Merge with Linux 2.3.6. Sorry, this isn't tested on silicon, I don't

have a MIPS box at hand.

13 files changed, 861 insertions, 729 deletions

diff --git a/arch/alpha/kernel/alpha_ksyms.c b/arch/alpha/kernel/alpha_ksyms.c
index 65975c168..ce949d722 100644
--- a/arch/alpha/kernel/alpha_ksyms.c
+++ b/arch/alpha/kernel/alpha_ksyms.c
@@ -52,6 +52,7 @@ EXPORT_SYMBOL(local_bh_count);
 EXPORT_SYMBOL(local_irq_count);
 EXPORT_SYMBOL(enable_irq);
 EXPORT_SYMBOL(disable_irq);
+EXPORT_SYMBOL(disable_irq_nosync);
 EXPORT_SYMBOL(screen_info);
 EXPORT_SYMBOL(perf_irq);
 
diff --git a/arch/alpha/kernel/fpreg.c b/arch/alpha/kernel/fpreg.c
index 6da94c0cb..0bd8b9f67 100644
--- a/arch/alpha/kernel/fpreg.c
+++ b/arch/alpha/kernel/fpreg.c
@@ -1,10 +1,10 @@
 /*
- * kernel/fpreg.c
+ * arch/alpha/kernel/fpreg.c
  *
  * (C) Copyright 1998 Linus Torvalds
  */
 
-#ifdef __alpha_cix__
+#if defined(__alpha_cix__) || defined(__alpha_fix__)
 #define STT(reg,val)  asm volatile ("ftoit $f"#reg",%0" : "=r"(val));
 #else
 #define STT(reg,val)  asm volatile ("stt $f"#reg",%0" : "=m"(val));
@@ -52,7 +52,7 @@ alpha_read_fp_reg (unsigned long reg)
 	return val;
 }
 
-#ifdef __alpha_cix__
+#if defined(__alpha_cix__) || defined(__alpha_fix__)
 #define LDT(reg,val)  asm volatile ("itoft %0,$f"#reg : : "r"(val));
 #else
 #define LDT(reg,val)  asm volatile ("ldt $f"#reg",%0" : : "m"(val));
diff --git a/arch/alpha/kernel/head.S b/arch/alpha/kernel/head.S
index 8ba50462b..3fcbdbcda 100644
--- a/arch/alpha/kernel/head.S
+++ b/arch/alpha/kernel/head.S
@@ -32,24 +32,26 @@ __start:
 
 #ifdef __SMP__
 	.align 3
-	.globl	__start_cpu
-	.ent	__start_cpu
-	/* On entry here from SRM console, the HWPCB of this processor
-	   has been loaded, and $27 contains the task pointer */
-__start_cpu:
-	.prologue 0
-	/* First order of business, load the GP */
-	br	$26,1f
-1:	ldgp	$29,0($26)
-	/* We need to get current loaded up with our first task...  */
-	mov	$27,$8
-	/* Set FEN */
-	lda	$16,1($31)
-	call_pal PAL_wrfen
-	/* ... and then we can start the processor.  */
-	jsr	$26,start_secondary
+	.globl	__smp_callin
+	.ent	__smp_callin
+	/* On entry here from SRM console, the HWPCB of the per-cpu
+	   slot for this processor has been loaded.  We've arranged
+	   for the UNIQUE value for this process to contain the PCBB
+	   of the target idle task.  */
+__smp_callin:
+	.prologue 1
+	ldgp	$29,0($27)	# First order of business, load the GP.
+
+	call_pal PAL_rduniq	# Grab the target PCBB.
+	mov	$0,$16		# Install it.
+	call_pal PAL_swpctx
+
+	lda	$8,0x3fff	# Find "current".
+	bic	$30,$8,$8
+	
+	jsr	$26,smp_callin
 	call_pal PAL_halt
-	.end __start_cpu
+	.end __smp_callin
 #endif /* __SMP__ */
 
 	.align 3
diff --git a/arch/alpha/kernel/irq.c b/arch/alpha/kernel/irq.c
index e1a53ab57..d53f7ffea 100644
--- a/arch/alpha/kernel/irq.c
+++ b/arch/alpha/kernel/irq.c
@@ -192,7 +192,7 @@ unmask_irq(unsigned long irq)
 }
 
 void
-disable_irq(unsigned int irq_nr)
+disable_irq_nosync(unsigned int irq_nr)
 {
 	unsigned long flags;
 
@@ -202,6 +202,14 @@ disable_irq(unsigned int irq_nr)
 }
 
 void
+disable_irq(unsigned int irq_nr)
+{
+	/* This works non-SMP, and SMP until we write code to distribute
+	   interrupts to more that cpu 0.  */
+	disable_irq_nosync(irq_nr);
+}
+
+void
 enable_irq(unsigned int irq_nr)
 {
 	unsigned long flags;
diff --git a/arch/alpha/kernel/process.c b/arch/alpha/kernel/process.c
index 93d8db602..1993ed3b4 100644
--- a/arch/alpha/kernel/process.c
+++ b/arch/alpha/kernel/process.c
@@ -58,10 +58,10 @@ static struct fs_struct init_fs = INIT_FS;
 static struct file * init_fd_array[NR_OPEN] = { NULL, };
 static struct files_struct init_files = INIT_FILES;
 static struct signal_struct init_signals = INIT_SIGNALS;
-struct mm_struct init_mm = INIT_MM;
+struct mm_struct init_mm = INIT_MM(init_mm);
 
 union task_union init_task_union __attribute__((section("init_task")))
-	 = { task: INIT_TASK };
+	 = { task: INIT_TASK(init_task_union.task) };
 
 /*
  * No need to acquire the kernel lock, we're entirely local..
@@ -75,33 +75,46 @@ sys_sethae(unsigned long hae, unsigned long a1, unsigned long a2,
 	return 0;
 }
 
-static void __attribute__((noreturn))
-do_cpu_idle(void)
+#ifdef __SMP__
+void
+cpu_idle(void *unused)
 {
 	/* An endless idle loop with no priority at all.  */
 	current->priority = 0;
+	current->counter = -100;
+
 	while (1) {
-		check_pgt_cache();
-		run_task_queue(&tq_scheduler);
-		current->counter = 0;
-		schedule();
+		/* FIXME -- EV6 and LCA45 know how to power down
+		   the CPU.  */
+
+		/* Although we are an idle CPU, we do not want to 
+		   get into the scheduler unnecessarily.  */
+		if (current->need_resched) {
+			schedule();
+			check_pgt_cache();
+		}
 	}
 }
-
-#ifdef __SMP__
-void
-cpu_idle(void *unused)
-{
-	do_cpu_idle();
-}
 #endif
 
 asmlinkage int
 sys_idle(void)
 {
-	if (current->pid == 0)
-        	do_cpu_idle();
-	return -EPERM;
+	if (current->pid != 0)
+		return -EPERM;
+
+	/* An endless idle loop with no priority at all.  */
+	current->priority = 0;
+	current->counter = -100;
+	init_idle();
+
+	while (1) {
+		/* FIXME -- EV6 and LCA45 know how to power down
+		   the CPU.  */
+
+		schedule();
+		check_pgt_cache();
+	}
 }
 
 void
diff --git a/arch/alpha/kernel/proto.h b/arch/alpha/kernel/proto.h
index 8a0efe52d..f7e54a982 100644
--- a/arch/alpha/kernel/proto.h
+++ b/arch/alpha/kernel/proto.h
@@ -151,6 +151,8 @@ extern unsigned long srm_hae;
 extern void setup_smp(void);
 extern int smp_info(char *buffer);
 extern void handle_ipi(struct pt_regs *);
+extern void smp_percpu_timer_interrupt(struct pt_regs *);
+extern int smp_boot_cpuid;
 
 /* bios32.c */
 extern void reset_for_srm(void);
@@ -178,7 +180,7 @@ extern unsigned long alpha_read_fp_reg (unsigned long reg);
 extern void wrmces(unsigned long mces);
 extern void cserve_ena(unsigned long);
 extern void cserve_dis(unsigned long);
-extern void __start_cpu(unsigned long);
+extern void __smp_callin(void);
 
 /* entry.S */
 extern void entArith(void);
diff --git a/arch/alpha/kernel/ptrace.c b/arch/alpha/kernel/ptrace.c
index 18c9a8b13..3e20e7091 100644
--- a/arch/alpha/kernel/ptrace.c
+++ b/arch/alpha/kernel/ptrace.c
@@ -246,26 +246,6 @@ put_long(struct task_struct * tsk, struct vm_area_struct * vma,
 	flush_tlb();
 }
 
-static struct vm_area_struct *
-find_extend_vma(struct task_struct * tsk, unsigned long addr)
-{
-	struct vm_area_struct * vma;
-
-	addr &= PAGE_MASK;
-	vma = find_vma(tsk->mm,addr);
-	if (!vma)
-		return NULL;
-	if (vma->vm_start <= addr)
-		return vma;
-	if (!(vma->vm_flags & VM_GROWSDOWN))
-		return NULL;
-	if (vma->vm_end - addr > tsk->rlim[RLIMIT_STACK].rlim_cur)
-		return NULL;
-	vma->vm_offset -= vma->vm_start - addr;
-	vma->vm_start = addr;
-	return vma;
-}
-
 /*
  * This routine checks the page boundaries, and that the offset is
  * within the task area. It then calls get_long() to read a long.
@@ -506,7 +486,8 @@ sys_ptrace(long request, long pid, long addr, long data,
 		     (current->uid != child->uid) ||
 		     (current->gid != child->egid) ||
 		     (current->gid != child->sgid) ||
-		     (current->gid != child->gid))
+		     (current->gid != child->gid) ||
+		     (!cap_issubset(child->cap_permitted, current->cap_permitted)))
 		    && !capable(CAP_SYS_PTRACE))
 			goto out;
 		/* the same process cannot be attached many times */
diff --git a/arch/alpha/kernel/signal.c b/arch/alpha/kernel/signal.c
index 99bd36785..a477c2344 100644
--- a/arch/alpha/kernel/signal.c
+++ b/arch/alpha/kernel/signal.c
@@ -24,6 +24,9 @@
 #include <asm/sigcontext.h>
 #include <asm/ucontext.h>
 
