[PATCH] xtensa: Architecture support for Tensilica Xtensa Part 3

The attached patches provides part 3 of an architecture implementation for the
Tensilica Xtensa CPU series.

Signed-off-by: Chris Zankel <chris@zankel.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

1 files changed, 1996 insertions, 0 deletions

diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
new file mode 100644
index 000000000000..c64a01f71de6
--- /dev/null
+++ b/arch/xtensa/kernel/entry.S
@@ -0,0 +1,1996 @@
+/*
+ * arch/xtensa/kernel/entry.S
+ *
+ * Low-level exception handling
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2004-2005 by Tensilica Inc.
+ *
+ * Chris Zankel <chris@zankel.net>
+ *
+ */
+
+#include <linux/linkage.h>
+#include <asm/offsets.h>
+#include <asm/processor.h>
+#include <asm/thread_info.h>
+#include <asm/uaccess.h>
+#include <asm/unistd.h>
+#include <asm/ptrace.h>
+#include <asm/current.h>
+#include <asm/pgtable.h>
+#include <asm/page.h>
+#include <asm/signal.h>
+#include <xtensa/coreasm.h>
+
+/* Unimplemented features. */
+
+#undef SIGNAL_HANDLING_IN_DOUBLE_EXCEPTION
+#undef KERNEL_STACK_OVERFLOW_CHECK
+#undef PREEMPTIBLE_KERNEL
+#undef ALLOCA_EXCEPTION_IN_IRAM
+
+/* Not well tested.
+ *
+ * - fast_coprocessor
+ */
+
+/*
+ * Macro to find first bit set in WINDOWBASE from the left + 1
+ *
+ * 100....0 -> 1
+ * 010....0 -> 2
+ * 000....1 -> WSBITS
+ */
+
+	.macro ffs_ws bit mask
+
+#if XCHAL_HAVE_NSA
+	nsau    \bit, \mask			# 32-WSBITS ... 31 (32 iff 0)
+	addi    \bit, \bit, WSBITS - 32 + 1   	# uppest bit set -> return 1
+#else
+	movi    \bit, WSBITS
+#if WSBITS > 16
+	_bltui  \mask, 0x10000, 99f
+	addi    \bit, \bit, -16
+	extui   \mask, \mask, 16, 16
+#endif
+#if WSBITS > 8
+99:	_bltui  \mask, 0x100, 99f
+	addi    \bit, \bit, -8
+	srli    \mask, \mask, 8
+#endif
+99:	_bltui  \mask, 0x10, 99f
+	addi    \bit, \bit, -4
+	srli    \mask, \mask, 4
+99:	_bltui  \mask, 0x4, 99f
+	addi    \bit, \bit, -2
+	srli    \mask, \mask, 2
+99:	_bltui  \mask, 0x2, 99f
+	addi    \bit, \bit, -1
+99:
+
+#endif
+	.endm
+
+/* ----------------- DEFAULT FIRST LEVEL EXCEPTION HANDLERS ----------------- */
+
+/*
+ * First-level exception handler for user exceptions.
+ * Save some special registers, extra states and all registers in the AR
+ * register file that were in use in the user task, and jump to the common
+ * exception code.
+ * We save SAR (used to calculate WMASK), and WB and WS (we don't have to
+ * save them for kernel exceptions).
+ *
+ * Entry condition for user_exception:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original value in depc
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ *
+ * Entry condition for _user_exception:
+ *
+ *   a0-a3 and depc have been saved to PT_AREG0...PT_AREG3 and PT_DEPC
+ *   excsave has been restored, and
+ *   stack pointer (a1) has been set.
+ *
+ * Note: _user_exception might be at an odd adress. Don't use call0..call12
+ */
+
+ENTRY(user_exception)
+
+	/* Save a2, a3, and depc, restore excsave_1 and set SP. */
+
+	xsr	a3, EXCSAVE_1
+	rsr	a0, DEPC
+	s32i	a1, a2, PT_AREG1
+	s32i	a0, a2, PT_AREG2
+	s32i	a3, a2, PT_AREG3
+	mov	a1, a2
+
+	.globl _user_exception
+_user_exception:
+
+	/* Save SAR and turn off single stepping */
+
+	movi	a2, 0
+	rsr	a3, SAR
+	wsr	a2, ICOUNTLEVEL
+	s32i	a3, a1, PT_SAR
+
+	/* Rotate ws so that the current windowbase is at bit0. */
+	/* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */
+
+	rsr	a2, WINDOWBASE
+	rsr	a3, WINDOWSTART
+	ssr	a2
+	s32i	a2, a1, PT_WINDOWBASE
+	s32i	a3, a1, PT_WINDOWSTART
+	slli	a2, a3, 32-WSBITS
+	src	a2, a3, a2
+	srli	a2, a2, 32-WSBITS
+	s32i	a2, a1, PT_WMASK	# needed for restoring registers
+
+	/* Save only live registers. */
+
+	_bbsi.l	a2, 1, 1f
+	s32i	a4, a1, PT_AREG4
+	s32i	a5, a1, PT_AREG5
+	s32i	a6, a1, PT_AREG6
+	s32i	a7, a1, PT_AREG7
+	_bbsi.l	a2, 2, 1f
+	s32i	a8, a1, PT_AREG8
+	s32i	a9, a1, PT_AREG9
+	s32i	a10, a1, PT_AREG10
+	s32i	a11, a1, PT_AREG11
+	_bbsi.l	a2, 3, 1f
+	s32i	a12, a1, PT_AREG12
+	s32i	a13, a1, PT_AREG13
+	s32i	a14, a1, PT_AREG14
+	s32i	a15, a1, PT_AREG15
+	_bnei	a2, 1, 1f		# only one valid frame?
+
+	/* Only one valid frame, skip saving regs. */
+
+	j	2f
+
+	/* Save the remaining registers.
+	 * We have to save all registers up to the first '1' from
+	 * the right, except the current frame (bit 0).
+	 * Assume a2 is:  001001000110001
+	 * All regiser frames starting from the top fiel to the marked '1'
+	 * must be saved.
+	 */
+
+1:	addi	a3, a2, -1		# eliminate '1' in bit 0: yyyyxxww0
+	neg	a3, a3			# yyyyxxww0 -> YYYYXXWW1+1
+	and	a3, a3, a2		# max. only one bit is set
+
+	/* Find number of frames to save */
+
+	ffs_ws	a0, a3			# number of frames to the '1' from left
+
+	/* Store information into WMASK:
+	 * bits 0..3: xxx1 masked lower 4 bits of the rotated windowstart,
+	 * bits 4...: number of valid 4-register frames
+	 */
+
+	slli	a3, a0, 4		# number of frames to save in bits 8..4
+	extui	a2, a2, 0, 4		# mask for the first 16 registers
+	or	a2, a3, a2
+	s32i	a2, a1, PT_WMASK	# needed when we restore the reg-file
+
+	/* Save 4 registers at a time */
+
+1:	rotw	-1
+	s32i	a0, a5, PT_AREG_END - 16
+	s32i	a1, a5, PT_AREG_END - 12
+	s32i	a2, a5, PT_AREG_END - 8
+	s32i	a3, a5, PT_AREG_END - 4
+	addi	a0, a4, -1
+	addi	a1, a5, -16
+	_bnez	a0, 1b
+
+	/* WINDOWBASE still in SAR! */
+
+	rsr	a2, SAR			# original WINDOWBASE
+	movi	a3, 1
+	ssl	a2
+	sll	a3, a3
+	wsr	a3, WINDOWSTART		# set corresponding WINDOWSTART bit
+	wsr	a2, WINDOWBASE		# and WINDOWSTART
+	rsync
+
+	/* We are back to the original stack pointer (a1) */
+
+2:
+#if XCHAL_EXTRA_SA_SIZE
+
+	/* For user exceptions, save the extra state into the user's TCB.
+	 * Note: We must assume that xchal_extra_store_funcbody destroys a2..a15
+	 */
+
+	GET_CURRENT(a2,a1)
+	addi	a2, a2, THREAD_CP_SAVE
+	xchal_extra_store_funcbody
+#endif
+
+	/* Now, jump to the common exception handler. */
+
+	j	common_exception
+
+
+/*
+ * First-level exit handler for kernel exceptions
+ * Save special registers and the live window frame.
+ * Note: Even though we changes the stack pointer, we don't have to do a
+ *	 MOVSP here, as we do that when we return from the exception.
+ *	 (See comment in the kernel exception exit code)
+ *
+ * Entry condition for kernel_exception:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ *
+ * Entry condition for _kernel_exception:
+ *
+ *   a0-a3 and depc have been saved to PT_AREG0...PT_AREG3 and PT_DEPC
+ *   excsave has been restored, and
+ *   stack pointer (a1) has been set.
