[XTENSA] Add support for configurable registers and coprocessors

The Xtensa architecture allows to define custom instructions and
registers. Registers that are bound to a coprocessor are only
accessible if the corresponding enable bit is set, which allows
to implement a 'lazy' context switch mechanism. Other registers
needs to be saved and restore at the time of the context switch
or during interrupt handling.

This patch adds support for these additional states:

- save and restore registers that are used by the compiler upon
  interrupt entry and exit.
- context switch additional registers unbound to any coprocessor
- 'lazy' context switch of registers bound to a coprocessor
- ptrace interface to provide access to additional registers
- update configuration files in include/asm-xtensa/variant-fsf

Signed-off-by: Chris Zankel <chris@zankel.net>

17 files changed, 1065 insertions, 874 deletions

diff --git a/arch/xtensa/kernel/asm-offsets.c b/arch/xtensa/kernel/asm-offsets.c
index 5d9ef51..ef63ada 100644
--- a/arch/xtensa/kernel/asm-offsets.c
+++ b/arch/xtensa/kernel/asm-offsets.c
@@ -63,6 +63,8 @@ int main(void)
 	DEFINE(PT_SIZE, sizeof(struct pt_regs));
 	DEFINE(PT_AREG_END, offsetof (struct pt_regs, areg[XCHAL_NUM_AREGS]));
 	DEFINE(PT_USER_SIZE, offsetof(struct pt_regs, areg[XCHAL_NUM_AREGS]));
+	DEFINE(PT_XTREGS_OPT, offsetof(struct pt_regs, xtregs_opt));
+	DEFINE(XTREGS_OPT_SIZE, sizeof(xtregs_opt_t));
 
 	/* struct task_struct */
 	DEFINE(TASK_PTRACE, offsetof (struct task_struct, ptrace));
@@ -76,7 +78,19 @@ int main(void)
 	/* struct thread_info (offset from start_struct) */
 	DEFINE(THREAD_RA, offsetof (struct task_struct, thread.ra));
 	DEFINE(THREAD_SP, offsetof (struct task_struct, thread.sp));
-	DEFINE(THREAD_CP_SAVE, offsetof (struct task_struct, thread.cp_save));
+	DEFINE(THREAD_CPENABLE, offsetof (struct thread_info, cpenable));
+#if XTENSA_HAVE_COPROCESSORS
+	DEFINE(THREAD_XTREGS_CP0, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP1, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP2, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP3, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP4, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP5, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP6, offsetof (struct thread_info, xtregs_cp));
+	DEFINE(THREAD_XTREGS_CP7, offsetof (struct thread_info, xtregs_cp));
+#endif
+	DEFINE(THREAD_XTREGS_USER, offsetof (struct thread_info, xtregs_user));
+	DEFINE(XTREGS_USER_SIZE, sizeof(xtregs_user_t));
 	DEFINE(THREAD_CURRENT_DS, offsetof (struct task_struct, thread.current_ds));
 
 	/* struct mm_struct */
diff --git a/arch/xtensa/kernel/coprocessor.S b/arch/xtensa/kernel/coprocessor.S
index 01bcb9f..2bc1e14 100644
--- a/arch/xtensa/kernel/coprocessor.S
+++ b/arch/xtensa/kernel/coprocessor.S
@@ -8,193 +8,328 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 2003 - 2005 Tensilica Inc.
- *
- * Marc Gauthier <marc@tensilica.com> <marc@alumni.uwaterloo.ca>
+ * Copyright (C) 2003 - 2007 Tensilica Inc.
  */
 
-/*
- * This module contains a table that describes the layout of the various
- * custom registers and states associated with each coprocessor, as well
- * as those not associated with any coprocessor ("extra state").
- * This table is included with core dumps and is available via the ptrace
- * interface, allowing the layout of such register/state information to
- * be modified in the kernel without affecting the debugger.  Each
- * register or state is identified using a 32-bit "libdb target number"
- * assigned when the Xtensa processor is generated.
- */
 
 #include <linux/linkage.h>
+#include <asm/asm-offsets.h>
 #include <asm/processor.h>
+#include <asm/coprocessor.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 <asm/tlbflush.h>
 
-#if XCHAL_HAVE_CP
+/*
+ * 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
+ */
 
-#define CP_LAST ((XCHAL_CP_MAX - 1) * COPROCESSOR_INFO_SIZE)
+/* IO protection is currently unsupported. */
 
-ENTRY(release_coprocessors)
+ENTRY(fast_io_protect)
+	wsr	a0, EXCSAVE_1
+	movi	a0, unrecoverable_exception
+	callx0	a0
 
-	entry	a1, 16
-						# a2: task
-	movi	a3, 1 << XCHAL_CP_MAX 		# a3: coprocessor-bit
-	movi	a4, coprocessor_info+CP_LAST	# a4: owner-table
-						# a5: tmp
-	movi	a6, 0				# a6: 0
-	rsil	a7, LOCKLEVEL			# a7: PS
+#if XTENSA_HAVE_COPROCESSORS
 
-1:	/* Check if task is coprocessor owner of coprocessor[i]. */
+/*
+ * Macros for lazy context switch. 
+ */
 
-	l32i	a5, a4, COPROCESSOR_INFO_OWNER
-	srli	a3, a3, 1
-	beqz	a3, 1f
-	addi	a4, a4, -8
-	beq	a2, a5, 1b
+#define SAVE_CP_REGS(x)							\
+	.align 4;							\
+	.Lsave_cp_regs_cp##x:						\
+	.if XTENSA_HAVE_COPROCESSOR(x);					\
+		xchal_cp##x##_store a2 a4 a5 a6 a7;			\
+	.endif;								\
+	jx	a0
 
-	/* Found an entry: Clear entry CPENABLE bit to disable CP. */
+#define SAVE_CP_REGS_TAB(x)						\
+	.if XTENSA_HAVE_COPROCESSOR(x);					\
+		.long .Lsave_cp_regs_cp##x - .Lsave_cp_regs_jump_table;	\
+	.else;								\
+		.long 0;						\
+	.endif;								\
+	.long THREAD_XTREGS_CP##x
 
-	rsr	a5, CPENABLE
-	s32i	a6, a4, COPROCESSOR_INFO_OWNER
-	xor	a5, a3, a5
-	wsr	a5, CPENABLE
 
-	bnez	a3, 1b
+#define LOAD_CP_REGS(x)							\
+	.align 4;							\
+	.Lload_cp_regs_cp##x:						\
+	.if XTENSA_HAVE_COPROCESSOR(x);					\
+		xchal_cp##x##_load a2 a4 a5 a6 a7;			\
+	.endif;								\
+	jx	a0
 
-1:	wsr	a7, PS
-	rsync
-	retw
+#define LOAD_CP_REGS_TAB(x)						\
+	.if XTENSA_HAVE_COPROCESSOR(x);					\
+		.long .Lload_cp_regs_cp##x - .Lload_cp_regs_jump_table; \
+	.else;								\
+		.long 0;						\
+	.endif;								\
+	.long THREAD_XTREGS_CP##x
 
