26 files changed, 992 insertions, 248 deletions

diff --git a/arch/tile/kernel/Makefile b/arch/tile/kernel/Makefile
index 112b1e248f0..b4c8e8ec45d 100644
--- a/arch/tile/kernel/Makefile
+++ b/arch/tile/kernel/Makefile
@@ -15,3 +15,4 @@ obj-$(CONFIG_SMP)		+= smpboot.o smp.o tlb.o
 obj-$(CONFIG_MODULES)		+= module.o
 obj-$(CONFIG_EARLY_PRINTK)	+= early_printk.o
 obj-$(CONFIG_KEXEC)		+= machine_kexec.o relocate_kernel.o
+obj-$(CONFIG_PCI)		+= pci.o
diff --git a/arch/tile/kernel/backtrace.c b/arch/tile/kernel/backtrace.c
index d3c41c1ff6b..55a6a74974b 100644
--- a/arch/tile/kernel/backtrace.c
+++ b/arch/tile/kernel/backtrace.c
@@ -369,6 +369,10 @@ static void find_caller_pc_and_caller_sp(CallerLocation *location,
 					/* Weird; reserved value, ignore it. */
 					continue;
 				}
+				if (info_operand & ENTRY_POINT_INFO_OP)	{
+					/* This info op is ignored by the backtracer. */
+					continue;
+				}
 
 				/* Skip info ops which are not in the
 				 * "one_ago" mode we want right now.
diff --git a/arch/tile/kernel/compat.c b/arch/tile/kernel/compat.c
index b1e06d04155..dbc213adf5e 100644
--- a/arch/tile/kernel/compat.c
+++ b/arch/tile/kernel/compat.c
@@ -21,7 +21,6 @@
 #include <linux/kdev_t.h>
 #include <linux/fs.h>
 #include <linux/fcntl.h>
-#include <linux/smp_lock.h>
 #include <linux/uaccess.h>
 #include <linux/signal.h>
 #include <asm/syscalls.h>
@@ -148,14 +147,20 @@ long tile_compat_sys_msgrcv(int msqid,
 #define compat_sys_readahead sys32_readahead
 #define compat_sys_sync_file_range compat_sys_sync_file_range2
 
-/* The native 64-bit "struct stat" matches the 32-bit "struct stat64". */
-#define compat_sys_stat64 sys_newstat
-#define compat_sys_lstat64 sys_newlstat
-#define compat_sys_fstat64 sys_newfstat
-#define compat_sys_fstatat64 sys_newfstatat
+/* We leverage the "struct stat64" type for 32-bit time_t/nsec. */
+#define compat_sys_stat64 sys_stat64
+#define compat_sys_lstat64 sys_lstat64
+#define compat_sys_fstat64 sys_fstat64
+#define compat_sys_fstatat64 sys_fstatat64
 
-/* Pass full 64-bit values through ptrace. */
-#define compat_sys_ptrace tile_compat_sys_ptrace
+/* The native sys_ptrace dynamically handles compat binaries. */
+#define compat_sys_ptrace sys_ptrace
+
+/* Call the trampolines to manage pt_regs where necessary. */
+#define compat_sys_execve _compat_sys_execve
+#define compat_sys_sigaltstack _compat_sys_sigaltstack
+#define compat_sys_rt_sigreturn _compat_sys_rt_sigreturn
+#define sys_clone _sys_clone
 
 /*
  * Note that we can't include <linux/unistd.h> here since the header
diff --git a/arch/tile/kernel/compat_signal.c b/arch/tile/kernel/compat_signal.c
index 9c710db43f1..dbb0dfc7bec 100644
--- a/arch/tile/kernel/compat_signal.c
+++ b/arch/tile/kernel/compat_signal.c
@@ -15,7 +15,6 @@
 #include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
-#include <linux/smp_lock.h>
 #include <linux/kernel.h>
 #include <linux/signal.h>
 #include <linux/errno.h>
@@ -256,9 +255,9 @@ int copy_siginfo_from_user32(siginfo_t *to, struct compat_siginfo __user *from)
 	return err;
 }
 
-long _compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
-			     struct compat_sigaltstack __user *uoss_ptr,
-			     struct pt_regs *regs)
+long compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
+			    struct compat_sigaltstack __user *uoss_ptr,
+			    struct pt_regs *regs)
 {
 	stack_t uss, uoss;
 	int ret;
@@ -291,12 +290,12 @@ long _compat_sys_sigaltstack(const struct compat_sigaltstack __user *uss_ptr,
 	return ret;
 }
 
-long _compat_sys_rt_sigreturn(struct pt_regs *regs)
+/* The assembly shim for this function arranges to ignore the return value. */
+long compat_sys_rt_sigreturn(struct pt_regs *regs)
 {
 	struct compat_rt_sigframe __user *frame =
 		(struct compat_rt_sigframe __user *) compat_ptr(regs->sp);
 	sigset_t set;
-	long r0;
 
 	if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
 		goto badframe;
@@ -309,13 +308,13 @@ long _compat_sys_rt_sigreturn(struct pt_regs *regs)
 	recalc_sigpending();
 	spin_unlock_irq(&current->sighand->siglock);
 
-	if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &r0))
+	if (restore_sigcontext(regs, &frame->uc.uc_mcontext))
 		goto badframe;
 
-	if (_compat_sys_sigaltstack(&frame->uc.uc_stack, NULL, regs) != 0)
+	if (compat_sys_sigaltstack(&frame->uc.uc_stack, NULL, regs) != 0)
 		goto badframe;
 
-	return r0;
+	return 0;
 
 badframe:
 	force_sig(SIGSEGV, current);
diff --git a/arch/tile/kernel/early_printk.c b/arch/tile/kernel/early_printk.c
index 2c54fd43a8a..493a0e66d91 100644
--- a/arch/tile/kernel/early_printk.c
+++ b/arch/tile/kernel/early_printk.c
@@ -54,7 +54,7 @@ void early_printk(const char *fmt, ...)
 void early_panic(const char *fmt, ...)
 {
 	va_list ap;
-	raw_local_irq_disable_all();
+	arch_local_irq_disable_all();
 	va_start(ap, fmt);
 	early_printk("Kernel panic - not syncing: ");
 	early_vprintk(fmt, ap);
diff --git a/arch/tile/kernel/entry.S b/arch/tile/kernel/entry.S
index 3d01383b1b0..fd8dc42abdc 100644
--- a/arch/tile/kernel/entry.S
+++ b/arch/tile/kernel/entry.S
@@ -15,7 +15,9 @@
 #include <linux/linkage.h>
 #include <linux/unistd.h>
 #include <asm/irqflags.h>
+#include <asm/processor.h>
 #include <arch/abi.h>
+#include <arch/spr_def.h>
 
 #ifdef __tilegx__
 #define bnzt bnezt
@@ -25,28 +27,6 @@ STD_ENTRY(current_text_addr)
 	{ move r0, lr; jrp lr }
 	STD_ENDPROC(current_text_addr)
 
-STD_ENTRY(_sim_syscall)
-	/*
-	 * Wait for r0-r9 to be ready (and lr on the off chance we
-	 * want the syscall to locate its caller), then make a magic
-	 * simulator syscall.
-	 *
-	 * We carefully stall until the registers are readable in case they
-	 * are the target of a slow load, etc. so that tile-sim will
-	 * definitely be able to read all of them inside the magic syscall.
-	 *
-	 * Technically this is wrong for r3-r9 and lr, since an interrupt
-	 * could come in and restore the registers with a slow load right
-	 * before executing the mtspr. We may need to modify tile-sim to
-	 * explicitly stall for this case, but we do not yet have
-	 * a way to implement such a stall.
-	 */
-	{ and zero, lr, r9 ; and zero, r8, r7 }
-	{ and zero, r6, r5 ; and zero, r4, r3 }
-	{ and zero, r2, r1 ; mtspr SIM_CONTROL, r0 }
-	{ jrp lr }
-	STD_ENDPROC(_sim_syscall)
-
 /*
  * Implement execve().  The i386 code has a note that forking from kernel
  * space results in no copy on write until the execve, so we should be
@@ -102,7 +82,7 @@ STD_ENTRY(KBacktraceIterator_init_current)
 STD_ENTRY(cpu_idle_on_new_stack)
 	{
 	 move sp, r1
-	 mtspr SYSTEM_SAVE_1_0, r2
+	 mtspr SPR_SYSTEM_SAVE_K_0, r2
 	}
 	jal free_thread_info
 	j cpu_idle
@@ -124,15 +104,15 @@ STD_ENTRY(smp_nap)
 STD_ENTRY(_cpu_idle)
 	{
 	 lnk r0
-	 movei r1, 1
+	 movei r1, KERNEL_PL
 	}
 	{
 	 addli r0, r0, _cpu_idle_nap - .
 	 mtspr INTERRUPT_CRITICAL_SECTION, r1
 	}
-	IRQ_ENABLE(r2, r3)         /* unmask, but still with ICS set */
-	mtspr EX_CONTEXT_1_1, r1   /* PL1, ICS clear */
-	mtspr EX_CONTEXT_1_0, r0
+	IRQ_ENABLE(r2, r3)             /* unmask, but still with ICS set */
+	mtspr SPR_EX_CONTEXT_K_1, r1   /* Kernel PL, ICS clear */
+	mtspr SPR_EX_CONTEXT_K_0, r0
 	iret
 	.global _cpu_idle_nap
 _cpu_idle_nap:
diff --git a/arch/tile/kernel/hardwall.c b/arch/tile/kernel/hardwall.c
index 584b965dc82..e910530436e 100644
--- a/arch/tile/kernel/hardwall.c
+++ b/arch/tile/kernel/hardwall.c
@@ -151,12 +151,12 @@ enum direction_protect {
 
 static void enable_firewall_interrupts(void)
 {
-	raw_local_irq_unmask_now(INT_UDN_FIREWALL);
+	arch_local_irq_unmask_now(INT_UDN_FIREWALL);
 }
 
 static void disable_firewall_interrupts(void)
 {
-	raw_local_irq_mask_now(INT_UDN_FIREWALL);
+	arch_local_irq_mask_now(INT_UDN_FIREWALL);
 }
 
 /* Set up hardwall on this cpu based on the passed hardwall_info. */
@@ -768,6 +768,7 @@ static int hardwall_release(struct inode *inode, struct file *file)
 }
 
 static const struct file_operations dev_hardwall_fops = {
+	.open           = nonseekable_open,
 	.unlocked_ioctl = hardwall_ioctl,
 #ifdef CONFIG_COMPAT
 	.compat_ioctl   = hardwall_compat_ioctl,
diff --git a/arch/tile/kernel/head_32.S b/arch/tile/kernel/head_32.S
index 2b4f6c09170..90e7c443569 100644
--- a/arch/tile/kernel/head_32.S
+++ b/arch/tile/kernel/head_32.S
@@ -23,6 +23,7 @@
 #include <asm/asm-offsets.h>
 #include <hv/hypervisor.h>
 #include <arch/chip.h>
+#include <arch/spr_def.h>
 