+#include "proto.h"
+
+
 #define DEBUG_SIG 0
 
 #define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
diff --git a/arch/alpha/kernel/smp.c b/arch/alpha/kernel/smp.c
index aa1eaf363..f01c0e55d 100644
--- a/arch/alpha/kernel/smp.c
+++ b/arch/alpha/kernel/smp.c
@@ -18,6 +18,7 @@
 #include <asm/ptrace.h>
 #include <asm/atomic.h>
 
+#include <asm/io.h>
 #include <asm/irq.h>
 #include <asm/bitops.h>
 #include <asm/pgtable.h>
@@ -29,6 +30,8 @@
 #include <asm/unistd.h>
 
 #include "proto.h"
+#include "irq.h"
+
 
 #define DEBUG_SMP 0
 #if DEBUG_SMP
@@ -37,62 +40,44 @@
 #define DBGS(args)
 #endif
 
-struct ipi_msg_flush_tb_struct {
-	volatile unsigned int flush_tb_mask;
-	union {
-		struct mm_struct *	flush_mm;
-		struct vm_area_struct *	flush_vma;
-	} p;
-	unsigned long flush_addr;
-	unsigned long flush_end;
-};
-
-static struct ipi_msg_flush_tb_struct ipi_msg_flush_tb __cacheline_aligned;
-static spinlock_t flush_tb_lock = SPIN_LOCK_UNLOCKED;
-
+/* A collection of per-processor data.  */
 struct cpuinfo_alpha cpu_data[NR_CPUS];
 
-spinlock_t ticker_lock = SPIN_LOCK_UNLOCKED;
-spinlock_t kernel_flag = SPIN_LOCK_UNLOCKED;
-
-unsigned int boot_cpu_id = 0;
-static int smp_activated = 0;
+/* A collection of single bit ipi messages.  */
+static struct {
+	unsigned long bits __cacheline_aligned;
+} ipi_data[NR_CPUS];
 
-int smp_found_config = 0; /* Have we found an SMP box */
-static int max_cpus = -1;
+enum ipi_message_type {
+        IPI_RESCHEDULE,
+        IPI_CALL_FUNC,
+        IPI_CPU_STOP,
+};
 
-unsigned int cpu_present_map = 0;
+spinlock_t kernel_flag __cacheline_aligned = SPIN_LOCK_UNLOCKED;
 
-int smp_num_cpus = 1;
-int smp_num_probed = 0; /* Internal processor count */
+/* Set to a secondary's cpuid when it comes online.  */
+static unsigned long smp_secondary_alive;
 
-int smp_threads_ready = 0;
-volatile unsigned long cpu_callin_map[NR_CPUS] = {0,};
-volatile unsigned long smp_spinning[NR_CPUS] = { 0, };
+unsigned long cpu_present_mask;	/* Which cpus ids came online.  */
 
+static int max_cpus = -1;	/* Command-line limitation.  */
+int smp_boot_cpuid;		/* Which processor we booted from.  */
+int smp_num_probed;		/* Internal processor count */
+int smp_num_cpus = 1;		/* Number that came online.  */
+int smp_threads_ready;		/* True once the per process idle is forked. */
 cycles_t cacheflush_time;
 
-unsigned int prof_multiplier[NR_CPUS];
-unsigned int prof_counter[NR_CPUS];
-
-volatile int ipi_bits[NR_CPUS] __cacheline_aligned;
-
-unsigned long boot_cpu_palrev;
-
-volatile int smp_commenced = 0;
-volatile int smp_processors_ready = 0;
-
-volatile int cpu_number_map[NR_CPUS];
-volatile int cpu_logical_map[NR_CPUS];
+int cpu_number_map[NR_CPUS];
+int __cpu_logical_map[NR_CPUS];
 
 extern void calibrate_delay(void);
-extern struct thread_struct * original_pcb_ptr;
+extern asmlinkage void entInt(void);
 
-static void smp_setup_percpu_timer(void);
-static void secondary_cpu_start(int, struct task_struct *);
-static void send_cpu_msg(char *, int);
-
-/* Process bootcommand SMP options, like "nosmp" and "maxcpus=" */
+
+/*
+ * Process bootcommand SMP options, like "nosmp" and "maxcpus=".
+ */
 void __init
 smp_setup(char *str, int *ints)
 {
@@ -102,100 +87,87 @@ smp_setup(char *str, int *ints)
 		max_cpus = 0;
 }
 
-static void __init
-smp_store_cpu_info(int id)
+/*
+ * Called by both boot and secondaries to move global data into
+ *  per-processor storage.
+ */
+static inline void __init
+smp_store_cpu_info(int cpuid)
 {
-	/* This is it on Alpha, so far. */
-	cpu_data[id].loops_per_sec = loops_per_sec;
+	cpu_data[cpuid].loops_per_sec = loops_per_sec;
 }
 
-void __init
-smp_commence(void)
+/*
+ * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
+ */
+static inline void __init
+smp_setup_percpu_timer(int cpuid)
 {
-	/* Lets the callin's below out of their loop. */
-	mb();
-	smp_commenced = 1;
+	cpu_data[cpuid].prof_counter = 1;
+	cpu_data[cpuid].prof_multiplier = 1;
+
+#ifdef NOT_YET_PROFILING
+	load_profile_irq(mid_xlate[cpu], lvl14_resolution);
+	if (cpu == smp_boot_cpuid)
+		enable_pil_irq(14);
+#endif
 }
 
+/*
+ * Where secondaries begin a life of C.
+ */
 void __init
 smp_callin(void)
 {
 	int cpuid = hard_smp_processor_id();
 
 	DBGS(("CALLIN %d state 0x%lx\n", cpuid, current->state));
-#ifdef HUH
-	local_flush_cache_all();
-	local_flush_tlb_all();
-#endif
-#if 0
-	set_irq_udt(mid_xlate[boot_cpu_id]);
-#endif
+
+	/* Turn on machine checks.  */
+	wrmces(7);
+
+	/* Set trap vectors.  */
+	trap_init();
+
+	/* Set interrupt vector.  */
+	wrent(entInt, 0);
+
+	/* Setup the scheduler for this processor.  */
+	init_idle();
 
 	/* Get our local ticker going. */
-	smp_setup_percpu_timer();
+	smp_setup_percpu_timer(cpuid);
 
-#if 0
+	/* Must have completely accurate bogos.  */
+	__sti();
 	calibrate_delay();
-#endif
 	smp_store_cpu_info(cpuid);
-#ifdef HUH
-	local_flush_cache_all();
-	local_flush_tlb_all();
-#endif
 
 	/* Allow master to continue. */
-	set_bit(cpuid, (unsigned long *)&cpu_callin_map[cpuid]);
-#ifdef HUH
-	local_flush_cache_all();
-	local_flush_tlb_all();
-#endif
-
-#ifdef NOT_YET
-	while(!task[cpuid] || current_set[cpuid] != task[cpuid])
-	        barrier();
-#endif
+	wmb();
+	smp_secondary_alive = cpuid;
 
-#ifdef HUH
-	local_flush_cache_all();
-	local_flush_tlb_all();
-#endif
-#if 0
-	__sti();
-#endif
-}
+	/* Wait for the go code.  */
+	while (!smp_threads_ready)
+		barrier();
 
-asmlinkage int __init
-start_secondary(void *unused)
-{
-	extern asmlinkage void entInt(void);
-	extern void paging_init_secondary(void);
+	printk(KERN_INFO "SMP: commencing CPU %d current %p\n",
+	       cpuid, current);
 
-	wrmces(7);
-	paging_init_secondary();
-	trap_init();
-	wrent(entInt, 0);
-
-	smp_callin();
-	while (!smp_commenced)
-		barrier();
-#if 1
-	printk("start_secondary: commencing CPU %d current %p\n",
-	       hard_smp_processor_id(), current);
-#endif
+	/* Do nothing.  */
 	cpu_idle(NULL);
 }
 
+
+/*
+ * Rough estimation for SMP scheduling, this is the number of cycles it
+ * takes for a fully memory-limited process to flush the SMP-local cache.
+ *
+ * We are not told how much cache there is, so we have to guess.
+ */
 static void __init
 smp_tune_scheduling (void)
 {
-	/*
-	 * Rough estimation for SMP scheduling, this is the number of
-	 * cycles it takes for a fully memory-limited process to flush
-	 * the SMP-local cache.
-	 *
-	 * We are not told how much cache there is, so we have to guess.
-	 */
-
 	struct percpu_struct *cpu;
 	unsigned long on_chip_cache;
 	unsigned long freq;
@@ -231,259 +203,159 @@ smp_tune_scheduling (void)
 	cacheflush_time = freq / 1024 * on_chip_cache / 5000;
 }
 
-
 /*
- *      Cycle through the processors sending START msgs to boot each.
+ * Send a message to a secondary's console.  "START" is one such
+ * interesting message.  ;-)
  */
-void __init
-smp_boot_cpus(void)
+static void
+send_secondary_console_msg(char *str, int cpuid)
 {
-	int cpucount = 0;
-	int i, first, prev;
-
-	printk("Entering SMP Mode.\n");
-
-#if 0
-	__sti();
-#endif
-
-	for(i=0; i < NR_CPUS; i++) {
-		cpu_number_map[i] = -1;
-		cpu_logical_map[i] = -1;
-	        prof_counter[i] = 1;
-	        prof_multiplier[i] = 1;
-		ipi_bits[i] = 0;
-	}
-
-	cpu_number_map[boot_cpu_id] = 0;
-	cpu_logical_map[0] = boot_cpu_id;
-	current->processor = boot_cpu_id; /* ??? */
-
-	smp_store_cpu_info(boot_cpu_id);
-	smp_tune_scheduling();
-#ifdef NOT_YET
-	printk("CPU%d: ", boot_cpu_id);
-	print_cpu_info(&cpu_data[boot_cpu_id]);
-	set_irq_udt(mid_xlate[boot_cpu_id]);
-#endif
-	smp_setup_percpu_timer();
-#ifdef HUH
-	local_flush_cache_all();
-#endif
-	if (smp_num_probed == 1)
-		return;  /* Not an MP box. */
-
-#if NOT_YET
-	/*
-	 * If SMP should be disabled, then really disable it!
-	 */
-	if (!max_cpus)
-	{
-		smp_found_config = 0;
-	        printk(KERN_INFO "SMP mode deactivated.\n");
-	}
-#endif
-
-	for (i = 0; i < NR_CPUS; i++) {
+	struct percpu_struct *cpu;
+	register char *cp1, *cp2;
+	unsigned long cpumask;
+	size_t len;
+	long timeout;
 