+	SAVE_CP_REGS(0)
+	SAVE_CP_REGS(1)
+	SAVE_CP_REGS(2)
+	SAVE_CP_REGS(3)
+	SAVE_CP_REGS(4)
+	SAVE_CP_REGS(5)
+	SAVE_CP_REGS(6)
+	SAVE_CP_REGS(7)
 
-ENTRY(disable_coprocessor)
-	entry	sp, 16
-	rsil	a7, LOCKLEVEL
-	rsr	a3, CPENABLE
-	movi	a4, 1
-	ssl	a2
-	sll	a4, a4
-	and	a4, a3, a4
-	xor	a3, a3, a4
-	wsr	a3, CPENABLE
-	wsr	a7, PS
-	rsync
-	retw
+	LOAD_CP_REGS(0)
+	LOAD_CP_REGS(1)
+	LOAD_CP_REGS(2)
+	LOAD_CP_REGS(3)
+	LOAD_CP_REGS(4)
+	LOAD_CP_REGS(5)
+	LOAD_CP_REGS(6)
+	LOAD_CP_REGS(7)
 
-ENTRY(enable_coprocessor)
-	entry	sp, 16
-	rsil	a7, LOCKLEVEL
-	rsr	a3, CPENABLE
-	movi	a4, 1
-	ssl	a2
-	sll	a4, a4
-	or	a3, a3, a4
-	wsr	a3, CPENABLE
-	wsr	a7, PS
-	rsync
-	retw
+	.align 4
+.Lsave_cp_regs_jump_table:
+	SAVE_CP_REGS_TAB(0)
+	SAVE_CP_REGS_TAB(1)
+	SAVE_CP_REGS_TAB(2)
+	SAVE_CP_REGS_TAB(3)
+	SAVE_CP_REGS_TAB(4)
+	SAVE_CP_REGS_TAB(5)
+	SAVE_CP_REGS_TAB(6)
+	SAVE_CP_REGS_TAB(7)
 
+.Lload_cp_regs_jump_table:
+	LOAD_CP_REGS_TAB(0)
+	LOAD_CP_REGS_TAB(1)
+	LOAD_CP_REGS_TAB(2)
+	LOAD_CP_REGS_TAB(3)
+	LOAD_CP_REGS_TAB(4)
+	LOAD_CP_REGS_TAB(5)
+	LOAD_CP_REGS_TAB(6)
+	LOAD_CP_REGS_TAB(7)
 
-ENTRY(save_coprocessor_extra)
-	entry	sp, 16
-	xchal_extra_store_funcbody
-	retw
+/*
+ * coprocessor_save(buffer, index) 
+ *                    a2      a3
+ * coprocessor_load(buffer, index)
+ *                    a2      a3
+ *
+ * Save or load coprocessor registers for coprocessor 'index'. 
+ * The register values are saved to or loaded from them 'buffer' address.
+ *
+ * Note that these functions don't update the coprocessor_owner information!
+ *
+ */
 
-ENTRY(restore_coprocessor_extra)
-	entry	sp, 16
-	xchal_extra_load_funcbody
+ENTRY(coprocessor_save)
+	entry	a1, 32
+	s32i	a0, a1, 0
+	movi	a0, .Lsave_cp_regs_jump_table
+	addx8	a3, a3, a0
+	l32i	a3, a3, 0
+	beqz	a3, 1f
+	add	a0, a0, a3
+	callx0	a0
+1:	l32i	a0, a1, 0
 	retw
 
-ENTRY(save_coprocessor_registers)
-	entry	sp, 16
-	xchal_cpi_store_funcbody
+ENTRY(coprocessor_load)
+	entry	a1, 32
+	s32i	a0, a1, 0
+	movi	a0, .Lload_cp_regs_jump_table
+	addx4	a3, a3, a0
+	l32i	a3, a3, 0
+	beqz	a3, 1f
+	add	a0, a0, a3
+	callx0	a0
+1:	l32i	a0, a1, 0
 	retw
 
-ENTRY(restore_coprocessor_registers)
-	entry	sp, 16
-	xchal_cpi_load_funcbody
+/*
+ * coprocessor_flush(struct task_info*, index) 
+ *                             a2        a3
+ * coprocessor_restore(struct task_info*, index)
+ *                              a2         a3
+ *
+ * Save or load coprocessor registers for coprocessor 'index'. 
+ * The register values are saved to or loaded from the coprocessor area 
+ * inside the task_info structure.
+ *
+ * Note that these functions don't update the coprocessor_owner information!
+ *
+ */
+
+
+ENTRY(coprocessor_flush)
+	entry	a1, 32
+	s32i	a0, a1, 0
+	movi	a0, .Lsave_cp_regs_jump_table
+	addx8	a3, a3, a0
+	l32i	a4, a3, 4
+	l32i	a3, a3, 0
+	add	a2, a2, a4
+	beqz	a3, 1f
+	add	a0, a0, a3
+	callx0	a0
+1:	l32i	a0, a1, 0
 	retw
 
+ENTRY(coprocessor_restore)
+	entry	a1, 32
+	s32i	a0, a1, 0
+	movi	a0, .Lload_cp_regs_jump_table
+	addx4	a3, a3, a0
+	l32i	a4, a3, 4
+	l32i	a3, a3, 0
+	add	a2, a2, a4
+	beqz	a3, 1f
+	add	a0, a0, a3
+	callx0	a0
+1:	l32i	a0, a1, 0
+	retw
 
 /*
- *  The Xtensa compile-time HAL (core.h) XCHAL_*_SA_CONTENTS_LIBDB macros
- *  describe the contents of coprocessor & extra save areas in terms of
- *  undefined CONTENTS_LIBDB_{SREG,UREG,REGF} macros.  We define these
- *  latter macros here; they expand into a table of the format we want.
- *  The general format is:
+ * Entry condition:
  *
- *	CONTENTS_LIBDB_SREG(libdbnum, offset, size, align, rsv1, name, sregnum,
- *			    bitmask, rsv2, rsv3)
- *	CONTENTS_LIBDB_UREG(libdbnum, offset, size, align, rsv1, name, uregnum,
- *			    bitmask, rsv2, rsv3)
- *	CONTENTS_LIBDB_REGF(libdbnum, offset, size, align, rsv1, name, index,
- *			    numentries, contentsize, regname_base,
- *			    regfile_name, rsv2, rsv3)
+ *   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
  *
- *  For this table, we only care about the <libdbnum>, <offset> and <size>
- *  fields.
+ *   PT_DEPC >= VALID_DOUBLE_EXCEPTION_ADDRESS: double exception, DEPC
+ *	     <  VALID_DOUBLE_EXCEPTION_ADDRESS: regular exception
  */
 