 /*
  * This module contains the entry code for kernel images. It performs the
@@ -76,7 +77,7 @@ ENTRY(_start)
 	}
 1:
 
-	/* Get our processor number and save it away in SAVE_1_0. */
+	/* Get our processor number and save it away in SAVE_K_0. */
 	jal hv_inquire_topology
 	mulll_uu r4, r1, r2        /* r1 == y, r2 == width */
 	add r4, r4, r0             /* r0 == x, so r4 == cpu == y*width + x */
@@ -124,7 +125,7 @@ ENTRY(_start)
 	lw r0, r0
 	lw sp, r1
 	or r4, sp, r4
-	mtspr SYSTEM_SAVE_1_0, r4  /* save ksp0 + cpu */
+	mtspr SPR_SYSTEM_SAVE_K_0, r4  /* save ksp0 + cpu */
 	addi sp, sp, -STACK_TOP_DELTA
 	{
 	  move lr, zero   /* stop backtraces in the called function */
diff --git a/arch/tile/kernel/intvec_32.S b/arch/tile/kernel/intvec_32.S
index 8f58bdff20d..5eed4a02bf6 100644
--- a/arch/tile/kernel/intvec_32.S
+++ b/arch/tile/kernel/intvec_32.S
@@ -32,8 +32,8 @@
 # error "No support for kernel preemption currently"
 #endif
 
-#if INT_INTCTRL_1 < 32 || INT_INTCTRL_1 >= 48
-# error INT_INTCTRL_1 coded to set high interrupt mask
+#if INT_INTCTRL_K < 32 || INT_INTCTRL_K >= 48
+# error INT_INTCTRL_K coded to set high interrupt mask
 #endif
 
 #define PTREGS_PTR(reg, ptreg) addli reg, sp, C_ABI_SAVE_AREA_SIZE + (ptreg)
@@ -132,8 +132,8 @@ intvec_\vecname:
 
 	/* Temporarily save a register so we have somewhere to work. */
 
-	mtspr   SYSTEM_SAVE_1_1, r0
-	mfspr   r0, EX_CONTEXT_1_1
+	mtspr   SPR_SYSTEM_SAVE_K_1, r0
+	mfspr   r0, SPR_EX_CONTEXT_K_1
 
 	/* The cmpxchg code clears sp to force us to reset it here on fault. */
 	{
@@ -167,18 +167,18 @@ intvec_\vecname:
 	 * The page_fault handler may be downcalled directly by the
 	 * hypervisor even when Linux is running and has ICS set.
 	 *
-	 * In this case the contents of EX_CONTEXT_1_1 reflect the
+	 * In this case the contents of EX_CONTEXT_K_1 reflect the
 	 * previous fault and can't be relied on to choose whether or
 	 * not to reinitialize the stack pointer.  So we add a test
-	 * to see whether SYSTEM_SAVE_1_2 has the high bit set,
+	 * to see whether SYSTEM_SAVE_K_2 has the high bit set,
 	 * and if so we don't reinitialize sp, since we must be coming
 	 * from Linux.  (In fact the precise case is !(val & ~1),
 	 * but any Linux PC has to have the high bit set.)
 	 *
-	 * Note that the hypervisor *always* sets SYSTEM_SAVE_1_2 for
+	 * Note that the hypervisor *always* sets SYSTEM_SAVE_K_2 for
 	 * any path that turns into a downcall to one of our TLB handlers.
 	 */
-	mfspr   r0, SYSTEM_SAVE_1_2
+	mfspr   r0, SPR_SYSTEM_SAVE_K_2
 	{
 	 blz    r0, 0f    /* high bit in S_S_1_2 is for a PC to use */
 	 move   r0, sp
@@ -187,12 +187,12 @@ intvec_\vecname:
 
 2:
 	/*
-	 * SYSTEM_SAVE_1_0 holds the cpu number in the low bits, and
+	 * SYSTEM_SAVE_K_0 holds the cpu number in the low bits, and
 	 * the current stack top in the higher bits.  So we recover
 	 * our stack top by just masking off the low bits, then
 	 * point sp at the top aligned address on the actual stack page.
 	 */
-	mfspr   r0, SYSTEM_SAVE_1_0
+	mfspr   r0, SPR_SYSTEM_SAVE_K_0
 	mm      r0, r0, zero, LOG2_THREAD_SIZE, 31
 
 0:
@@ -254,7 +254,7 @@ intvec_\vecname:
 	 sw     sp, r3
 	 addli  sp, sp, PTREGS_OFFSET_PC - PTREGS_OFFSET_REG(3)
 	}
-	mfspr   r0, EX_CONTEXT_1_0
+	mfspr   r0, SPR_EX_CONTEXT_K_0
 	.ifc \processing,handle_syscall
 	/*
 	 * Bump the saved PC by one bundle so that when we return, we won't
@@ -267,7 +267,7 @@ intvec_\vecname:
 	 sw     sp, r0
 	 addli  sp, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
 	}
-	mfspr   r0, EX_CONTEXT_1_1
+	mfspr   r0, SPR_EX_CONTEXT_K_1
 	{
 	 sw     sp, r0
 	 addi   sp, sp, PTREGS_OFFSET_FAULTNUM - PTREGS_OFFSET_EX1
@@ -289,7 +289,7 @@ intvec_\vecname:
 	 .endif
 	 addli  sp, sp, PTREGS_OFFSET_REG(0) - PTREGS_OFFSET_FAULTNUM
 	}
-	mfspr   r0, SYSTEM_SAVE_1_1    /* Original r0 */
+	mfspr   r0, SPR_SYSTEM_SAVE_K_1    /* Original r0 */
 	{
 	 sw     sp, r0
 	 addi   sp, sp, -PTREGS_OFFSET_REG(0) - 4
@@ -309,12 +309,12 @@ intvec_\vecname:
 	 * See discussion below at "finish_interrupt_save".
 	 */
 	.ifc \c_routine, do_page_fault
-	mfspr   r2, SYSTEM_SAVE_1_3   /* address of page fault */
-	mfspr   r3, SYSTEM_SAVE_1_2   /* info about page fault */
+	mfspr   r2, SPR_SYSTEM_SAVE_K_3   /* address of page fault */
+	mfspr   r3, SPR_SYSTEM_SAVE_K_2   /* info about page fault */
 	.else
 	.ifc \vecnum, INT_DOUBLE_FAULT
 	{
-	 mfspr  r2, SYSTEM_SAVE_1_2   /* double fault info from HV */
+	 mfspr  r2, SPR_SYSTEM_SAVE_K_2   /* double fault info from HV */
 	 movei  r3, 0
 	}
 	.else
@@ -467,7 +467,7 @@ intvec_\vecname:
 	/* Load tp with our per-cpu offset. */
 #ifdef CONFIG_SMP
 	{
-	 mfspr  r20, SYSTEM_SAVE_1_0
+	 mfspr  r20, SPR_SYSTEM_SAVE_K_0
 	 moveli r21, lo16(__per_cpu_offset)
 	}
 	{
@@ -487,7 +487,7 @@ intvec_\vecname:
 	 * We load flags in r32 here so we can jump to .Lrestore_regs
 	 * directly after do_page_fault_ics() if necessary.
 	 */
-	mfspr   r32, EX_CONTEXT_1_1
+	mfspr   r32, SPR_EX_CONTEXT_K_1
 	{
 	 andi   r32, r32, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
 	 PTREGS_PTR(r21, PTREGS_OFFSET_FLAGS)
@@ -957,11 +957,11 @@ STD_ENTRY(interrupt_return)
 	pop_reg_zero r21, r3, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
 	pop_reg_zero lr, r4, sp, PTREGS_OFFSET_REG(52) - PTREGS_OFFSET_EX1
 	{
-	 mtspr  EX_CONTEXT_1_0, r21
+	 mtspr  SPR_EX_CONTEXT_K_0, r21
 	 move   r5, zero
 	}
 	{
-	 mtspr  EX_CONTEXT_1_1, lr
+	 mtspr  SPR_EX_CONTEXT_K_1, lr
 	 andi   lr, lr, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
 	}
 
@@ -1020,7 +1020,7 @@ STD_ENTRY(interrupt_return)
 
 	/* Set r1 to errno if we are returning an error, otherwise zero. */
 	{
-	 moveli r29, 1024
+	 moveli r29, 4096
 	 sub    r1, zero, r0
 	}
 	slt_u   r29, r1, r29
@@ -1199,7 +1199,7 @@ STD_ENTRY(interrupt_return)
 	STD_ENDPROC(interrupt_return)
 
 	/*
-	 * This interrupt variant clears the INT_INTCTRL_1 interrupt mask bit
+	 * This interrupt variant clears the INT_INTCTRL_K interrupt mask bit
 	 * before returning, so we can properly get more downcalls.
 	 */
 	.pushsection .text.handle_interrupt_downcall,"ax"
@@ -1208,11 +1208,11 @@ handle_interrupt_downcall:
 	check_single_stepping normal, .Ldispatch_downcall
 .Ldispatch_downcall:
 
-	/* Clear INTCTRL_1 from the set of interrupts we ever enable. */
+	/* Clear INTCTRL_K from the set of interrupts we ever enable. */
 	GET_INTERRUPTS_ENABLED_MASK_PTR(r30)
 	{
 	 addi   r30, r30, 4
-	 movei  r31, INT_MASK(INT_INTCTRL_1)
+	 movei  r31, INT_MASK(INT_INTCTRL_K)
 	}
 	{
 	 lw     r20, r30
@@ -1227,7 +1227,7 @@ handle_interrupt_downcall:
 	}
 	FEEDBACK_REENTER(handle_interrupt_downcall)
 
-	/* Allow INTCTRL_1 to be enabled next time we enable interrupts. */
+	/* Allow INTCTRL_K to be enabled next time we enable interrupts. */
 	lw      r20, r30
 	or      r20, r20, r31
 	sw      r30, r20
@@ -1342,8 +1342,8 @@ handle_syscall:
 	lw      r20, r20
 
 	/* Jump to syscall handler. */
-	jalr    r20; .Lhandle_syscall_link:
-	FEEDBACK_REENTER(handle_syscall)
+	jalr    r20
+.Lhandle_syscall_link: /* value of "lr" after "jalr r20" above */
 
 	/*
 	 * Write our r0 onto the stack so it gets restored instead
@@ -1352,6 +1352,9 @@ handle_syscall:
 	PTREGS_PTR(r29, PTREGS_OFFSET_REG(0))
 	sw      r29, r0
 
+.Lsyscall_sigreturn_skip:
+	FEEDBACK_REENTER(handle_syscall)
+
 	/* Do syscall trace again, if requested. */
 	lw	r30, r31
 	andi    r30, r30, _TIF_SYSCALL_TRACE
@@ -1472,7 +1475,12 @@ handle_ill:
 	lw      r26, r24
 	sw      r28, r26
 