+ *
+ * Note: _kernel_exception might be at an odd adress. Don't use call0..call12
+ */
+
+ENTRY(kernel_exception)
+
+	/* Save a0, a2, a3, DEPC and set SP. */
+
+	xsr	a3, EXCSAVE_1		# restore a3, excsave_1
+	rsr	a0, DEPC		# get a2
+	s32i	a1, a2, PT_AREG1
+	s32i	a0, a2, PT_AREG2
+	s32i	a3, a2, PT_AREG3
+	mov	a1, a2
+
+	.globl _kernel_exception
+_kernel_exception:
+
+	/* Save SAR and turn off single stepping */
+
+	movi	a2, 0
+	rsr	a3, SAR
+	wsr	a2, ICOUNTLEVEL
+	s32i	a3, a1, PT_SAR
+
+	/* Rotate ws so that the current windowbase is at bit0. */
+	/* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */
+
+	rsr	a2, WINDOWBASE		# don't need to save these, we only
+	rsr	a3, WINDOWSTART		# need shifted windowstart: windowmask
+	ssr	a2
+	slli	a2, a3, 32-WSBITS
+	src	a2, a3, a2
+	srli	a2, a2, 32-WSBITS
+	s32i	a2, a1, PT_WMASK	# needed for kernel_exception_exit
+
+	/* Save only the live window-frame */
+
+	_bbsi.l	a2, 1, 1f
+	s32i	a4, a1, PT_AREG4
+	s32i	a5, a1, PT_AREG5
+	s32i	a6, a1, PT_AREG6
+	s32i	a7, a1, PT_AREG7
+	_bbsi.l	a2, 2, 1f
+	s32i	a8, a1, PT_AREG8
+	s32i	a9, a1, PT_AREG9
+	s32i	a10, a1, PT_AREG10
+	s32i	a11, a1, PT_AREG11
+	_bbsi.l	a2, 3, 1f
+	s32i	a12, a1, PT_AREG12
+	s32i	a13, a1, PT_AREG13
+	s32i	a14, a1, PT_AREG14
+	s32i	a15, a1, PT_AREG15
+
+1:
+
+#ifdef KERNEL_STACK_OVERFLOW_CHECK
+
+	/*  Stack overflow check, for debugging  */
+	extui	a2, a1, TASK_SIZE_BITS,XX
+	movi	a3, SIZE??
+	_bge	a2, a3, out_of_stack_panic
+
+#endif
+
+/*
+ * This is the common exception handler.
+ * We get here from the user exception handler or simply by falling through
+ * from the kernel exception handler.
+ * Save the remaining special registers, switch to kernel mode, and jump
+ * to the second-level exception handler.
+ *
+ */
+
+common_exception:
+
+	/* Save EXCVADDR, DEBUGCAUSE, and PC, and clear LCOUNT */
+
+	rsr	a2, DEBUGCAUSE
+	rsr	a3, EPC_1
+	s32i	a2, a1, PT_DEBUGCAUSE
+	s32i	a3, a1, PT_PC
+
+	rsr	a3, EXCVADDR
+	movi	a2, 0
+	s32i	a3, a1, PT_EXCVADDR
+	xsr	a2, LCOUNT
+	s32i	a2, a1, PT_LCOUNT
+
+	/* It is now save to restore the EXC_TABLE_FIXUP variable. */
+
+	rsr	a0, EXCCAUSE
+	movi	a3, 0
+	rsr	a2, EXCSAVE_1
+	s32i	a0, a1, PT_EXCCAUSE
+	s32i	a3, a2, EXC_TABLE_FIXUP
+
+	/* All unrecoverable states are saved on stack, now, and a1 is valid,
+	 * so we can allow exceptions and interrupts (*) again.
+	 * Set PS(EXCM = 0, UM = 0, RING = 0, OWB = 0, WOE = 1, INTLEVEL = X)
+	 *
+	 * (*) We only allow interrupts if PS.INTLEVEL was not set to 1 before
+	 *     (interrupts disabled) and if this exception is not an interrupt.
+	 */
+
+	rsr	a3, PS
+	addi	a0, a0, -4
+	movi	a2, 1
+	extui	a3, a3, 0, 1		# a3 = PS.INTLEVEL[0]
+	moveqz	a3, a2, a0		# a3 = 1 iff interrupt exception
+	movi	a2, PS_WOE_MASK
+	or	a3, a3, a2
+	rsr	a0, EXCCAUSE
+	xsr	a3, PS
+
+	s32i	a3, a1, PT_PS		# save ps
+
+	/* Save LBEG, LEND */
+
+	rsr	a2, LBEG
+	rsr	a3, LEND
+	s32i	a2, a1, PT_LBEG
+	s32i	a3, a1, PT_LEND
+
+	/* Go to second-level dispatcher. Set up parameters to pass to the
+	 * exception handler and call the exception handler.
+	 */
+
+	movi	a4, exc_table
+	mov	a6, a1			# pass stack frame
+	mov	a7, a0			# pass EXCCAUSE
+	addx4	a4, a0, a4
+	l32i	a4, a4, EXC_TABLE_DEFAULT		# load handler
+
+	/* Call the second-level handler */
+
+	callx4	a4
+
+	/* Jump here for exception exit */
+
+common_exception_return:
+
+	/* Jump if we are returning from kernel exceptions. */
+
+1:	l32i	a3, a1, PT_PS
+	_bbsi.l	a3, PS_UM_SHIFT, 2f
+	j	kernel_exception_exit
+
+	/* Specific to a user exception exit:
+	 * We need to check some flags for signal handling and rescheduling,
+	 * and have to restore WB and WS, extra states, and all registers
+	 * in the register file that were in use in the user task.
+	 */
+
+2:	wsr	a3, PS		/* disable interrupts */
+
+	/* Check for signals (keep interrupts disabled while we read TI_FLAGS)
+	 * Note: PS.INTLEVEL = 0, PS.EXCM = 1
+	 */
+
+	GET_THREAD_INFO(a2,a1)
+	l32i	a4, a2, TI_FLAGS
+
+	/* Enable interrupts again.
+	 * Note: When we get here, we certainly have handled any interrupts.
+	 *       (Hint: There is only one user exception frame on stack)
+	 */
+
+	movi	a3, PS_WOE_MASK
+
+	_bbsi.l	a4, TIF_NEED_RESCHED, 3f
+	_bbci.l	a4, TIF_SIGPENDING, 4f
+
+#ifndef SIGNAL_HANDLING_IN_DOUBLE_EXCEPTION
+	l32i	a4, a1, PT_DEPC
+	bgeui	a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 4f
+#endif
+
+	/* Reenable interrupts and call do_signal() */
+
+	wsr	a3, PS
+	movi	a4, do_signal	# int do_signal(struct pt_regs*, sigset_t*)
+	mov	a6, a1
+	movi	a7, 0
+	callx4	a4
+	j	1b
+
+3:	/* Reenable interrupts and reschedule */
+
+	wsr	a3, PS
+	movi	a4, schedule	# void schedule (void)
+	callx4	a4
+	j	1b
+
+	/* Restore the state of the task and return from the exception. */
+
+
+	/* If we are returning from a user exception, and the process
+	 * to run next has PT_SINGLESTEP set, we want to setup
+	 * ICOUNT and ICOUNTLEVEL to step one instruction.
+	 * PT_SINGLESTEP is set by sys_ptrace (ptrace.c)
+	 */
+
+4:	/* a2 holds GET_CURRENT(a2,a1)  */
+
+	l32i	a3, a2, TI_TASK
+	l32i	a3, a3, TASK_PTRACE
+	bbci.l	a3, PT_SINGLESTEP_BIT, 1f # jump if single-step flag is not set
+
+	movi	a3, -2			# PT_SINGLESTEP flag is set,
+	movi	a4, 1			# icountlevel of 1 means it won't
+	wsr	a3, ICOUNT		# start counting until after rfe
+	wsr	a4, ICOUNTLEVEL		# so setup icount & icountlevel.