-		if (i == boot_cpu_id)
-			continue;
+	cpu = (struct percpu_struct *)
+		((char*)hwrpb
+		 + hwrpb->processor_offset
+		 + cpuid * hwrpb->processor_size);
 
-	        if (cpu_present_map & (1 << i)) {
-	                struct task_struct *idle;
-	                int timeout;
+	cpumask = (1L << cpuid);
+	if (hwrpb->txrdy & cpumask)
+		goto delay1;
+	ready1:
 
-	                /* Cook up an idler for this guy. */
-	                kernel_thread(start_secondary, NULL, CLONE_PID);
-	                idle = task[++cpucount];
-			if (!idle)
-				panic("No idle process for CPU %d", i);
-	                idle->processor = i;
+	cp2 = str;
+	len = strlen(cp2);
+	*(unsigned int *)&cpu->ipc_buffer[0] = len;
+	cp1 = (char *) &cpu->ipc_buffer[1];
+	memcpy(cp1, cp2, len);
 
-			DBGS(("smp_boot_cpus: CPU %d state 0x%lx flags 0x%lx\n",
-			      i, idle->state, idle->flags));
+	/* atomic test and set */
+	wmb();
+	set_bit(cpuid, &hwrpb->rxrdy);
 
-	                /* whirrr, whirrr, whirrrrrrrrr... */
-#ifdef HUH
-	                local_flush_cache_all();
-#endif
-	                secondary_cpu_start(i, idle);
+	if (hwrpb->txrdy & cpumask)
+		goto delay2;
+	ready2:
+	return;
 
-	                /* wheee... it's going... wait for 5 secs...*/
-	                for (timeout = 0; timeout < 50000; timeout++) {
-				if (cpu_callin_map[i])
-					break;
-	                        udelay(100);
-	                }
-	                if (cpu_callin_map[i]) {
-				/* Another "Red Snapper". */
-				cpu_number_map[i] = cpucount;
-	                        cpu_logical_map[cpucount] = i;
-	                } else {
-				cpucount--;
-	                        printk("smp_boot_cpus: Processor %d"
-				       " is stuck 0x%lx.\n", i, idle->flags);
-	                }
-	        }
-	        if (!(cpu_callin_map[i])) {
-			cpu_present_map &= ~(1 << i);
-	                cpu_number_map[i] = -1;
-	        }
-	}
-#ifdef HUH
-	local_flush_cache_all();
-#endif
-	if (cpucount == 0) {
-		printk("smp_boot_cpus: ERROR - only one Processor found.\n");
-	        cpu_present_map = (1 << smp_processor_id());
-	} else {
-		unsigned long bogosum = 0;
-	        for (i = 0; i < NR_CPUS; i++) {
-			if (cpu_present_map & (1 << i))
-				bogosum += cpu_data[i].loops_per_sec;
-	        }
-	        printk("smp_boot_cpus: Total of %d Processors activated"
-		       " (%lu.%02lu BogoMIPS).\n",
-	               cpucount + 1,
-	               (bogosum + 2500)/500000,
-	               ((bogosum + 2500)/5000)%100);
-	        smp_activated = 1;
-	        smp_num_cpus = cpucount + 1;
+delay1:
+	/* Wait one second.  Note that jiffies aren't ticking yet.  */
+	for (timeout = 100000; timeout > 0; --timeout) {
+		if (!(hwrpb->txrdy & cpumask))
+			goto ready1;
+		udelay(10);
+		barrier();
 	}
+	goto timeout;
 
-	/* Setup CPU list for IRQ distribution scheme. */
-	first = prev = -1;
-	for (i = 0; i < NR_CPUS; i++) {
-		if (cpu_present_map & (1 << i)) {
-			if (first == -1)
-				first = i;
-			if (prev != -1)
-				cpu_data[i].next = i;
-	                prev = i;
-	        }
+delay2:
+	/* Wait one second.  */
+	for (timeout = 100000; timeout > 0; --timeout) {
+		if (!(hwrpb->txrdy & cpumask))
+			goto ready2;
+		udelay(10);
+		barrier();
 	}
-	cpu_data[prev].next = first;
+	goto timeout;
 
-	/* Ok, they are spinning and ready to go. */
-	smp_processors_ready = 1;
+timeout:
+	printk("Processor %x not ready\n", cpuid);
+	return;
 }
 
-static void __init
-smp_setup_percpu_timer(void)
+/*
+ * A secondary console wants to send a message.  Receive it.
+ */
+static void
+recv_secondary_console_msg(void)
 {
-	int cpu = smp_processor_id();
-
-	prof_counter[cpu] = prof_multiplier[cpu] = 1;
-#ifdef NOT_YET
-	load_profile_irq(mid_xlate[cpu], lvl14_resolution);
-	if (cpu == boot_cpu_id)
-		enable_pil_irq(14);
-#endif
-}
-
-extern void update_one_process(struct task_struct *p, unsigned long ticks,
-	                       unsigned long user, unsigned long system,
-			       int cpu);
+	int mycpu, i, cnt;
+	unsigned long txrdy = hwrpb->txrdy;
+	char *cp1, *cp2, buf[80];
+	struct percpu_struct *cpu;
 
-void
-smp_percpu_timer_interrupt(struct pt_regs *regs)
-{
-	int cpu = smp_processor_id();
+	DBGS(("recv_secondary_console_msg: TXRDY 0x%lx.\n", txrdy));
 
-#ifdef NOT_YET
-	clear_profile_irq(mid_xlate[cpu]);
-	if(!user_mode(regs))
-		alpha_do_profile(regs->pc);
-#endif
+	mycpu = hard_smp_processor_id();
 
-	if (!--prof_counter[cpu]) {
-		int user = user_mode(regs);
-	        if (current->pid) {
-			update_one_process(current, 1, user, !user, cpu);
+	for (i = 0; i < NR_CPUS; i++) {
+		if (!(txrdy & (1L << i)))
+			continue;
 
-	                if (--current->counter < 0) {
-				current->counter = 0;
-	                        current->need_resched = 1;
-	                }
+		DBGS(("recv_secondary_console_msg: "
+		      "TXRDY contains CPU %d.\n", i));
 
-	                spin_lock(&ticker_lock);
-	                if (user) {
-				if (current->priority < DEF_PRIORITY) {
-					kstat.cpu_nice++;
-					kstat.per_cpu_nice[cpu]++;
-				} else {
-					kstat.cpu_user++;
-					kstat.per_cpu_user[cpu]++;
-				}
-	                } else {
-				kstat.cpu_system++;
-				kstat.per_cpu_system[cpu]++;
-	                }
-	                spin_unlock(&ticker_lock);
-	        }
-	        prof_counter[cpu] = prof_multiplier[cpu];
-	}
-}
+		cpu = (struct percpu_struct *)
+		  ((char*)hwrpb
+		   + hwrpb->processor_offset
+		   + i * hwrpb->processor_size);
 
-int __init
-setup_profiling_timer(unsigned int multiplier)
-{
-#ifdef NOT_YET
-	int i;
-	unsigned long flags;
+ 		printk(KERN_INFO "recv_secondary_console_msg: on %d from %d"
+		       " HALT_REASON 0x%lx FLAGS 0x%lx\n",
+		       mycpu, i, cpu->halt_reason, cpu->flags);
 
-	/* Prevent level14 ticker IRQ flooding. */
-	if((!multiplier) || (lvl14_resolution / multiplier) < 500)
-	        return -EINVAL;
+		cnt = cpu->ipc_buffer[0] >> 32;
+		if (cnt <= 0 || cnt >= 80)
+			strcpy(buf, "<<< BOGUS MSG >>>");
+		else {
+			cp1 = (char *) &cpu->ipc_buffer[11];
+			cp2 = buf;
+			strcpy(cp2, cp1);
+			
+			while ((cp2 = strchr(cp2, '\r')) != 0) {
+				*cp2 = ' ';
+				if (cp2[1] == '\n')
+					cp2[1] = ' ';
+			}
+		}
 
-	save_and_cli(flags);
-	for(i = 0; i < NR_CPUS; i++) {
-	        if(cpu_present_map & (1 << i)) {
-	                load_profile_irq(mid_xlate[i], lvl14_resolution / multip
-lier);
-	                prof_multiplier[i] = multiplier;
-	        }
+		printk(KERN_INFO "recv_secondary_console_msg: on %d "
+		       "message is '%s'\n", mycpu, buf);
 	}
-	restore_flags(flags);
-
-	return 0;
-
-#endif
-  return -EINVAL;
-}
-
-/* Only broken Intel needs this, thus it should not even be
-   referenced globally.  */
 
-void __init
-initialize_secondary(void)
-{
+	hwrpb->txrdy = 0;
 }
 
-static void __init
+/*
+ * Convince the console to have a secondary cpu begin execution.
+ */
+static int __init
 secondary_cpu_start(int cpuid, struct task_struct *idle)
 {
 	struct percpu_struct *cpu;
-	int timeout;
+	struct pcb_struct *hwpcb;
+	long timeout;
 	  
 	cpu = (struct percpu_struct *)
 		((char*)hwrpb
 		 + hwrpb->processor_offset
 		 + cpuid * hwrpb->processor_size);
-
-	/* Set context to idle thread this CPU will use when running
-	   assumption is that the idle thread is all set to go... ??? */
-	memcpy(&cpu->hwpcb[0], &idle->tss, sizeof(struct pcb_struct));
-	cpu->hwpcb[4] = cpu->hwpcb[0]; /* UNIQUE set to KSP ??? */
-
-	DBGS(("KSP 0x%lx PTBR 0x%lx VPTBR 0x%lx\n",
-	      cpu->hwpcb[0], cpu->hwpcb[2], hwrpb->vptb));
+	hwpcb = (struct pcb_struct *) cpu->hwpcb;
+
+	/* Initialize the CPU's HWPCB to something just good enough for
+	   us to get started.  Immediately after starting, we'll swpctx
+	   to the target idle task's tss.  Reuse the stack in the mean
+	   time.  Precalculate the target PCBB.  */
+	hwpcb->ksp = (unsigned long) idle + sizeof(union task_union) - 16;
+	hwpcb->usp = 0;
+	hwpcb->ptbr = idle->tss.ptbr;
+	hwpcb->pcc = 0;
+	hwpcb->asn = 0;
+	hwpcb->unique = virt_to_phys(&idle->tss);
+	hwpcb->flags = idle->tss.pal_flags;
+	hwpcb->res1 = hwpcb->res2 = 0;
+
+	DBGS(("KSP 0x%lx PTBR 0x%lx VPTBR 0x%lx UNIQUE 0x%lx\n",
+	      hwpcb->ksp, hwpcb->ptbr, hwrpb->vptb, hwcpb->unique));
 	DBGS(("Starting secondary cpu %d: state 0x%lx pal_flags 0x%lx\n",
 	      cpuid, idle->state, idle->tss.pal_flags));
 