-	/* Clear TIF_SINGLESTEP */
+	/*
+	 * Clear TIF_SINGLESTEP to prevent recursion if we execute an ill.
+	 * The normal non-arch flow redundantly clears TIF_SINGLESTEP, but we
+	 * need to clear it here and can't really impose on all other arches.
+	 * So what's another write between friends?
+	 */
 	GET_THREAD_INFO(r0)
 
 	addi    r1, r0, THREAD_INFO_FLAGS_OFFSET
@@ -1509,7 +1517,7 @@ handle_ill:
 /* Various stub interrupt handlers and syscall handlers */
 
 STD_ENTRY_LOCAL(_kernel_double_fault)
-	mfspr   r1, EX_CONTEXT_1_0
+	mfspr   r1, SPR_EX_CONTEXT_K_0
 	move    r2, lr
 	move    r3, sp
 	move    r4, r52
@@ -1518,34 +1526,44 @@ STD_ENTRY_LOCAL(_kernel_double_fault)
 	STD_ENDPROC(_kernel_double_fault)
 
 STD_ENTRY_LOCAL(bad_intr)
-	mfspr   r2, EX_CONTEXT_1_0
+	mfspr   r2, SPR_EX_CONTEXT_K_0
 	panic   "Unhandled interrupt %#x: PC %#lx"
 	STD_ENDPROC(bad_intr)
 
 /* Put address of pt_regs in reg and jump. */
 #define PTREGS_SYSCALL(x, reg)                          \
-	STD_ENTRY(x);                                   \
+	STD_ENTRY(_##x);                                \
 	{                                               \
 	 PTREGS_PTR(reg, PTREGS_OFFSET_BASE);           \
-	 j      _##x                                    \
+	 j      x                                       \
 	};                                              \
-	STD_ENDPROC(x)
+	STD_ENDPROC(_##x)
+
+/*
+ * Special-case sigreturn to not write r0 to the stack on return.
+ * This is technically more efficient, but it also avoids difficulties
+ * in the 64-bit OS when handling 32-bit compat code, since we must not
+ * sign-extend r0 for the sigreturn return-value case.
+ */
+#define PTREGS_SYSCALL_SIGRETURN(x, reg)                \
+	STD_ENTRY(_##x);                                \
+	addli   lr, lr, .Lsyscall_sigreturn_skip - .Lhandle_syscall_link; \
+	{                                               \
+	 PTREGS_PTR(reg, PTREGS_OFFSET_BASE);           \
+	 j      x                                       \
+	};                                              \
+	STD_ENDPROC(_##x)
 
 PTREGS_SYSCALL(sys_execve, r3)
 PTREGS_SYSCALL(sys_sigaltstack, r2)
-PTREGS_SYSCALL(sys_rt_sigreturn, r0)
+PTREGS_SYSCALL_SIGRETURN(sys_rt_sigreturn, r0)
+PTREGS_SYSCALL(sys_cmpxchg_badaddr, r1)
 
-/* Save additional callee-saves to pt_regs, put address in reg and jump. */
-#define PTREGS_SYSCALL_ALL_REGS(x, reg)                 \
-	STD_ENTRY(x);                                   \
-	push_extra_callee_saves reg;                    \
-	j       _##x;                                   \
-	STD_ENDPROC(x)
-
-PTREGS_SYSCALL_ALL_REGS(sys_fork, r0)
-PTREGS_SYSCALL_ALL_REGS(sys_vfork, r0)
-PTREGS_SYSCALL_ALL_REGS(sys_clone, r4)
-PTREGS_SYSCALL_ALL_REGS(sys_cmpxchg_badaddr, r1)
+/* Save additional callee-saves to pt_regs, put address in r4 and jump. */
+STD_ENTRY(_sys_clone)
+	push_extra_callee_saves r4
+	j       sys_clone
+	STD_ENDPROC(_sys_clone)
 
 /*
  * This entrypoint is taken for the cmpxchg and atomic_update fast
@@ -1558,12 +1576,14 @@ PTREGS_SYSCALL_ALL_REGS(sys_cmpxchg_badaddr, r1)
  * to be available to it on entry.  It does not modify any callee-save
  * registers (including "lr").  It does not check what PL it is being
  * called at, so you'd better not call it other than at PL0.
+ * The <atomic.h> wrapper assumes it only clobbers r20-r29, so if
+ * it ever is necessary to use more registers, be aware.
  *
  * It does not use the stack, but since it might be re-interrupted by
  * a page fault which would assume the stack was valid, it does
  * save/restore the stack pointer and zero it out to make sure it gets reset.
  * Since we always keep interrupts disabled, the hypervisor won't
- * clobber our EX_CONTEXT_1_x registers, so we don't save/restore them
+ * clobber our EX_CONTEXT_K_x registers, so we don't save/restore them
  * (other than to advance the PC on return).
  *
  * We have to manually validate the user vs kernel address range
@@ -1769,7 +1789,7 @@ ENTRY(sys_cmpxchg)
 	/* Do slow mtspr here so the following "mf" waits less. */
 	{
 	 move   sp, r27
-	 mtspr  EX_CONTEXT_1_0, r28
+	 mtspr  SPR_EX_CONTEXT_K_0, r28
 	}
 	mf
 
@@ -1788,7 +1808,7 @@ ENTRY(sys_cmpxchg)
 	}
 	{
 	 move   sp, r27
-	 mtspr  EX_CONTEXT_1_0, r28
+	 mtspr  SPR_EX_CONTEXT_K_0, r28
 	}
 	iret
 
@@ -1816,7 +1836,7 @@ ENTRY(sys_cmpxchg)
 #endif
 
 	/* Issue the slow SPR here while the tns result is in flight. */
-	mfspr   r28, EX_CONTEXT_1_0
+	mfspr   r28, SPR_EX_CONTEXT_K_0
 
 	{
 	 addi   r28, r28, 8    /* return to the instruction after the swint1 */
@@ -1904,7 +1924,7 @@ ENTRY(sys_cmpxchg)
 .Lcmpxchg64_mismatch:
 	{
 	 move   sp, r27
-	 mtspr  EX_CONTEXT_1_0, r28
+	 mtspr  SPR_EX_CONTEXT_K_0, r28
 	}
 	mf
 	{
@@ -1985,8 +2005,13 @@ int_unalign:
 	int_hand     INT_PERF_COUNT, PERF_COUNT, \
 		     op_handle_perf_interrupt, handle_nmi
 	int_hand     INT_INTCTRL_3, INTCTRL_3, bad_intr
+#if CONFIG_KERNEL_PL == 2
+	dc_dispatch  INT_INTCTRL_2, INTCTRL_2
+	int_hand     INT_INTCTRL_1, INTCTRL_1, bad_intr
+#else
 	int_hand     INT_INTCTRL_2, INTCTRL_2, bad_intr
 	dc_dispatch  INT_INTCTRL_1, INTCTRL_1
+#endif
 	int_hand     INT_INTCTRL_0, INTCTRL_0, bad_intr
 	int_hand     INT_MESSAGE_RCV_DWNCL, MESSAGE_RCV_DWNCL, \
 		     hv_message_intr, handle_interrupt_downcall
diff --git a/arch/tile/kernel/irq.c b/arch/tile/kernel/irq.c
index 9a27d563fc3..128805ef8f2 100644
--- a/arch/tile/kernel/irq.c
+++ b/arch/tile/kernel/irq.c
@@ -26,7 +26,7 @@
 #define IS_HW_CLEARED 1
 
 /*
- * The set of interrupts we enable for raw_local_irq_enable().
+ * The set of interrupts we enable for arch_local_irq_enable().
  * This is initialized to have just a single interrupt that the kernel
  * doesn't actually use as a sentinel.  During kernel init,
  * interrupts are added as the kernel gets prepared to support them.
@@ -61,9 +61,9 @@ static DEFINE_SPINLOCK(available_irqs_lock);
 
 #if CHIP_HAS_IPI()
 /* Use SPRs to manipulate device interrupts. */
-#define mask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_SET_1, irq_mask)
-#define unmask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_RESET_1, irq_mask)
-#define clear_irqs(irq_mask) __insn_mtspr(SPR_IPI_EVENT_RESET_1, irq_mask)
+#define mask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_SET_K, irq_mask)
+#define unmask_irqs(irq_mask) __insn_mtspr(SPR_IPI_MASK_RESET_K, irq_mask)
+#define clear_irqs(irq_mask) __insn_mtspr(SPR_IPI_EVENT_RESET_K, irq_mask)
 #else
 /* Use HV to manipulate device interrupts. */
 #define mask_irqs(irq_mask) hv_disable_intr(irq_mask)
@@ -89,16 +89,16 @@ void tile_dev_intr(struct pt_regs *regs, int intnum)
 	 * masked by a previous interrupt.  Then, mask out the ones
 	 * we're going to handle.
 	 */
-	unsigned long masked = __insn_mfspr(SPR_IPI_MASK_1);
-	original_irqs = __insn_mfspr(SPR_IPI_EVENT_1) & ~masked;
-	__insn_mtspr(SPR_IPI_MASK_SET_1, original_irqs);
+	unsigned long masked = __insn_mfspr(SPR_IPI_MASK_K);
+	original_irqs = __insn_mfspr(SPR_IPI_EVENT_K) & ~masked;
+	__insn_mtspr(SPR_IPI_MASK_SET_K, original_irqs);
 #else
 	/*
 	 * Hypervisor performs the equivalent of the Gx code above and
 	 * then puts the pending interrupt mask into a system save reg
 	 * for us to find.
 	 */
-	original_irqs = __insn_mfspr(SPR_SYSTEM_SAVE_1_3);
+	original_irqs = __insn_mfspr(SPR_SYSTEM_SAVE_K_3);
 #endif
 	remaining_irqs = original_irqs;
 
@@ -225,7 +225,7 @@ void __cpuinit setup_irq_regs(void)
 	/* Enable interrupt delivery. */
 	unmask_irqs(~0UL);
 #if CHIP_HAS_IPI()
-	raw_local_irq_unmask(INT_IPI_1);
+	arch_local_irq_unmask(INT_IPI_K);
 #endif
 }
 
diff --git a/arch/tile/kernel/machine_kexec.c b/arch/tile/kernel/machine_kexec.c
index ba7a265d617..0d8b9e93348 100644
--- a/arch/tile/kernel/machine_kexec.c
+++ b/arch/tile/kernel/machine_kexec.c
@@ -182,13 +182,13 @@ static void kexec_find_and_set_command_line(struct kimage *image)
 
 		if ((entry & IND_SOURCE)) {
 			void *va =
-				kmap_atomic_pfn(entry >> PAGE_SHIFT, KM_USER0);
+				kmap_atomic_pfn(entry >> PAGE_SHIFT);
 			r = kexec_bn2cl(va);
 			if (r) {
 				command_line = r;
 				break;
 			}
-			kunmap_atomic(va, KM_USER0);
+			kunmap_atomic(va);
 		}
 	}
 
