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

21 files changed, 1646 insertions, 1333 deletions

diff --git a/arch/ia64/hp/sim/boot/boot_head.S b/arch/ia64/hp/sim/boot/boot_head.S
index 1c8c7e6a9a5..a9bd71ac78e 100644
--- a/arch/ia64/hp/sim/boot/boot_head.S
+++ b/arch/ia64/hp/sim/boot/boot_head.S
@@ -4,6 +4,7 @@
  */
 
 #include <asm/asmmacro.h>
+#include <asm/pal.h>
 
 	.bss
 	.align 16
@@ -49,7 +50,11 @@ GLOBAL_ENTRY(jmp_to_kernel)
 	br.sptk.few b7
 END(jmp_to_kernel)
 
-
+/*
+ * r28 contains the index of the PAL function
+ * r29--31 the args
+ * Return values in ret0--3 (r8--11)
+ */
 GLOBAL_ENTRY(pal_emulator_static)
 	mov r8=-1
 	mov r9=256
@@ -62,7 +67,7 @@ GLOBAL_ENTRY(pal_emulator_static)
 	cmp.gtu p6,p7=r9,r28
 (p6)	br.cond.sptk.few stacked
 	;;
-static:	cmp.eq p6,p7=6,r28		/* PAL_PTCE_INFO */
+static:	cmp.eq p6,p7=PAL_PTCE_INFO,r28
 (p7)	br.cond.sptk.few 1f
 	;;
 	mov r8=0			/* status = 0 */
@@ -70,21 +75,21 @@ static:	cmp.eq p6,p7=6,r28		/* PAL_PTCE_INFO */
 	movl r10=0x0000000200000003	/* count[0], count[1] */
 	movl r11=0x1000000000002000	/* stride[0], stride[1] */
 	br.cond.sptk.few rp
-1:	cmp.eq p6,p7=14,r28		/* PAL_FREQ_RATIOS */
+1:	cmp.eq p6,p7=PAL_FREQ_RATIOS,r28
 (p7)	br.cond.sptk.few 1f
 	mov r8=0			/* status = 0 */
 	movl r9 =0x100000064		/* proc_ratio (1/100) */
 	movl r10=0x100000100		/* bus_ratio<<32 (1/256) */
 	movl r11=0x100000064		/* itc_ratio<<32 (1/100) */
 	;;
-1:	cmp.eq p6,p7=19,r28		/* PAL_RSE_INFO */
+1:	cmp.eq p6,p7=PAL_RSE_INFO,r28
 (p7)	br.cond.sptk.few 1f
 	mov r8=0			/* status = 0 */
 	mov r9=96			/* num phys stacked */
 	mov r10=0			/* hints */
 	mov r11=0
 	br.cond.sptk.few rp
-1:	cmp.eq p6,p7=1,r28		/* PAL_CACHE_FLUSH */
+1:	cmp.eq p6,p7=PAL_CACHE_FLUSH,r28		/* PAL_CACHE_FLUSH */
 (p7)	br.cond.sptk.few 1f
 	mov r9=ar.lc
 	movl r8=524288			/* flush 512k million cache lines (16MB) */
@@ -102,7 +107,7 @@ static:	cmp.eq p6,p7=6,r28		/* PAL_PTCE_INFO */
 	mov ar.lc=r9
 	mov r8=r0
 	;;
-1:	cmp.eq p6,p7=15,r28		/* PAL_PERF_MON_INFO */
+1:	cmp.eq p6,p7=PAL_PERF_MON_INFO,r28
 (p7)	br.cond.sptk.few 1f
 	mov r8=0			/* status = 0 */
 	movl r9 =0x08122f04		/* generic=4 width=47 retired=8 cycles=18 */
@@ -138,6 +143,20 @@ static:	cmp.eq p6,p7=6,r28		/* PAL_PTCE_INFO */
 	st8 [r29]=r0,16			/* clear remaining bits  */
 	st8 [r18]=r0,16			/* clear remaining bits  */
 	;;
+1:	cmp.eq p6,p7=PAL_VM_SUMMARY,r28
+(p7)	br.cond.sptk.few 1f
+	mov	r8=0			/* status = 0  */
+	movl	r9=0x2044040020F1865	/* num_tc_levels=2, num_unique_tcs=4 */
+					/* max_itr_entry=64, max_dtr_entry=64 */
+					/* hash_tag_id=2, max_pkr=15 */
+					/* key_size=24, phys_add_size=50, vw=1 */
+	movl	r10=0x183C		/* rid_size=24, impl_va_msb=60 */
+	;;
+1:	cmp.eq p6,p7=PAL_MEM_ATTRIB,r28
+(p7)	br.cond.sptk.few 1f
+	mov	r8=0			/* status = 0 */
+	mov	r9=0x80|0x01		/* NatPage|WB */
+	;;
 1:	br.cond.sptk.few rp
 stacked:
 	br.ret.sptk.few rp
diff --git a/arch/ia64/kernel/asm-offsets.c b/arch/ia64/kernel/asm-offsets.c
index 7d1ae2982c5..f6a23428934 100644
--- a/arch/ia64/kernel/asm-offsets.c
+++ b/arch/ia64/kernel/asm-offsets.c
@@ -211,17 +211,41 @@ void foo(void)
 #endif
 