+	isync
+
+1:
+
+#if XCHAL_EXTRA_SA_SIZE
+
+	/* For user exceptions, restore the extra state from the user's TCB. */
+
+	/* Note: a2 still contains GET_CURRENT(a2,a1) */
+	addi	a2, a2, THREAD_CP_SAVE
+	xchal_extra_load_funcbody
+
+	/* We must assume that xchal_extra_store_funcbody destroys
+	 * registers a2..a15.  FIXME, this list can eventually be
+	 * reduced once real register requirements of the macro are
+	 * finalized. */
+
+#endif /* XCHAL_EXTRA_SA_SIZE */
+
+
+	/* Switch to the user thread WINDOWBASE. Save SP temporarily in DEPC */
+
+	l32i	a2, a1, PT_WINDOWBASE
+	l32i	a3, a1, PT_WINDOWSTART
+	wsr	a1, DEPC		# use DEPC as temp storage
+	wsr	a3, WINDOWSTART		# restore WINDOWSTART
+	ssr	a2			# preserve user's WB in the SAR
+	wsr	a2, WINDOWBASE		# switch to user's saved WB
+	rsync
+	rsr	a1, DEPC		# restore stack pointer
+	l32i	a2, a1, PT_WMASK	# register frames saved (in bits 4...9)
+	rotw	-1			# we restore a4..a7
+	_bltui	a6, 16, 1f		# only have to restore current window?
+
+	/* The working registers are a0 and a3.  We are restoring to
+	 * a4..a7.  Be careful not to destroy what we have just restored.
+	 * Note: wmask has the format YYYYM:
+	 *       Y: number of registers saved in groups of 4
+	 *       M: 4 bit mask of first 16 registers
+	 */
+
+	mov	a2, a6
+	mov	a3, a5
+
+2:	rotw	-1			# a0..a3 become a4..a7
+	addi	a3, a7, -4*4		# next iteration
+	addi	a2, a6, -16		# decrementing Y in WMASK
+	l32i	a4, a3, PT_AREG_END + 0
+	l32i	a5, a3, PT_AREG_END + 4
+	l32i	a6, a3, PT_AREG_END + 8
+	l32i	a7, a3, PT_AREG_END + 12
+	_bgeui	a2, 16, 2b
+
+	/* Clear unrestored registers (don't leak anything to user-land */
+
+1:	rsr	a0, WINDOWBASE
+	rsr	a3, SAR
+	sub	a3, a0, a3
+	beqz	a3, 2f
+	extui	a3, a3, 0, WBBITS
+
+1:	rotw	-1
+	addi	a3, a7, -1
+	movi	a4, 0
+	movi	a5, 0
+	movi	a6, 0
+	movi	a7, 0
+	bgei	a3, 1, 1b
+
+	/* We are back were we were when we started.
+	 * Note: a2 still contains WMASK (if we've returned to the original
+	 *	 frame where we had loaded a2), or at least the lower 4 bits
+	 *	 (if we have restored WSBITS-1 frames).
+	 */
+
+2:	j	common_exception_exit
+
+	/* This is the kernel exception exit.
+	 * We avoided to do a MOVSP when we entered the exception, but we
+	 * have to do it here.
+	 */
+
+kernel_exception_exit:
+
+	/* Disable interrupts (a3 holds PT_PS) */
+
+	wsr	a3, PS
+
+#ifdef PREEMPTIBLE_KERNEL
+
+#ifdef CONFIG_PREEMPT
+
+	/*
+	 * Note: We've just returned from a call4, so we have
+	 * at least 4 addt'l regs.
+	 */
+
+	/* Check current_thread_info->preempt_count */
+
+	GET_THREAD_INFO(a2)
+	l32i	a3, a2, TI_PREEMPT
+	bnez	a3, 1f
+
+	l32i	a2, a2, TI_FLAGS
+
+1:
+
+#endif
+
+#endif
+
+	/* Check if we have to do a movsp.
+	 *
+	 * We only have to do a movsp if the previous window-frame has
+	 * been spilled to the *temporary* exception stack instead of the
+	 * task's stack. This is the case if the corresponding bit in
+	 * WINDOWSTART for the previous window-frame was set before
+	 * (not spilled) but is zero now (spilled).
+	 * If this bit is zero, all other bits except the one for the
+	 * current window frame are also zero. So, we can use a simple test:
+	 * 'and' WINDOWSTART and WINDOWSTART-1:
+	 *
+	 *  (XXXXXX1[0]* - 1) AND XXXXXX1[0]* = XXXXXX0[0]*
+	 *
+	 * The result is zero only if one bit was set.
+	 *
+	 * (Note: We might have gone through several task switches before
+	 *        we come back to the current task, so WINDOWBASE might be
+	 *        different from the time the exception occurred.)
+	 */
+
+	/* Test WINDOWSTART before and after the exception.
+	 * We actually have WMASK, so we only have to test if it is 1 or not.
+	 */
+
+	l32i	a2, a1, PT_WMASK
+	_beqi	a2, 1, common_exception_exit	# Spilled before exception,jump
+
+	/* Test WINDOWSTART now. If spilled, do the movsp */
+
+	rsr     a3, WINDOWSTART
+	addi	a0, a3, -1
+	and     a3, a3, a0
+	_bnez	a3, common_exception_exit
+
+	/* Do a movsp (we returned from a call4, so we have at least a0..a7) */
+
+	addi    a0, a1, -16
+	l32i    a3, a0, 0
+	l32i    a4, a0, 4
+	s32i    a3, a1, PT_SIZE+0
+	s32i    a4, a1, PT_SIZE+4
+	l32i    a3, a0, 8
+	l32i    a4, a0, 12
+	s32i    a3, a1, PT_SIZE+8
+	s32i    a4, a1, PT_SIZE+12
+
+	/* Common exception exit.
+	 * We restore the special register and the current window frame, and
+	 * return from the exception.
+	 *
+	 * Note: We expect a2 to hold PT_WMASK
+	 */
+
+common_exception_exit:
+
+	_bbsi.l	a2, 1, 1f
+	l32i	a4,  a1, PT_AREG4
+	l32i	a5,  a1, PT_AREG5
+	l32i	a6,  a1, PT_AREG6
+	l32i	a7,  a1, PT_AREG7
+	_bbsi.l	a2, 2, 1f
+	l32i	a8,  a1, PT_AREG8
+	l32i	a9,  a1, PT_AREG9
+	l32i	a10, a1, PT_AREG10
+	l32i	a11, a1, PT_AREG11
+	_bbsi.l	a2, 3, 1f
+	l32i	a12, a1, PT_AREG12
+	l32i	a13, a1, PT_AREG13
+	l32i	a14, a1, PT_AREG14
+	l32i	a15, a1, PT_AREG15
+
+	/* Restore PC, SAR */
+
+1:	l32i	a2, a1, PT_PC
+	l32i	a3, a1, PT_SAR
+	wsr	a2, EPC_1
+	wsr	a3, SAR
+
+	/* Restore LBEG, LEND, LCOUNT */
+
+	l32i	a2, a1, PT_LBEG
+	l32i	a3, a1, PT_LEND
+	wsr	a2, LBEG
+	l32i	a2, a1, PT_LCOUNT
+	wsr	a3, LEND
+	wsr	a2, LCOUNT
+
+	/* Check if it was double exception. */
+
+	l32i	a0, a1, PT_DEPC
+	l32i	a3, a1, PT_AREG3
+	l32i	a2, a1, PT_AREG2
+	_bgeui	a0, VALID_DOUBLE_EXCEPTION_ADDRESS, 1f
+
+	/* Restore a0...a3 and return */
+
+	l32i	a0, a1, PT_AREG0
+	l32i	a1, a1, PT_AREG1
+	rfe
+
+1:	wsr	a0, DEPC
+	l32i	a0, a1, PT_AREG0
+	l32i	a1, a1, PT_AREG1
+	rfde
+
+/*
+ * Debug exception handler.
+ *
+ * Currently, we don't support KGDB, so only user application can be debugged.
+ *
+ * When we get here,  a0 is trashed and saved to excsave[debuglevel]
+ */
+
+ENTRY(debug_exception)
+
+	rsr	a0, EPS + XCHAL_DEBUGLEVEL
+	bbsi.l	a0, PS_EXCM_SHIFT, 1f	# exception mode
+
+	/* Set EPC_1 and EXCCAUSE */
+
+	wsr	a2, DEPC		# save a2 temporarily
+	rsr	a2, EPC + XCHAL_DEBUGLEVEL
+	wsr	a2, EPC_1
+
+	movi	a2, EXCCAUSE_MAPPED_DEBUG
+	wsr	a2, EXCCAUSE
+
+	/* Restore PS to the value before the debug exc but with PS.EXCM set.*/
+
+	movi	a2, 1 << PS_EXCM_SHIFT
+	or	a2, a0, a2
+	movi	a0, debug_exception	# restore a3, debug jump vector
+	wsr	a2, PS
+	xsr	a0, EXCSAVE + XCHAL_DEBUGLEVEL
+
+	/* Switch to kernel/user stack, restore jump vector, and save a0 */
+
+	bbsi.l	a2, PS_UM_SHIFT, 2f	# jump if user mode
+
+	addi	a2, a1, -16-PT_SIZE	# assume kernel stack
+	s32i	a0, a2, PT_AREG0
+	movi	a0, 0
+	s32i	a1, a2, PT_AREG1
+	s32i	a0, a2, PT_DEPC		# mark it as a regular exception
+	xsr	a0, DEPC
+	s32i	a3, a2, PT_AREG3
+	s32i	a0, a2, PT_AREG2
+	mov	a1, a2
+	j	_kernel_exception
+
+2:	rsr	a2, EXCSAVE_1
+	l32i	a2, a2, EXC_TABLE_KSTK	# load kernel stack pointer
+	s32i	a0, a2, PT_AREG0
+	movi	a0, 0
+	s32i	a1, a2, PT_AREG1
+	s32i	a0, a2, PT_DEPC
+	xsr	a0, DEPC
+	s32i	a3, a2, PT_AREG3
+	s32i	a0, a2, PT_AREG2
+	mov	a1, a2
+	j	_user_exception
+
+	/* Debug exception while in exception mode. */
+1:	j	1b	// FIXME!!