@@ -198,7 +198,7 @@ static void kexec_find_and_set_command_line(struct kimage *image)
 
 		hverr = hv_set_command_line(
 			(HV_VirtAddr) command_line, strlen(command_line));
-		kunmap_atomic(command_line, KM_USER0);
+		kunmap_atomic(command_line);
 	} else {
 		pr_info("%s: no command line found; making empty\n",
 		       __func__);
diff --git a/arch/tile/kernel/messaging.c b/arch/tile/kernel/messaging.c
index 6d23ed271d1..0858ee6b520 100644
--- a/arch/tile/kernel/messaging.c
+++ b/arch/tile/kernel/messaging.c
@@ -34,7 +34,7 @@ void __cpuinit init_messaging(void)
 		panic("hv_register_message_state: error %d", rc);
 
 	/* Make sure downcall interrupts will be enabled. */
-	raw_local_irq_unmask(INT_INTCTRL_1);
+	arch_local_irq_unmask(INT_INTCTRL_K);
 }
 
 void hv_message_intr(struct pt_regs *regs, int intnum)
diff --git a/arch/tile/kernel/pci.c b/arch/tile/kernel/pci.c
new file mode 100644
index 00000000000..a1ee25be9ad
--- /dev/null
+++ b/arch/tile/kernel/pci.c
@@ -0,0 +1,621 @@
+/*
+ * Copyright 2010 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/capability.h>
+#include <linux/sched.h>
+#include <linux/errno.h>
+#include <linux/bootmem.h>
+#include <linux/irq.h>
+#include <linux/io.h>
+#include <linux/uaccess.h>
+
+#include <asm/processor.h>
+#include <asm/sections.h>
+#include <asm/byteorder.h>
+#include <asm/hv_driver.h>
+#include <hv/drv_pcie_rc_intf.h>
+
+
+/*
+ * Initialization flow and process
+ * -------------------------------
+ *
+ * This files containes the routines to search for PCI buses,
+ * enumerate the buses, and configure any attached devices.
+ *
+ * There are two entry points here:
+ * 1) tile_pci_init
+ *    This sets up the pci_controller structs, and opens the
+ *    FDs to the hypervisor.  This is called from setup_arch() early
+ *    in the boot process.
+ * 2) pcibios_init
+ *    This probes the PCI bus(es) for any attached hardware.  It's
+ *    called by subsys_initcall.  All of the real work is done by the
+ *    generic Linux PCI layer.
+ *
+ */
+
+/*
+ * This flag tells if the platform is TILEmpower that needs
+ * special configuration for the PLX switch chip.
+ */
+int __write_once tile_plx_gen1;
+
+static struct pci_controller controllers[TILE_NUM_PCIE];
+static int num_controllers;
+
+static struct pci_ops tile_cfg_ops;
+
+
+/*
+ * We don't need to worry about the alignment of resources.
+ */
+resource_size_t pcibios_align_resource(void *data, const struct resource *res,
+			    resource_size_t size, resource_size_t align)
+{
+	return res->start;
+}
+EXPORT_SYMBOL(pcibios_align_resource);
+
+/*
+ * Open a FD to the hypervisor PCI device.
+ *
+ * controller_id is the controller number, config type is 0 or 1 for
+ * config0 or config1 operations.
+ */
+static int __init tile_pcie_open(int controller_id, int config_type)
+{
+	char filename[32];
+	int fd;
+
+	sprintf(filename, "pcie/%d/config%d", controller_id, config_type);
+
+	fd = hv_dev_open((HV_VirtAddr)filename, 0);
+
+	return fd;
+}
+
+
+/*
+ * Get the IRQ numbers from the HV and set up the handlers for them.
+ */
+static int __init tile_init_irqs(int controller_id,
+				 struct pci_controller *controller)
+{
+	char filename[32];
+	int fd;
+	int ret;
+	int x;
+	struct pcie_rc_config rc_config;
+
+	sprintf(filename, "pcie/%d/ctl", controller_id);
+	fd = hv_dev_open((HV_VirtAddr)filename, 0);
+	if (fd < 0) {
+		pr_err("PCI: hv_dev_open(%s) failed\n", filename);
+		return -1;
+	}
+	ret = hv_dev_pread(fd, 0, (HV_VirtAddr)(&rc_config),
+			   sizeof(rc_config), PCIE_RC_CONFIG_MASK_OFF);
+	hv_dev_close(fd);
+	if (ret != sizeof(rc_config)) {
+		pr_err("PCI: wanted %zd bytes, got %d\n",
+		       sizeof(rc_config), ret);
+		return -1;
+	}
+	/* Record irq_base so that we can map INTx to IRQ # later. */
+	controller->irq_base = rc_config.intr;
+
+	for (x = 0; x < 4; x++)
+		tile_irq_activate(rc_config.intr + x,
+				  TILE_IRQ_HW_CLEAR);
+
+	if (rc_config.plx_gen1)
+		controller->plx_gen1 = 1;
+
+	return 0;
+}
+
+/*
+ * First initialization entry point, called from setup_arch().
+ *
+ * Find valid controllers and fill in pci_controller structs for each
+ * of them.
+ *
+ * Returns the number of controllers discovered.
+ */
+int __init tile_pci_init(void)
+{
+	int i;
+
+	pr_info("PCI: Searching for controllers...\n");
+
+	/* Do any configuration we need before using the PCIe */
+
+	for (i = 0; i < TILE_NUM_PCIE; i++) {
+		int hv_cfg_fd0 = -1;
+		int hv_cfg_fd1 = -1;
+		int hv_mem_fd = -1;
+		char name[32];
+		struct pci_controller *controller;
+
+		/*
+		 * Open the fd to the HV.  If it fails then this
+		 * device doesn't exist.
+		 */
+		hv_cfg_fd0 = tile_pcie_open(i, 0);
+		if (hv_cfg_fd0 < 0)
+			continue;
+		hv_cfg_fd1 = tile_pcie_open(i, 1);
+		if (hv_cfg_fd1 < 0) {
+			pr_err("PCI: Couldn't open config fd to HV "
+			    "for controller %d\n", i);
+			goto err_cont;
+		}
+
+		sprintf(name, "pcie/%d/mem", i);
+		hv_mem_fd = hv_dev_open((HV_VirtAddr)name, 0);
+		if (hv_mem_fd < 0) {
+			pr_err("PCI: Could not open mem fd to HV!\n");
+			goto err_cont;
+		}
+
+		pr_info("PCI: Found PCI controller #%d\n", i);
+
+		controller = &controllers[num_controllers];
+
+		if (tile_init_irqs(i, controller)) {
+			pr_err("PCI: Could not initialize "
+			       "IRQs, aborting.\n");
+			goto err_cont;
+		}
+
+		controller->index = num_controllers;
+		controller->hv_cfg_fd[0] = hv_cfg_fd0;
+		controller->hv_cfg_fd[1] = hv_cfg_fd1;
+		controller->hv_mem_fd = hv_mem_fd;
+		controller->first_busno = 0;
+		controller->last_busno = 0xff;
+		controller->ops = &tile_cfg_ops;
+
+		num_controllers++;
+		continue;
+
+err_cont:
+		if (hv_cfg_fd0 >= 0)
+			hv_dev_close(hv_cfg_fd0);
+		if (hv_cfg_fd1 >= 0)
+			hv_dev_close(hv_cfg_fd1);
+		if (hv_mem_fd >= 0)
+			hv_dev_close(hv_mem_fd);
+		continue;
+	}
+
+	/*
+	 * Before using the PCIe, see if we need to do any platform-specific
+	 * configuration, such as the PLX switch Gen 1 issue on TILEmpower.
+	 */
+	for (i = 0; i < num_controllers; i++) {
+		struct pci_controller *controller = &controllers[i];
+
+		if (controller->plx_gen1)
+			tile_plx_gen1 = 1;
+	}
+
+	return num_controllers;
+}
+
+/*
+ * (pin - 1) converts from the PCI standard's [1:4] convention to
+ * a normal [0:3] range.
+ */
+static int tile_map_irq(struct pci_dev *dev, u8 slot, u8 pin)
+{
+	struct pci_controller *controller =
+		(struct pci_controller *)dev->sysdata;
+	return (pin - 1) + controller->irq_base;
+}
+
+
+static void __init fixup_read_and_payload_sizes(void)
+{
+	struct pci_dev *dev = NULL;
+	int smallest_max_payload = 0x1; /* Tile maxes out at 256 bytes. */
+	int max_read_size = 0x2; /* Limit to 512 byte reads. */
+	u16 new_values;
+
+	/* Scan for the smallest maximum payload size. */
+	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
+		int pcie_caps_offset;
+		u32 devcap;
+		int max_payload;
+
+		pcie_caps_offset = pci_find_capability(dev, PCI_CAP_ID_EXP);
+		if (pcie_caps_offset == 0)
+			continue;
+
+		pci_read_config_dword(dev, pcie_caps_offset + PCI_EXP_DEVCAP,
+				      &devcap);
+		max_payload = devcap & PCI_EXP_DEVCAP_PAYLOAD;
+		if (max_payload < smallest_max_payload)
+			smallest_max_payload = max_payload;
+	}
+
+	/* Now, set the max_payload_size for all devices to that value. */
+	new_values = (max_read_size << 12) | (smallest_max_payload << 5);
+	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
+		int pcie_caps_offset;
+		u16 devctl;
+
+		pcie_caps_offset = pci_find_capability(dev, PCI_CAP_ID_EXP);
+		if (pcie_caps_offset == 0)
+			continue;
+
+		pci_read_config_word(dev, pcie_caps_offset + PCI_EXP_DEVCTL,
+				     &devctl);
+		devctl &= ~(PCI_EXP_DEVCTL_PAYLOAD | PCI_EXP_DEVCTL_READRQ);
+		devctl |= new_values;
+		pci_write_config_word(dev, pcie_caps_offset + PCI_EXP_DEVCTL,
+				      devctl);
+	}
+}
+
+
+/*
+ * Second PCI initialization entry point, called by subsys_initcall.
+ *
+ * The controllers have been set up by the time we get here, by a call to
+ * tile_pci_init.
+ */
+static int __init pcibios_init(void)
+{
+	int i;
+
+	pr_info("PCI: Probing PCI hardware\n");
+
+	/*
+	 * Delay a bit in case devices aren't ready.  Some devices are
+	 * known to require at least 20ms here, but we use a more
+	 * conservative value.
+	 */
+	mdelay(250);
+
+	/* Scan all of the recorded PCI controllers.  */
+	for (i = 0; i < num_controllers; i++) {
+		struct pci_controller *controller = &controllers[i];
+		struct pci_bus *bus;
+
+		pr_info("PCI: initializing controller #%d\n", i);
+
+		/*
+		 * This comes from the generic Linux PCI driver.
+		 *
+		 * It reads the PCI tree for this bus into the Linux
+		 * data structures.
+		 *
+		 * This is inlined in linux/pci.h and calls into