 	BLANK();
-	DEFINE(IA64_MCA_CPU_PROC_STATE_DUMP_OFFSET,
-	       offsetof (struct ia64_mca_cpu, proc_state_dump));
-	DEFINE(IA64_MCA_CPU_STACK_OFFSET,
-	       offsetof (struct ia64_mca_cpu, stack));
-	DEFINE(IA64_MCA_CPU_STACKFRAME_OFFSET,
-	       offsetof (struct ia64_mca_cpu, stackframe));
-	DEFINE(IA64_MCA_CPU_RBSTORE_OFFSET,
-	       offsetof (struct ia64_mca_cpu, rbstore));
+	DEFINE(IA64_MCA_CPU_MCA_STACK_OFFSET,
+	       offsetof (struct ia64_mca_cpu, mca_stack));
 	DEFINE(IA64_MCA_CPU_INIT_STACK_OFFSET,
 	       offsetof (struct ia64_mca_cpu, init_stack));
 	BLANK();
+	DEFINE(IA64_SAL_OS_STATE_COMMON_OFFSET,
+	       offsetof (struct ia64_sal_os_state, sal_ra));
+	DEFINE(IA64_SAL_OS_STATE_OS_GP_OFFSET,
+	       offsetof (struct ia64_sal_os_state, os_gp));
+	DEFINE(IA64_SAL_OS_STATE_PAL_MIN_STATE_OFFSET,
+	       offsetof (struct ia64_sal_os_state, pal_min_state));
+	DEFINE(IA64_SAL_OS_STATE_PROC_STATE_PARAM_OFFSET,
+	       offsetof (struct ia64_sal_os_state, proc_state_param));
+	DEFINE(IA64_SAL_OS_STATE_SIZE,
+	       sizeof (struct ia64_sal_os_state));
+	DEFINE(IA64_PMSA_GR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_gr));
+	DEFINE(IA64_PMSA_BANK1_GR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_bank1_gr));
+	DEFINE(IA64_PMSA_PR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_pr));
+	DEFINE(IA64_PMSA_BR0_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_br0));
+	DEFINE(IA64_PMSA_RSC_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_rsc));
+	DEFINE(IA64_PMSA_IIP_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_iip));
+	DEFINE(IA64_PMSA_IPSR_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_ipsr));
+	DEFINE(IA64_PMSA_IFS_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_ifs));
+	DEFINE(IA64_PMSA_XIP_OFFSET,
+	       offsetof (struct pal_min_state_area_s, pmsa_xip));
+	BLANK();
+
 	/* used by fsys_gettimeofday in arch/ia64/kernel/fsys.S */
 	DEFINE(IA64_TIME_INTERPOLATOR_ADDRESS_OFFSET, offsetof (struct time_interpolator, addr));
 	DEFINE(IA64_TIME_INTERPOLATOR_SOURCE_OFFSET, offsetof (struct time_interpolator, source));
diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S
index 3ba8384cb43..c13ca0d49c4 100644
--- a/arch/ia64/kernel/ivt.S
+++ b/arch/ia64/kernel/ivt.S
@@ -69,7 +69,6 @@
 # define DBG_FAULT(i)
 #endif
 
-#define MINSTATE_VIRT	/* needed by minstate.h */
 #include "minstate.h"
 
 #define FAULT(n)									\
diff --git a/arch/ia64/kernel/mca.c b/arch/ia64/kernel/mca.c
index 8d484204a3f..6dc726ad713 100644
--- a/arch/ia64/kernel/mca.c
+++ b/arch/ia64/kernel/mca.c
@@ -48,6 +48,9 @@
  *            Delete dead variables and functions.
  *            Reorder to remove the need for forward declarations and to consolidate
  *            related code.
+ *
+ * 2005-08-12 Keith Owens <kaos@sgi.com>
+ *	      Convert MCA/INIT handlers to use per event stacks and SAL/OS state.
  */
 #include <linux/config.h>
 #include <linux/types.h>
@@ -77,6 +80,8 @@
 #include <asm/irq.h>
 #include <asm/hw_irq.h>
 
+#include "entry.h"
+
 #if defined(IA64_MCA_DEBUG_INFO)
 # define IA64_MCA_DEBUG(fmt...)	printk(fmt)
 #else
@@ -84,9 +89,7 @@
 #endif
 
 /* Used by mca_asm.S */
-ia64_mca_sal_to_os_state_t	ia64_sal_to_os_handoff_state;
-ia64_mca_os_to_sal_state_t	ia64_os_to_sal_handoff_state;
-u64				ia64_mca_serialize;
+u32				ia64_mca_serialize;
 DEFINE_PER_CPU(u64, ia64_mca_data); /* == __per_cpu_mca[smp_processor_id()] */
 DEFINE_PER_CPU(u64, ia64_mca_per_cpu_pte); /* PTE to map per-CPU area */
 DEFINE_PER_CPU(u64, ia64_mca_pal_pte);	    /* PTE to map PAL code */
@@ -95,8 +98,10 @@ DEFINE_PER_CPU(u64, ia64_mca_pal_base);    /* vaddr PAL code granule */
 unsigned long __per_cpu_mca[NR_CPUS];
 
 /* In mca_asm.S */
-extern void			ia64_monarch_init_handler (void);
-extern void			ia64_slave_init_handler (void);
+extern void			ia64_os_init_dispatch_monarch (void);
+extern void			ia64_os_init_dispatch_slave (void);
+
+static int monarch_cpu = -1;
 
 static ia64_mc_info_t		ia64_mc_info;
 