+
+
+/*
+ * We get here in case of an unrecoverable exception.
+ * The only thing we can do is to be nice and print a panic message.
+ * We only produce a single stack frame for panic, so ???
+ *
+ *
+ * Entry conditions:
+ *
+ *   - a0 contains the caller address; original value saved in excsave1.
+ *   - the original a0 contains a valid return address (backtrace) or 0.
+ *   - a2 contains a valid stackpointer
+ *
+ * Notes:
+ *
+ *   - If the stack pointer could be invalid, the caller has to setup a
+ *     dummy stack pointer (e.g. the stack of the init_task)
+ *
+ *   - If the return address could be invalid, the caller has to set it
+ *     to 0, so the backtrace would stop.
+ *
+ */
+	.align 4
+unrecoverable_text:
+	.ascii "Unrecoverable error in exception handler\0"
+
+ENTRY(unrecoverable_exception)
+
+	movi	a0, 1
+	movi	a1, 0
+
+	wsr	a0, WINDOWSTART
+	wsr	a1, WINDOWBASE
+	rsync
+
+	movi	a1, PS_WOE_MASK | 1
+	wsr	a1, PS
+	rsync
+
+	movi	a1, init_task
+	movi	a0, 0
+	addi	a1, a1, PT_REGS_OFFSET
+
+	movi	a4, panic
+	movi	a6, unrecoverable_text
+
+	callx4	a4
+
+1:	j	1b
+
+
+/* -------------------------- FAST EXCEPTION HANDLERS ----------------------- */
+
+/*
+ * Fast-handler for alloca exceptions
+ *
+ *  The ALLOCA handler is entered when user code executes the MOVSP
+ *  instruction and the caller's frame is not in the register file.
+ *  In this case, the caller frame's a0..a3 are on the stack just
+ *  below sp (a1), and this handler moves them.
+ *
+ *  For "MOVSP <ar>,<as>" without destination register a1, this routine
+ *  simply moves the value from <as> to <ar> without moving the save area.
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ */
+
+#if XCHAL_HAVE_BE
+#define _EXTUI_MOVSP_SRC(ar)	extui ar, ar, 4, 4
+#define _EXTUI_MOVSP_DST(ar)	extui ar, ar, 0, 4
+#else
+#define _EXTUI_MOVSP_SRC(ar)	extui ar, ar, 0, 4
+#define _EXTUI_MOVSP_DST(ar)	extui ar, ar, 4, 4
+#endif
+
+ENTRY(fast_alloca)
+
+	/* We shouldn't be in a double exception. */
+
+	l32i	a0, a2, PT_DEPC
+	_bgeui	a0, VALID_DOUBLE_EXCEPTION_ADDRESS, .Lunhandled_double
+
+	rsr	a0, DEPC		# get a2
+	s32i	a4, a2, PT_AREG4	# save a4 and
+	s32i	a0, a2, PT_AREG2	# a2 to stack
+
+	/* Exit critical section. */
+
+	movi	a0, 0
+	s32i	a0, a3, EXC_TABLE_FIXUP
+
+	/* Restore a3, excsave_1 */
+
+	xsr	a3, EXCSAVE_1		# make sure excsave_1 is valid for dbl.
+	rsr	a4, EPC_1		# get exception address
+	s32i	a3, a2, PT_AREG3	# save a3 to stack
+
+#ifdef ALLOCA_EXCEPTION_IN_IRAM
+#error	iram not supported
+#else
+	/* Note: l8ui not allowed in IRAM/IROM!! */
+	l8ui	a0, a4, 1		# read as(src) from MOVSP instruction
+#endif
+	movi	a3, .Lmovsp_src
+	_EXTUI_MOVSP_SRC(a0)		# extract source register number
+	addx8	a3, a0, a3
+	jx	a3
+
+.Lunhandled_double:
+	wsr	a0, EXCSAVE_1
+	movi	a0, unrecoverable_exception
+	callx0	a0
+
+	.align 8
+.Lmovsp_src:
+	l32i	a3, a2, PT_AREG0;	_j 1f;	.align 8
+	mov	a3, a1;			_j 1f;	.align 8
+	l32i	a3, a2, PT_AREG2;	_j 1f;	.align 8
+	l32i	a3, a2, PT_AREG3;	_j 1f;	.align 8
+	l32i	a3, a2, PT_AREG4;	_j 1f;	.align 8
+	mov	a3, a5;			_j 1f;	.align 8
+	mov	a3, a6;			_j 1f;	.align 8
+	mov	a3, a7;			_j 1f;	.align 8
+	mov	a3, a8;			_j 1f;	.align 8
+	mov	a3, a9;			_j 1f;	.align 8
+	mov	a3, a10;		_j 1f;	.align 8
+	mov	a3, a11;		_j 1f;	.align 8
+	mov	a3, a12;		_j 1f;	.align 8
+	mov	a3, a13;		_j 1f;	.align 8
+	mov	a3, a14;		_j 1f;	.align 8
+	mov	a3, a15;		_j 1f;	.align 8
+
+1:
+
+#ifdef ALLOCA_EXCEPTION_IN_IRAM
+#error	iram not supported
+#else
+	l8ui	a0, a4, 0		# read ar(dst) from MOVSP instruction
+#endif
+	addi	a4, a4, 3		# step over movsp
+	_EXTUI_MOVSP_DST(a0)		# extract destination register
+	wsr	a4, EPC_1		# save new epc_1
+
+	_bnei	a0, 1, 1f		# no 'movsp a1, ax': jump
+
+        /* Move the save area. This implies the use of the L32E
+	 * and S32E instructions, because this move must be done with
+	 * the user's PS.RING privilege levels, not with ring 0
+	 * (kernel's) privileges currently active with PS.EXCM
+	 * set. Note that we have stil registered a fixup routine with the
+	 * double exception vector in case a double exception occurs.
+	 */
+
+	/* a0,a4:avail a1:old user stack a2:exc. stack a3:new user stack. */
+
+	l32e	a0, a1, -16
+	l32e	a4, a1, -12
+	s32e	a0, a3, -16
+	s32e	a4, a3, -12
+	l32e	a0, a1, -8
+	l32e	a4, a1, -4
+	s32e	a0, a3, -8
+	s32e	a4, a3, -4
+
+	/* Restore stack-pointer and all the other saved registers. */
+
+	mov	a1, a3
+
+	l32i	a4, a2, PT_AREG4
+	l32i	a3, a2, PT_AREG3
+	l32i	a0, a2, PT_AREG0
+	l32i	a2, a2, PT_AREG2
+	rfe
+
+	/*  MOVSP <at>,<as>  was invoked with <at> != a1.
+	 *  Because the stack pointer is not being modified,
+	 *  we should be able to just modify the pointer
+	 *  without moving any save area.
+	 *  The processor only traps these occurrences if the
+	 *  caller window isn't live, so unfortunately we can't
+	 *  use this as an alternate trap mechanism.
+	 *  So we just do the move.  This requires that we
+	 *  resolve the destination register, not just the source,
+	 *  so there's some extra work.
+	 *  (PERHAPS NOT REALLY NEEDED, BUT CLEANER...)