+		 * pci_scan_bus_parented() in probe.c.
+		 */
+		bus = pci_scan_bus(0, controller->ops, controller);
+		controller->root_bus = bus;
+		controller->last_busno = bus->subordinate;
+
+	}
+
+	/* Do machine dependent PCI interrupt routing */
+	pci_fixup_irqs(pci_common_swizzle, tile_map_irq);
+
+	/*
+	 * This comes from the generic Linux PCI driver.
+	 *
+	 * It allocates all of the resources (I/O memory, etc)
+	 * associated with the devices read in above.
+	 */
+
+	pci_assign_unassigned_resources();
+
+	/* Configure the max_read_size and max_payload_size values. */
+	fixup_read_and_payload_sizes();
+
+	/* Record the I/O resources in the PCI controller structure. */
+	for (i = 0; i < num_controllers; i++) {
+		struct pci_bus *root_bus = controllers[i].root_bus;
+		struct pci_bus *next_bus;
+		struct pci_dev *dev;
+
+		list_for_each_entry(dev, &root_bus->devices, bus_list) {
+			/* Find the PCI host controller, ie. the 1st bridge. */
+			if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI &&
+				(PCI_SLOT(dev->devfn) == 0)) {
+				next_bus = dev->subordinate;
+				controllers[i].mem_resources[0] =
+					*next_bus->resource[0];
+				controllers[i].mem_resources[1] =
+					 *next_bus->resource[1];
+				controllers[i].mem_resources[2] =
+					 *next_bus->resource[2];
+
+				break;
+			}
+		}
+
+	}
+
+	return 0;
+}
+subsys_initcall(pcibios_init);
+
+/*
+ * No bus fixups needed.
+ */
+void __devinit pcibios_fixup_bus(struct pci_bus *bus)
+{
+	/* Nothing needs to be done. */
+}
+
+/*
+ * This can be called from the generic PCI layer, but doesn't need to
+ * do anything.
+ */
+char __devinit *pcibios_setup(char *str)
+{
+	/* Nothing needs to be done. */
+	return str;
+}
+
+/*
+ * This is called from the generic Linux layer.
+ */
+void __init pcibios_update_irq(struct pci_dev *dev, int irq)
+{
+	pci_write_config_byte(dev, PCI_INTERRUPT_LINE, irq);
+}
+
+/*
+ * Enable memory and/or address decoding, as appropriate, for the
+ * device described by the 'dev' struct.
+ *
+ * This is called from the generic PCI layer, and can be called
+ * for bridges or endpoints.
+ */
+int pcibios_enable_device(struct pci_dev *dev, int mask)
+{
+	u16 cmd, old_cmd;
+	u8 header_type;
+	int i;
+	struct resource *r;
+
+	pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type);
+
+	pci_read_config_word(dev, PCI_COMMAND, &cmd);
+	old_cmd = cmd;
+	if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) {
+		/*
+		 * For bridges, we enable both memory and I/O decoding
+		 * in call cases.
+		 */
+		cmd |= PCI_COMMAND_IO;
+		cmd |= PCI_COMMAND_MEMORY;
+	} else {
+		/*
+		 * For endpoints, we enable memory and/or I/O decoding
+		 * only if they have a memory resource of that type.
+		 */
+		for (i = 0; i < 6; i++) {
+			r = &dev->resource[i];
+			if (r->flags & IORESOURCE_UNSET) {
+				pr_err("PCI: Device %s not available "
+				       "because of resource collisions\n",
+				       pci_name(dev));
+				return -EINVAL;
+			}
+			if (r->flags & IORESOURCE_IO)
+				cmd |= PCI_COMMAND_IO;
+			if (r->flags & IORESOURCE_MEM)
+				cmd |= PCI_COMMAND_MEMORY;
+		}
+	}
+
+	/*
+	 * We only write the command if it changed.
+	 */
+	if (cmd != old_cmd)
+		pci_write_config_word(dev, PCI_COMMAND, cmd);
+	return 0;
+}
+
+void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long max)
+{
+	unsigned long start = pci_resource_start(dev, bar);
+	unsigned long len = pci_resource_len(dev, bar);
+	unsigned long flags = pci_resource_flags(dev, bar);
+
+	if (!len)
+		return NULL;
+	if (max && len > max)
+		len = max;
+
+	if (!(flags & IORESOURCE_MEM)) {
+		pr_info("PCI: Trying to map invalid resource %#lx\n", flags);
+		start = 0;
+	}
+
+	return (void __iomem *)start;
+}
+EXPORT_SYMBOL(pci_iomap);
+
+
+/****************************************************************
+ *
+ * Tile PCI config space read/write routines
+ *
+ ****************************************************************/
+
+/*
+ * These are the normal read and write ops
+ * These are expanded with macros from  pci_bus_read_config_byte() etc.
+ *
+ * devfn is the combined PCI slot & function.
+ *
+ * offset is in bytes, from the start of config space for the
+ * specified bus & slot.
+ */
+
+static int __devinit tile_cfg_read(struct pci_bus *bus,
+				   unsigned int devfn,
+				   int offset,
+				   int size,
+				   u32 *val)
+{
+	struct pci_controller *controller = bus->sysdata;
+	int busnum = bus->number & 0xff;
+	int slot = (devfn >> 3) & 0x1f;
+	int function = devfn & 0x7;
+	u32 addr;
+	int config_mode = 1;
+
+	/*
+	 * There is no bridge between the Tile and bus 0, so we
+	 * use config0 to talk to bus 0.
+	 *
+	 * If we're talking to a bus other than zero then we
+	 * must have found a bridge.
+	 */
+	if (busnum == 0) {
+		/*
+		 * We fake an empty slot for (busnum == 0) && (slot > 0),
+		 * since there is only one slot on bus 0.
+		 */
+		if (slot) {
+			*val = 0xFFFFFFFF;
+			return 0;
+		}
+		config_mode = 0;
+	}
+
+	addr = busnum << 20;		/* Bus in 27:20 */
+	addr |= slot << 15;		/* Slot (device) in 19:15 */
+	addr |= function << 12;		/* Function is in 14:12 */
+	addr |= (offset & 0xFFF);	/* byte address in 0:11 */
+
+	return hv_dev_pread(controller->hv_cfg_fd[config_mode], 0,
+			    (HV_VirtAddr)(val), size, addr);
+}
+
+
+/*
+ * See tile_cfg_read() for relevent comments.
+ * Note that "val" is the value to write, not a pointer to that value.
+ */
+static int __devinit tile_cfg_write(struct pci_bus *bus,
+				    unsigned int devfn,
+				    int offset,
+				    int size,
+				    u32 val)
+{
+	struct pci_controller *controller = bus->sysdata;
+	int busnum = bus->number & 0xff;
+	int slot = (devfn >> 3) & 0x1f;
+	int function = devfn & 0x7;
+	u32 addr;
+	int config_mode = 1;
+	HV_VirtAddr valp = (HV_VirtAddr)&val;
+
+	/*
+	 * For bus 0 slot 0 we use config 0 accesses.
+	 */
+	if (busnum == 0) {
+		/*
+		 * We fake an empty slot for (busnum == 0) && (slot > 0),
+		 * since there is only one slot on bus 0.
+		 */
+		if (slot)
+			return 0;
+		config_mode = 0;
+	}
+
+	addr = busnum << 20;		/* Bus in 27:20 */
+	addr |= slot << 15;		/* Slot (device) in 19:15 */
+	addr |= function << 12;		/* Function is in 14:12 */
+	addr |= (offset & 0xFFF);	/* byte address in 0:11 */
+
+#ifdef __BIG_ENDIAN
+	/* Point to the correct part of the 32-bit "val". */
+	valp += 4 - size;
+#endif
+
+	return hv_dev_pwrite(controller->hv_cfg_fd[config_mode], 0,
+			     valp, size, addr);
+}
+
+
+static struct pci_ops tile_cfg_ops = {
+	.read =         tile_cfg_read,
+	.write =        tile_cfg_write,
+};
+
+
+/*
+ * In the following, each PCI controller's mem_resources[1]
+ * represents its (non-prefetchable) PCI memory resource.
+ * mem_resources[0] and mem_resources[2] refer to its PCI I/O and
+ * prefetchable PCI memory resources, respectively.
+ * For more details, see pci_setup_bridge() in setup-bus.c.
+ * By comparing the target PCI memory address against the
+ * end address of controller 0, we can determine the controller
+ * that should accept the PCI memory access.
+ */
+#define TILE_READ(size, type)						\
+type _tile_read##size(unsigned long addr)				\
+{									\
+	type val;							\
+	int idx = 0;							\
+	if (addr > controllers[0].mem_resources[1].end &&		\
+	    addr > controllers[0].mem_resources[2].end)			\
+		idx = 1;                                                \
+	if (hv_dev_pread(controllers[idx].hv_mem_fd, 0,			\
+			 (HV_VirtAddr)(&val), sizeof(type), addr))	\
+		pr_err("PCI: read %zd bytes at 0x%lX failed\n",		\
+		       sizeof(type), addr);				\
+	return val;							\
+}									\
+EXPORT_SYMBOL(_tile_read##size)
+
+TILE_READ(b, u8);
+TILE_READ(w, u16);
+TILE_READ(l, u32);
+TILE_READ(q, u64);
+
+#define TILE_WRITE(size, type)						\
+void _tile_write##size(type val, unsigned long addr)			\
+{									\
+	int idx = 0;							\
+	if (addr > controllers[0].mem_resources[1].end &&		\
+	    addr > controllers[0].mem_resources[2].end)			\
+		idx = 1;                                                \
+	if (hv_dev_pwrite(controllers[idx].hv_mem_fd, 0,		\
+			  (HV_VirtAddr)(&val), sizeof(type), addr))	\
+		pr_err("PCI: write %zd bytes at 0x%lX failed\n",	\
+		       sizeof(type), addr);				\
+}									\
+EXPORT_SYMBOL(_tile_write##size)
+
+TILE_WRITE(b, u8);
+TILE_WRITE(w, u16);
+TILE_WRITE(l, u32);
+TILE_WRITE(q, u64);
diff --git a/arch/tile/kernel/process.c b/arch/tile/kernel/process.c
index 84c29111756..e90eb53173b 100644
--- a/arch/tile/kernel/process.c
+++ b/arch/tile/kernel/process.c
@@ -212,11 +212,19 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
 	childregs->sp = sp;  /* override with new user stack pointer */
 