@@ -234,7 +239,8 @@ ia64_log_get(int sal_info_type, u8 **buffer, int irq_safe)
  *  This function retrieves a specified error record type from SAL
  *  and wakes up any processes waiting for error records.
  *
- *  Inputs  :   sal_info_type   (Type of error record MCA/CMC/CPE/INIT)
+ *  Inputs  :   sal_info_type   (Type of error record MCA/CMC/CPE)
+ *              FIXME: remove MCA and irq_safe.
  */
 static void
 ia64_mca_log_sal_error_record(int sal_info_type)
@@ -242,7 +248,7 @@ ia64_mca_log_sal_error_record(int sal_info_type)
 	u8 *buffer;
 	sal_log_record_header_t *rh;
 	u64 size;
-	int irq_safe = sal_info_type != SAL_INFO_TYPE_MCA && sal_info_type != SAL_INFO_TYPE_INIT;
+	int irq_safe = sal_info_type != SAL_INFO_TYPE_MCA;
 #ifdef IA64_MCA_DEBUG_INFO
 	static const char * const rec_name[] = { "MCA", "INIT", "CMC", "CPE" };
 #endif
@@ -330,182 +336,6 @@ ia64_mca_cpe_int_handler (int cpe_irq, void *arg, struct pt_regs *ptregs)
 