+	 */
+
+	/* a0 dst-reg, a1 user-stack, a2 stack, a3 value of src reg. */
+
+1:	movi	a4, .Lmovsp_dst
+	addx8	a4, a0, a4
+	jx	a4
+
+	.align 8
+.Lmovsp_dst:
+	s32i	a3, a2, PT_AREG0;	_j 1f;	.align 8
+	mov	a1, a3;			_j 1f;	.align 8
+	s32i	a3, a2, PT_AREG2;	_j 1f;	.align 8
+	s32i	a3, a2, PT_AREG3;	_j 1f;	.align 8
+	s32i	a3, a2, PT_AREG4;	_j 1f;	.align 8
+	mov	a5, a3;			_j 1f;	.align 8
+	mov	a6, a3;			_j 1f;	.align 8
+	mov	a7, a3;			_j 1f;	.align 8
+	mov	a8, a3;			_j 1f;	.align 8
+	mov	a9, a3;			_j 1f;	.align 8
+	mov	a10, a3;		_j 1f;	.align 8
+	mov	a11, a3;		_j 1f;	.align 8
+	mov	a12, a3;		_j 1f;	.align 8
+	mov	a13, a3;		_j 1f;	.align 8
+	mov	a14, a3;		_j 1f;	.align 8
+	mov	a15, a3;		_j 1f;	.align 8
+
+1:	l32i	a4, a2, PT_AREG4
+	l32i	a3, a2, PT_AREG3
+	l32i	a0, a2, PT_AREG0
+	l32i	a2, a2, PT_AREG2
+	rfe
+
+
+/*
+ * fast system calls.
+ *
+ * WARNING:  The kernel doesn't save the entire user context before
+ * handling a fast system call.  These functions are small and short,
+ * usually offering some functionality not available to user tasks.
+ *
+ * BE CAREFUL TO PRESERVE THE USER'S CONTEXT.
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ */
+
+ENTRY(fast_syscall_kernel)
+
+	/* Skip syscall. */
+
+	rsr	a0, EPC_1
+	addi	a0, a0, 3
+	wsr	a0, EPC_1
+
+	l32i	a0, a2, PT_DEPC
+	bgeui	a0, VALID_DOUBLE_EXCEPTION_ADDRESS, fast_syscall_unrecoverable
+
+	rsr	a0, DEPC			# get syscall-nr
+	_beqz	a0, fast_syscall_spill_registers
+
+	addi	a0, a0, -__NR_sysxtensa
+	_beqz	a0, fast_syscall_sysxtensa
+
+	j	kernel_exception
+
+
+ENTRY(fast_syscall_user)
+
+	/* Skip syscall. */
+
+	rsr	a0, EPC_1
+	addi	a0, a0, 3
+	wsr	a0, EPC_1
+
+	l32i	a0, a2, PT_DEPC
+	bgeui	a0, VALID_DOUBLE_EXCEPTION_ADDRESS, fast_syscall_unrecoverable
+
+	rsr	a0, DEPC			# get syscall-nr
+	_beqz	a0, fast_syscall_spill_registers
+
+	addi	a0, a0, -__NR_sysxtensa
+	_beqz	a0, fast_syscall_sysxtensa
+
+	j	user_exception
+
+ENTRY(fast_syscall_unrecoverable)
+
+        /* Restore all states. */
+
+        l32i    a0, a2, PT_AREG0        # restore a0
+        xsr     a2, DEPC                # restore a2, depc
+        rsr     a3, EXCSAVE_1
+
+        wsr     a0, EXCSAVE_1
+        movi    a0, unrecoverable_exception
+        callx0  a0
+
+
+
+/*
+ * sysxtensa syscall handler
+ *
+ * int sysxtensa (XTENSA_ATOMIC_SET, ptr, val, unused);
+ * int sysxtensa (XTENSA_ATOMIC_ADD, ptr, val, unused);
+ * int sysxtensa (XTENSA_ATOMIC_EXG_ADD, ptr, val, unused);
+ * int sysxtensa (XTENSA_ATOMIC_CMP_SWP, ptr, oldval, newval);
+ * a2                    a6              a3    a4      a5
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ *
+ * Note: we don't have to save a2; a2 holds the return value
+ *
+ * We use the two macros TRY and CATCH:
+ *
+ * TRY	 adds an entry to the __ex_table fixup table for the immediately
+ *	 following instruction.
+ *
+ * CATCH catches any exception that occurred at one of the preceeding TRY
+ *       statements and continues from there
+ *
+ * Usage TRY	l32i	a0, a1, 0
+ *		<other code>
+ *	 done:	rfe
+ *	 CATCH	<set return code>
+ *		j done
+ */
+
+#define TRY								\
+	.section __ex_table, "a";					\
+	.word	66f, 67f;						\
+	.text;								\
+66:
+
+#define CATCH								\
+67:
+
+ENTRY(fast_syscall_sysxtensa)
+
+	_beqz	a6, 1f
+	_blti	a6, SYSXTENSA_COUNT, 2f
+
+1:	j	user_exception
+
+2:	xsr	a3, EXCSAVE_1		# restore a3, excsave1
+	s32i	a7, a2, PT_AREG7
+
+	movi	a7, 4			# sizeof(unsigned int)
+	verify_area a3, a7, a0, a2, .Leac
+
+	_beqi	a6, SYSXTENSA_ATOMIC_SET, .Lset
+	_beqi	a6, SYSXTENSA_ATOMIC_EXG_ADD, .Lexg
+	_beqi	a6, SYSXTENSA_ATOMIC_ADD, .Ladd
+
+	/* Fall through for SYSXTENSA_ATOMIC_CMP_SWP */
+
+.Lswp:	/* Atomic compare and swap */
+
+TRY	l32i	a7, a3, 0		# read old value
+	bne	a7, a4, 1f		# same as old value? jump
+	s32i	a5, a3, 0		# different, modify value
+	movi	a7, 1			# and return 1
+	j	.Lret
+
+1:	movi	a7, 0			# same values: return 0
+	j	.Lret
+
+.Ladd:	/* Atomic add */
+.Lexg:	/* Atomic (exchange) add */
+
+TRY	l32i	a7, a3, 0
+	add	a4, a4, a7
+	s32i	a4, a3, 0
+	j	.Lret
+
+.Lset:	/* Atomic set */
+
+TRY	l32i	a7, a3, 0		# read old value as return value
+	s32i	a4, a3, 0		# write new value
+
+.Lret:	mov	a0, a2
+	mov	a2, a7
+	l32i	a7, a0, PT_AREG7
+	l32i	a3, a0, PT_AREG3
+	l32i	a0, a0, PT_AREG0
+	rfe
+
+CATCH
+.Leac:	movi	a7, -EFAULT
+	j	.Lret
+
+
+
+/* fast_syscall_spill_registers.
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ * Note: We assume the stack pointer is EXC_TABLE_KSTK in the fixup handler.
+ * Note: We don't need to save a2 in depc (return value)
+ */
+
+ENTRY(fast_syscall_spill_registers)
+
+	/* Register a FIXUP handler (pass current wb as a parameter) */
+
+	movi	a0, fast_syscall_spill_registers_fixup
+	s32i	a0, a3, EXC_TABLE_FIXUP
+	rsr	a0, WINDOWBASE
+	s32i	a0, a3, EXC_TABLE_PARAM
+
+	/* Save a3 and SAR on stack. */
+
+	rsr	a0, SAR
+	xsr	a3, EXCSAVE_1		# restore a3 and excsave_1
+	s32i	a0, a2, PT_AREG4	# store SAR to PT_AREG4
+	s32i	a3, a2, PT_AREG3
+
+	/* The spill routine might clobber a7, a11, and a15. */
+
+	s32i	a7, a2, PT_AREG5
+	s32i	a11, a2, PT_AREG6
+	s32i	a15, a2, PT_AREG7
+
+	call0	_spill_registers	# destroys a3, DEPC, and SAR
+
+	/* Advance PC, restore registers and SAR, and return from exception. */
+
+	l32i	a3, a2, PT_AREG4
+	l32i	a0, a2, PT_AREG0
+	wsr	a3, SAR
+	l32i	a3, a2, PT_AREG3
+
+	/* Restore clobbered registers. */
+
+	l32i	a7, a2, PT_AREG5
+	l32i	a11, a2, PT_AREG6
+	l32i	a15, a2, PT_AREG7
+
+	movi	a2, 0
+	rfe
+
+/* Fixup handler.
+ *
+ * We get here if the spill routine causes an exception, e.g. tlb miss.
+ * We basically restore WINDOWBASE and WINDOWSTART to the condition when
+ * we entered the spill routine and jump to the user exception handler.
+ *
+ * a0: value of depc, original value in depc
+ * a2: trashed, original value in EXC_TABLE_DOUBLE_SAVE
+ * a3: exctable, original value in excsave1
+ */
+
+fast_syscall_spill_registers_fixup:
+
+	rsr	a2, WINDOWBASE	# get current windowbase (a2 is saved)
+	xsr	a0, DEPC	# restore depc and a0
+	ssl	a2		# set shift (32 - WB)
+
+	/* We need to make sure the current registers (a0-a3) are preserved.