 	/* Setup HWRPB fields that SRM uses to activate secondary CPU */
-	hwrpb->CPU_restart = __start_cpu;
-	hwrpb->CPU_restart_data = (unsigned long) idle;
+	hwrpb->CPU_restart = __smp_callin;
+	hwrpb->CPU_restart_data = (unsigned long) __smp_callin;
 
 	/* Recalculate and update the HWRPB checksum */
 	hwrpb_update_checksum(hwrpb);
@@ -495,99 +367,97 @@ secondary_cpu_start(int cpuid, struct task_struct *idle)
 	/* SRM III 3.4.1.3 */
 	cpu->flags |= 0x22;	/* turn on Context Valid and Restart Capable */
 	cpu->flags &= ~1;	/* turn off Bootstrap In Progress */
-	mb();
+	wmb();
 
-	send_cpu_msg("START\r\n", cpuid);
+	send_secondary_console_msg("START\r\n", cpuid);
 
-	/* now, we wait... */
-	for (timeout = 10000; !(cpu->flags & 1); timeout--) {
-		if (timeout <= 0) {
-			printk("Processor %d failed to start\n", cpuid);
-	                        /* needed for pset_info to work */
-#if 0
-			ipc_processor_enable(cpu_to_processor(cpunum));
-#endif
-			return;
-		}
-		mdelay(1);
+	/* Wait 1 second for an ACK from the console.  Note that jiffies 
+	   aren't ticking yet.  */
+	for (timeout = 100000; timeout > 0; timeout--) {
+		if (cpu->flags & 1)
+			goto started;
+		udelay(10);
 		barrier();
 	}
+	printk(KERN_ERR "SMP: Processor %d failed to start.\n", cpuid);
+	return -1;
+
+started:
 	DBGS(("secondary_cpu_start: SUCCESS for CPU %d!!!\n", cpuid));
+	return 0;
 }
 
-static void
-send_cpu_msg(char *str, int cpuid)
+/*
+ * Bring one cpu online.
+ */
+static int __init
+smp_boot_one_cpu(int cpuid, int cpunum)
 {
-	struct percpu_struct *cpu;
-	register char *cp1, *cp2;
-	unsigned long cpumask;
-	size_t len;
-	int timeout;
-
-	cpu = (struct percpu_struct *)
-		((char*)hwrpb
-		 + hwrpb->processor_offset
-		 + cpuid * hwrpb->processor_size);
-
-	cpumask = (1L << cpuid);
-	if (hwrpb->txrdy & cpumask)
-		goto delay1;
-	ready1:
-
-	cp2 = str;
-	len = strlen(cp2);
-	*(unsigned int *)&cpu->ipc_buffer[0] = len;
-	cp1 = (char *) &cpu->ipc_buffer[1];
-	memcpy(cp1, cp2, len);
-
-	/* atomic test and set */
-	set_bit(cpuid, &hwrpb->rxrdy);
-
-	if (hwrpb->txrdy & cpumask)
-		goto delay2;
-	ready2:
-	return;
-
-delay1:
-	for (timeout = 10000; timeout > 0; --timeout) {
-		if (!(hwrpb->txrdy & cpumask))
-			goto ready1;
-		udelay(100);
+	struct task_struct *idle;
+	long timeout;
+
+	/* Cook up an idler for this guy.  Note that the address we give
+	   to kernel_thread is irrelevant -- it's going to start where
+	   HWRPB.CPU_restart says to start.  But this gets all the other
+	   task-y sort of data structures set up like we wish.  */
+	kernel_thread((void *)__smp_callin, NULL, CLONE_PID|CLONE_VM);
+	idle = task[cpunum];
+	if (!idle)
+		panic("No idle process for CPU %d", cpuid);
+	idle->processor = cpuid;
+
+	/* Schedule the first task manually.  */
+	/* ??? Ingo, what is this?  */
+	idle->has_cpu = 1;
+
+	DBGS(("smp_boot_one_cpu: CPU %d state 0x%lx flags 0x%lx\n",
+	      cpuid, idle->state, idle->flags));
+
+	/* The secondary will change this once it is happy.  Note that
+	   secondary_cpu_start contains the necessary memory barrier.  */
+	smp_secondary_alive = -1;
+
+	/* Whirrr, whirrr, whirrrrrrrrr... */
+	if (secondary_cpu_start(cpuid, idle))
+		return -1;
+
+	/* We've been acked by the console; wait one second for the task
+	   to start up for real.  Note that jiffies aren't ticking yet.  */
+	for (timeout = 0; timeout < 100000; timeout++) {
+		if (smp_secondary_alive != -1)
+			goto alive;
+		udelay(10);
 		barrier();
 	}
-	goto timeout;
 
-delay2:
-	for (timeout = 10000; timeout > 0; --timeout) {
-		if (!(hwrpb->txrdy & cpumask))
-			goto ready2;
-		udelay(100);
-		barrier();
-	}
-	goto timeout;
+	printk(KERN_ERR "SMP: Processor %d is stuck.\n", cpuid);
+	return -1;
 
-timeout:
-	printk("Processor %x not ready\n", cpuid);
-	return;
+alive:
+	/* Another "Red Snapper". */
+	cpu_number_map[cpuid] = cpunum;
+	__cpu_logical_map[cpunum] = cpuid;
+	return 0;
 }
 
 /*
- * setup_smp()
- *
- * called from arch/alpha/kernel/setup.c:setup_arch() when __SMP__ defined
+ * Called from setup_arch.  Detect an SMP system and which processors
+ * are present.
  */
 void __init
 setup_smp(void)
 {
 	struct percpu_struct *cpubase, *cpu;
 	int i;
-	  
-	boot_cpu_id = hard_smp_processor_id();
-	if (boot_cpu_id != 0) {
-		printk("setup_smp: boot_cpu_id != 0 (%d).\n", boot_cpu_id);
+
+	smp_boot_cpuid = hard_smp_processor_id();
+	if (smp_boot_cpuid != 0) {
+		printk(KERN_WARNING "SMP: Booting off cpu %d instead of 0?\n",
+		       smp_boot_cpuid);
 	}
 
 	if (hwrpb->nr_processors > 1) {
+		int boot_cpu_palrev;
 
 		DBGS(("setup_smp: nr_processors %ld\n",
 		      hwrpb->nr_processors));
@@ -601,10 +471,9 @@ setup_smp(void)
 				((char *)cpubase + i*hwrpb->processor_size);
 			if ((cpu->flags & 0x1cc) == 0x1cc) {
 				smp_num_probed++;
-				/* assume here that "whami" == index */
-				cpu_present_map |= (1 << i);
-				if (i != boot_cpu_id)
-				  cpu->pal_revision = boot_cpu_palrev;
+				/* Assume here that "whami" == index */
+				cpu_present_mask |= (1L << i);
+				cpu->pal_revision = boot_cpu_palrev;
 			}
 
 			DBGS(("setup_smp: CPU %d: flags 0x%lx type 0x%lx\n",
@@ -614,76 +483,249 @@ setup_smp(void)
 		}
 	} else {
 		smp_num_probed = 1;
-		cpu_present_map = (1 << boot_cpu_id);
+		cpu_present_mask = (1L << smp_boot_cpuid);
 	}
-	printk("setup_smp: %d CPUs probed, cpu_present_map 0x%x,"
-	       " boot_cpu_id %d\n",
-	       smp_num_probed, cpu_present_map, boot_cpu_id);
+
+	printk(KERN_INFO "SMP: %d CPUs probed -- cpu_present_mask = %lx\n",
+	       smp_num_probed, cpu_present_mask);
 }
 
-static void
-secondary_console_message(void)
+/*
+ * Called by smp_init bring all the secondaries online and hold them.
+ */
+void __init
+smp_boot_cpus(void)
 {
-	int mycpu, i, cnt;
-	unsigned long txrdy = hwrpb->txrdy;
-	char *cp1, *cp2, buf[80];
-	struct percpu_struct *cpu;
+	int cpu_count, i;
+	unsigned long bogosum;
 
-	DBGS(("secondary_console_message: TXRDY 0x%lx.\n", txrdy));
+	/* Take care of some initial bookkeeping.  */
+	memset(cpu_number_map, -1, sizeof(cpu_number_map));
+	memset(__cpu_logical_map, -1, sizeof(__cpu_logical_map));
+	memset(ipi_data, 0, sizeof(ipi_data));
 
-	mycpu = hard_smp_processor_id();
+	cpu_number_map[smp_boot_cpuid] = 0;
+	__cpu_logical_map[0] = smp_boot_cpuid;
+	current->processor = smp_boot_cpuid;
+
+	smp_store_cpu_info(smp_boot_cpuid);
+	smp_tune_scheduling();
+	smp_setup_percpu_timer(smp_boot_cpuid);
+
+	init_idle();
 
+	/* Nothing to do on a UP box, or when told not to.  */
+	if (smp_num_probed == 1 || max_cpus == 0) {
+	        printk(KERN_INFO "SMP mode deactivated.\n");
+		return;
+	}
+
+	printk(KERN_INFO "SMP starting up secondaries.\n");
+
+	cpu_count = 1;
 	for (i = 0; i < NR_CPUS; i++) {
-		if (!(txrdy & (1L << i)))
+		if (i == smp_boot_cpuid)
 			continue;
 