 #endif /* CONFIG_ACPI */
 
-static void
-show_min_state (pal_min_state_area_t *minstate)
-{
-	u64 iip = minstate->pmsa_iip + ((struct ia64_psr *)(&minstate->pmsa_ipsr))->ri;
-	u64 xip = minstate->pmsa_xip + ((struct ia64_psr *)(&minstate->pmsa_xpsr))->ri;
-
-	printk("NaT bits\t%016lx\n", minstate->pmsa_nat_bits);
-	printk("pr\t\t%016lx\n", minstate->pmsa_pr);
-	printk("b0\t\t%016lx ", minstate->pmsa_br0); print_symbol("%s\n", minstate->pmsa_br0);
-	printk("ar.rsc\t\t%016lx\n", minstate->pmsa_rsc);
-	printk("cr.iip\t\t%016lx ", iip); print_symbol("%s\n", iip);
-	printk("cr.ipsr\t\t%016lx\n", minstate->pmsa_ipsr);
-	printk("cr.ifs\t\t%016lx\n", minstate->pmsa_ifs);
-	printk("xip\t\t%016lx ", xip); print_symbol("%s\n", xip);
-	printk("xpsr\t\t%016lx\n", minstate->pmsa_xpsr);
-	printk("xfs\t\t%016lx\n", minstate->pmsa_xfs);
-	printk("b1\t\t%016lx ", minstate->pmsa_br1);
-	print_symbol("%s\n", minstate->pmsa_br1);
-
-	printk("\nstatic registers r0-r15:\n");
-	printk(" r0- 3 %016lx %016lx %016lx %016lx\n",
-	       0UL, minstate->pmsa_gr[0], minstate->pmsa_gr[1], minstate->pmsa_gr[2]);
-	printk(" r4- 7 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_gr[3], minstate->pmsa_gr[4],
-	       minstate->pmsa_gr[5], minstate->pmsa_gr[6]);
-	printk(" r8-11 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_gr[7], minstate->pmsa_gr[8],
-	       minstate->pmsa_gr[9], minstate->pmsa_gr[10]);
-	printk("r12-15 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_gr[11], minstate->pmsa_gr[12],
-	       minstate->pmsa_gr[13], minstate->pmsa_gr[14]);
-
-	printk("\nbank 0:\n");
-	printk("r16-19 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank0_gr[0], minstate->pmsa_bank0_gr[1],
-	       minstate->pmsa_bank0_gr[2], minstate->pmsa_bank0_gr[3]);
-	printk("r20-23 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank0_gr[4], minstate->pmsa_bank0_gr[5],
-	       minstate->pmsa_bank0_gr[6], minstate->pmsa_bank0_gr[7]);
-	printk("r24-27 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank0_gr[8], minstate->pmsa_bank0_gr[9],
-	       minstate->pmsa_bank0_gr[10], minstate->pmsa_bank0_gr[11]);
-	printk("r28-31 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank0_gr[12], minstate->pmsa_bank0_gr[13],
-	       minstate->pmsa_bank0_gr[14], minstate->pmsa_bank0_gr[15]);
-
-	printk("\nbank 1:\n");
-	printk("r16-19 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank1_gr[0], minstate->pmsa_bank1_gr[1],
-	       minstate->pmsa_bank1_gr[2], minstate->pmsa_bank1_gr[3]);
-	printk("r20-23 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank1_gr[4], minstate->pmsa_bank1_gr[5],
-	       minstate->pmsa_bank1_gr[6], minstate->pmsa_bank1_gr[7]);
-	printk("r24-27 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank1_gr[8], minstate->pmsa_bank1_gr[9],
-	       minstate->pmsa_bank1_gr[10], minstate->pmsa_bank1_gr[11]);
-	printk("r28-31 %016lx %016lx %016lx %016lx\n",
-	       minstate->pmsa_bank1_gr[12], minstate->pmsa_bank1_gr[13],
-	       minstate->pmsa_bank1_gr[14], minstate->pmsa_bank1_gr[15]);
-}
-
-static void
-fetch_min_state (pal_min_state_area_t *ms, struct pt_regs *pt, struct switch_stack *sw)
-{
-	u64 *dst_banked, *src_banked, bit, shift, nat_bits;
-	int i;
-
-	/*
-	 * First, update the pt-regs and switch-stack structures with the contents stored
-	 * in the min-state area:
-	 */
-	if (((struct ia64_psr *) &ms->pmsa_ipsr)->ic == 0) {
-		pt->cr_ipsr = ms->pmsa_xpsr;
-		pt->cr_iip = ms->pmsa_xip;
-		pt->cr_ifs = ms->pmsa_xfs;
-	} else {
-		pt->cr_ipsr = ms->pmsa_ipsr;
-		pt->cr_iip = ms->pmsa_iip;
-		pt->cr_ifs = ms->pmsa_ifs;
-	}
-	pt->ar_rsc = ms->pmsa_rsc;
-	pt->pr = ms->pmsa_pr;
-	pt->r1 = ms->pmsa_gr[0];
-	pt->r2 = ms->pmsa_gr[1];
-	pt->r3 = ms->pmsa_gr[2];
-	sw->r4 = ms->pmsa_gr[3];
-	sw->r5 = ms->pmsa_gr[4];
-	sw->r6 = ms->pmsa_gr[5];
-	sw->r7 = ms->pmsa_gr[6];
-	pt->r8 = ms->pmsa_gr[7];
-	pt->r9 = ms->pmsa_gr[8];
-	pt->r10 = ms->pmsa_gr[9];
-	pt->r11 = ms->pmsa_gr[10];
-	pt->r12 = ms->pmsa_gr[11];
-	pt->r13 = ms->pmsa_gr[12];
-	pt->r14 = ms->pmsa_gr[13];
-	pt->r15 = ms->pmsa_gr[14];
-	dst_banked = &pt->r16;		/* r16-r31 are contiguous in struct pt_regs */
-	src_banked = ms->pmsa_bank1_gr;
-	for (i = 0; i < 16; ++i)
-		dst_banked[i] = src_banked[i];
-	pt->b0 = ms->pmsa_br0;
-	sw->b1 = ms->pmsa_br1;
-
-	/* construct the NaT bits for the pt-regs structure: */
-#	define PUT_NAT_BIT(dst, addr)					\
-	do {								\
-		bit = nat_bits & 1; nat_bits >>= 1;			\
-		shift = ((unsigned long) addr >> 3) & 0x3f;		\
-		dst = ((dst) & ~(1UL << shift)) | (bit << shift);	\
-	} while (0)
-
-	/* Rotate the saved NaT bits such that bit 0 corresponds to pmsa_gr[0]: */
-	shift = ((unsigned long) &ms->pmsa_gr[0] >> 3) & 0x3f;
-	nat_bits = (ms->pmsa_nat_bits >> shift) | (ms->pmsa_nat_bits << (64 - shift));
-
-	PUT_NAT_BIT(sw->caller_unat, &pt->r1);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r2);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r3);
-	PUT_NAT_BIT(sw->ar_unat, &sw->r4);
-	PUT_NAT_BIT(sw->ar_unat, &sw->r5);
-	PUT_NAT_BIT(sw->ar_unat, &sw->r6);
-	PUT_NAT_BIT(sw->ar_unat, &sw->r7);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r8);	PUT_NAT_BIT(sw->caller_unat, &pt->r9);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r10);	PUT_NAT_BIT(sw->caller_unat, &pt->r11);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r12);	PUT_NAT_BIT(sw->caller_unat, &pt->r13);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r14);	PUT_NAT_BIT(sw->caller_unat, &pt->r15);
-	nat_bits >>= 16;	/* skip over bank0 NaT bits */
-	PUT_NAT_BIT(sw->caller_unat, &pt->r16);	PUT_NAT_BIT(sw->caller_unat, &pt->r17);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r18);	PUT_NAT_BIT(sw->caller_unat, &pt->r19);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r20);	PUT_NAT_BIT(sw->caller_unat, &pt->r21);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r22);	PUT_NAT_BIT(sw->caller_unat, &pt->r23);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r24);	PUT_NAT_BIT(sw->caller_unat, &pt->r25);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r26);	PUT_NAT_BIT(sw->caller_unat, &pt->r27);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r28);	PUT_NAT_BIT(sw->caller_unat, &pt->r29);
-	PUT_NAT_BIT(sw->caller_unat, &pt->r30);	PUT_NAT_BIT(sw->caller_unat, &pt->r31);
-}
-
-static void
-init_handler_platform (pal_min_state_area_t *ms,
-		       struct pt_regs *pt, struct switch_stack *sw)
-{
-	struct unw_frame_info info;
-
-	/* if a kernel debugger is available call it here else just dump the registers */
-
-	/*
-	 * Wait for a bit.  On some machines (e.g., HP's zx2000 and zx6000, INIT can be
-	 * generated via the BMC's command-line interface, but since the console is on the
-	 * 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");
-	udelay(5*1000000);
-	show_min_state(ms);
-
-	printk("Backtrace of current task (pid %d, %s)\n", current->pid, current->comm);
-	fetch_min_state(ms, pt, sw);
-	unw_init_from_interruption(&info, current, pt, sw);
-	ia64_do_show_stack(&info, NULL);
-
-	if (read_trylock(&tasklist_lock)) {
-		struct task_struct *g, *t;
-		do_each_thread (g, t) {
-			if (t == current)
-				continue;
-
-			printk("\nBacktrace of pid %d (%s)\n", t->pid, t->comm);
-			show_stack(t, NULL);
-		} while_each_thread (g, t);
-	}
-
-	printk("\nINIT dump complete.  Please reboot now.\n");
-	while (1);			/* hang city if no debugger */
-}
-
 #ifdef CONFIG_ACPI
 /*
  * ia64_mca_register_cpev
@@ -648,42 +478,6 @@ ia64_mca_cmc_vector_enable_keventd(void *unused)
 }
 