 	/*
+	 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
+	 * which is passed in as arg #5 to sys_clone().
+	 */
+	if (clone_flags & CLONE_SETTLS)
+		childregs->tp = regs->regs[4];
+
+	/*
 	 * Copy the callee-saved registers from the passed pt_regs struct
 	 * into the context-switch callee-saved registers area.
-	 * We have to restore the callee-saved registers since we may
-	 * be cloning a userspace task with userspace register state,
-	 * and we won't be unwinding the same kernel frames to restore them.
+	 * This way when we start the interrupt-return sequence, the
+	 * callee-save registers will be correctly in registers, which
+	 * is how we assume the compiler leaves them as we start doing
+	 * the normal return-from-interrupt path after calling C code.
 	 * Zero out the C ABI save area to mark the top of the stack.
 	 */
 	ksp = (unsigned long) childregs;
@@ -304,15 +312,25 @@ int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 /* Allow user processes to access the DMA SPRs */
 void grant_dma_mpls(void)
 {
+#if CONFIG_KERNEL_PL == 2
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
+#else
 	__insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1);
 	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);
+#endif
 }
 
 /* Forbid user processes from accessing the DMA SPRs */
 void restrict_dma_mpls(void)
 {
+#if CONFIG_KERNEL_PL == 2
+	__insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1);
+	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);
+#else
 	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1);
 	__insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);
+#endif
 }
 
 /* Pause the DMA engine, then save off its state registers. */
@@ -523,19 +541,15 @@ struct task_struct *__sched _switch_to(struct task_struct *prev,
 	 * Switch kernel SP, PC, and callee-saved registers.
 	 * In the context of the new task, return the old task pointer
 	 * (i.e. the task that actually called __switch_to).
-	 * Pass the value to use for SYSTEM_SAVE_1_0 when we reset our sp.
+	 * Pass the value to use for SYSTEM_SAVE_K_0 when we reset our sp.
 	 */
 	return __switch_to(prev, next, next_current_ksp0(next));
 }
 
-long _sys_fork(struct pt_regs *regs)
-{
-	return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
-}
-
-long _sys_clone(unsigned long clone_flags, unsigned long newsp,
-		void __user *parent_tidptr, void __user *child_tidptr,
-		struct pt_regs *regs)
+/* Note there is an implicit fifth argument if (clone_flags & CLONE_SETTLS). */
+SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
+		void __user *, parent_tidptr, void __user *, child_tidptr,
+		struct pt_regs *, regs)
 {
 	if (!newsp)
 		newsp = regs->sp;
@@ -543,18 +557,13 @@ long _sys_clone(unsigned long clone_flags, unsigned long newsp,
 		       parent_tidptr, child_tidptr);
 }
 
-long _sys_vfork(struct pt_regs *regs)
-{
-	return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp,
-		       regs, 0, NULL, NULL);
-}
-
 /*
  * sys_execve() executes a new program.
  */
-long _sys_execve(const char __user *path,
-		 const char __user *const __user *argv,
-		 const char __user *const __user *envp, struct pt_regs *regs)
+SYSCALL_DEFINE4(execve, const char __user *, path,
+		const char __user *const __user *, argv,
+		const char __user *const __user *, envp,
+		struct pt_regs *, regs)
 {
 	long error;
 	char *filename;
@@ -570,9 +579,10 @@ out:
 }
 
 #ifdef CONFIG_COMPAT
-long _compat_sys_execve(const char __user *path,
-			const compat_uptr_t __user *argv,
-			const compat_uptr_t __user *envp, struct pt_regs *regs)
+long compat_sys_execve(const char __user *path,
+		       const compat_uptr_t __user *argv,
+		       const compat_uptr_t __user *envp,
+		       struct pt_regs *regs)
 {
 	long error;
 	char *filename;
diff --git a/arch/tile/kernel/ptrace.c b/arch/tile/kernel/ptrace.c
index 7161bd03d2f..e92e40527d6 100644
--- a/arch/tile/kernel/ptrace.c
+++ b/arch/tile/kernel/ptrace.c
@@ -32,25 +32,6 @@ void user_disable_single_step(struct task_struct *child)
 }
 
 /*
- * This routine will put a word on the process's privileged stack.
- */
-static void putreg(struct task_struct *task,
-		   unsigned long addr, unsigned long value)
-{
-	unsigned int regno = addr / sizeof(unsigned long);
-	struct pt_regs *childregs = task_pt_regs(task);
-	childregs->regs[regno] = value;
-	childregs->flags |= PT_FLAGS_RESTORE_REGS;
-}
-
-static unsigned long getreg(struct task_struct *task, unsigned long addr)
-{
-	unsigned int regno = addr / sizeof(unsigned long);
-	struct pt_regs *childregs = task_pt_regs(task);
-	return childregs->regs[regno];
-}
-
-/*
  * Called by kernel/ptrace.c when detaching..
  */
 void ptrace_disable(struct task_struct *child)
@@ -64,61 +45,80 @@ void ptrace_disable(struct task_struct *child)
 	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 }
 
-long arch_ptrace(struct task_struct *child, long request, long addr, long data)
+long arch_ptrace(struct task_struct *child, long request,
+		 unsigned long addr, unsigned long data)
 {
-	unsigned long __user *datap;
+	unsigned long __user *datap = (long __user __force *)data;
 	unsigned long tmp;
-	int i;
 	long ret = -EIO;
-
-#ifdef CONFIG_COMPAT
-	if (task_thread_info(current)->status & TS_COMPAT)
-		data = (u32)data;
-	if (task_thread_info(child)->status & TS_COMPAT)
-		addr = (u32)addr;
-#endif
-	datap = (unsigned long __user __force *)data;
+	char *childreg;
+	struct pt_regs copyregs;
+	int ex1_offset;
 
 	switch (request) {
 
 	case PTRACE_PEEKUSR:  /* Read register from pt_regs. */
-		if (addr & (sizeof(data)-1))
-			break;
-		if (addr < 0 || addr >= PTREGS_SIZE)
+		if (addr >= PTREGS_SIZE)
 			break;
-		tmp = getreg(child, addr);   /* Read register */
-		ret = put_user(tmp, datap);
+		childreg = (char *)task_pt_regs(child) + addr;
+#ifdef CONFIG_COMPAT
+		if (is_compat_task()) {
+			if (addr & (sizeof(compat_long_t)-1))
+				break;
+			ret = put_user(*(compat_long_t *)childreg,
+				       (compat_long_t __user *)datap);
+		} else
+#endif
+		{
+			if (addr & (sizeof(long)-1))
+				break;
+			ret = put_user(*(long *)childreg, datap);
+		}
 		break;
 
 	case PTRACE_POKEUSR:  /* Write register in pt_regs. */
-		if (addr & (sizeof(data)-1))
+		if (addr >= PTREGS_SIZE)
 			break;
-		if (addr < 0 || addr >= PTREGS_SIZE)
-			break;
-		putreg(child, addr, data);   /* Write register */
+		childreg = (char *)task_pt_regs(child) + addr;
+
+		/* Guard against overwrites of the privilege level. */
+		ex1_offset = PTREGS_OFFSET_EX1;
+#if defined(CONFIG_COMPAT) && defined(__BIG_ENDIAN)
+		if (is_compat_task())   /* point at low word */
+			ex1_offset += sizeof(compat_long_t);
+#endif
+		if (addr == ex1_offset)
+			data = PL_ICS_EX1(USER_PL, EX1_ICS(data));
+
+#ifdef CONFIG_COMPAT
+		if (is_compat_task()) {
+			if (addr & (sizeof(compat_long_t)-1))
+				break;
+			*(compat_long_t *)childreg = data;
+		} else
+#endif
+		{
+			if (addr & (sizeof(long)-1))
+				break;
+			*(long *)childreg = data;
+		}
 		ret = 0;
 		break;
 
 	case PTRACE_GETREGS:  /* Get all registers from the child. */
-		if (!access_ok(VERIFY_WRITE, datap, PTREGS_SIZE))
-			break;
-		for (i = 0; i < PTREGS_SIZE; i += sizeof(long)) {
-			ret = __put_user(getreg(child, i), datap);
-			if (ret != 0)
-				break;
-			datap++;
+		if (copy_to_user(datap, task_pt_regs(child),
+				 sizeof(struct pt_regs)) == 0) {
+			ret = 0;
 		}
 		break;
 
 	case PTRACE_SETREGS:  /* Set all registers in the child. */
-		if (!access_ok(VERIFY_READ, datap, PTREGS_SIZE))
-			break;
-		for (i = 0; i < PTREGS_SIZE; i += sizeof(long)) {
-			ret = __get_user(tmp, datap);
-			if (ret != 0)
-				break;
-			putreg(child, i, tmp);
-			datap++;
+		if (copy_from_user(&copyregs, datap,
+				   sizeof(struct pt_regs)) == 0) {
+			copyregs.ex1 =
+				PL_ICS_EX1(USER_PL, EX1_ICS(copyregs.ex1));
+			*task_pt_regs(child) = copyregs;
+			ret = 0;
 		}
 		break;
 
diff --git a/arch/tile/kernel/reboot.c b/arch/tile/kernel/reboot.c
index acd86d20beb..baa3d905fee 100644
--- a/arch/tile/kernel/reboot.c
+++ b/arch/tile/kernel/reboot.c
@@ -27,7 +27,7 @@
 void machine_halt(void)
 {
 	warn_early_printk();
-	raw_local_irq_disable_all();
+	arch_local_irq_disable_all();
 	smp_send_stop();
 	hv_halt();
 }
@@ -35,14 +35,14 @@ void machine_halt(void)
 void machine_power_off(void)
 {
 	warn_early_printk();
-	raw_local_irq_disable_all();
+	arch_local_irq_disable_all();
 	smp_send_stop();
 	hv_power_off();
 }
 
 void machine_restart(char *cmd)
 {
-	raw_local_irq_disable_all();
+	arch_local_irq_disable_all();
 	smp_send_stop();
 	hv_restart((HV_VirtAddr) "vmlinux", (HV_VirtAddr) cmd);
 }
diff --git a/arch/tile/kernel/regs_32.S b/arch/tile/kernel/regs_32.S
index e88d6e12278..caa13101c26 100644
--- a/arch/tile/kernel/regs_32.S
+++ b/arch/tile/kernel/regs_32.S
@@ -85,7 +85,7 @@ STD_ENTRY_SECTION(__switch_to, .sched.text)
 	{
 	  /* Update sp and ksp0 simultaneously to avoid backtracer warnings. */
 	  move sp, r13
-	  mtspr SYSTEM_SAVE_1_0, r2
+	  mtspr SPR_SYSTEM_SAVE_K_0, r2
 	}
 	FOR_EACH_CALLEE_SAVED_REG(LOAD_REG)
 .L__switch_to_pc:
diff --git a/arch/tile/kernel/setup.c b/arch/tile/kernel/setup.c
index e7d54c73d5c..f18573643ed 100644
--- a/arch/tile/kernel/setup.c
+++ b/arch/tile/kernel/setup.c
@@ -30,8 +30,6 @@
 #include <linux/timex.h>
 #include <asm/setup.h>
 #include <asm/sections.h>
-#include <asm/sections.h>
-#include <asm/cacheflush.h>
 #include <asm/cacheflush.h>
 #include <asm/pgalloc.h>
 #include <asm/mmu_context.h>
@@ -187,11 +185,11 @@ early_param("vmalloc", parse_vmalloc);
 
