types.h>
 #include <linux/init.h>
@@ -136,11 +139,175 @@ extern void salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe);
 
 static int mca_init __initdata;
 
+/*
+ * limited & delayed printing support for MCA/INIT handler
+ */
+
+#define mprintk(fmt...) ia64_mca_printk(fmt)
+
+#define MLOGBUF_SIZE (512+256*NR_CPUS)
+#define MLOGBUF_MSGMAX 256
+static char mlogbuf[MLOGBUF_SIZE];
+static DEFINE_SPINLOCK(mlogbuf_wlock);	/* mca context only */
+static DEFINE_SPINLOCK(mlogbuf_rlock);	/* normal context only */
+static unsigned long mlogbuf_start;
+static unsigned long mlogbuf_end;
+static unsigned int mlogbuf_finished = 0;
+static unsigned long mlogbuf_timestamp = 0;
+
+static int loglevel_save = -1;
+#define BREAK_LOGLEVEL(__console_loglevel)		\
+	oops_in_progress = 1;				\
+	if (loglevel_save < 0)				\
+		loglevel_save = __console_loglevel;	\
+	__console_loglevel = 15;
+
+#define RESTORE_LOGLEVEL(__console_loglevel)		\
+	if (loglevel_save >= 0) {			\
+		__console_loglevel = loglevel_save;	\
+		loglevel_save = -1;			\
+	}						\
+	mlogbuf_finished = 0;				\
+	oops_in_progress = 0;
+
+/*
+ * Push messages into buffer, print them later if not urgent.
+ */
+void ia64_mca_printk(const char *fmt, ...)
+{
+	va_list args;
+	int printed_len;
+	char temp_buf[MLOGBUF_MSGMAX];
+	char *p;
+
+	va_start(args, fmt);
+	printed_len = vscnprintf(temp_buf, sizeof(temp_buf), fmt, args);
+	va_end(args);
+
+	/* Copy the output into mlogbuf */
+	if (oops_in_progress) {
+		/* mlogbuf was abandoned, use printk directly instead. */
+		printk(temp_buf);
+	} else {
+		spin_lock(&mlogbuf_wlock);
+		for (p = temp_buf; *p; p++) {
+			unsigned long next = (mlogbuf_end + 1) % MLOGBUF_SIZE;
+			if (next != mlogbuf_start) {
+				mlogbuf[mlogbuf_end] = *p;
+				mlogbuf_end = next;
+			} else {
+				/* buffer full */
+				break;
+			}
+		}
+		mlogbuf[mlogbuf_end] = '\0';
+		spin_unlock(&mlogbuf_wlock);
+	}
+}
+EXPORT_SYMBOL(ia64_mca_printk);
+
+/*
+ * Print buffered messages.
+ *  NOTE: call this after returning normal context. (ex. from salinfod)
+ */
+void ia64_mlogbuf_dump(void)
+{
+	char temp_buf[MLOGBUF_MSGMAX];
+	char *p;
+	unsigned long index;
+	unsigned long flags;
+	unsigned int printed_len;
+
+	/* Get output from mlogbuf */
+	while (mlogbuf_start != mlogbuf_end) {
+		temp_buf[0] = '\0';
+		p = temp_buf;
+		printed_len = 0;
+
+		spin_lock_irqsave(&mlogbuf_rlock, flags);
+
+		index = mlogbuf_start;
+		while (index != mlogbuf_end) {
+			*p = mlogbuf[index];
+			index = (index + 1) % MLOGBUF_SIZE;
+			if (!*p)
+				break;
+			p++;
+			if (++printed_len >= MLOGBUF_MSGMAX - 1)
+				break;
+		}
+		*p = '\0';
+		if (temp_buf[0])
+			printk(temp_buf);
+		mlogbuf_start = index;
+
+		mlogbuf_timestamp = 0;
+		spin_unlock_irqrestore(&mlogbuf_rlock, flags);
+	}
+}
+EXPORT_SYMBOL(ia64_mlogbuf_dump);
+
+/*
+ * Call this if system is going to down or if immediate flushing messages to
+ * console is required. (ex. recovery was failed, crash dump is going to be
+ * invoked, long-wait rendezvous etc.)
+ *  NOTE: this should be called from monarch.
+ */
+static void ia64_mlogbuf_finish(int wait)
+{
+	BREAK_LOGLEVEL(console_loglevel);
+
+	spin_lock_init(&mlogbuf_rlock);
+	ia64_mlogbuf_dump();
+	printk(KERN_EMERG "mlogbuf_finish: printing switched to urgent mode, "
+		"MCA/INIT might be dodgy or fail.\n");
+
+	if (!wait)
+		return;
+
+	/* wait for console */
+	printk("Delaying for 5 seconds...\n");
+	udelay(5*1000000);
+
+	mlogbuf_finished = 1;
+}
+EXPORT_SYMBOL(ia64_mlogbuf_finish);
+
+/*
+ * Print buffered messages from INIT context.
+ */
+static void ia64_mlogbuf_dump_from_init(void)
+{
+	if (mlogbuf_finished)
+		return;
+
+	if (mlogbuf_timestamp && (mlogbuf_timestamp + 30*HZ > jiffies)) {
+		printk(KERN_ERR "INIT: mlogbuf_dump is interrupted by INIT "
+			" and the system seems to be messed up.\n");
+		ia64_mlogbuf_finish(0);
+		return;
+	}
+
+	if (!spin_trylock(&mlogbuf_rlock)) {
+		printk(KERN_ERR "INIT: mlogbuf_dump is interrupted by INIT. "
+			"Generated messages other than stack dump will be "
+			"buffered to mlogbuf and will be printed later.\n");
+		printk(KERN_ERR "INIT: If messages would not printed after "
+			"this INIT, wait 30sec and assert INIT again.\n");
+		if (!mlogbuf_timestamp)
+			mlogbuf_timestamp = jiffies;
+		return;
+	}
+	spin_unlock(&mlogbuf_rlock);
+	ia64_mlogbuf_dump();
+}
 