-		DBGS(("secondary_console_message: "
-		      "TXRDY contains CPU %d.\n", i));
+	        if (((cpu_present_mask >> i) & 1) == 0)
+			continue;
 
-		cpu = (struct percpu_struct *)
-		  ((char*)hwrpb
-		   + hwrpb->processor_offset
-		   + i * hwrpb->processor_size);
+		if (smp_boot_one_cpu(i, cpu_count))
+			continue;
 
- 		printk("secondary_console_message: on %d from %d"
-		       " HALT_REASON 0x%lx FLAGS 0x%lx\n",
-		       mycpu, i, cpu->halt_reason, cpu->flags);
+		cpu_count++;
+	}
 
-		cnt = cpu->ipc_buffer[0] >> 32;
-		if (cnt <= 0 || cnt >= 80)
-			strcpy(buf, "<<< BOGUS MSG >>>");
-		else {
-			cp1 = (char *) &cpu->ipc_buffer[11];
-			cp2 = buf;
-			strcpy(cp2, cp1);
-			
-			while ((cp2 = strchr(cp2, '\r')) != 0) {
-				*cp2 = ' ';
-				if (cp2[1] == '\n')
-					cp2[1] = ' ';
-			}
-		}
+	if (cpu_count == 1) {
+		printk(KERN_ERR "SMP: Only one lonely processor alive.\n");
+		return;
+	}
+
+	bogosum = 0;
+        for (i = 0; i < NR_CPUS; i++) {
+		if (cpu_present_mask & (1L << i))
+			bogosum += cpu_data[i].loops_per_sec;
+        }
+	printk(KERN_INFO "SMP: Total of %d processors activated "
+	       "(%lu.%02lu BogoMIPS).\n",
+	       cpu_count, (bogosum + 2500) / 500000,
+	       ((bogosum + 2500) / 5000) % 100);
+
+	smp_num_cpus = cpu_count;
+}
+
+/*
+ * Called by smp_init to release the blocking online cpus once they 
+ * are all started.
+ */
+void __init
+smp_commence(void)
+{
+	/* smp_init sets smp_threads_ready -- that's enough.  */
+	mb();
+}
+
+/*
+ * Only broken Intel needs this, thus it should not even be
+ * referenced globally.
+ */
+
+void __init
+initialize_secondary(void)
+{
+}
+
+
+extern void update_one_process(struct task_struct *p, unsigned long ticks,
+	                       unsigned long user, unsigned long system,
+			       int cpu);
 
-		printk("secondary_console_message: on %d message is '%s'\n",
-		       mycpu, buf);
+void
+smp_percpu_timer_interrupt(struct pt_regs *regs)
+{
+	int cpu = smp_processor_id();
+	int user = user_mode(regs);
+	struct cpuinfo_alpha *data = &cpu_data[cpu];
+
+#ifdef NOT_YET_PROFILING
+	clear_profile_irq(mid_xlate[cpu]);
+	if (!user)
+		alpha_do_profile(regs->pc);
+#endif
+
+	if (!--data->prof_counter) {
+		/* We need to make like a normal interrupt -- otherwise
+		   timer interrupts ignore the global interrupt lock,
+		   which would be a Bad Thing.  */
+		irq_enter(cpu, TIMER_IRQ);
+
+		update_one_process(current, 1, user, !user, cpu);
+	        if (current->pid) {
+	                if (--current->counter < 0) {
+				current->counter = 0;
+	                        current->need_resched = 1;
+	                }
+
+	                if (user) {
+				if (current->priority < DEF_PRIORITY) {
+					kstat.cpu_nice++;
+					kstat.per_cpu_nice[cpu]++;
+				} else {
+					kstat.cpu_user++;
+					kstat.per_cpu_user[cpu]++;
+				}
+	                } else {
+				kstat.cpu_system++;
+				kstat.per_cpu_system[cpu]++;
+	                }
+	        }
+
+		data->prof_counter = data->prof_multiplier;
+		irq_exit(cpu, TIMER_IRQ);
 	}
+}
 
-	hwrpb->txrdy = 0;
+int __init
+setup_profiling_timer(unsigned int multiplier)
+{
+#ifdef NOT_YET_PROFILING
+	int i;
+	unsigned long flags;
+
+	/* Prevent level14 ticker IRQ flooding. */
+	if((!multiplier) || (lvl14_resolution / multiplier) < 500)
+	        return -EINVAL;
+
+	save_and_cli(flags);
+	for (i = 0; i < NR_CPUS; i++) {
+	        if (cpu_present_mask & (1L << i)) {
+	                load_profile_irq(mid_xlate[i],
+					 lvl14_resolution / multiplier);
+	                prof_multiplier[i] = multiplier;
+	        }
+	}
+	restore_flags(flags);
+
+	return 0;
+#else
+	return -EINVAL;
+#endif
 }
 
-enum ipi_message_type {
-	IPI_TLB_ALL,
-	IPI_TLB_MM,
-	IPI_TLB_PAGE,
-	IPI_RESCHEDULE,
-	IPI_CPU_STOP
+
+static void
+send_ipi_message(unsigned long to_whom, enum ipi_message_type operation)
+{
+	long i, j;
+
+	/* Reduce the number of memory barriers by doing two loops,
+	   one to set the bits, one to invoke the interrupts.  */
+
+	mb();	/* Order out-of-band data and bit setting. */
+
+	for (i = 0, j = 1; i < NR_CPUS; ++i, j <<= 1) {
+		if (to_whom & j)
+			set_bit(operation, &ipi_data[i].bits);
+	}
+
+	mb();	/* Order bit setting and interrupt. */
+
+	for (i = 0, j = 1; i < NR_CPUS; ++i, j <<= 1) {
+		if (to_whom & j)
+			wripir(i);
+	}
+}
+
+/* Structure and data for smp_call_function.  This is designed to 
+   minimize static memory requirements.  Plus it looks cleaner.  */
+
+struct smp_call_struct {
+	void (*func) (void *info);
+	void *info;
+	long wait;
+	atomic_t unstarted_count;
+	atomic_t unfinished_count;
 };
 
+static struct smp_call_struct *smp_call_function_data;
+
+/* Atomicly drop data into a shared pointer.  The pointer is free if
+   it is initially locked.  If retry, spin until free.  */
+
+static inline int
+pointer_lock (void *lock, void *data, int retry)
+{
+	void *old, *tmp;
+
+	mb();
+again:
+	/* Compare and swap with zero.  */
+	asm volatile (
+	"1:	ldq_l	%0,%1\n"
+	"	mov	%3,%2\n"
+	"	bne	%0,2f\n"
+	"	stq_c	%2,%1\n"
+	"	beq	%2,1b\n"
+	"2:"
+	: "=&r"(old), "=m"(*(void **)lock), "=&r"(tmp)
+	: "r"(data)
+	: "memory");
+
+	if (old == 0)
+		return 0;
+	if (! retry)
+		return -EBUSY;
+
+	while (*(void **)lock)
+		schedule();
+	goto again;
+}
+
 void
 handle_ipi(struct pt_regs *regs)
 {
 	int this_cpu = smp_processor_id();
-	volatile int * pending_ipis = &ipi_bits[this_cpu];
+	unsigned long *pending_ipis = &ipi_data[this_cpu].bits;
 	unsigned long ops;
 
 	DBGS(("handle_ipi: on CPU %d ops 0x%x PC 0x%lx\n",
@@ -699,190 +741,189 @@ handle_ipi(struct pt_regs *regs)
 		ops &= ~which;
 		which = ffz(~which);
 
-		if (which < IPI_RESCHEDULE) {
-			if (which == IPI_TLB_ALL)
-				tbia();
-			else if (which == IPI_TLB_MM) {
-				struct mm_struct * mm;
-				mm = ipi_msg_flush_tb.p.flush_mm;
-				if (mm == current->mm)
-					flush_tlb_current(mm);
-			}
-			else /* IPI_TLB_PAGE */ {
-				struct vm_area_struct * vma;
-				struct mm_struct * mm;
-				unsigned long addr;
-
-				vma = ipi_msg_flush_tb.p.flush_vma;
-				mm = vma->vm_mm;
-				addr = ipi_msg_flush_tb.flush_addr;
-
-				if (mm == current->mm)
-					flush_tlb_current_page(mm, vma, addr);
-			}
-			clear_bit(this_cpu, &ipi_msg_flush_tb.flush_tb_mask);
-		}
-		else if (which == IPI_RESCHEDULE) {
+		if (which == IPI_RESCHEDULE) {
 			/* Reschedule callback.  Everything to be done
 			   is done by the interrupt return path.  */
 		}
+		else if (which == IPI_CALL_FUNC) {
+			struct smp_call_struct *data;
+			void (*func)(void *info);
+			void *info;
+			int wait;
+
+			data = smp_call_function_data;
+			func = data->func;
+			info = data->info;
+			wait = data->wait;
+
+			/* Notify the sending CPU that the data has been
+			   received, and execution is about to begin.  */
+			mb();
+			atomic_dec (&data->unstarted_count);
+
+			/* At this point the structure may be gone unless
+			   wait is true.  */
+			(*func)(info);
+
+			/* Notify the sending CPU that the task is done.  */
+			mb();
+			if (wait) atomic_dec (&data->unfinished_count);
+		}
 		else if (which == IPI_CPU_STOP) {
 			halt();
 		}
 		else {
-			printk(KERN_CRIT "unknown_ipi() on CPU %d: %lu\n",
+			printk(KERN_CRIT "Unknown IPI on CPU %d: %lu\n",
 			       this_cpu, which);
 		}
 	  } while (ops);
+
 	  mb();	/* Order data access and bit testing. */
 	}
 
 	cpu_data[this_cpu].ipi_count++;
 
 	if (hwrpb->txrdy)
-		secondary_console_message();
+		recv_secondary_console_msg();
 }
 