-/*  Map all XCHAL CONTENTS macros to the reg_entry asm macro defined below:  */
-
-#define CONTENTS_LIBDB_SREG(libdbnum,offset,size,align,rsv1,name,sregnum,     \
-			    bitmask, rsv2, rsv3)			      \
-		reg_entry libdbnum, offset, size ;
-#define CONTENTS_LIBDB_UREG(libdbnum,offset,size,align,rsv1,name,uregnum,     \
-			    bitmask, rsv2, rsv3)			      \
-		reg_entry libdbnum, offset, size ;
-#define CONTENTS_LIBDB_REGF(libdbnum, offset, size, align, rsv1, name, index, \
-			    numentries, contentsize, regname_base,	      \
-			    regfile_name, rsv2, rsv3)			      \
-		reg_entry libdbnum, offset, size ;
-
-/* A single table entry: */
-	.macro	reg_entry	libdbnum, offset, size
-	 .ifne	(__last_offset-(__last_group_offset+\offset))
-	  /* padding entry */
-	  .word	(0xFC000000+__last_offset-(__last_group_offset+\offset))
-	 .endif
-	 .word	\libdbnum				/* actual entry */
-	 .set	__last_offset, __last_group_offset+\offset+\size
-	.endm	/* reg_entry */
-
-
-/* Table entry that marks the beginning of a group (coprocessor or "extra"): */
-	.macro	reg_group	cpnum, num_entries, align
-	 .set	__last_group_offset, (__last_offset + \align- 1) & -\align
-	 .ifne	\num_entries
-	  .word	0xFD000000+(\cpnum<<16)+\num_entries
-	 .endif
-	.endm	/* reg_group */
+ENTRY(fast_coprocessor_double)
+	wsr	a0, EXCSAVE_1
+	movi	a0, unrecoverable_exception
+	callx0	a0
 
-/*
- * Register info tables.
- */
 
-	.section .rodata, "a"
-	.globl	_xtensa_reginfo_tables
-	.globl	_xtensa_reginfo_table_size
-	.align	4
-_xtensa_reginfo_table_size:
-	.word	_xtensa_reginfo_table_end - _xtensa_reginfo_tables
-
-_xtensa_reginfo_tables:
-	.set	__last_offset, 0
-	reg_group 0xFF, XCHAL_EXTRA_SA_CONTENTS_LIBDB_NUM, XCHAL_EXTRA_SA_ALIGN
-	XCHAL_EXTRA_SA_CONTENTS_LIBDB
-	reg_group 0, XCHAL_CP0_SA_CONTENTS_LIBDB_NUM, XCHAL_CP0_SA_ALIGN
-	XCHAL_CP0_SA_CONTENTS_LIBDB
-	reg_group 1, XCHAL_CP1_SA_CONTENTS_LIBDB_NUM, XCHAL_CP1_SA_ALIGN
-	XCHAL_CP1_SA_CONTENTS_LIBDB
-	reg_group 2, XCHAL_CP2_SA_CONTENTS_LIBDB_NUM, XCHAL_CP2_SA_ALIGN
-	XCHAL_CP2_SA_CONTENTS_LIBDB
-	reg_group 3, XCHAL_CP3_SA_CONTENTS_LIBDB_NUM, XCHAL_CP3_SA_ALIGN
-	XCHAL_CP3_SA_CONTENTS_LIBDB
-	reg_group 4, XCHAL_CP4_SA_CONTENTS_LIBDB_NUM, XCHAL_CP4_SA_ALIGN
-	XCHAL_CP4_SA_CONTENTS_LIBDB
-	reg_group 5, XCHAL_CP5_SA_CONTENTS_LIBDB_NUM, XCHAL_CP5_SA_ALIGN
-	XCHAL_CP5_SA_CONTENTS_LIBDB
-	reg_group 6, XCHAL_CP6_SA_CONTENTS_LIBDB_NUM, XCHAL_CP6_SA_ALIGN
-	XCHAL_CP6_SA_CONTENTS_LIBDB
-	reg_group 7, XCHAL_CP7_SA_CONTENTS_LIBDB_NUM, XCHAL_CP7_SA_ALIGN
-	XCHAL_CP7_SA_CONTENTS_LIBDB
-	.word	0xFC000000	/* invalid register number,marks end of table*/
-_xtensa_reginfo_table_end:
-#endif
+ENTRY(fast_coprocessor)
+
+	/* Save remaining registers a1-a3 and SAR */
+
+	xsr	a3, EXCSAVE_1
+	s32i	a3, a2, PT_AREG3
+	rsr	a3, SAR
+	s32i	a1, a2, PT_AREG1
+	s32i	a3, a2, PT_SAR
+	mov	a1, a2
+	rsr	a2, DEPC
+	s32i	a2, a1, PT_AREG2
+
+	/*
+	 * The hal macros require up to 4 temporary registers. We use a3..a6.
+	 */
+
+	s32i	a4, a1, PT_AREG4
+	s32i	a5, a1, PT_AREG5
+	s32i	a6, a1, PT_AREG6
+
+	/* Find coprocessor number. Subtract first CP EXCCAUSE from EXCCAUSE */
+
+	rsr	a3, EXCCAUSE
+	addi	a3, a3, -EXCCAUSE_COPROCESSOR0_DISABLED
+
+	/* Set corresponding CPENABLE bit -> (sar:cp-index, a3: 1<<cp-index)*/
+
+	ssl	a3			# SAR: 32 - coprocessor_number
+	movi	a2, 1
+	rsr	a0, CPENABLE
+	sll	a2, a2
+	or	a0, a0, a2
+	wsr	a0, CPENABLE
+	rsync
+
+	/* Retrieve previous owner. (a3 still holds CP number) */
+
+	movi	a0, coprocessor_owner	# list of owners
+	addx4	a0, a3, a0		# entry for CP
+	l32i	a4, a0, 0
+
+	beqz	a4, 1f			# skip 'save' if no previous owner
+
+	/* Disable coprocessor for previous owner. (a2 = 1 << CP number) */
+
+	l32i	a5, a4, THREAD_CPENABLE
+	xor	a5, a5, a2		# (1 << cp-id) still in a2
+	s32i	a5, a4, THREAD_CPENABLE
+
+	/*
+	 * Get context save area and 'call' save routine. 
+	 * (a4 still holds previous owner (thread_info), a3 CP number)
+	 */
+
+	movi	a5, .Lsave_cp_regs_jump_table
+	movi	a0, 2f			# a0: 'return' address
+	addx8	a3, a3, a5		# a3: coprocessor number
+	l32i	a2, a3, 4		# a2: xtregs offset
+	l32i	a3, a3, 0		# a3: jump offset
+	add	a2, a2, a4
+	add	a4, a3, a5		# a4: address of save routine
+	jx	a4
+
+	/* Note that only a0 and a1 were preserved. */
+
+2:	rsr	a3, EXCCAUSE
+	addi	a3, a3, -EXCCAUSE_COPROCESSOR0_DISABLED
+	movi	a0, coprocessor_owner
+	addx4	a0, a3, a0
+
+	/* Set new 'owner' (a0 points to the CP owner, a3 contains the CP nr) */
+
+1:	GET_THREAD_INFO (a4, a1)
+	s32i	a4, a0, 0
+
+	/* Get context save area and 'call' load routine. */
+
+	movi	a5, .Lload_cp_regs_jump_table
+	movi	a0, 1f
+	addx8	a3, a3, a5
+	l32i	a2, a3, 4		# a2: xtregs offset
+	l32i	a3, a3, 0		# a3: jump offset
+	add	a2, a2, a4
+	add	a4, a3, a5
+	jx	a4
+
+	/* Restore all registers and return from exception handler. */
+
+1:	l32i	a6, a1, PT_AREG6
+	l32i	a5, a1, PT_AREG5
+	l32i	a4, a1, PT_AREG4
+
+	l32i	a0, a1, PT_SAR
+	l32i	a3, a1, PT_AREG3
+	l32i	a2, a1, PT_AREG2
+	wsr	a0, SAR
+	l32i	a0, a1, PT_AREG0
+	l32i	a1, a1, PT_AREG1
+
+	rfe
+
+	.data
+ENTRY(coprocessor_owner)
+	.fill XCHAL_CP_MAX, 4, 0
+
+#endif /* XTENSA_HAVE_COPROCESSORS */
 
diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
index b51ddb0..24770b6 100644
--- a/arch/xtensa/kernel/entry.S
+++ b/arch/xtensa/kernel/entry.S
@@ -25,6 +25,7 @@
 #include <asm/page.h>
 #include <asm/signal.h>
 #include <asm/tlbflush.h>
+#include <asm/variant/tie-asm.h>
 
 /* Unimplemented features. */
 
@@ -213,19 +214,7 @@ _user_exception:
 
 	/* 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. */
+2:	/* Now, jump to the common exception handler. */
 
 	j	common_exception
 
@@ -381,6 +370,10 @@ common_exception:
 	s32i	a2, a1, PT_LBEG
 	s32i	a3, a1, PT_LEND
 
+	/* Save optional registers. */
+
+	save_xtregs_opt a1 a2 a4 a5 a6 a7 PT_XTREGS_OPT
+	
 	/* Go to second-level dispatcher. Set up parameters to pass to the
 	 * exception handler and call the exception handler.
 	 */
@@ -452,22 +445,6 @@ common_exception_return:
 
 4:	/* a2 holds GET_CURRENT(a2,a1)  */
 
-#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
@@ -614,6 +591,12 @@ kernel_exception_exit:
 
 common_exception_exit:
 
+	/* Restore optional registers. */
+
+	load_xtregs_opt a1 a3 a4 a5 a6 a7 PT_XTREGS_OPT
+
+	/* Restore address registers. */
+
 	_bbsi.l	a2, 1, 1f
 	l32i	a4,  a1, PT_AREG4
 	l32i	a5,  a1, PT_AREG5
@@ -1146,7 +1129,6 @@ CATCH
  *   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)
@@ -1162,29 +1144,31 @@ ENTRY(fast_syscall_spill_registers)
 
 	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
+	s32i	a4, a2, PT_AREG4
+	s32i	a0, a2, PT_AREG5	# store SAR to PT_AREG5
 
 	/* The spill routine might clobber a7, a11, and a15. */
 
-	s32i	a7, a2, PT_AREG5
-	s32i	a11, a2, PT_AREG6
-	s32i	a15, a2, PT_AREG7
+	s32i	a7, a2, PT_AREG7
+	s32i	a11, a2, PT_AREG11
+	s32i	a15, a2, PT_AREG15
 
-	call0	_spill_registers	# destroys a3, DEPC, and SAR
+	call0	_spill_registers	# destroys a3, a4, and SAR
 
 	/* Advance PC, restore registers and SAR, and return from exception. */
 
-	l32i	a3, a2, PT_AREG4
+	l32i	a3, a2, PT_AREG5
+	l32i	a4, 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
+	l32i	a7, a2, PT_AREG7
+	l32i	a11, a2, PT_AREG11
+	l32i	a15, a2, PT_AREG15
 
 	movi	a2, 0
 	rfe
@@ -1257,9 +1241,9 @@ fast_syscall_spill_registers_fixup:
 
 	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
+        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:
 
@@ -1297,7 +1281,7 @@ fast_syscall_spill_registers_fixup_return:
  * This is not a real function. The following conditions must be met:
  *
  *  - must be called with call0.
- *  - uses DEPC, a3 and SAR.
+ *  - uses a3, a4 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)
@@ -1309,41 +1293,39 @@ 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.
+	 * Rotate ws right so that a4 = yyxxxwww1.
 	 */
 
-	wsr	a2, DEPC		# preserve a2
-	rsr	a2, WINDOWBASE
+	rsr	a4, WINDOWBASE
 	rsr	a3, WINDOWSTART		# a3 = xxxwww1yy
-	ssr	a2			# holds WB
-	slli	a2, a3, WSBITS
-	or	a3, a3, a2		# a3 = xxxwww1yyxxxwww1yy
+	ssr	a4			# holds WB
+	slli	a4, a3, WSBITS
+	or	a3, a3, a4		# a3 = xxxwww1yyxxxwww1yy
 	srl	a3, a3			# a3 = 00xxxwww1yyxxxwww1
 
 	/* We are done if there are no more than the current register frame. */
 
 	extui	a3, a3, 1, WSBITS-1	# a3 = 0yyxxxwww
-	movi	a2, (1 << (WSBITS-1))
+	movi	a4, (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
+	or	a3, a3, a4		# 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
+	neg	a4, a3
+	and	a3, a4, a3		# first bit set from right: 000010000
 
-	ffs_ws	a2, a3			# a2: shifts to skip empty frames
+	ffs_ws	a4, a3			# a4: 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.
+	sub	a4, a3, a4		# WSBITS-a4:number of 0-bits from right
+	ssr	a4			# save in SAR for later.
 
 	rsr	a3, WINDOWBASE
-	add	a3, a3, a2
-	rsr	a2, DEPC		# restore a2
+	add	a3, a3, a4
 	wsr	a3, WINDOWBASE
 	rsync
 
@@ -1373,7 +1355,6 @@ ENTRY(_spill_registers)
 	j	.Lc12c
 
 .Lnospill:
-	rsr	a2, DEPC
 	ret
 
 .Lloop: _bbsi.l	a3, 1, .Lc4
@@ -1810,154 +1791,6 @@ ENTRY(fast_store_prohibited)
 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 */
-
 /*
  * System Calls.
  *
@@ -2066,20 +1899,36 @@ ENTRY(_switch_to)
 
 	entry	a1, 16
 
-	mov	a4, a3			# preserve a3
+	mov	a12, a2			# preserve 'prev' (a2)
+	mov	a13, a3			# and 'next' (a3)
 
-	s32i	a0, a2, THREAD_RA	# save return address
-	s32i	a1, a2, THREAD_SP	# save stack pointer
+	l32i	a4, a2, TASK_THREAD_INFO
+	l32i	a5, a3, TASK_THREAD_INFO
 
-	/* Disable ints while we manipulate the stack pointer; spill regs. */
+	save_xtregs_user a4 a6 a8 a9 a10 a11 THREAD_XTREGS_USER
 
-	movi	a5, (1 << PS_EXCM_BIT) | LOCKLEVEL
-	xsr	a5, PS
+	s32i	a0, a12, THREAD_RA	# save return address
+	s32i	a1, a12, THREAD_SP	# save stack pointer
+
+	/* Disable ints while we manipulate the stack pointer. */
+
+	movi	a14, (1 << PS_EXCM_BIT) | LOCKLEVEL
+	xsr	a14, PS
 	rsr	a3, EXCSAVE_1
 	rsync
 	s32i	a3, a3, EXC_TABLE_FIXUP	/* enter critical section */
 