 /*
- * ia64_mca_wakeup_ipi_wait
- *
- *	Wait for the inter-cpu interrupt to be sent by the
- *	monarch processor once it is done with handling the
- *	MCA.
- *
- *  Inputs  :   None
- *  Outputs :   None
- */
-static void
-ia64_mca_wakeup_ipi_wait(void)
-{
-	int	irr_num = (IA64_MCA_WAKEUP_VECTOR >> 6);
-	int	irr_bit = (IA64_MCA_WAKEUP_VECTOR & 0x3f);
-	u64	irr = 0;
-
-	do {
-		switch(irr_num) {
-		      case 0:
-			irr = ia64_getreg(_IA64_REG_CR_IRR0);
-			break;
-		      case 1:
-			irr = ia64_getreg(_IA64_REG_CR_IRR1);
-			break;
-		      case 2:
-			irr = ia64_getreg(_IA64_REG_CR_IRR2);
-			break;
-		      case 3:
-			irr = ia64_getreg(_IA64_REG_CR_IRR3);
-			break;
-		}
-		cpu_relax();
-	} while (!(irr & (1UL << irr_bit))) ;
-}
-
-/*
  * ia64_mca_wakeup
  *
  *	Send an inter-cpu interrupt to wake-up a particular cpu
@@ -748,11 +542,9 @@ ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs)
 	 */
 	ia64_sal_mc_rendez();
 
-	/* Wait for the wakeup IPI from the monarch
-	 * This waiting is done by polling on the wakeup-interrupt
-	 * vector bit in the processor's IRRs
-	 */
-	ia64_mca_wakeup_ipi_wait();
+	/* Wait for the monarch cpu to exit. */
+	while (monarch_cpu != -1)
+	       cpu_relax();	/* spin until monarch leaves */
 
 	/* Enable all interrupts */
 	local_irq_restore(flags);
@@ -780,53 +572,13 @@ ia64_mca_wakeup_int_handler(int wakeup_irq, void *arg, struct pt_regs *ptregs)
 	return IRQ_HANDLED;
 }
 
-/*
- * ia64_return_to_sal_check
- *
- *	This is function called before going back from the OS_MCA handler
- *	to the OS_MCA dispatch code which finally takes the control back
- *	to the SAL.
- *	The main purpose of this routine is to setup the OS_MCA to SAL
- *	return state which can be used by the OS_MCA dispatch code
- *	just before going back to SAL.
- *
- *  Inputs  :   None
- *  Outputs :   None
- */
-
-static void
-ia64_return_to_sal_check(int recover)
-{
-
-	/* Copy over some relevant stuff from the sal_to_os_mca_handoff
-	 * so that it can be used at the time of os_mca_to_sal_handoff
-	 */
-	ia64_os_to_sal_handoff_state.imots_sal_gp =
-		ia64_sal_to_os_handoff_state.imsto_sal_gp;
-
-	ia64_os_to_sal_handoff_state.imots_sal_check_ra =
-		ia64_sal_to_os_handoff_state.imsto_sal_check_ra;
-
-	if (recover)
-		ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_CORRECTED;
-	else
-		ia64_os_to_sal_handoff_state.imots_os_status = IA64_MCA_COLD_BOOT;
-
-	/* Default = tell SAL to return to same context */
-	ia64_os_to_sal_handoff_state.imots_context = IA64_MCA_SAME_CONTEXT;
-
-	ia64_os_to_sal_handoff_state.imots_new_min_state =
-		(u64 *)ia64_sal_to_os_handoff_state.pal_min_state;
-
-}
-
 /* Function pointer for extra MCA recovery */
 int (*ia64_mca_ucmc_extension)
-	(void*,ia64_mca_sal_to_os_state_t*,ia64_mca_os_to_sal_state_t*)
+	(void*,struct ia64_sal_os_state*)
 	= NULL;
 
 int
-ia64_reg_MCA_extension(void *fn)
+ia64_reg_MCA_extension(int (*fn)(void *, struct ia64_sal_os_state *))
 {
 	if (ia64_mca_ucmc_extension)
 		return 1;
@@ -845,8 +597,321 @@ ia64_unreg_MCA_extension(void)
 EXPORT_SYMBOL(ia64_reg_MCA_extension);
 EXPORT_SYMBOL(ia64_unreg_MCA_extension);
 