-static void
-send_ipi_message(unsigned long to_whom, enum ipi_message_type operation)
+void
+smp_send_reschedule(int cpu)
 {
-	long i, j;
-
-	/* Reduce the number of memory barriers by doing two loops,
-	   one to set the bits, one to invoke the interrupts.  */
-
-	mb();	/* Order out-of-band data and bit setting. */
-
-	for (i = 0, j = 1; i < NR_CPUS; ++i, j <<= 1) {
-		if (to_whom & j)
-			set_bit(operation, &ipi_bits[i]);
-	}
-
-	mb();	/* Order bit setting and interrupt. */
+	send_ipi_message(1L << cpu, IPI_RESCHEDULE);
+}
 
-	for (i = 0, j = 1; i < NR_CPUS; ++i, j <<= 1) {
-		if (to_whom & j)
-			wripir(i);
-	}
+void
+smp_send_stop(void)
+{
+	unsigned long to_whom = cpu_present_mask ^ (1L << smp_processor_id());
+	send_ipi_message(to_whom, IPI_CPU_STOP);
 }
 
+/*
+ * Run a function on all other CPUs.
+ *  <func>	The function to run. This must be fast and non-blocking.
+ *  <info>	An arbitrary pointer to pass to the function.
+ *  <retry>	If true, keep retrying until ready.
+ *  <wait>	If true, wait until function has completed on other CPUs.
+ *  [RETURNS]   0 on success, else a negative status code.
+ *
+ * Does not return until remote CPUs are nearly ready to execute <func>
+ * or are or have executed.
+ */
+
 int
-smp_info(char *buffer)
+smp_call_function (void (*func) (void *info), void *info, int retry, int wait)
 {
-	long i;
-	unsigned long sum = 0;
-	for (i = 0; i < NR_CPUS; i++)
-		sum += cpu_data[i].ipi_count;
+	unsigned long to_whom = cpu_present_mask ^ (1L << smp_processor_id());
+	struct smp_call_struct data;
+	long timeout;
+	
+	data.func = func;
+	data.info = info;
+	data.wait = wait;
+	atomic_set(&data.unstarted_count, smp_num_cpus - 1);
+	atomic_set(&data.unfinished_count, smp_num_cpus - 1);
+
+	/* Aquire the smp_call_function_data mutex.  */
+	if (pointer_lock(&smp_call_function_data, &data, retry))
+		return -EBUSY;
+
+	/* Send a message to all other CPUs.  */
+	send_ipi_message(to_whom, IPI_CALL_FUNC);
+
+	/* Wait for a minimal response.  */
+	timeout = jiffies + HZ;
+	while (atomic_read (&data.unstarted_count) > 0
+	       && time_before (jiffies, timeout))
+		barrier();
 
-	return sprintf(buffer, "CPUs probed %d active %d map 0x%x IPIs %ld\n",
-		       smp_num_probed, smp_num_cpus, cpu_present_map, sum);
-}
+	/* We either got one or timed out -- clear the lock.  */
+	mb();
+	smp_call_function_data = 0;
+	if (atomic_read (&data.unstarted_count) > 0)
+		return -ETIMEDOUT;
+
+	/* Wait for a complete response, if needed.  */
+	if (wait) {
+		while (atomic_read (&data.unfinished_count) > 0)
+			barrier();
+	}
 
-void
-smp_send_reschedule(int cpu)
-{
-	send_ipi_message(1 << cpu, IPI_RESCHEDULE);
+	return 0;
 }
 
-void
-smp_send_stop(void)
+static void
+ipi_flush_tlb_all(void *ignored)
 {
-	unsigned long to_whom = cpu_present_map ^ (1 << smp_processor_id());
-	send_ipi_message(to_whom, IPI_CPU_STOP);
+	tbia();
 }
 
 void
 flush_tlb_all(void)
 {
-	unsigned long to_whom = cpu_present_map ^ (1 << smp_processor_id());
-	long timeout = 1000000;
-
-	spin_lock(&flush_tb_lock);
-
-	ipi_msg_flush_tb.flush_tb_mask = to_whom;
-	send_ipi_message(to_whom, IPI_TLB_ALL);
 	tbia();
 
-	while (ipi_msg_flush_tb.flush_tb_mask && --timeout) {
-		udelay(1);
-		barrier();
-	}
-
-	if (timeout == 0) {
-		printk("flush_tlb_all: STUCK on CPU %d mask 0x%x\n",
-		       smp_processor_id(),
-		       ipi_msg_flush_tb.flush_tb_mask);
-		ipi_msg_flush_tb.flush_tb_mask = 0;
+	/* Although we don't have any data to pass, we do want to
+	   synchronize with the other processors.  */
+	if (smp_call_function(ipi_flush_tlb_all, NULL, 1, 1)) {
+		printk(KERN_CRIT "flush_tlb_all: timed out\n");
 	}
+}
 
-	spin_unlock(&flush_tb_lock);
+static void
+ipi_flush_tlb_mm(void *x)
+{
+	struct mm_struct *mm = (struct mm_struct *) x;
+	if (mm == current->mm)
+		flush_tlb_current(mm);
 }
 
 void
 flush_tlb_mm(struct mm_struct *mm)
 {
-	unsigned long to_whom = cpu_present_map ^ (1 << smp_processor_id());
-	long timeout = 1000000;
-
-	spin_lock(&flush_tb_lock);
-
-	ipi_msg_flush_tb.flush_tb_mask = to_whom;
-	ipi_msg_flush_tb.p.flush_mm = mm;
-	send_ipi_message(to_whom, IPI_TLB_MM);
-
-	if (mm != current->mm)
-		flush_tlb_other(mm);
-	else
+	if (mm == current->mm)
 		flush_tlb_current(mm);
+	else
+		flush_tlb_other(mm);
 
-	while (ipi_msg_flush_tb.flush_tb_mask && --timeout) {
-		udelay(1);
-		barrier();
+	if (smp_call_function(ipi_flush_tlb_mm, mm, 1, 1)) {
+		printk(KERN_CRIT "flush_tlb_mm: timed out\n");
 	}
+}
 
-	if (timeout == 0) {
-		printk("flush_tlb_mm: STUCK on CPU %d mask 0x%x\n",
-		       smp_processor_id(),
-		       ipi_msg_flush_tb.flush_tb_mask);
-		ipi_msg_flush_tb.flush_tb_mask = 0;
-	}
+struct flush_tlb_page_struct {
+	struct vm_area_struct *vma;
+	struct mm_struct *mm;
+	unsigned long addr;
+};
 
-	spin_unlock(&flush_tb_lock);
+static void
+ipi_flush_tlb_page(void *x)
+{
+	struct flush_tlb_page_struct *data = (struct flush_tlb_page_struct *)x;
+	if (data->mm == current->mm)
+		flush_tlb_current_page(data->mm, data->vma, data->addr);
 }
 
 void
 flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
 {
-	int cpu = smp_processor_id();
-	unsigned long to_whom = cpu_present_map ^ (1 << cpu);
-	struct mm_struct * mm = vma->vm_mm;
-	int timeout = 1000000;
-
-	spin_lock(&flush_tb_lock);
+	struct flush_tlb_page_struct data;
+	struct mm_struct *mm = vma->vm_mm;
 
-	ipi_msg_flush_tb.flush_tb_mask = to_whom;
-	ipi_msg_flush_tb.p.flush_vma = vma;
-	ipi_msg_flush_tb.flush_addr = addr;
-	send_ipi_message(to_whom, IPI_TLB_PAGE);
+	data.vma = vma;
+	data.mm = mm;
+	data.addr = addr;
 
-	if (mm != current->mm)
-		flush_tlb_other(mm);
-	else
+	if (mm == current->mm)
 		flush_tlb_current_page(mm, vma, addr);
-
-	while (ipi_msg_flush_tb.flush_tb_mask && --timeout) {
-		udelay(1);
-		barrier();
-	}
-
-	if (timeout == 0) {
-		printk("flush_tlb_page: STUCK on CPU %d mask 0x%x\n",
-		       smp_processor_id(),
-		       ipi_msg_flush_tb.flush_tb_mask);
-		ipi_msg_flush_tb.flush_tb_mask = 0;
+	else
+		flush_tlb_other(mm);
+	
+	if (smp_call_function(ipi_flush_tlb_page, &data, 1, 1)) {
+		printk(KERN_CRIT "flush_tlb_page: timed out\n");
 	}
-
-	spin_unlock(&flush_tb_lock);
 }
 
 void
@@ -892,6 +933,20 @@ flush_tlb_range(struct mm_struct *mm, unsigned long start, unsigned long end)
 	flush_tlb_mm(mm);
 }
 
+
+int
+smp_info(char *buffer)
+{
+	long i;
+	unsigned long sum = 0;
+	for (i = 0; i < NR_CPUS; i++)
+		sum += cpu_data[i].ipi_count;
+
+	return sprintf(buffer, "CPUs probed %d active %d map 0x%lx IPIs %ld\n",
+		       smp_num_probed, smp_num_cpus, cpu_present_mask, sum);
+}
+
+
 #if DEBUG_SPINLOCK
 
 #ifdef MANAGE_SPINLOCK_IPL
@@ -932,17 +987,16 @@ void
 spin_lock(spinlock_t * lock)
 {
 	long tmp;
-	long stuck = 1<<27;
+	long stuck;
 	void *inline_pc = __builtin_return_address(0);
 	unsigned long started = jiffies;
 	int printed = 0;
 	int cpu = smp_processor_id();
 	long old_ipl = spinlock_raise_ipl(lock);
 
+	stuck = 1L << 28;
  try_again:
 
-	stuck = 0x10000000; /* was 4G, now 256M */
-
 	/* Use sub-sections to put the actual loop at the end
 	   of this object file's text section so as to perfect
 	   branch prediction.  */
@@ -961,19 +1015,16 @@ spin_lock(spinlock_t * lock)
 	"	blbs	%0,2b\n"
 	"	br	1b\n"
 	".previous"
-	: "=r" (tmp),
-	  "=m" (__dummy_lock(lock)),
-	  "=r" (stuck)
-	: "2" (stuck));
+	: "=r" (tmp), "=m" (__dummy_lock(lock)), "=r" (stuck)
+	: "1" (__dummy_lock(lock)), "2" (stuck));
 
 	if (stuck < 0) {
-		if (!printed) {
-			printk("spinlock stuck at %p(%d) owner %s at %p\n",
-			       inline_pc, cpu, lock->task->comm,
-			       lock->previous);
-			printed = 1;
-		}
-		stuck = 1<<30;
+		printk(KERN_WARNING
+		       "spinlock stuck at %p(%d) owner %s at %p(%d) st %ld\n",
+		       inline_pc, cpu, lock->task->comm, lock->previous,
+		       lock->task->processor, lock->task->state);
+		stuck = 1L << 36;
+		printed = 1;
 		goto try_again;
 	}
 