+	 * To do this, we simply set the bit for the current window frame
+	 * in WS, so that the exception handlers save them to the task stack.
+	 */
+
+	rsr	a3, EXCSAVE_1	# get spill-mask
+	slli	a2, a3, 1	# shift left by one
+
+	slli	a3, a2, 32-WSBITS
+	src	a2, a2, a3	# a1 = xxwww1yyxxxwww1yy......
+	wsr	a2, WINDOWSTART	# set corrected windowstart
+
+	movi	a3, exc_table
+	l32i	a2, a3, EXC_TABLE_DOUBLE_SAVE	# restore a2
+	l32i	a3, a3, EXC_TABLE_PARAM	# original WB (in user task)
+
+	/* Return to the original (user task) WINDOWBASE.
+	 * We leave the following frame behind:
+	 * a0, a1, a2	same
+	 * a3:		trashed (saved in excsave_1)
+	 * depc:	depc (we have to return to that address)
+	 * excsave_1:	a3
+	 */
+
+	wsr	a3, WINDOWBASE
+	rsync
+
+	/* We are now in the original frame when we entered _spill_registers:
+	 *  a0: return address
+	 *  a1: used, stack pointer
+	 *  a2: kernel stack pointer
+	 *  a3: available, saved in EXCSAVE_1
+	 *  depc: exception address
+	 *  excsave: a3
+	 * Note: This frame might be the same as above.
+	 */
+
+#ifdef SIGNAL_HANDLING_IN_DOUBLE_EXCEPTION
+	/* Restore registers we precautiously saved.
+	 * We have the value of the 'right' a3
+	 */
+
+	l32i	a7, a2, PT_AREG5
+	l32i	a11, a2, PT_AREG6
+	l32i	a15, a2, PT_AREG7
+#endif
+
+	/* Setup stack pointer. */
+
+	addi	a2, a2, -PT_USER_SIZE
+	s32i	a0, a2, PT_AREG0
+
+	/* Make sure we return to this fixup handler. */
+
+	movi	a3, fast_syscall_spill_registers_fixup_return
+	s32i	a3, a2, PT_DEPC		# setup depc
+
+	/* Jump to the exception handler. */
+
+	movi	a3, exc_table
+	rsr	a0, EXCCAUSE
+        addx4   a0, a0, a3              	# find entry in table
+        l32i    a0, a0, EXC_TABLE_FAST_USER     # load handler
+        jx      a0
+
+fast_syscall_spill_registers_fixup_return:
+
+	/* When we return here, all registers have been restored (a2: DEPC) */
+
+	wsr	a2, DEPC		# exception address
+
+	/* Restore fixup handler. */
+
+	xsr	a3, EXCSAVE_1
+	movi	a2, fast_syscall_spill_registers_fixup
+	s32i	a2, a3, EXC_TABLE_FIXUP
+	rsr	a2, WINDOWBASE
+	s32i	a2, a3, EXC_TABLE_PARAM
+	l32i	a2, a3, EXC_TABLE_KSTK
+
+#ifdef SIGNAL_HANDLING_IN_DOUBLE_EXCEPTION
+	/* Save registers again that might be clobbered. */
+
+	s32i	a7, a2, PT_AREG5
+	s32i	a11, a2, PT_AREG6
+	s32i	a15, a2, PT_AREG7
+#endif
+
+	/* Load WB at the time the exception occurred. */
+
+	rsr	a3, SAR			# WB is still in SAR
+	neg	a3, a3
+	wsr	a3, WINDOWBASE
+	rsync
+
+	/* Restore a3 and return. */
+
+	movi	a3, exc_table
+	xsr	a3, EXCSAVE_1
+
+	rfde
+
+
+/*
+ * spill all registers.
+ *
+ * This is not a real function. The following conditions must be met:
+ *
+ *  - must be called with call0.
+ *  - uses DEPC, a3 and SAR.
+ *  - the last 'valid' register of each frame are clobbered.
+ *  - the caller must have registered a fixup handler
+ *    (or be inside a critical section)
+ *  - PS_EXCM must be set (PS_WOE cleared?)
+ */
+
+ENTRY(_spill_registers)
+
+	/*
+	 * Rotate ws so that the current windowbase is at bit 0.
+	 * Assume ws = xxxwww1yy (www1 current window frame).
+	 * Rotate ws right so that a2 = yyxxxwww1.
+	 */
+
+	wsr	a2, DEPC		# preserve a2
+	rsr	a2, WINDOWBASE
+	rsr	a3, WINDOWSTART
+	ssr	a2			# holds WB
+	slli	a2, a3, WSBITS
+	or	a3, a3, a2		# a2 = xxxwww1yyxxxwww1yy
+	srl	a3, a3
+
+	/* We are done if there are no more than the current register frame. */
+
+	extui	a3, a3, 1, WSBITS-2	# a3 = 0yyxxxwww
+	movi	a2, (1 << (WSBITS-1))
+	_beqz	a3, .Lnospill		# only one active frame? jump
+
+	/* We want 1 at the top, so that we return to the current windowbase */
+
+	or	a3, a3, a2		# 1yyxxxwww
+
+	/* Skip empty frames - get 'oldest' WINDOWSTART-bit. */
+
+	wsr	a3, WINDOWSTART		# save shifted windowstart
+	neg	a2, a3
+	and	a3, a2, a3		# first bit set from right: 000010000
+
+	ffs_ws	a2, a3			# a2: shifts to skip empty frames
+	movi	a3, WSBITS
+	sub	a2, a3, a2		# WSBITS-a2:number of 0-bits from right
+	ssr	a2			# save in SAR for later.
+
+	rsr	a3, WINDOWBASE
+	add	a3, a3, a2
+	rsr	a2, DEPC		# restore a2
+	wsr	a3, WINDOWBASE
+	rsync
+
+	rsr	a3, WINDOWSTART
+	srl	a3, a3			# shift windowstart
+
+	/* WB is now just one frame below the oldest frame in the register
+	   window. WS is shifted so the oldest frame is in bit 0, thus, WB
+	   and WS differ by one 4-register frame. */
+
+	/* Save frames. Depending what call was used (call4, call8, call12),
+	 * we have to save 4,8. or 12 registers.
+	 */
+
+	_bbsi.l	a3, 1, .Lc4
+	_bbsi.l	a3, 2, .Lc8
+
+	/* Special case: we have a call12-frame starting at a4. */
+
+	_bbci.l	a3, 3, .Lc12	# bit 3 shouldn't be zero! (Jump to Lc12 first)
+
+	s32e	a4, a1, -16	# a1 is valid with an empty spill area
+	l32e	a4, a5, -12
+	s32e	a8, a4, -48
+	mov	a8, a4
+	l32e	a4, a1, -16
+	j	.Lc12c
+
+.Lloop: _bbsi.l	a3, 1, .Lc4
+	_bbci.l	a3, 2, .Lc12
+
+.Lc8:	s32e	a4, a13, -16
+	l32e	a4, a5, -12
+	s32e	a8, a4, -32
+	s32e	a5, a13, -12
+	s32e	a6, a13, -8
+	s32e	a7, a13, -4
+	s32e	a9, a4, -28
+	s32e	a10, a4, -24
+	s32e	a11, a4, -20
+
+	srli	a11, a3, 2		# shift windowbase by 2
+	rotw	2
+	_bnei	a3, 1, .Lloop
+
+.Lexit: /* Done. Do the final rotation, set WS, and return. */
+
+	rotw	1
+	rsr	a3, WINDOWBASE
+	ssl	a3
+	movi	a3, 1
+	sll	a3, a3
+	wsr	a3, WINDOWSTART
+
+.Lnospill:
+	jx	a0
+
+.Lc4:	s32e	a4, a9, -16
+	s32e	a5, a9, -12
+	s32e	a6, a9, -8
+	s32e	a7, a9, -4
+
+	srli	a7, a3, 1
+	rotw	1
+	_bnei	a3, 1, .Lloop
+	j	.Lexit
+
+.Lc12:	_bbci.l	a3, 3, .Linvalid_mask	# bit 2 shouldn't be zero!
+
+	/* 12-register frame (call12) */
+
+	l32e	a2, a5, -12
+	s32e	a8, a2, -48
+	mov	a8, a2
+
+.Lc12c: s32e	a9, a8, -44
+	s32e	a10, a8, -40
+	s32e	a11, a8, -36
+	s32e	a12, a8, -32
+	s32e	a13, a8, -28
+	s32e	a14, a8, -24
+	s32e	a15, a8, -20
+	srli	a15, a3, 3
+
+	/* The stack pointer for a4..a7 is out of reach, so we rotate the
+	 * window, grab the stackpointer, and rotate back.