 static void inline
 ia64_mca_spin(const char *func)
 {
-	printk(KERN_EMERG "%s: spinning here, not returning to SAL\n", func);
+	if (monarch_cpu == smp_processor_id())
+		ia64_mlogbuf_finish(0);
+	mprintk(KERN_EMERG "%s: spinning here, not returning to SAL\n", func);
 	while (1)
 		cpu_relax();
 }
@@ -344,9 +511,6 @@ ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
 	/* SAL spec states this should run w/ interrupts enabled */
 	local_irq_enable();
 
-	/* Get the CPE error record and log it */
-	ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CPE);
-
 	spin_lock(&cpe_history_lock);
 	if (!cpe_poll_enabled && cpe_vector >= 0) {
 
@@ -375,7 +539,7 @@ ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
 			mod_timer(&cpe_poll_timer, jiffies + MIN_CPE_POLL_INTERVAL);
 
 			/* lock already released, get out now */
-			return IRQ_HANDLED;
+			goto out;
 		} else {
 			cpe_history[index++] = now;
 			if (index == CPE_HISTORY_LENGTH)
@@ -383,6 +547,10 @@ ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
 		}
 	}
 	spin_unlock(&cpe_history_lock);
+out:
+	/* Get the CPE error record and log it */
+	ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CPE);
+
 	return IRQ_HANDLED;
 }
 
@@ -988,18 +1156,22 @@ ia64_wait_for_slaves(int monarch, const char *type)
 	}
 	if (!missing)
 		goto all_in;
-	printk(KERN_INFO "OS %s slave did not rendezvous on cpu", type);
+	/*
+	 * Maybe slave(s) dead. Print buffered messages immediately.
+	 */
+	ia64_mlogbuf_finish(0);
+	mprintk(KERN_INFO "OS %s slave did not rendezvous on cpu", type);
 	for_each_online_cpu(c) {
 		if (c == monarch)
 			continue;
 		if (ia64_mc_info.imi_rendez_checkin[c] == IA64_MCA_RENDEZ_CHECKIN_NOTDONE)
-			printk(" %d", c);
+			mprintk(" %d", c);
 	}
-	printk("\n");
+	mprintk("\n");
 	return;
 
 all_in:
-	printk(KERN_INFO "All OS %s slaves have reached rendezvous\n", type);
+	mprintk(KERN_INFO "All OS %s slaves have reached rendezvous\n", type);
 	return;
 }
 