 #ifdef CONFIG_HIGHMEM
 /*
- * Determine for each controller where its lowmem is mapped and how
- * much of it is mapped there.  On controller zero, the first few
- * megabytes are mapped at 0xfd000000 as code, so in principle we
- * could start our data mappings higher up, but for now we don't
- * bother, to avoid additional confusion.
+ * Determine for each controller where its lowmem is mapped and how much of
+ * it is mapped there.  On controller zero, the first few megabytes are
+ * already mapped in as code at MEM_SV_INTRPT, so in principle we could
+ * start our data mappings higher up, but for now we don't bother, to avoid
+ * additional confusion.
  *
  * One question is whether, on systems with more than 768 Mb and
  * controllers of different sizes, to map in a proportionate amount of
@@ -311,7 +309,7 @@ static void __init setup_memory(void)
 #endif
 
 	/* We are using a char to hold the cpu_2_node[] mapping */
-	BUG_ON(MAX_NUMNODES > 127);
+	BUILD_BUG_ON(MAX_NUMNODES > 127);
 
 	/* Discover the ranges of memory available to us */
 	for (i = 0; ; ++i) {
@@ -842,7 +840,7 @@ static int __init topology_init(void)
 	for_each_online_node(i)
 		register_one_node(i);
 
-	for_each_present_cpu(i)
+	for (i = 0; i < smp_height * smp_width; ++i)
 		register_cpu(&cpu_devices[i], i);
 
 	return 0;
@@ -870,11 +868,14 @@ void __cpuinit setup_cpu(int boot)
 
 	/* Allow asynchronous TLB interrupts. */
 #if CHIP_HAS_TILE_DMA()
-	raw_local_irq_unmask(INT_DMATLB_MISS);
-	raw_local_irq_unmask(INT_DMATLB_ACCESS);
+	arch_local_irq_unmask(INT_DMATLB_MISS);
+	arch_local_irq_unmask(INT_DMATLB_ACCESS);
 #endif
 #if CHIP_HAS_SN_PROC()
-	raw_local_irq_unmask(INT_SNITLB_MISS);
+	arch_local_irq_unmask(INT_SNITLB_MISS);
+#endif
+#ifdef __tilegx__
+	arch_local_irq_unmask(INT_SINGLE_STEP_K);
 #endif
 
 	/*
@@ -893,11 +894,12 @@ void __cpuinit setup_cpu(int boot)
 #endif
 
 	/*
-	 * Set the MPL for interrupt control 0 to user level.
-	 * This includes access to the SYSTEM_SAVE and EX_CONTEXT SPRs,
-	 * as well as the PL 0 interrupt mask.
+	 * Set the MPL for interrupt control 0 & 1 to the corresponding
+	 * values.  This includes access to the SYSTEM_SAVE and EX_CONTEXT
+	 * SPRs, as well as the interrupt mask.
 	 */
 	__insn_mtspr(SPR_MPL_INTCTRL_0_SET_0, 1);
+	__insn_mtspr(SPR_MPL_INTCTRL_1_SET_1, 1);
 
 	/* Initialize IRQ support for this cpu. */
 	setup_irq_regs();
@@ -1033,7 +1035,7 @@ static void __init validate_va(void)
 	 * In addition, make sure we CAN'T use the end of memory, since
 	 * we use the last chunk of each pgd for the pgd_list.
 	 */
-	int i, fc_fd_ok = 0;
+	int i, user_kernel_ok = 0;
 	unsigned long max_va = 0;
 	unsigned long list_va =
 		((PGD_LIST_OFFSET / sizeof(pgd_t)) << PGDIR_SHIFT);
@@ -1044,13 +1046,13 @@ static void __init validate_va(void)
 			break;
 		if (range.start <= MEM_USER_INTRPT &&
 		    range.start + range.size >= MEM_HV_INTRPT)
-			fc_fd_ok = 1;
+			user_kernel_ok = 1;
 		if (range.start == 0)
 			max_va = range.size;
 		BUG_ON(range.start + range.size > list_va);
 	}
-	if (!fc_fd_ok)
-		early_panic("Hypervisor not configured for VAs 0xfc/0xfd\n");
+	if (!user_kernel_ok)
+		early_panic("Hypervisor not configured for user/kernel VAs\n");
 	if (max_va == 0)
 		early_panic("Hypervisor not configured for low VAs\n");
 	if (max_va < KERNEL_HIGH_VADDR)
@@ -1334,6 +1336,10 @@ static void __init pcpu_fc_populate_pte(unsigned long addr)
 	pte_t *pte;
 
 	BUG_ON(pgd_addr_invalid(addr));
+	if (addr < VMALLOC_START || addr >= VMALLOC_END)
+		panic("PCPU addr %#lx outside vmalloc range %#lx..%#lx;"
+		      " try increasing CONFIG_VMALLOC_RESERVE\n",
+		      addr, VMALLOC_START, VMALLOC_END);
 
 	pgd = swapper_pg_dir + pgd_index(addr);
 	pud = pud_offset(pgd, addr);
diff --git a/arch/tile/kernel/signal.c b/arch/tile/kernel/signal.c
index ce183aa1492..1260321155f 100644
--- a/arch/tile/kernel/signal.c
+++ b/arch/tile/kernel/signal.c
@@ -16,7 +16,6 @@
 #include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
-#include <linux/smp_lock.h>
 #include <linux/kernel.h>
 #include <linux/signal.h>
 #include <linux/errno.h>
@@ -41,8 +40,8 @@
 #define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
 
 
-long _sys_sigaltstack(const stack_t __user *uss,
-		      stack_t __user *uoss, struct pt_regs *regs)
+SYSCALL_DEFINE3(sigaltstack, const stack_t __user *, uss,
+		stack_t __user *, uoss, struct pt_regs *, regs)
 {
 	return do_sigaltstack(uss, uoss, regs->sp);
 }
@@ -53,7 +52,7 @@ long _sys_sigaltstack(const stack_t __user *uss,
  */
 
 int restore_sigcontext(struct pt_regs *regs,
-		       struct sigcontext __user *sc, long *pr0)
+		       struct sigcontext __user *sc)
 {
 	int err = 0;
 	int i;
@@ -71,19 +70,20 @@ int restore_sigcontext(struct pt_regs *regs,
 	for (i = 0; i < sizeof(struct pt_regs)/sizeof(long); ++i)
 		err |= __get_user(regs->regs[i], &sc->gregs[i]);
 
+	/* Ensure that the PL is always set to USER_PL. */
+	regs->ex1 = PL_ICS_EX1(USER_PL, EX1_ICS(regs->ex1));
+
 	regs->faultnum = INT_SWINT_1_SIGRETURN;
 
-	err |= __get_user(*pr0, &sc->gregs[0]);
 	return err;
 }
 
-/* sigreturn() returns long since it restores r0 in the interrupted code. */
-long _sys_rt_sigreturn(struct pt_regs *regs)
+/* The assembly shim for this function arranges to ignore the return value. */
+SYSCALL_DEFINE1(rt_sigreturn, struct pt_regs *, regs)
 {
 	struct rt_sigframe __user *frame =
 		(struct rt_sigframe __user *)(regs->sp);
 	sigset_t set;
-	long r0;
 
 	if (!access_ok(VERIFY_READ, frame, sizeof(*frame)))
 		goto badframe;
@@ -96,13 +96,13 @@ long _sys_rt_sigreturn(struct pt_regs *regs)
 	recalc_sigpending();
 	spin_unlock_irq(&current->sighand->siglock);
 
-	if (restore_sigcontext(regs, &frame->uc.uc_mcontext, &r0))
+	if (restore_sigcontext(regs, &frame->uc.uc_mcontext))
 		goto badframe;
 
 	if (do_sigaltstack(&frame->uc.uc_stack, NULL, regs->sp) == -EFAULT)
 		goto badframe;
 
-	return r0;
+	return 0;
 
 badframe:
 	force_sig(SIGSEGV, current);
@@ -330,7 +330,7 @@ void do_signal(struct pt_regs *regs)
 			current_thread_info()->status &= ~TS_RESTORE_SIGMASK;
 		}
 
-		return;
+		goto done;
 	}
 
 	/* Did we come from a system call? */
@@ -358,4 +358,8 @@ void do_signal(struct pt_regs *regs)
 		current_thread_info()->status &= ~TS_RESTORE_SIGMASK;
 		sigprocmask(SIG_SETMASK, &current->saved_sigmask, NULL);
 	}
+
+done:
+	/* Avoid double syscall restart if there are nested signals. */
+	regs->faultnum = INT_SWINT_1_SIGRETURN;
 }
diff --git a/arch/tile/kernel/single_step.c b/arch/tile/kernel/single_step.c
index 5ec4b9c651f..1eb3b39e36c 100644
--- a/arch/tile/kernel/single_step.c
+++ b/arch/tile/kernel/single_step.c
@@ -15,7 +15,7 @@
  * Derived from iLib's single-stepping code.
  */
 
-#ifndef __tilegx__   /* No support for single-step yet. */
+#ifndef __tilegx__   /* Hardware support for single step unavailable. */
 
 /* These functions are only used on the TILE platform */
 #include <linux/slab.h>
@@ -660,4 +660,75 @@ void single_step_once(struct pt_regs *regs)
 		regs->pc += 8;
 }
 