@@ -984,7 +1035,7 @@ spin_lock(spinlock_t * lock)
 	lock->task = current;
 
 	if (printed) {
-		printk("spinlock grabbed at %p(%d) %ld ticks\n",
+		printk(KERN_WARNING "spinlock grabbed at %p(%d) %ld ticks\n",
 		       inline_pc, cpu, jiffies - started);
 	}
 }
@@ -1006,7 +1057,7 @@ spin_trylock(spinlock_t * lock)
 	return ret;
 }
 #endif /* DEBUG_SPINLOCK */
-
+
 #if DEBUG_RWLOCK
 void write_lock(rwlock_t * lock)
 {
@@ -1038,18 +1089,17 @@ void write_lock(rwlock_t * lock)
 	"	blt	%1,8b\n"
 	"	br	1b\n"
 	".previous"
-	: "=m" (__dummy_lock(lock)), "=&r" (regx), "=&r" (regy)
-	, "=&r" (stuck_lock), "=&r" (stuck_reader)
-	: "0" (__dummy_lock(lock))
-	, "3" (stuck_lock), "4" (stuck_reader)
-	);
+	: "=m" (__dummy_lock(lock)), "=&r" (regx), "=&r" (regy),
+	  "=&r" (stuck_lock), "=&r" (stuck_reader)
+	: "0" (__dummy_lock(lock)), "3" (stuck_lock), "4" (stuck_reader));
 
 	if (stuck_lock < 0) {
-		printk("write_lock stuck at %p\n", inline_pc);
+		printk(KERN_WARNING "write_lock stuck at %p\n", inline_pc);
 		goto try_again;
 	}
 	if (stuck_reader < 0) {
-		printk("write_lock stuck on readers at %p\n", inline_pc);
+		printk(KERN_WARNING "write_lock stuck on readers at %p\n",
+		       inline_pc);
 		goto try_again;
 	}
 }
@@ -1079,11 +1129,10 @@ void read_lock(rwlock_t * lock)
 	"	br	1b\n"
 	".previous"
 	: "=m" (__dummy_lock(lock)), "=&r" (regx), "=&r" (stuck_lock)
-	: "0" (__dummy_lock(lock)), "2" (stuck_lock)
-	);
+	: "0" (__dummy_lock(lock)), "2" (stuck_lock));
 
 	if (stuck_lock < 0) {
-		printk("read_lock stuck at %p\n", inline_pc);
+		printk(KERN_WARNING "read_lock stuck at %p\n", inline_pc);
 		goto try_again;
 	}
 }
diff --git a/arch/alpha/kernel/time.c b/arch/alpha/kernel/time.c
index a84378926..927ca201f 100644
--- a/arch/alpha/kernel/time.c
+++ b/arch/alpha/kernel/time.c
@@ -42,6 +42,9 @@
 #include "proto.h"
 #include "irq.h"
 
+extern rwlock_t xtime_lock;
+extern volatile unsigned long lost_ticks;	/*kernel/sched.c*/
+
 static int set_rtc_mmss(unsigned long);
 
 
@@ -86,15 +89,15 @@ void timer_interrupt(int irq, void *dev, struct pt_regs * regs)
 	long nticks;
 
 #ifdef __SMP__
-	extern void smp_percpu_timer_interrupt(struct pt_regs *);
-	extern unsigned int boot_cpu_id;
-	/* when SMP, do this for *all* CPUs, 
-	   but only do the rest for the boot CPU */
+	/* When SMP, do this for *all* CPUs, but only do the rest for
+           the boot CPU.  */
 	smp_percpu_timer_interrupt(regs);
-	if (smp_processor_id() != boot_cpu_id)
-	  return;
+	if (smp_processor_id() != smp_boot_cpuid)
+		return;
 #endif
 
+	write_lock(&xtime_lock);
+
 	/*
 	 * Calculate how many ticks have passed since the last update,
 	 * including any previous partial leftover.  Save any resulting
@@ -124,6 +127,8 @@ void timer_interrupt(int irq, void *dev, struct pt_regs * regs)
 		int tmp = set_rtc_mmss(xtime.tv_sec);
 		state.last_rtc_update = xtime.tv_sec - (tmp ? 600 : 0);
 	}
+
+	write_unlock(&xtime_lock);
 }
 
 /*
@@ -226,7 +231,8 @@ time_init(void)
 {
 	void (*irq_handler)(int, void *, struct pt_regs *);
 	unsigned int year, mon, day, hour, min, sec, cc1, cc2;
-	unsigned long cycle_freq, diff, one_percent;
+	unsigned long cycle_freq, one_percent;
+	long diff;
 
 	/*
 	 * The Linux interpretation of the CMOS clock register contents:
@@ -242,7 +248,7 @@ time_init(void)
 
 	if (!est_cycle_freq) {
 		/* Sometimes the hwrpb->cycle_freq value is bogus. 
-	   	Go another round to check up on it and see.  */
+		   Go another round to check up on it and see.  */
 		do { } while (!(CMOS_READ(RTC_FREQ_SELECT) & RTC_UIP));
 		do { } while (CMOS_READ(RTC_FREQ_SELECT) & RTC_UIP);
 		cc2 = rpcc();
@@ -279,8 +285,7 @@ time_init(void)
 	mon = CMOS_READ(RTC_MONTH);
 	year = CMOS_READ(RTC_YEAR);
 
-	if (!(CMOS_READ(RTC_CONTROL) & RTC_DM_BINARY) || RTC_ALWAYS_BCD)
-	{
+	if (!(CMOS_READ(RTC_CONTROL) & RTC_DM_BINARY) || RTC_ALWAYS_BCD) {
 		BCD_TO_BIN(sec);
 		BCD_TO_BIN(min);
 		BCD_TO_BIN(hour);
@@ -328,18 +333,24 @@ time_init(void)
 void
 do_gettimeofday(struct timeval *tv)
 {
-	unsigned long flags, delta_cycles, delta_usec;
-	unsigned long sec, usec;
-	__u32 now;
-	extern volatile unsigned long lost_ticks;	/*kernel/sched.c*/
+	unsigned long sec, usec, lost, flags;
+	unsigned long delta_cycles, delta_usec, partial_tick;
 
-	now = rpcc();
-	save_and_cli(flags);
+	read_lock_irqsave(&xtime_lock, flags);
+
+	delta_cycles = rpcc() - state.last_time;
 	sec = xtime.tv_sec;
 	usec = xtime.tv_usec;
-	delta_cycles = now - state.last_time;
-	restore_flags(flags);
+	partial_tick = state.partial_tick;
+	lost = lost_ticks;
+
+	read_unlock_irqrestore(&xtime_lock, flags);
 
+#ifdef __SMP__
+	/* Until and unless we figure out how to get cpu cycle counters
+	   in sync and keep them there, we can't use the rpcc tricks.  */
+	delta_usec = lost * (1000000 / HZ);
+#else
 	/*
 	 * usec = cycles * ticks_per_cycle * 2**48 * 1e6 / (2**48 * ticks)
 	 *	= cycles * (s_t_p_c) * 1e6 / (2**48 * ticks)
@@ -354,13 +365,10 @@ do_gettimeofday(struct timeval *tv)
 	 */
 
 	delta_usec = (delta_cycles * state.scaled_ticks_per_cycle 
-			+ state.partial_tick
-			+ (lost_ticks << FIX_SHIFT) ) * 15625;
+		      + partial_tick
+		      + (lost << FIX_SHIFT)) * 15625;
 	delta_usec = ((delta_usec / ((1UL << (FIX_SHIFT-6-1)) * HZ)) + 1) / 2;
-
-	/* the 'lost_tics' term above implements this:	
-	 * delta_usec += lost_ticks * (1000000 / HZ);
-	 */
+#endif
 
 	usec += delta_usec;
 	if (usec >= 1000000) {
@@ -375,13 +383,41 @@ do_gettimeofday(struct timeval *tv)
 void
 do_settimeofday(struct timeval *tv)
 {
-	cli();
-	xtime = *tv;
+	unsigned long delta_usec;
+	long sec, usec;
+	
+	write_lock_irq(&xtime_lock);
+
+	/* The offset that is added into time in do_gettimeofday above
+	   must be subtracted out here to keep a coherent view of the
+	   time.  Without this, a full-tick error is possible.  */
+
+#ifdef __SMP__
+	delta_usec = lost_ticks * (1000000 / HZ);
+#else
+	delta_usec = rpcc() - state.last_time;
+	delta_usec = (delta_usec * state.scaled_ticks_per_cycle 
+		      + state.partial_tick
+		      + (lost_ticks << FIX_SHIFT)) * 15625;
+	delta_usec = ((delta_usec / ((1UL << (FIX_SHIFT-6-1)) * HZ)) + 1) / 2;
+#endif
+
+	sec = tv->tv_sec;
+	usec = tv->tv_usec;
+	usec -= delta_usec;
+	if (usec < 0) {
+		usec += 1000000;
+		sec -= 1;
+	}
+
+	xtime.tv_sec = sec;
+	xtime.tv_usec = usec;
 	time_adjust = 0;		/* stop active adjtime() */
 	time_status |= STA_UNSYNC;
 	time_maxerror = NTP_PHASE_LIMIT;
 	time_esterror = NTP_PHASE_LIMIT;
-	sti();
+
+	write_unlock_irq(&xtime_lock);
 }
 
 
diff --git a/arch/alpha/kernel/traps.c b/arch/alpha/kernel/traps.c
index 2548f0914..1b490b9c1 100644
--- a/arch/alpha/kernel/traps.c
+++ b/arch/alpha/kernel/traps.c
@@ -1,5 +1,5 @@
 /*
- * kernel/traps.c
+ * arch/alpha/kernel/traps.c
  *
  * (C) Copyright 1994 Linus Torvalds
  */
@@ -95,6 +95,9 @@ die_if_kernel(char * str, struct pt_regs *regs, long err, unsigned long *r9_15)
 {
 	if (regs->ps & 8)
 		return;
+#ifdef __SMP__
+	printk("CPU %d ", hard_smp_processor_id());
+#endif
 	printk("%s(%d): %s %ld\n", current->comm, current->pid, str, err);
 	dik_show_regs(regs, r9_15);
 	dik_show_code((unsigned int *)regs->pc);
@@ -128,8 +131,8 @@ do_entArith(unsigned long summary, unsigned long write_mask,
 	if (summary & 1) {
 		/* Software-completion summary bit is set, so try to
 		   emulate the instruction.  */
-		if (implver() == IMPLVER_EV6) {
-			/* Whee!  EV6 has precice exceptions.  */
+		if (!amask(AMASK_PRECISE_TRAP)) {
+			/* 21264 (except pass 1) has precise exceptions.  */
 			if (alpha_fp_emul(regs.pc - 4))
 				return;
 		} else {
@@ -138,14 +141,12 @@ do_entArith(unsigned long summary, unsigned long write_mask,
 		}
 	}
 