@@ -1027,10 +1199,8 @@ ia64_mca_handler(struct pt_regs *regs, struct switch_stack *sw,
 	struct ia64_mca_notify_die nd =
 		{ .sos = sos, .monarch_cpu = &monarch_cpu };
 
-	oops_in_progress = 1;	/* FIXME: make printk NMI/MCA/INIT safe */
-	console_loglevel = 15;	/* make sure printks make it to console */
-	printk(KERN_INFO "Entered OS MCA handler. PSP=%lx cpu=%d monarch=%ld\n",
-		sos->proc_state_param, cpu, sos->monarch);
+	mprintk(KERN_INFO "Entered OS MCA handler. PSP=%lx cpu=%d "
+		"monarch=%ld\n", sos->proc_state_param, cpu, sos->monarch);
 
 	previous_current = ia64_mca_modify_original_stack(regs, sw, sos, "MCA");
 	monarch_cpu = cpu;
@@ -1066,6 +1236,9 @@ ia64_mca_handler(struct pt_regs *regs, struct switch_stack *sw,
 		rh->severity = sal_log_severity_corrected;
 		ia64_sal_clear_state_info(SAL_INFO_TYPE_MCA);
 		sos->os_status = IA64_MCA_CORRECTED;
+	} else {
+		/* Dump buffered message to console */
+		ia64_mlogbuf_finish(1);
 	}
 	if (notify_die(DIE_MCA_MONARCH_LEAVE, "MCA", regs, (long)&nd, 0, recover)
 			== NOTIFY_STOP)
@@ -1106,9 +1279,6 @@ ia64_mca_cmc_int_handler(int cmc_irq, void *arg, struct pt_regs *ptregs)
 	/* SAL spec states this should run w/ interrupts enabled */
 	local_irq_enable();
 
-	/* Get the CMC error record and log it */
-	ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CMC);
-
 	spin_lock(&cmc_history_lock);
 	if (!cmc_polling_enabled) {
 		int i, count = 1; /* we know 1 happened now */
@@ -1141,7 +1311,7 @@ ia64_mca_cmc_int_handler(int cmc_irq, void *arg, struct pt_regs *ptregs)
 			mod_timer(&cmc_poll_timer, jiffies + CMC_POLL_INTERVAL);
 
 			/* lock already released, get out now */
-			return IRQ_HANDLED;
+			goto out;
 		} else {
 			cmc_history[index++] = now;
 			if (index == CMC_HISTORY_LENGTH)
@@ -1149,6 +1319,10 @@ ia64_mca_cmc_int_handler(int cmc_irq, void *arg, struct pt_regs *ptregs)
 		}
 	}
 	spin_unlock(&cmc_history_lock);
+out:
+	/* Get the CMC error record and log it */
+	ia64_mca_log_sal_error_record(SAL_INFO_TYPE_CMC);
+
 	return IRQ_HANDLED;
 }
 
@@ -1305,6 +1479,15 @@ default_monarch_init_process(struct notifier_block *self, unsigned long val, voi
 	struct task_struct *g, *t;
 	if (val != DIE_INIT_MONARCH_PROCESS)
 		return NOTIFY_DONE;
+
+	/*
+	 * FIXME: mlogbuf will brim over with INIT stack dumps.
+	 * To enable show_stack from INIT, we use oops_in_progress which should
+	 * be used in real oops. This would cause something wrong after INIT.
+	 */
+	BREAK_LOGLEVEL(console_loglevel);
+	ia64_mlogbuf_dump_from_init();
+
 	printk(KERN_ERR "Processes interrupted by INIT -");
 	for_each_online_cpu(c) {
 		struct ia64_sal_os_state *s;
@@ -1326,6 +1509,8 @@ default_monarch_init_process(struct notifier_block *self, unsigned long val, voi
 		} while_each_thread (g, t);
 		read_unlock(&tasklist_lock);
 	}
+	/* FIXME: This will not restore zapped printk locks. */
+	RESTORE_LOGLEVEL(console_loglevel);
 	return NOTIFY_DONE;
 }
 