-	call0	_spill_registers
+	/* Switch CPENABLE */
+
+#if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
+	l32i	a3, a5, THREAD_CPENABLE
+	xsr	a3, CPENABLE
+	s32i	a3, a4, THREAD_CPENABLE
+#endif
+
+	/* Flush register file. */
+
+	call0	_spill_registers	# destroys a3, a4, and SAR
 
 	/* Set kernel stack (and leave critical section)
 	 * Note: It's save to set it here. The stack will not be overwritten
@@ -2087,19 +1936,21 @@ ENTRY(_switch_to)
 	 *       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
+	movi	a6, 0
+	addi	a7, a5, PT_REGS_OFFSET
+	s32i	a6, a3, EXC_TABLE_FIXUP
+	s32i	a7, 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 */
+	l32i	a0, a13, THREAD_RA	# restore return address
+	l32i	a1, a13, THREAD_SP	# restore stack pointer
+
+	load_xtregs_user a5 a6 a8 a9 a10 a11 THREAD_XTREGS_USER
 
-	wsr	a5, PS
+	wsr	a14, PS
+	mov	a2, a12			# return 'prev'
 	rsync
 
 	retw
diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
index 026138d..9185597 100644
--- a/arch/xtensa/kernel/process.c
+++ b/arch/xtensa/kernel/process.c
@@ -52,6 +52,55 @@ void (*pm_power_off)(void) = NULL;
 EXPORT_SYMBOL(pm_power_off);
 
 
+#if XTENSA_HAVE_COPROCESSORS
+
+void coprocessor_release_all(struct thread_info *ti)
+{
+	unsigned long cpenable;
+	int i;
+
+	/* Make sure we don't switch tasks during this operation. */
+
+	preempt_disable();
+
+	/* Walk through all cp owners and release it for the requested one. */
+
+	cpenable = ti->cpenable;
+
+	for (i = 0; i < XCHAL_CP_MAX; i++) {
+		if (coprocessor_owner[i] == ti) {
+			coprocessor_owner[i] = 0;
+			cpenable &= ~(1 << i);
+		}
+	}
+
+	ti->cpenable = cpenable;
+	coprocessor_clear_cpenable();
+
+	preempt_enable();
+}
+
+void coprocessor_flush_all(struct thread_info *ti)
+{
+	unsigned long cpenable;
+	int i;
+
+	preempt_disable();
+
+	cpenable = ti->cpenable;
+
+	for (i = 0; i < XCHAL_CP_MAX; i++) {
+		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
+			coprocessor_flush(ti, i);
+		cpenable >>= 1;
+	}
+
+	preempt_enable();
+}
+
+#endif
+
+
 /*
  * Powermanagement idle function, if any is provided by the platform.
  */
@@ -71,15 +120,36 @@ void cpu_idle(void)
 }
 
 /*
- * Free current thread data structures etc..
+ * This is called when the thread calls exit().
  */
-
 void exit_thread(void)
 {
+#if XTENSA_HAVE_COPROCESSORS
+	coprocessor_release_all(current_thread_info());
+#endif
 }
 
+/*
+ * Flush thread state. This is called when a thread does an execve()
+ * Note that we flush coprocessor registers for the case execve fails.
+ */
 void flush_thread(void)
 {
+#if XTENSA_HAVE_COPROCESSORS
+	struct thread_info *ti = current_thread_info();
+	coprocessor_flush_all(ti);
+	coprocessor_release_all(ti);
+#endif
+}
+
+/*
+ * This is called before the thread is copied. 
+ */
+void prepare_to_copy(struct task_struct *tsk)
+{
+#if XTENSA_HAVE_COPROCESSORS
+	coprocessor_flush_all(task_thread_info(tsk));
+#endif
 }
 
 /*
@@ -107,6 +177,7 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
                 struct task_struct * p, struct pt_regs * regs)
 {
 	struct pt_regs *childregs;
+	struct thread_info *ti;
 	unsigned long tos;
 	int user_mode = user_mode(regs);
 
@@ -128,13 +199,14 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 	p->set_child_tid = p->clear_child_tid = NULL;
 	p->thread.ra = MAKE_RA_FOR_CALL((unsigned long)ret_from_fork, 0x1);
 	p->thread.sp = (unsigned long)childregs;
+
 	if (user_mode(regs)) {
 
 		int len = childregs->wmask & ~0xf;
 		childregs->areg[1] = usp;
 		memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
 		       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
-
+// FIXME: we need to set THREADPTR in thread_info...
 		if (clone_flags & CLONE_SETTLS)
 			childregs->areg[2] = childregs->areg[6];
 
@@ -142,6 +214,12 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 		/* In kernel space, we start a new thread with a new stack. */
 		childregs->wmask = 1;
 	}
+
+#if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
+	ti = task_thread_info(p);
+	ti->cpenable = 0;
+#endif
+
 	return 0;
 }
 
@@ -179,10 +257,6 @@ unsigned long get_wchan(struct task_struct *p)
 }
 
 /*
- * do_copy_regs() gathers information from 'struct pt_regs' and
- * 'current->thread.areg[]' to fill in the xtensa_gregset_t
- * structure.
- *
  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
  * of processor registers.  Besides different ordering,
  * xtensa_gregset_t contains non-live register information that
@@ -191,9 +265,20 @@ unsigned long get_wchan(struct task_struct *p)
  *
  */
 
-void do_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs,
-		   struct task_struct *tsk)
+void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
 {
+	unsigned long wb, ws, wm;
+	int live, last;
+
+	wb = regs->windowbase;
+	ws = regs->windowstart;
+	wm = regs->wmask;
+	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
+
+	/* Don't leak any random bits. */
+
+	memset(elfregs, 0, sizeof (elfregs));
+
 	/* Note:  PS.EXCM is not set while user task is running; its
 	 * being set in regs->ps is for exception handling convenience.
 	 */
@@ -204,159 +289,18 @@ void do_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs,
 	elfregs->lend		= regs->lend;
 	elfregs->lcount		= regs->lcount;
 	elfregs->sar		= regs->sar;
+	elfregs->windowstart	= ws;
 