-	lock_kernel();
 #if 0
 	printk("%s: arithmetic trap at %016lx: %02lx %016lx\n",
 		current->comm, regs.pc, summary, write_mask);
 #endif
 	die_if_kernel("Arithmetic fault", &regs, 0, 0);
 	send_sig(SIGFPE, current, 1);
-	unlock_kernel();
 }
 
 asmlinkage void
@@ -235,10 +236,8 @@ do_entDbg(unsigned long type, unsigned long a1,
 	  unsigned long a2, unsigned long a3, unsigned long a4,
 	  unsigned long a5, struct pt_regs regs)
 {
-	lock_kernel();
 	die_if_kernel("Instruction fault", &regs, type, 0);
 	force_sig(SIGILL, current);
-	unlock_kernel();
 }
 
 
@@ -453,10 +452,8 @@ got_exception:
 		unsigned long newpc;
 		newpc = fixup_exception(una_reg, fixup, pc);
 
-		lock_kernel();
 		printk("Forwarding unaligned exception at %lx (%lx)\n",
 		       pc, newpc);
-		unlock_kernel();
 
 		(&regs)->pc = newpc;
 		return;
@@ -610,11 +607,9 @@ do_entUnaUser(void * va, unsigned long opcode,
 			cnt = 0;
 		}
 		if (++cnt < 5) {
-			lock_kernel();
 			printk("%s(%d): unaligned trap at %016lx: %p %lx %ld\n",
 			       current->comm, current->pid,
 			       regs->pc - 4, va, opcode, reg);
-			unlock_kernel();
 		}
 		last_time = jiffies;
 	}
@@ -868,16 +863,12 @@ do_entUnaUser(void * va, unsigned long opcode,
 
 give_sigsegv:
 	regs->pc -= 4;  /* make pc point to faulting insn */
-	lock_kernel();
 	send_sig(SIGSEGV, current, 1);
-	unlock_kernel();
 	return;
 
 give_sigbus:
 	regs->pc -= 4;
-	lock_kernel();
 	send_sig(SIGBUS, current, 1);
-	unlock_kernel();
 	return;
 }
 
diff --git a/arch/alpha/lib/memcpy.c b/arch/alpha/lib/memcpy.c
index dc708c73e..d715f0219 100644
--- a/arch/alpha/lib/memcpy.c
+++ b/arch/alpha/lib/memcpy.c
@@ -21,30 +21,44 @@
  * This should be done in one go with ldq_u*2/mask/stq_u. Do it
  * with a macro so that we can fix it up later..
  */
-#define ALIGN_DEST_TO8(d,s,n) \
+#define ALIGN_DEST_TO8_UP(d,s,n) \
 	while (d & 7) { \
 		if (n <= 0) return; \
 		n--; \
 		*(char *) d = *(char *) s; \
 		d++; s++; \
 	}
+#define ALIGN_DEST_TO8_DN(d,s,n) \
+	while (d & 7) { \
+		if (n <= 0) return; \
+		n--; \
+		d--; s--; \
+		*(char *) d = *(char *) s; \
+	}
 
 /*
  * This should similarly be done with ldq_u*2/mask/stq. The destination
  * is aligned, but we don't fill in a full quad-word
  */
-#define DO_REST(d,s,n) \
+#define DO_REST_UP(d,s,n) \
 	while (n > 0) { \
 		n--; \
 		*(char *) d = *(char *) s; \
 		d++; s++; \
 	}
+#define DO_REST_DN(d,s,n) \
+	while (n > 0) { \
+		n--; \
+		d--; s--; \
+		*(char *) d = *(char *) s; \
+	}
 
 /*
  * This should be done with ldq/mask/stq. The source and destination are
  * aligned, but we don't fill in a full quad-word
  */
-#define DO_REST_ALIGNED(d,s,n) DO_REST(d,s,n)
+#define DO_REST_ALIGNED_UP(d,s,n) DO_REST_UP(d,s,n)
+#define DO_REST_ALIGNED_DN(d,s,n) DO_REST_DN(d,s,n)
 
 /*
  * This does unaligned memory copies. We want to avoid storing to
@@ -53,9 +67,10 @@
  *
  * Note the ordering to try to avoid load (and address generation) latencies.
  */
-static inline void __memcpy_unaligned(unsigned long d, unsigned long s, long n)
+static inline void __memcpy_unaligned_up (unsigned long d, unsigned long s,
+					  long n)
 {
-	ALIGN_DEST_TO8(d,s,n);
+	ALIGN_DEST_TO8_UP(d,s,n);
 	n -= 8;			/* to avoid compare against 8 in the loop */
 	if (n >= 0) {
 		unsigned long low_word, high_word;
@@ -77,7 +92,17 @@ static inline void __memcpy_unaligned(unsigned long d, unsigned long s, long n)
 		} while (n >= 0);
 	}
 	n += 8;
-	DO_REST(d,s,n);
+	DO_REST_UP(d,s,n);
+}
+
+static inline void __memcpy_unaligned_dn (unsigned long d, unsigned long s,
+					  long n)
+{
+	/* I don't understand AXP assembler well enough for this. -Tim */
+	s += n;
+	d += n;
+	while (n--)
+		* (char *) --d = * (char *) --s;
 }
 
 /*
@@ -88,9 +113,10 @@ static inline void __memcpy_unaligned(unsigned long d, unsigned long s, long n)
  *
  * Note the ordering to try to avoid load (and address generation) latencies.
  */
-static inline void __memcpy_aligned(unsigned long d, unsigned long s, long n)
+static inline void __memcpy_aligned_up (unsigned long d, unsigned long s,
+					long n)
 {
-	ALIGN_DEST_TO8(d,s,n);
+	ALIGN_DEST_TO8_UP(d,s,n);
 	n -= 8;
 	while (n >= 0) {
 		unsigned long tmp;
@@ -101,18 +127,58 @@ static inline void __memcpy_aligned(unsigned long d, unsigned long s, long n)
 		d += 8;
 	}
 	n += 8;
-	DO_REST_ALIGNED(d,s,n);
+	DO_REST_ALIGNED_UP(d,s,n);
+}
+static inline void __memcpy_aligned_dn (unsigned long d, unsigned long s,
+					long n)
+{
+	s += n;
+	d += n;
+	ALIGN_DEST_TO8_DN(d,s,n);
+	n -= 8;
+	while (n >= 0) {
+		unsigned long tmp;
+		s -= 8;
+		__asm__("ldq %0,%1":"=r" (tmp):"m" (*(unsigned long *) s));
+		n -= 8;
+		d -= 8;
+		*(unsigned long *) d = tmp;
+	}
+	n += 8;
+	DO_REST_ALIGNED_DN(d,s,n);
 }
 
 void * memcpy(void * dest, const void *src, size_t n)
 {
 	if (!(((unsigned long) dest ^ (unsigned long) src) & 7)) {
-		__memcpy_aligned((unsigned long) dest, (unsigned long) src, n);
+		__memcpy_aligned_up ((unsigned long) dest, (unsigned long) src,
+				     n);
 		return dest;
 	}
-	__memcpy_unaligned((unsigned long) dest, (unsigned long) src, n);
+	__memcpy_unaligned_up ((unsigned long) dest, (unsigned long) src, n);
 	return dest;
 }
 
 /* For backward modules compatibility, define __memcpy.  */
 asm("__memcpy = memcpy; .globl __memcpy");
+
+void *memmove (void *dest, const void *src, size_t n)
+{
+	if (dest <= src) {
+		if (!(((unsigned long) dest ^ (unsigned long) src) & 7))
+			__memcpy_aligned_up ((unsigned long) dest,
+					     (unsigned long) src, n);
+		else
+			__memcpy_unaligned_up ((unsigned long) dest,
+					       (unsigned long) src, n);
+	}
+	else {
+		if (!(((unsigned long) dest ^ (unsigned long) src) & 7))
+			__memcpy_aligned_dn ((unsigned long) dest,
+					     (unsigned long) src, n);
+		else
+			__memcpy_unaligned_dn ((unsigned long) dest,
+					       (unsigned long) src, n);
+	}
+	return dest;
+}
diff --git a/arch/alpha/mm/init.c b/arch/alpha/mm/init.c
index 2e67deb8a..fc5a964bb 100644
--- a/arch/alpha/mm/init.c
+++ b/arch/alpha/mm/init.c
@@ -256,26 +256,6 @@ paging_init(unsigned long start_mem, unsigned long end_mem)
 	return start_mem;
 }
 
-#ifdef __SMP__
-/*
- * paging_init_secondary(), called ONLY by secondary CPUs,
- * sets up current->tss contents appropriately and does a load_PCB.
- * note that current should be pointing at the idle thread task struct
- * for this CPU.
- */
-void
-paging_init_secondary(void)
-{
-	current->tss.ptbr = init_task.tss.ptbr;
-	current->tss.pal_flags = 1;
-	current->tss.flags = 0;
-	load_PCB(&current->tss);
-	tbia();
-
-	return;
-}
-#endif /* __SMP__ */
-
 #if defined(CONFIG_ALPHA_GENERIC) || defined(CONFIG_ALPHA_SRM)
 void
 srm_paging_stop (void)