+
+static inline void
+copy_reg(const u64 *fr, u64 fnat, u64 *tr, u64 *tnat)
+{
+	u64 fslot, tslot, nat;
+	*tr = *fr;
+	fslot = ((unsigned long)fr >> 3) & 63;
+	tslot = ((unsigned long)tr >> 3) & 63;
+	*tnat &= ~(1UL << tslot);
+	nat = (fnat >> fslot) & 1;
+	*tnat |= (nat << tslot);
+}
+
+/* On entry to this routine, we are running on the per cpu stack, see
+ * mca_asm.h.  The original stack has not been touched by this event.  Some of
+ * the original stack's registers will be in the RBS on this stack.  This stack
+ * also contains a partial pt_regs and switch_stack, the rest of the data is in
+ * PAL minstate.
+ *
+ * The first thing to do is modify the original stack to look like a blocked
+ * task so we can run backtrace on the original task.  Also mark the per cpu
+ * stack as current to ensure that we use the correct task state, it also means
+ * that we can do backtrace on the MCA/INIT handler code itself.
+ */
+
+static task_t *
+ia64_mca_modify_original_stack(struct pt_regs *regs,
+		const struct switch_stack *sw,
+		struct ia64_sal_os_state *sos,
+		const char *type)
+{
+	char *p, comm[sizeof(current->comm)];
+	ia64_va va;
+	extern char ia64_leave_kernel[];	/* Need asm address, not function descriptor */
+	const pal_min_state_area_t *ms = sos->pal_min_state;
+	task_t *previous_current;
+	struct pt_regs *old_regs;
+	struct switch_stack *old_sw;
+	unsigned size = sizeof(struct pt_regs) +
+			sizeof(struct switch_stack) + 16;
+	u64 *old_bspstore, *old_bsp;
+	u64 *new_bspstore, *new_bsp;
+	u64 old_unat, old_rnat, new_rnat, nat;
+	u64 slots, loadrs = regs->loadrs;
+	u64 r12 = ms->pmsa_gr[12-1], r13 = ms->pmsa_gr[13-1];
+	u64 ar_bspstore = regs->ar_bspstore;
+	u64 ar_bsp = regs->ar_bspstore + (loadrs >> 16);
+	const u64 *bank;
+	const char *msg;
+	int cpu = smp_processor_id();
+
+	previous_current = curr_task(cpu);
+	set_curr_task(cpu, current);
+	if ((p = strchr(current->comm, ' ')))
+		*p = '\0';
+
+	/* Best effort attempt to cope with MCA/INIT delivered while in
+	 * physical mode.
+	 */
+	regs->cr_ipsr = ms->pmsa_ipsr;
+	if (ia64_psr(regs)->dt == 0) {
+		va.l = r12;
+		if (va.f.reg == 0) {
+			va.f.reg = 7;
+			r12 = va.l;
+		}
+		va.l = r13;
+		if (va.f.reg == 0) {
+			va.f.reg = 7;
+			r13 = va.l;
+		}
+	}
+	if (ia64_psr(regs)->rt == 0) {
+		va.l = ar_bspstore;
+		if (va.f.reg == 0) {
+			va.f.reg = 7;
+			ar_bspstore = va.l;
+		}
+		va.l = ar_bsp;
+		if (va.f.reg == 0) {
+			va.f.reg = 7;
+			ar_bsp = va.l;
+		}
+	}
+
+	/* mca_asm.S ia64_old_stack() cannot assume that the dirty registers
+	 * have been copied to the old stack, the old stack may fail the
+	 * validation tests below.  So ia64_old_stack() must restore the dirty
+	 * registers from the new stack.  The old and new bspstore probably
+	 * have different alignments, so loadrs calculated on the old bsp
+	 * cannot be used to restore from the new bsp.  Calculate a suitable
+	 * loadrs for the new stack and save it in the new pt_regs, where
+	 * ia64_old_stack() can get it.
+	 */
+	old_bspstore = (u64 *)ar_bspstore;
+	old_bsp = (u64 *)ar_bsp;
+	slots = ia64_rse_num_regs(old_bspstore, old_bsp);
+	new_bspstore = (u64 *)((u64)current + IA64_RBS_OFFSET);
+	new_bsp = ia64_rse_skip_regs(new_bspstore, slots);
+	regs->loadrs = (new_bsp - new_bspstore) * 8 << 16;
+
+	/* Verify the previous stack state before we change it */
+	if (user_mode(regs)) {
+		msg = "occurred in user space";
+		goto no_mod;
+	}
+	if (r13 != sos->prev_IA64_KR_CURRENT) {
+		msg = "inconsistent previous current and r13";
+		goto no_mod;
+	}
+	if ((r12 - r13) >= KERNEL_STACK_SIZE) {
+		msg = "inconsistent r12 and r13";
+		goto no_mod;
+	}
+	if ((ar_bspstore - r13) >= KERNEL_STACK_SIZE) {
+		msg = "inconsistent ar.bspstore and r13";
+		goto no_mod;
+	}
+	va.p = old_bspstore;
+	if (va.f.reg < 5) {
+		msg = "old_bspstore is in the wrong region";
+		goto no_mod;
+	}
+	if ((ar_bsp - r13) >= KERNEL_STACK_SIZE) {