@@ -1357,12 +1542,9 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 	struct ia64_mca_notify_die nd =
 		{ .sos = sos, .monarch_cpu = &monarch_cpu };
 
-	oops_in_progress = 1;	/* FIXME: make printk NMI/MCA/INIT safe */
-	console_loglevel = 15;	/* make sure printks make it to console */
-
 	(void) notify_die(DIE_INIT_ENTER, "INIT", regs, (long)&nd, 0, 0);
 
-	printk(KERN_INFO "Entered OS INIT handler. PSP=%lx cpu=%d monarch=%ld\n",
+	mprintk(KERN_INFO "Entered OS INIT handler. PSP=%lx cpu=%d monarch=%ld\n",
 		sos->proc_state_param, cpu, sos->monarch);
 	salinfo_log_wakeup(SAL_INFO_TYPE_INIT, NULL, 0, 0);
 
@@ -1375,7 +1557,7 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 	 * fix their proms and get their customers updated.
 	 */
 	if (!sos->monarch && atomic_add_return(1, &slaves) == num_online_cpus()) {
-		printk(KERN_WARNING "%s: Promoting cpu %d to monarch.\n",
+		mprintk(KERN_WARNING "%s: Promoting cpu %d to monarch.\n",
 		       __FUNCTION__, cpu);
 		atomic_dec(&slaves);
 		sos->monarch = 1;
@@ -1387,7 +1569,7 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 	 * fix their proms and get their customers updated.
 	 */
 	if (sos->monarch && atomic_add_return(1, &monarchs) > 1) {
-		printk(KERN_WARNING "%s: Demoting cpu %d to slave.\n",
+		mprintk(KERN_WARNING "%s: Demoting cpu %d to slave.\n",
 			       __FUNCTION__, cpu);
 		atomic_dec(&monarchs);
 		sos->monarch = 0;
@@ -1408,7 +1590,7 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 		if (notify_die(DIE_INIT_SLAVE_LEAVE, "INIT", regs, (long)&nd, 0, 0)
 				== NOTIFY_STOP)
 			ia64_mca_spin(__FUNCTION__);
-		printk("Slave on cpu %d returning to normal service.\n", cpu);
+		mprintk("Slave on cpu %d returning to normal service.\n", cpu);
 		set_curr_task(cpu, previous_current);
 		ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
 		atomic_dec(&slaves);
@@ -1426,7 +1608,7 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 	 * same serial line, the user will need some time to switch out of the BMC before
 	 * the dump begins.
 	 */
-	printk("Delaying for 5 seconds...\n");
+	mprintk("Delaying for 5 seconds...\n");
 	udelay(5*1000000);
 	ia64_wait_for_slaves(cpu, "INIT");
 	/* If nobody intercepts DIE_INIT_MONARCH_PROCESS then we drop through
@@ -1439,7 +1621,7 @@ ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
 	if (notify_die(DIE_INIT_MONARCH_LEAVE, "INIT", regs, (long)&nd, 0, 0)
 			== NOTIFY_STOP)
 		ia64_mca_spin(__FUNCTION__);
-	printk("\nINIT dump complete.  Monarch on cpu %d returning to normal service.\n", cpu);
+	mprintk("\nINIT dump complete.  Monarch on cpu %d returning to normal service.\n", cpu);
 	atomic_dec(&monarchs);
 	set_curr_task(cpu, previous_current);
 	monarch_cpu = -1;
diff --git a/arch/ia64/kernel/mca_asm.S b/arch/ia64/kernel/mca_asm.S
index 96047491d1b..c6b607c00de 100644
--- a/arch/ia64/kernel/mca_asm.S
+++ b/arch/ia64/kernel/mca_asm.S
@@ -1025,18 +1025,13 @@ ia64_old_stack:
 
 ia64_set_kernel_registers:
 	add temp3=MCA_SP_OFFSET, r3
-	add temp4=MCA_SOS_OFFSET+SOS(OS_GP), r3
 	mov b0=r2		// save return address