+	 * Alternatively, we could also use the following approach, but that
+	 * makes the fixup routine much more complicated:
+	 * rotw	1
+	 * s32e	a0, a13, -16
+	 * ...
+	 * rotw 2
+	 */
+
+	rotw	1
+	mov	a5, a13
+	rotw	-1
+
+	s32e	a4, a9, -16
+	s32e	a5, a9, -12
+	s32e	a6, a9, -8
+	s32e	a7, a9, -4
+
+	rotw	3
+
+	_beqi	a3, 1, .Lexit
+	j	.Lloop
+
+.Linvalid_mask:
+
+	/* We get here because of an unrecoverable error in the window
+	 * registers. If we are in user space, we kill the application,
+	 * however, this condition is unrecoverable in kernel space.
+	 */
+
+	rsr	a0, PS
+	_bbci.l	a0, PS_UM_SHIFT, 1f
+
+ 	/* User space: Setup a dummy frame and kill application.
+	 * Note: We assume EXC_TABLE_KSTK contains a valid stack pointer.
+	 */
+
+	movi	a0, 1
+	movi	a1, 0
+
+	wsr	a0, WINDOWSTART
+	wsr	a1, WINDOWBASE
+	rsync
+
+	movi	a0, 0
+
+	movi	a3, exc_table
+	l32i	a1, a3, EXC_TABLE_KSTK
+	wsr	a3, EXCSAVE_1
+
+	movi	a4, PS_WOE_MASK | 1
+	wsr	a4, PS
+	rsync
+
+	movi	a6, SIGSEGV
+	movi	a4, do_exit
+	callx4	a4
+
+1:	/* Kernel space: PANIC! */
+
+	wsr	a0, EXCSAVE_1
+	movi	a0, unrecoverable_exception
+	callx0	a0		# should not return
+1:	j	1b
+
+/*
+ * We should never get here. Bail out!
+ */
+
+ENTRY(fast_second_level_miss_double_kernel)
+
+1:	movi	a0, unrecoverable_exception
+	callx0	a0		# should not return
+1:	j	1b
+
+/* First-level entry handler for user, kernel, and double 2nd-level
+ * TLB miss exceptions.  Note that for now, user and kernel miss
+ * exceptions share the same entry point and are handled identically.
+ *
+ * An old, less-efficient C version of this function used to exist.
+ * We include it below, interleaved as comments, for reference.
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ */
+
+ENTRY(fast_second_level_miss)
+
+	/* Save a1. Note: we don't expect a double exception. */
+
+	s32i	a1, a2, PT_AREG1
+
+	/* We need to map the page of PTEs for the user task.  Find
+	 * the pointer to that page.  Also, it's possible for tsk->mm
+	 * to be NULL while tsk->active_mm is nonzero if we faulted on
+	 * a vmalloc address.  In that rare case, we must use
+	 * active_mm instead to avoid a fault in this handler.  See
+	 *
+	 * http://mail.nl.linux.org/linux-mm/2002-08/msg00258.html
+	 *   (or search Internet on "mm vs. active_mm")
+	 *
+	 *	if (!mm)
+	 *		mm = tsk->active_mm;
+	 *	pgd = pgd_offset (mm, regs->excvaddr);
+	 *	pmd = pmd_offset (pgd, regs->excvaddr);
+	 *	pmdval = *pmd;
+	 */
+
+	GET_CURRENT(a1,a2)
+	l32i	a0, a1, TASK_MM		# tsk->mm
+	beqz	a0, 9f
+
+8:	rsr	a1, EXCVADDR		# fault address
+	_PGD_OFFSET(a0, a1, a1)
+	l32i	a0, a0, 0		# read pmdval
+	//beqi	a0, _PAGE_USER, 2f
+	beqz	a0, 2f
+
+	/* Read ptevaddr and convert to top of page-table page.
+	 *
+	 * 	vpnval = read_ptevaddr_register() & PAGE_MASK;
+	 * 	vpnval += DTLB_WAY_PGTABLE;
+	 *	pteval = mk_pte (virt_to_page(pmd_val(pmdval)), PAGE_KERNEL);
+	 *	write_dtlb_entry (pteval, vpnval);
+	 *
+	 * The messy computation for 'pteval' above really simplifies
+	 * into the following:
+	 *
+	 * pteval = ((pmdval - PAGE_OFFSET) & PAGE_MASK) | PAGE_KERNEL
+	 */
+
+	movi	a1, -PAGE_OFFSET
+	add	a0, a0, a1		# pmdval - PAGE_OFFSET
+	extui	a1, a0, 0, PAGE_SHIFT	# ... & PAGE_MASK
+	xor	a0, a0, a1
+
+
+	movi	a1, PAGE_DIRECTORY
+	or	a0, a0, a1		# ... | PAGE_DIRECTORY
+
+	rsr	a1, PTEVADDR
+	srli	a1, a1, PAGE_SHIFT
+	slli	a1, a1, PAGE_SHIFT	# ptevaddr & PAGE_MASK
+	addi	a1, a1, DTLB_WAY_PGTABLE	# ... + way_number
+
+	wdtlb	a0, a1
+	dsync
+
+	/* Exit critical section. */
+
+	movi	a0, 0
+	s32i	a0, a3, EXC_TABLE_FIXUP
+
+	/* Restore the working registers, and return. */
+
+	l32i	a0, a2, PT_AREG0
+	l32i	a1, a2, PT_AREG1
+	l32i	a2, a2, PT_DEPC
+	xsr	a3, EXCSAVE_1
+
+	bgeui	a2, VALID_DOUBLE_EXCEPTION_ADDRESS, 1f
+
+	/* Restore excsave1 and return. */
+
+	rsr	a2, DEPC
+	rfe
+
+	/* Return from double exception. */
+
+1:	xsr	a2, DEPC
+	esync
+	rfde
+
+9:	l32i	a0, a1, TASK_ACTIVE_MM	# unlikely case mm == 0
+	j	8b
+
+2:	/* Invalid PGD, default exception handling */
+
+	rsr	a1, DEPC
+	xsr	a3, EXCSAVE_1
+	s32i	a1, a2, PT_AREG2
+	s32i	a3, a2, PT_AREG3
+	mov	a1, a2
+
+	rsr	a2, PS
+	bbsi.l	a2, PS_UM_SHIFT, 1f
+	j	_kernel_exception
+1:	j	_user_exception
+
+
+/*
+ * StoreProhibitedException
+ *
+ * Update the pte and invalidate the itlb mapping for this pte.
+ *
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ */
+
+ENTRY(fast_store_prohibited)
+
+	/* Save a1 and a4. */
+
+	s32i	a1, a2, PT_AREG1
+	s32i	a4, a2, PT_AREG4
+
+	GET_CURRENT(a1,a2)
+	l32i	a0, a1, TASK_MM		# tsk->mm
+	beqz	a0, 9f
+
+8:	rsr	a1, EXCVADDR		# fault address
+	_PGD_OFFSET(a0, a1, a4)
+	l32i	a0, a0, 0
+	//beqi	a0, _PAGE_USER, 2f	# FIXME use _PAGE_INVALID
+	beqz	a0, 2f
+
+	_PTE_OFFSET(a0, a1, a4)
+	l32i	a4, a0, 0		# read pteval
+	movi	a1, _PAGE_VALID | _PAGE_RW
+	bnall	a4, a1, 2f
+
+	movi	a1, _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_WRENABLE
+	or	a4, a4, a1
+	rsr	a1, EXCVADDR
+	s32i	a4, a0, 0
+
+	/* We need to flush the cache if we have page coloring. */
+#if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK
+	dhwb	a0, 0
+#endif
+	pdtlb	a0, a1
+	beqz	a0, 1f
+	idtlb	a0		// FIXME do we need this?