+		msg = "inconsistent ar.bsp and r13";
+		goto no_mod;
+	}
+	size += (ia64_rse_skip_regs(old_bspstore, slots) - old_bspstore) * 8;
+	if (ar_bspstore + size > r12) {
+		msg = "no room for blocked state";
+		goto no_mod;
+	}
+
+	/* Change the comm field on the MCA/INT task to include the pid that
+	 * was interrupted, it makes for easier debugging.  If that pid was 0
+	 * (swapper or nested MCA/INIT) then use the start of the previous comm
+	 * field suffixed with its cpu.
+	 */
+	if (previous_current->pid)
+		snprintf(comm, sizeof(comm), "%s %d",
+			current->comm, previous_current->pid);
+	else {
+		int l;
+		if ((p = strchr(previous_current->comm, ' ')))
+			l = p - previous_current->comm;
+		else
+			l = strlen(previous_current->comm);
+		snprintf(comm, sizeof(comm), "%s %*s %d",
+			current->comm, l, previous_current->comm,
+			previous_current->thread_info->cpu);
+	}
+	memcpy(current->comm, comm, sizeof(current->comm));
+
+	/* Make the original task look blocked.  First stack a struct pt_regs,
+	 * describing the state at the time of interrupt.  mca_asm.S built a
+	 * partial pt_regs, copy it and fill in the blanks using minstate.
+	 */
+	p = (char *)r12 - sizeof(*regs);
+	old_regs = (struct pt_regs *)p;
+	memcpy(old_regs, regs, sizeof(*regs));
+	/* If ipsr.ic then use pmsa_{iip,ipsr,ifs}, else use
+	 * pmsa_{xip,xpsr,xfs}
+	 */
+	if (ia64_psr(regs)->ic) {
+		old_regs->cr_iip = ms->pmsa_iip;
+		old_regs->cr_ipsr = ms->pmsa_ipsr;
+		old_regs->cr_ifs = ms->pmsa_ifs;
+	} else {
+		old_regs->cr_iip = ms->pmsa_xip;
+		old_regs->cr_ipsr = ms->pmsa_xpsr;
+		old_regs->cr_ifs = ms->pmsa_xfs;
+	}
+	old_regs->pr = ms->pmsa_pr;
+	old_regs->b0 = ms->pmsa_br0;
+	old_regs->loadrs = loadrs;
+	old_regs->ar_rsc = ms->pmsa_rsc;
+	old_unat = old_regs->ar_unat;
+	copy_reg(&ms->pmsa_gr[1-1], ms->pmsa_nat_bits, &old_regs->r1, &old_unat);
+	copy_reg(&ms->pmsa_gr[2-1], ms->pmsa_nat_bits, &old_regs->r2, &old_unat);
+	copy_reg(&ms->pmsa_gr[3-1], ms->pmsa_nat_bits, &old_regs->r3, &old_unat);
+	copy_reg(&ms->pmsa_gr[8-1], ms->pmsa_nat_bits, &old_regs->r8, &old_unat);
+	copy_reg(&ms->pmsa_gr[9-1], ms->pmsa_nat_bits, &old_regs->r9, &old_unat);
+	copy_reg(&ms->pmsa_gr[10-1], ms->pmsa_nat_bits, &old_regs->r10, &old_unat);
+	copy_reg(&ms->pmsa_gr[11-1], ms->pmsa_nat_bits, &old_regs->r11, &old_unat);
+	copy_reg(&ms->pmsa_gr[12-1], ms->pmsa_nat_bits, &old_regs->r12, &old_unat);
+	copy_reg(&ms->pmsa_gr[13-1], ms->pmsa_nat_bits, &old_regs->r13, &old_unat);
+	copy_reg(&ms->pmsa_gr[14-1], ms->pmsa_nat_bits, &old_regs->r14, &old_unat);
+	copy_reg(&ms->pmsa_gr[15-1], ms->pmsa_nat_bits, &old_regs->r15, &old_unat);
+	if (ia64_psr(old_regs)->bn)
+		bank = ms->pmsa_bank1_gr;
+	else
+		bank = ms->pmsa_bank0_gr;
+	copy_reg(&bank[16-16], ms->pmsa_nat_bits, &old_regs->r16, &old_unat);
+	copy_reg(&bank[17-16], ms->pmsa_nat_bits, &old_regs->r17, &old_unat);
+	copy_reg(&bank[18-16], ms->pmsa_nat_bits, &old_regs->r18, &old_unat);
+	copy_reg(&bank[19-16], ms->pmsa_nat_bits, &old_regs->r19, &old_unat);
+	copy_reg(&bank[20-16], ms->pmsa_nat_bits, &old_regs->r20, &old_unat);
+	copy_reg(&bank[21-16], ms->pmsa_nat_bits, &old_regs->r21, &old_unat);
+	copy_reg(&bank[22-16], ms->pmsa_nat_bits, &old_regs->r22, &old_unat);
+	copy_reg(&bank[23-16], ms->pmsa_nat_bits, &old_regs->r23, &old_unat);
+	copy_reg(&bank[24-16], ms->pmsa_nat_bits, &old_regs->r24, &old_unat);
+	copy_reg(&bank[25-16], ms->pmsa_nat_bits, &old_regs->r25, &old_unat);
+	copy_reg(&bank[26-16], ms->pmsa_nat_bits, &old_regs->r26, &old_unat);
+	copy_reg(&bank[27-16], ms->pmsa_nat_bits, &old_regs->r27, &old_unat);
+	copy_reg(&bank[28-16], ms->pmsa_nat_bits, &old_regs->r28, &old_unat);
+	copy_reg(&bank[29-16], ms->pmsa_nat_bits, &old_regs->r29, &old_unat);
+	copy_reg(&bank[30-16], ms->pmsa_nat_bits, &old_regs->r30, &old_unat);
+	copy_reg(&bank[31-16], ms->pmsa_nat_bits, &old_regs->r31, &old_unat);