-	memcpy (elfregs->a, regs->areg, sizeof(elfregs->a));
-}
-
-void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
-{
-	do_copy_regs ((xtensa_gregset_t *)elfregs, regs, current);
-}
-
-
-/* The inverse of do_copy_regs().  No error or sanity checking. */
-
-void do_restore_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs,
-		      struct task_struct *tsk)
-{
-	const unsigned long ps_mask = PS_CALLINC_MASK | PS_OWB_MASK;
-	unsigned long ps;
-
-	/* Note:  PS.EXCM is not set while user task is running; it
-	 * needs to be set in regs->ps is for exception handling convenience.
-	 */
-
-	ps = (regs->ps & ~ps_mask) | (elfregs->ps & ps_mask) | (1<<PS_EXCM_BIT);
-	regs->ps		= ps;
-	regs->pc		= elfregs->pc;
-	regs->lbeg		= elfregs->lbeg;
-	regs->lend		= elfregs->lend;
-	regs->lcount		= elfregs->lcount;
-	regs->sar		= elfregs->sar;
-
-	memcpy (regs->areg, elfregs->a, sizeof(regs->areg));
-}
-
-/*
- * do_save_fpregs() gathers information from 'struct pt_regs' and
- * 'current->thread' to fill in the elf_fpregset_t structure.
- *
- * Core files and ptrace use elf_fpregset_t.
- */
-
-void do_save_fpregs (elf_fpregset_t *fpregs, struct pt_regs *regs,
-		     struct task_struct *tsk)
-{
-#if XCHAL_HAVE_CP
-
-	extern unsigned char	_xtensa_reginfo_tables[];
-	extern unsigned		_xtensa_reginfo_table_size;
-	int i;
-	unsigned long flags;
-
-	/* Before dumping coprocessor state from memory,
-	 * ensure any live coprocessor contents for this
-	 * task are first saved to memory:
-	 */
-	local_irq_save(flags);
-
-	for (i = 0; i < XCHAL_CP_MAX; i++) {
-		if (tsk == coprocessor_info[i].owner) {
-			enable_coprocessor(i);
-			save_coprocessor_registers(
-			    tsk->thread.cp_save+coprocessor_info[i].offset,i);
-			disable_coprocessor(i);
-		}
-	}
-
-	local_irq_restore(flags);
-
-	/* Now dump coprocessor & extra state: */
-	memcpy((unsigned char*)fpregs,
-		_xtensa_reginfo_tables, _xtensa_reginfo_table_size);
-	memcpy((unsigned char*)fpregs + _xtensa_reginfo_table_size,
-		tsk->thread.cp_save, XTENSA_CP_EXTRA_SIZE);
-#endif
+	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
+	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
+	memcpy(elfregs->a, regs->areg, live * 4);
+	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
 }
 
-/*
- * The inverse of do_save_fpregs().
- * Copies coprocessor and extra state from fpregs into regs and tsk->thread.
- * Returns 0 on success, non-zero if layout doesn't match.
- */
-
-int  do_restore_fpregs (elf_fpregset_t *fpregs, struct pt_regs *regs,
-		        struct task_struct *tsk)
+int dump_fpu(void)
 {
-#if XCHAL_HAVE_CP
-
-	extern unsigned char	_xtensa_reginfo_tables[];
-	extern unsigned		_xtensa_reginfo_table_size;
-	int i;
-	unsigned long flags;
-
-	/* Make sure save area layouts match.
-	 * FIXME:  in the future we could allow restoring from
-	 * a different layout of the same registers, by comparing
-	 * fpregs' table with _xtensa_reginfo_tables and matching
-	 * entries and copying registers one at a time.
-	 * Not too sure yet whether that's very useful.
-	 */
-
-	if( memcmp((unsigned char*)fpregs,
-		_xtensa_reginfo_tables, _xtensa_reginfo_table_size) ) {
-	    return -1;
-	}
-
-	/* Before restoring coprocessor state from memory,
-	 * ensure any live coprocessor contents for this
-	 * task are first invalidated.
-	 */
-
-	local_irq_save(flags);
-
-	for (i = 0; i < XCHAL_CP_MAX; i++) {
-		if (tsk == coprocessor_info[i].owner) {
-			enable_coprocessor(i);
-			save_coprocessor_registers(
-			    tsk->thread.cp_save+coprocessor_info[i].offset,i);
-			coprocessor_info[i].owner = 0;
-			disable_coprocessor(i);
-		}
-	}
-
-	local_irq_restore(flags);
-
-	/*  Now restore coprocessor & extra state:  */
-
-	memcpy(tsk->thread.cp_save,
-		(unsigned char*)fpregs + _xtensa_reginfo_table_size,
-		XTENSA_CP_EXTRA_SIZE);
-#endif
 	return 0;
 }
-/*
- * Fill in the CP structure for a core dump for a particular task.
- */
-
-int
-dump_task_fpu(struct pt_regs *regs, struct task_struct *task, elf_fpregset_t *r)
-{
-	return 0;	/* no coprocessors active on this processor */
-}
-
-/*
- * Fill in the CP structure for a core dump.
- * This includes any FPU coprocessor.
- * Here, we dump all coprocessors, and other ("extra") custom state.
- *
- * This function is called by elf_core_dump() in fs/binfmt_elf.c
- * (in which case 'regs' comes from calls to do_coredump, see signals.c).
- */
-int  dump_fpu(struct pt_regs *regs, elf_fpregset_t *r)
-{
-	return dump_task_fpu(regs, current, r);
-}
 
 asmlinkage
 long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
@@ -370,8 +314,8 @@ long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
 }
 
 /*
- *  * xtensa_execve() executes a new program.
- *   */
+ * xtensa_execve() executes a new program.
+ */
 
 asmlinkage
 long xtensa_execve(char __user *name, char __user * __user *argv,
@@ -386,7 +330,6 @@ long xtensa_execve(char __user *name, char __user * __user *argv,
 	error = PTR_ERR(filename);
 	if (IS_ERR(filename))
 		goto out;
-	// FIXME: release coprocessor??
 	error = do_execve(filename, argv, envp, regs);
 	if (error == 0) {
 		task_lock(current);
diff --git a/arch/xtensa/kernel/ptrace.c b/arch/xtensa/kernel/ptrace.c
index 5533c78..f6669d6 100644
--- a/arch/xtensa/kernel/ptrace.c
+++ b/arch/xtensa/kernel/ptrace.c
@@ -4,7 +4,7 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 2001 - 2005  Tensilica Inc.
+ * Copyright (C) 2001 - 2007  Tensilica Inc.
  *
  * Joe Taylor	<joe@tensilica.com, joetylr@yahoo.com>
  * Chris Zankel <chris@zankel.net>
@@ -28,14 +28,10 @@
 #include <asm/uaccess.h>
 #include <asm/ptrace.h>
 #include <asm/elf.h>
-
-#define TEST_KERNEL	// verify kernel operations FIXME: remove
-
+#include <asm/coprocessor.h>
 
 /*
- * Called by kernel/ptrace.c when detaching..
- *
- * Make sure single step bits etc are not set.
+ * Called by kernel/ptrace.c when detaching to disable single stepping.
  */
 
 void ptrace_disable(struct task_struct *child)
@@ -43,136 +39,233 @@ void ptrace_disable(struct task_struct *child)
 	/* Nothing to do.. */
 }
 