+	wdtlb	a4, a0
+1:
+
+	/* Exit critical section. */
+
+	movi	a0, 0
+	s32i	a0, a3, EXC_TABLE_FIXUP
+
+	/* Restore the working registers, and return. */
+
+	l32i	a4, a2, PT_AREG4
+	l32i	a1, a2, PT_AREG1
+	l32i	a0, a2, PT_AREG0
+	l32i	a2, a2, PT_DEPC
+
+	/* Restore excsave1 and a3. */
+
+	xsr	a3, EXCSAVE_1
+	bgeui	a2, VALID_DOUBLE_EXCEPTION_ADDRESS, 1f
+
+	rsr	a2, DEPC
+	rfe
+
+	/* Double exception. Restore FIXUP handler and return. */
+
+1:	xsr	a2, DEPC
+	esync
+	rfde
+
+9:	l32i	a0, a1, TASK_ACTIVE_MM	# unlikely case mm == 0
+	j	8b
+
+2:	/* If there was a problem, handle fault in C */
+
+	rsr	a4, DEPC	# still holds a2
+	xsr	a3, EXCSAVE_1
+	s32i	a4, a2, PT_AREG2
+	s32i	a3, a2, PT_AREG3
+	l32i	a4, a2, PT_AREG4
+	mov	a1, a2
+
+	rsr	a2, PS
+	bbsi.l	a2, PS_UM_SHIFT, 1f
+	j	_kernel_exception
+1:	j	_user_exception
+
+
+#if XCHAL_EXTRA_SA_SIZE
+
+#warning fast_coprocessor untested
+
+/*
+ * Entry condition:
+ *
+ *   a0:	trashed, original value saved on stack (PT_AREG0)
+ *   a1:	a1
+ *   a2:	new stack pointer, original in DEPC
+ *   a3:	dispatch table
+ *   depc:	a2, original value saved on stack (PT_DEPC)
+ *   excsave_1:	a3
+ *
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
+ */
+
+ENTRY(fast_coprocessor_double)
+	wsr	a0, EXCSAVE_1
+	movi	a0, unrecoverable_exception
+	callx0	a0
+
+ENTRY(fast_coprocessor)
+
+	/* Fatal if we are in a double exception. */
+
+	l32i	a0, a2, PT_DEPC
+	_bgeui	a0, VALID_DOUBLE_EXCEPTION_ADDRESS, fast_coprocessor_double
+
+	/* Save some registers a1, a3, a4, SAR */
+
+	xsr	a3, EXCSAVE_1
+	s32i	a3, a2, PT_AREG3
+	rsr	a3, SAR
+	s32i	a4, a2, PT_AREG4
+	s32i	a1, a2, PT_AREG1
+	s32i	a5, a1, PT_AREG5
+	s32i	a3, a2, PT_SAR
+	mov	a1, a2
+
+	/* Currently, the HAL macros only guarantee saving a0 and a1.
+	 * These can and will be refined in the future, but for now,
+	 * just save the remaining registers of a2...a15.
+	 */
+	s32i	a6, a1, PT_AREG6
+	s32i	a7, a1, PT_AREG7
+	s32i	a8, a1, PT_AREG8
+	s32i	a9, a1, PT_AREG9
+	s32i	a10, a1, PT_AREG10
+	s32i	a11, a1, PT_AREG11
+	s32i	a12, a1, PT_AREG12
+	s32i	a13, a1, PT_AREG13
+	s32i	a14, a1, PT_AREG14
+	s32i	a15, a1, PT_AREG15
+
+	/* Find coprocessor number. Subtract first CP EXCCAUSE from EXCCAUSE */
+
+	rsr	a0, EXCCAUSE
+	addi	a3, a0, -XCHAL_EXCCAUSE_COPROCESSOR0_DISABLED
+
+	/* Set corresponding CPENABLE bit */
+
+	movi	a4, 1
+	ssl	a3			# SAR: 32 - coprocessor_number
+	rsr	a5, CPENABLE
+	sll	a4, a4
+	or	a4, a5, a4
+	wsr	a4, CPENABLE
+	rsync
+	movi	a5, coprocessor_info	# list of owner and offset into cp_save
+	addx8	a0, a4, a5		# entry for CP
+
+	bne	a4, a5, .Lload		# bit wasn't set before, cp not in use
+
+	/* Now compare the current task with the owner of the coprocessor.
+	 * If they are the same, there is no reason to save or restore any
+	 * coprocessor state. Having already enabled the coprocessor,
+	 * branch ahead to return.
+	 */
+	GET_CURRENT(a5,a1)
+	l32i	a4, a0, COPROCESSOR_INFO_OWNER	# a4: current owner for this CP
+	beq	a4, a5, .Ldone
+
+	/* Find location to dump current coprocessor state:
+	 *  task_struct->task_cp_save_offset + coprocessor_offset[coprocessor]
+	 *
+	 * Note: a0 pointer to the entry in the coprocessor owner table,
+	 *	 a3 coprocessor number,
+         *	 a4 current owner of coprocessor.
+	 */
+	l32i	a5, a0, COPROCESSOR_INFO_OFFSET
+	addi	a2, a4, THREAD_CP_SAVE
+	add	a2, a2, a5
+
+	/* Store current coprocessor states. (a5 still has CP number) */
+
+	xchal_cpi_store_funcbody
+
+	/* The macro might have destroyed a3 (coprocessor number), but
+	 * SAR still has 32 - coprocessor_number!
+	 */
+	movi	a3, 32
+	rsr	a4, SAR
+	sub	a3, a3, a4
+
+.Lload:	/* A new task now owns the corpocessors. Save its TCB pointer into
+	 * the coprocessor owner table.
+	 *
+	 * Note: a0 pointer to the entry in the coprocessor owner table,
+	 *	 a3 coprocessor number.
+	 */
+	GET_CURRENT(a4,a1)
+	s32i	a4, a0, 0
+
+	/* Find location from where to restore the current coprocessor state.*/
+
+	l32i	a5, a0, COPROCESSOR_INFO_OFFSET
+	addi	a2, a4, THREAD_CP_SAVE
+	add	a2, a2, a4
+
+	xchal_cpi_load_funcbody
+
+	/* We must assume that the xchal_cpi_store_funcbody macro destroyed
+	 * registers a2..a15.
+	 */
+
+.Ldone:	l32i	a15, a1, PT_AREG15
+	l32i	a14, a1, PT_AREG14
+	l32i	a13, a1, PT_AREG13
+	l32i	a12, a1, PT_AREG12
+	l32i	a11, a1, PT_AREG11
+	l32i	a10, a1, PT_AREG10
+	l32i	a9, a1, PT_AREG9
+	l32i	a8, a1, PT_AREG8
+	l32i	a7, a1, PT_AREG7
+	l32i	a6, a1, PT_AREG6
+	l32i	a5, a1, PT_AREG5
+	l32i	a4, a1, PT_AREG4
+	l32i	a3, a1, PT_AREG3
+	l32i	a2, a1, PT_AREG2
+	l32i	a0, a1, PT_AREG0
+	l32i	a1, a1, PT_AREG1
+
+	rfe
+
+#endif /* XCHAL_EXTRA_SA_SIZE */
+
+/*
+ * Task switch.
+ *
+ * struct task*  _switch_to (struct task* prev, struct task* next)
+ *         a2                              a2                 a3
+ */
+
+ENTRY(_switch_to)
+
+	entry	a1, 16
+
+	mov	a4, a3			# preserve a3
+
+	s32i	a0, a2, THREAD_RA	# save return address
+	s32i	a1, a2, THREAD_SP	# save stack pointer
+
+	/* Disable ints while we manipulate the stack pointer; spill regs. */
+
+	movi	a5, PS_EXCM_MASK | LOCKLEVEL
+	xsr	a5, PS
+	rsr	a3, EXCSAVE_1
+	rsync
+	s32i	a3, a3, EXC_TABLE_FIXUP	/* enter critical section */
+
+	call0	_spill_registers
+
+	/* Set kernel stack (and leave critical section)
+	 * Note: It's save to set it here. The stack will not be overwritten
+	 *       because the kernel stack will only be loaded again after
+	 *       we return from kernel space.
+	 */
+
+	l32i	a0, a4, TASK_THREAD_INFO
+	rsr	a3, EXCSAVE_1		# exc_table
+	movi	a1, 0
+	addi	a0, a0, PT_REGS_OFFSET
+	s32i	a1, a3, EXC_TABLE_FIXUP
+	s32i	a0, a3, EXC_TABLE_KSTK
+
+	/* restore context of the task that 'next' addresses */
+
+	l32i	a0, a4, THREAD_RA	/* restore return address */
+	l32i	a1, a4, THREAD_SP	/* restore stack pointer */
+
+	wsr	a5, PS
+	rsync
+
+	retw
+
+
+ENTRY(ret_from_fork)
+
+	/* void schedule_tail (struct task_struct *prev)
+	 * Note: prev is still in a6 (return value from fake call4 frame)
+	 */
+	movi	a4, schedule_tail
+	callx4	a4
+
+	movi	a4, do_syscall_trace
+	callx4	a4
+
+	j	common_exception_return
+
+
+
+/*
+ * Table of syscalls
+ */
+
+.data
+.align  4
+.global sys_call_table
+sys_call_table:
+
+#define SYSCALL(call, narg) .word call
+#include "syscalls.h"
+
+/*
+ * Number of arguments of each syscall
+ */
+
+.global sys_narg_table
+sys_narg_table:
+
+#undef SYSCALL
+#define SYSCALL(call, narg) .byte narg
+#include "syscalls.h"
+