+
+	/* Next stack a struct switch_stack.  mca_asm.S built a partial
+	 * switch_stack, copy it and fill in the blanks using pt_regs and
+	 * minstate.
+	 *
+	 * In the synthesized switch_stack, b0 points to ia64_leave_kernel,
+	 * ar.pfs is set to 0.
+	 *
+	 * unwind.c::unw_unwind() does special processing for interrupt frames.
+	 * It checks if the PRED_NON_SYSCALL predicate is set, if the predicate
+	 * is clear then unw_unwind() does _not_ adjust bsp over pt_regs.  Not
+	 * that this is documented, of course.  Set PRED_NON_SYSCALL in the
+	 * switch_stack on the original stack so it will unwind correctly when
+	 * unwind.c reads pt_regs.
+	 *
+	 * thread.ksp is updated to point to the synthesized switch_stack.
+	 */
+	p -= sizeof(struct switch_stack);
+	old_sw = (struct switch_stack *)p;
+	memcpy(old_sw, sw, sizeof(*sw));
+	old_sw->caller_unat = old_unat;
+	old_sw->ar_fpsr = old_regs->ar_fpsr;
+	copy_reg(&ms->pmsa_gr[4-1], ms->pmsa_nat_bits, &old_sw->r4, &old_unat);
+	copy_reg(&ms->pmsa_gr[5-1], ms->pmsa_nat_bits, &old_sw->r5, &old_unat);
+	copy_reg(&ms->pmsa_gr[6-1], ms->pmsa_nat_bits, &old_sw->r6, &old_unat);
+	copy_reg(&ms->pmsa_gr[7-1], ms->pmsa_nat_bits, &old_sw->r7, &old_unat);
+	old_sw->b0 = (u64)ia64_leave_kernel;
+	old_sw->b1 = ms->pmsa_br1;
+	old_sw->ar_pfs = 0;
+	old_sw->ar_unat = old_unat;
+	old_sw->pr = old_regs->pr | (1UL << PRED_NON_SYSCALL);
+	previous_current->thread.ksp = (u64)p - 16;
+
+	/* Finally copy the original stack's registers back to its RBS.
+	 * Registers from ar.bspstore through ar.bsp at the time of the event
+	 * are in the current RBS, copy them back to the original stack.  The
+	 * copy must be done register by register because the original bspstore
+	 * and the current one have different alignments, so the saved RNAT
+	 * data occurs at different places.
+	 *
+	 * mca_asm does cover, so the old_bsp already includes all registers at
+	 * the time of MCA/INIT.  It also does flushrs, so all registers before
+	 * this function have been written to backing store on the MCA/INIT
+	 * stack.
+	 */
+	new_rnat = ia64_get_rnat(ia64_rse_rnat_addr(new_bspstore));
+	old_rnat = regs->ar_rnat;
+	while (slots--) {
+		if (ia64_rse_is_rnat_slot(new_bspstore)) {
+			new_rnat = ia64_get_rnat(new_bspstore++);
+		}
+		if (ia64_rse_is_rnat_slot(old_bspstore)) {
+			*old_bspstore++ = old_rnat;
+			old_rnat = 0;
+		}
+		nat = (new_rnat >> ia64_rse_slot_num(new_bspstore)) & 1UL;
+		old_rnat &= ~(1UL << ia64_rse_slot_num(old_bspstore));
+		old_rnat |= (nat << ia64_rse_slot_num(old_bspstore));
+		*old_bspstore++ = *new_bspstore++;
+	}
+	old_sw->ar_bspstore = (unsigned long)old_bspstore;
+	old_sw->ar_rnat = old_rnat;
+
+	sos->prev_task = previous_current;
+	return previous_current;
+
+no_mod:
+	printk(KERN_INFO "cpu %d, %s %s, original stack not modified\n",
+			smp_processor_id(), type, msg);
+	return previous_current;
+}
+
+/* The monarch/slave interaction is based on monarch_cpu and requires that all
+ * slaves have entered rendezvous before the monarch leaves.  If any cpu has
+ * not entered rendezvous yet then wait a bit.  The assumption is that any
+ * slave that has not rendezvoused after a reasonable time is never going to do
+ * so.  In this context, slave includes cpus that respond to the MCA rendezvous
+ * interrupt, as well as cpus that receive the INIT slave event.
+ */
+
+static void
+ia64_wait_for_slaves(int monarch)
+{
+	int c, wait = 0;
+	for_each_online_cpu(c) {
+		if (c == monarch)
+			continue;
+		if (ia64_mc_info.imi_rendez_checkin[c] == IA64_MCA_RENDEZ_CHECKIN_NOTDONE) {
+			udelay(1000);		/* short wait first */
+			wait = 1;
+			break;
+		}
+	}
+	if (!wait)
+		return;
+	for_each_online_cpu(c) {
+		if (c == monarch)
+			continue;
+		if (ia64_mc_info.imi_rendez_checkin[c] == IA64_MCA_RENDEZ_CHECKIN_NOTDONE) {
+			udelay(5*1000000);	/* wait 5 seconds for slaves (arbitrary) */
+			break;
+