-long arch_ptrace(struct task_struct *child, long request, long addr, long data)
+int ptrace_getregs(struct task_struct *child, void __user *uregs)
 {
-	int ret = -EPERM;
+	struct pt_regs *regs = task_pt_regs(child);
+	xtensa_gregset_t __user *gregset = uregs;
+	unsigned long wb = regs->windowbase;
+	unsigned long ws = regs->windowstart;
+	unsigned long wm = regs->wmask;
+	int ret = 0;
+	int live, last;
+
+	if (!access_ok(VERIFY_WRITE, uregs, sizeof(xtensa_gregset_t)))
+		return -EIO;
+
+	/* Norm windowstart to a windowbase of 0. */
+
+	ws = ((ws>>wb) | (ws<<(WSBITS-wb))) & ((1<<WSBITS)-1);
+
+	ret |= __put_user(regs->pc, &gregset->pc);
+	ret |= __put_user(regs->ps & ~(1 << PS_EXCM_BIT), &gregset->ps);
+	ret |= __put_user(regs->lbeg, &gregset->lbeg);
+	ret |= __put_user(regs->lend, &gregset->lend);
+	ret |= __put_user(regs->lcount, &gregset->lcount);
+	ret |= __put_user(ws, &gregset->windowstart);
+
+	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
+	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
+	ret |= __copy_to_user(gregset->a, regs->areg, live * 4);
+	ret |= __copy_to_user(gregset->a + last, regs->areg + last, (wm>>4)*16);
+
+	return ret ? -EFAULT : 0;
+}
 
-	switch (request) {
-	case PTRACE_PEEKTEXT: /* read word at location addr. */
-	case PTRACE_PEEKDATA:
-		ret = generic_ptrace_peekdata(child, addr, data);
-		goto out;
+int ptrace_setregs(struct task_struct *child, void __user *uregs)
+{
+	struct pt_regs *regs = task_pt_regs(child);
+	xtensa_gregset_t *gregset = uregs;
+	const unsigned long ps_mask = PS_CALLINC_MASK | PS_OWB_MASK;
+	unsigned long wm = regs->wmask;
+	unsigned long ps;
+	int ret = 0;
+	int live, last;
+
+	if (!access_ok(VERIFY_WRITE, uregs, sizeof(xtensa_gregset_t)))
+		return -EIO;
+
+	ret |= __get_user(regs->pc, &gregset->pc);
+	ret |= __get_user(ps, &gregset->ps);
+	ret |= __get_user(regs->lbeg, &gregset->lbeg);
+	ret |= __get_user(regs->lend, &gregset->lend);
+	ret |= __get_user(regs->lcount, &gregset->lcount);
+
+	regs->ps = (regs->ps & ~ps_mask) | (ps & ps_mask) | (1 << PS_EXCM_BIT);
+
+	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
+	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
+	ret |= __copy_from_user(regs->areg, gregset->a, live * 4);
+	ret |= __copy_from_user(regs->areg+last, gregset->a+last, (wm>>4)*16);
+
+	return ret ? -EFAULT : 0;
+}
 
-	/* Read the word at location addr in the USER area.  */
 
-	case PTRACE_PEEKUSR:
-		{
-		struct pt_regs *regs;
-		unsigned long tmp;
+int ptrace_getxregs(struct task_struct *child, void __user *uregs)
+{
+	struct pt_regs *regs = task_pt_regs(child);
+	struct thread_info *ti = task_thread_info(child);
+	elf_xtregs_t __user *xtregs = uregs;
+	int ret = 0;
+
+	if (!access_ok(VERIFY_WRITE, uregs, sizeof(elf_xtregs_t)))
+		return -EIO;
+
+#if XTENSA_HAVE_COPROCESSORS
+	/* Flush all coprocessor registers to memory. */
+	coprocessor_flush_all(ti);
+	ret |= __copy_to_user(&xtregs->cp0, &ti->xtregs_cp,
+			      sizeof(xtregs_coprocessor_t));
+#endif
+	ret |= __copy_to_user(&xtregs->opt, &regs->xtregs_opt,
+			      sizeof(xtregs->opt));
+	ret |= __copy_to_user(&xtregs->user,&ti->xtregs_user,
+			      sizeof(xtregs->user));
 
-		regs = task_pt_regs(child);
-		tmp = 0;  /* Default return value. */
+	return ret ? -EFAULT : 0;
+}
+
+int ptrace_setxregs(struct task_struct *child, void __user *uregs)
+{
+	struct thread_info *ti = task_thread_info(child);
+	struct pt_regs *regs = task_pt_regs(child);
+	elf_xtregs_t *xtregs = uregs;
+	int ret = 0;
+
+#if XTENSA_HAVE_COPROCESSORS
+	/* Flush all coprocessors before we overwrite them. */
+	coprocessor_flush_all(ti);
+	coprocessor_release_all(ti);
+
+	ret |= __copy_from_user(&ti->xtregs_cp, &xtregs->cp0, 
+				sizeof(xtregs_coprocessor_t));
+#endif
+	ret |= __copy_from_user(&regs->xtregs_opt, &xtregs->opt,
+				sizeof(xtregs->opt));
+	ret |= __copy_from_user(&ti->xtregs_user, &xtregs->user,
+				sizeof(xtregs->user));
+
+	return ret ? -EFAULT : 0;
+}
+
+int ptrace_peekusr(struct task_struct *child, long regno, long __user *ret)
+{
+	struct pt_regs *regs;
+	unsigned long tmp;
+
+	regs = task_pt_regs(child);
+	tmp = 0;  /* Default return value. */
 
-		switch(addr) {
+	switch(regno) {
 
 		case REG_AR_BASE ... REG_AR_BASE + XCHAL_NUM_AREGS - 1:
-			{
-			int ar = addr - REG_AR_BASE - regs->windowbase * 4;
-			ar &= (XCHAL_NUM_AREGS - 1);
-			if (ar < 16 && ar + (regs->wmask >> 4) * 4 >= 0)
-				tmp = regs->areg[ar];
-			else
-				ret = -EIO;
+			tmp = regs->areg[regno - REG_AR_BASE];
 			break;
-			}
+
 		case REG_A_BASE ... REG_A_BASE + 15:
-			tmp = regs->areg[addr - REG_A_BASE];
+			tmp = regs->areg[regno - REG_A_BASE];
 			break;
+
 		case REG_PC:
 			tmp = regs->pc;
 			break;
+
 		case REG_PS:
 			/* Note:  PS.EXCM is not set while user task is running;
 			 * its being set in regs is for exception handling
 			 * convenience.  */
 			tmp = (regs->ps & ~(1 << PS_EXCM_BIT));
 			break;
+
 		case REG_WB:
-			tmp = regs->windowbase;
-			break;
+			break;		/* tmp = 0 */
+
 		case REG_WS:
-			tmp = regs->windowstart;
+		{
+			unsigned long wb = regs->windowbase;
+			unsigned long ws = regs->windowstart;
+			tmp = ((ws>>wb) | (ws<<(WSBITS-wb))) & ((1<<WSBITS)-1);
 			break;
+		}
 		case REG_LBEG:
 			tmp = regs->lbeg;
 			break;
+
 		case REG_LEND:
 			tmp = regs->lend;
 			break;
+
 		case REG_LCOUNT:
 			tmp = regs->lcount;
 			break;
+
 		case REG_SAR:
 			tmp = regs->sar;
 			break;
-		case REG_DEPC:
-			tmp = regs->depc;
-			break;
-		case REG_EXCCAUSE:
-			tmp = regs->exccause;
-			break;
-		case REG_EXCVADDR:
-			tmp = regs->excvaddr;
-			break;
+
 		case SYSCALL_NR:
 			tmp = regs->syscall;
 			break;
-		default:
-			tmp = 0;
-			ret = -EIO;
-			goto out;
-		}
-		ret = put_user(tmp, (unsigned long *) data);
-		goto out;
-		}
 