+#else
+#include <linux/smp.h>
+#include <linux/ptrace.h>
+#include <arch/spr_def.h>
+
+static DEFINE_PER_CPU(unsigned long, ss_saved_pc);
+
+
+/*
+ * Called directly on the occasion of an interrupt.
+ *
+ * If the process doesn't have single step set, then we use this as an
+ * opportunity to turn single step off.
+ *
+ * It has been mentioned that we could conditionally turn off single stepping
+ * on each entry into the kernel and rely on single_step_once to turn it
+ * on for the processes that matter (as we already do), but this
+ * implementation is somewhat more efficient in that we muck with registers
+ * once on a bum interrupt rather than on every entry into the kernel.
+ *
+ * If SINGLE_STEP_CONTROL_K has CANCELED set, then an interrupt occurred,
+ * so we have to run through this process again before we can say that an
+ * instruction has executed.
+ *
+ * swint will set CANCELED, but it's a legitimate instruction.  Fortunately
+ * it changes the PC.  If it hasn't changed, then we know that the interrupt
+ * wasn't generated by swint and we'll need to run this process again before
+ * we can say an instruction has executed.
+ *
+ * If either CANCELED == 0 or the PC's changed, we send out SIGTRAPs and get
+ * on with our lives.
+ */
+
+void gx_singlestep_handle(struct pt_regs *regs, int fault_num)
+{
+	unsigned long *ss_pc = &__get_cpu_var(ss_saved_pc);
+	struct thread_info *info = (void *)current_thread_info();
+	int is_single_step = test_ti_thread_flag(info, TIF_SINGLESTEP);
+	unsigned long control = __insn_mfspr(SPR_SINGLE_STEP_CONTROL_K);
+
+	if (is_single_step == 0) {
+		__insn_mtspr(SPR_SINGLE_STEP_EN_K_K, 0);
+
+	} else if ((*ss_pc != regs->pc) ||
+		   (!(control & SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK))) {
+
+		ptrace_notify(SIGTRAP);
+		control |= SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK;
+		control |= SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK;
+		__insn_mtspr(SPR_SINGLE_STEP_CONTROL_K, control);
+	}
+}
+
+
+/*
+ * Called from need_singlestep.  Set up the control registers and the enable
+ * register, then return back.
+ */
+
+void single_step_once(struct pt_regs *regs)
+{
+	unsigned long *ss_pc = &__get_cpu_var(ss_saved_pc);
+	unsigned long control = __insn_mfspr(SPR_SINGLE_STEP_CONTROL_K);
+
+	*ss_pc = regs->pc;
+	control |= SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK;
+	control |= SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK;
+	__insn_mtspr(SPR_SINGLE_STEP_CONTROL_K, control);
+	__insn_mtspr(SPR_SINGLE_STEP_EN_K_K, 1 << USER_PL);
+}
+
 #endif /* !__tilegx__ */
diff --git a/arch/tile/kernel/smp.c b/arch/tile/kernel/smp.c
index 1cb5ec79de0..9575b37a8b7 100644
--- a/arch/tile/kernel/smp.c
+++ b/arch/tile/kernel/smp.c
@@ -115,7 +115,7 @@ static void smp_start_cpu_interrupt(void)
 static void smp_stop_cpu_interrupt(void)
 {
 	set_cpu_online(smp_processor_id(), 0);
-	raw_local_irq_disable_all();
+	arch_local_irq_disable_all();
 	for (;;)
 		asm("nap");
 }
@@ -212,7 +212,7 @@ void __init ipi_init(void)
 
 		tile.x = cpu_x(cpu);
 		tile.y = cpu_y(cpu);
-		if (hv_get_ipi_pte(tile, 1, &pte) != 0)
+		if (hv_get_ipi_pte(tile, KERNEL_PL, &pte) != 0)
 			panic("Failed to initialize IPI for cpu %d\n", cpu);
 
 		offset = hv_pte_get_pfn(pte) << PAGE_SHIFT;
diff --git a/arch/tile/kernel/smpboot.c b/arch/tile/kernel/smpboot.c
index 74d62d098ed..b949edcec20 100644
--- a/arch/tile/kernel/smpboot.c
+++ b/arch/tile/kernel/smpboot.c
@@ -18,7 +18,6 @@
 #include <linux/mm.h>
 #include <linux/sched.h>
 #include <linux/kernel_stat.h>
-#include <linux/smp_lock.h>
 #include <linux/bootmem.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
diff --git a/arch/tile/kernel/stack.c b/arch/tile/kernel/stack.c
index ea2e0ce2838..0d54106be3d 100644
--- a/arch/tile/kernel/stack.c
+++ b/arch/tile/kernel/stack.c
@@ -30,6 +30,10 @@
 #include <arch/abi.h>
 #include <arch/interrupts.h>
 
+#define KBT_ONGOING	0  /* Backtrace still ongoing */
+#define KBT_DONE	1  /* Backtrace cleanly completed */
+#define KBT_RUNNING	2  /* Can't run backtrace on a running task */
+#define KBT_LOOP	3  /* Backtrace entered a loop */
 
 /* Is address on the specified kernel stack? */
 static int in_kernel_stack(struct KBacktraceIterator *kbt, VirtualAddress sp)
@@ -207,11 +211,11 @@ static int KBacktraceIterator_next_item_inclusive(
 	for (;;) {
 		do {
 			if (!KBacktraceIterator_is_sigreturn(kbt))
-				return 1;
+				return KBT_ONGOING;
 		} while (backtrace_next(&kbt->it));
 
 		if (!KBacktraceIterator_restart(kbt))
-			return 0;
+			return KBT_DONE;
 	}
 }
 
@@ -264,7 +268,7 @@ void KBacktraceIterator_init(struct KBacktraceIterator *kbt,
 	kbt->pgtable = NULL;
 	kbt->verbose = 0;   /* override in caller if desired */
 	kbt->profile = 0;   /* override in caller if desired */
-	kbt->end = 0;
+	kbt->end = KBT_ONGOING;
 	kbt->new_context = 0;
 	if (is_current) {
 		HV_PhysAddr pgdir_pa = hv_inquire_context().page_table;
@@ -290,7 +294,7 @@ void KBacktraceIterator_init(struct KBacktraceIterator *kbt,
 	if (regs == NULL) {
 		if (is_current || t->state == TASK_RUNNING) {
 			/* Can't do this; we need registers */
-			kbt->end = 1;
+			kbt->end = KBT_RUNNING;
 			return;
 		}
 		pc = get_switch_to_pc();
@@ -305,26 +309,29 @@ void KBacktraceIterator_init(struct KBacktraceIterator *kbt,
 	}
 
 	backtrace_init(&kbt->it, read_memory_func, kbt, pc, lr, sp, r52);
-	kbt->end = !KBacktraceIterator_next_item_inclusive(kbt);
+	kbt->end = KBacktraceIterator_next_item_inclusive(kbt);
 }
 EXPORT_SYMBOL(KBacktraceIterator_init);
 
 int KBacktraceIterator_end(struct KBacktraceIterator *kbt)
 {
-	return kbt->end;
+	return kbt->end != KBT_ONGOING;
 }
 EXPORT_SYMBOL(KBacktraceIterator_end);
 
 void KBacktraceIterator_next(struct KBacktraceIterator *kbt)
 {
+	VirtualAddress old_pc = kbt->it.pc, old_sp = kbt->it.sp;
 	kbt->new_context = 0;
-	if (!backtrace_next(&kbt->it) &&
-	    !KBacktraceIterator_restart(kbt)) {
-			kbt->end = 1;
-			return;
-		}
-
-	kbt->end = !KBacktraceIterator_next_item_inclusive(kbt);
+	if (!backtrace_next(&kbt->it) && !KBacktraceIterator_restart(kbt)) {
+		kbt->end = KBT_DONE;
+		return;
+	}
+	kbt->end = KBacktraceIterator_next_item_inclusive(kbt);
+	if (old_pc == kbt->it.pc && old_sp == kbt->it.sp) {
+		/* Trapped in a loop; give up. */
+		kbt->end = KBT_LOOP;
+	}
 }
 EXPORT_SYMBOL(KBacktraceIterator_next);
 
@@ -387,6 +394,8 @@ void tile_show_stack(struct KBacktraceIterator *kbt, int headers)
 			break;
 		}
 	}
+	if (kbt->end == KBT_LOOP)
+		pr_err("Stack dump stopped; next frame identical to this one\n");
 	if (headers)
 		pr_err("Stack dump complete\n");
 }
diff --git a/arch/tile/kernel/sys.c b/arch/tile/kernel/sys.c
index f0f87eab8c3..e2187d24a9b 100644
--- a/arch/tile/kernel/sys.c
+++ b/arch/tile/kernel/sys.c
@@ -20,7 +20,6 @@
 #include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
-#include <linux/smp_lock.h>
 #include <linux/syscalls.h>
 #include <linux/mman.h>
 #include <linux/file.h>
@@ -110,6 +109,15 @@ SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
 #define sys_sync_file_range sys_sync_file_range2
 #endif
 
+/* Call the trampolines to manage pt_regs where necessary. */
+#define sys_execve _sys_execve
+#define sys_sigaltstack _sys_sigaltstack
+#define sys_rt_sigreturn _sys_rt_sigreturn
+#define sys_clone _sys_clone
+#ifndef __tilegx__
+#define sys_cmpxchg_badaddr _sys_cmpxchg_badaddr
+#endif
+
 /*
  * Note that we can't include <linux/unistd.h> here since the header
  * guard will defeat us; <asm/unistd.h> checks for __SYSCALL as well.
diff --git a/arch/tile/kernel/time.c b/arch/tile/kernel/time.c
index 6bed820e142..f2e156e4469 100644
--- a/arch/tile/kernel/time.c
+++ b/arch/tile/kernel/time.c
@@ -132,7 +132,7 @@ static int tile_timer_set_next_event(unsigned long ticks,
 {
 	BUG_ON(ticks > MAX_TICK);
 	__insn_mtspr(SPR_TILE_TIMER_CONTROL, ticks);
-	raw_local_irq_unmask_now(INT_TILE_TIMER);
+	arch_local_irq_unmask_now(INT_TILE_TIMER);
 	return 0;
 }
 
@@ -143,7 +143,7 @@ static int tile_timer_set_next_event(unsigned long ticks,
 static void tile_timer_set_mode(enum clock_event_mode mode,
 				struct clock_event_device *evt)
 {
-	raw_local_irq_mask_now(INT_TILE_TIMER);
+	arch_local_irq_mask_now(INT_TILE_TIMER);
 }
 
 /*
@@ -172,7 +172,7 @@ void __cpuinit setup_tile_timer(void)
 	evt->cpumask = cpumask_of(smp_processor_id());
 
 	/* Start out with timer not firing. */
-	raw_local_irq_mask_now(INT_TILE_TIMER);
+	arch_local_irq_mask_now(INT_TILE_TIMER);
 
 	/* Register tile timer. */
 	clockevents_register_device(evt);
@@ -188,7 +188,7 @@ void do_timer_interrupt(struct pt_regs *regs, int fault_num)
 	 * Mask the timer interrupt here, since we are a oneshot timer
 	 * and there are now by definition no events pending.
 	 */
-	raw_local_irq_mask(INT_TILE_TIMER);
+	arch_local_irq_mask(INT_TILE_TIMER);
 
 	/* Track time spent here in an interrupt context */
 	irq_enter();
diff --git a/arch/tile/kernel/traps.c b/arch/tile/kernel/traps.c
index 0f362dc2c57..5474fc2e77e 100644
--- a/arch/tile/kernel/traps.c
+++ b/arch/tile/kernel/traps.c
@@ -260,7 +260,7 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
 		address = regs->pc;
 		break;
 	case INT_UNALIGN_DATA:
-#ifndef __tilegx__  /* FIXME: GX: no single-step yet */
+#ifndef __tilegx__  /* Emulated support for single step debugging */
 		if (unaligned_fixup >= 0) {
 			struct single_step_state *state =
 				current_thread_info()->step_state;
@@ -278,7 +278,7 @@ void __kprobes do_trap(struct pt_regs *regs, int fault_num,
 	case INT_DOUBLE_FAULT:
 		/*
 		 * For double fault, "reason" is actually passed as
-		 * SYSTEM_SAVE_1_2, the hypervisor's double-fault info, so
+		 * SYSTEM_SAVE_K_2, the hypervisor's double-fault info, so
 		 * we can provide the original fault number rather than
 		 * the uninteresting "INT_DOUBLE_FAULT" so the user can
 		 * learn what actually struck while PL0 ICS was set.