-	case PTRACE_POKETEXT: /* write the word at location addr. */
-	case PTRACE_POKEDATA:
-		ret = generic_ptrace_pokedata(child, addr, data);
-		goto out;
+		default:
+			return -EIO;
+	}
+	return put_user(tmp, ret);
+}
 
-	case PTRACE_POKEUSR:
-		{
-		struct pt_regs *regs;
-		regs = task_pt_regs(child);
+int ptrace_pokeusr(struct task_struct *child, long regno, long val)
+{
+	struct pt_regs *regs;
+	regs = task_pt_regs(child);
 
-		switch (addr) {
+	switch (regno) {
 		case REG_AR_BASE ... REG_AR_BASE + XCHAL_NUM_AREGS - 1:
-			{
-			int ar = addr - REG_AR_BASE - regs->windowbase * 4;
-			if (ar < 16 && ar + (regs->wmask >> 4) * 4 >= 0)
-				regs->areg[ar & (XCHAL_NUM_AREGS - 1)] = data;
-			else
-				ret = -EIO;
+			regs->areg[regno - REG_AR_BASE] = val;
 			break;
-			}
+
 		case REG_A_BASE ... REG_A_BASE + 15:
-			regs->areg[addr - REG_A_BASE] = data;
+			regs->areg[regno - REG_A_BASE] = val;
 			break;
+
 		case REG_PC:
-			regs->pc = data;
+			regs->pc = val;
 			break;
+
 		case SYSCALL_NR:
-			regs->syscall = data;
-			break;
-#ifdef TEST_KERNEL
-		case REG_WB:
-			regs->windowbase = data;
+			regs->syscall = val;
 			break;
-		case REG_WS:
-			regs->windowstart = data;
-			break;
-#endif
 
 		default:
-			/* The rest are not allowed. */
-			ret = -EIO;
-			break;
-		}
+			return -EIO;
+	}
+	return 0;
+}
+
+long arch_ptrace(struct task_struct *child, long request, long addr, long data)
+{
+	int ret = -EPERM;
+
+	switch (request) {
+	case PTRACE_PEEKTEXT:	/* read word at location addr. */
+	case PTRACE_PEEKDATA:
+		ret = generic_ptrace_peekdata(child, addr, data);
+		break;
+
+	case PTRACE_PEEKUSR:	/* read register specified by addr. */
+		ret = ptrace_peekusr(child, addr, (void __user *) data);
+		break;
+
+	case PTRACE_POKETEXT:	/* write the word at location addr. */
+	case PTRACE_POKEDATA:
+		ret = generic_ptrace_pokedata(child, addr, data);
+		break;
+
+	case PTRACE_POKEUSR:	/* write register specified by addr. */
+		ret = ptrace_pokeusr(child, addr, data);
 		break;
-		}
 
 	/* continue and stop at next (return from) syscall */
+
 	case PTRACE_SYSCALL:
 	case PTRACE_CONT: /* restart after signal. */
 	{
@@ -217,98 +310,26 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 		break;
 
 	case PTRACE_GETREGS:
-	{
-		/* 'data' points to user memory in which to write.
-		 * Mainly due to the non-live register values, we
-		 * reformat the register values into something more
-		 * standard.  For convenience, we use the handy
-		 * elf_gregset_t format. */
-
-		xtensa_gregset_t format;
-		struct pt_regs *regs = task_pt_regs(child);
-
-		do_copy_regs (&format, regs, child);
-
-		/* Now, copy to user space nice and easy... */
-		ret = 0;
-		if (copy_to_user((void *)data, &format, sizeof(elf_gregset_t)))
-			ret = -EFAULT;
+		ret = ptrace_getregs(child, (void __user *) data);
 		break;
-	}
 
 	case PTRACE_SETREGS:
-	{
-		/* 'data' points to user memory that contains the new
-		 * values in the elf_gregset_t format. */
-
-		xtensa_gregset_t format;
-		struct pt_regs *regs = task_pt_regs(child);
-
-		if (copy_from_user(&format,(void *)data,sizeof(elf_gregset_t))){
-			ret = -EFAULT;
-			break;
-		}
-
-		/* FIXME: Perhaps we want some sanity checks on
-		 * these user-space values?  See ARM version.  Are
-		 * debuggers a security concern? */
-
-		do_restore_regs (&format, regs, child);
-
-		ret = 0;
-		break;
-	}
-
-	case PTRACE_GETFPREGS:
-	{
-		/* 'data' points to user memory in which to write.
-		 * For convenience, we use the handy
-		 * elf_fpregset_t format. */
-
-		elf_fpregset_t fpregs;
-		struct pt_regs *regs = task_pt_regs(child);
-
-		do_save_fpregs (&fpregs, regs, child);
-
-		/* Now, copy to user space nice and easy... */
-		ret = 0;
-		if (copy_to_user((void *)data, &fpregs, sizeof(elf_fpregset_t)))
-			ret = -EFAULT;
-
+		ret = ptrace_setregs(child, (void __user *) data);
 		break;
-	}
-
-	case PTRACE_SETFPREGS:
-	{
-		/* 'data' points to user memory that contains the new
-		 * values in the elf_fpregset_t format.
-		 */
-		elf_fpregset_t fpregs;
-		struct pt_regs *regs = task_pt_regs(child);
 
-		ret = 0;
-		if (copy_from_user(&fpregs, (void *)data, sizeof(elf_fpregset_t))) {
-			ret = -EFAULT;
-			break;
-		}
-
-		if (do_restore_fpregs (&fpregs, regs, child))
-			ret = -EIO;
+	case PTRACE_GETXTREGS:
+		ret = ptrace_getxregs(child, (void __user *) data);
 		break;
-	}
 
-	case PTRACE_GETFPREGSIZE:
-		/* 'data' points to 'unsigned long' set to the size
-		 * of elf_fpregset_t
-		 */
-		ret = put_user(sizeof(elf_fpregset_t), (unsigned long *) data);
+	case PTRACE_SETXTREGS:
+		ret = ptrace_setxregs(child, (void __user *) data);
 		break;
 
 	default:
 		ret = ptrace_request(child, request, addr, data);
-		goto out;
+		break;
 	}
- out:
+
 	return ret;
 }
 
diff --git a/arch/xtensa/kernel/signal.c b/arch/xtensa/kernel/signal.c
index 42d9fd8..299be42 100644
--- a/arch/xtensa/kernel/signal.c
+++ b/arch/xtensa/kernel/signal.c
@@ -35,13 +35,17 @@ asmlinkage int do_signal(struct pt_regs *regs, sigset_t *oldset);
 
 extern struct task_struct *coproc_owners[];
 
-extern void release_all_cp (struct task_struct *);
-
 struct rt_sigframe
 {
 	struct siginfo info;
 	struct ucontext uc;
-	cp_state_t cpstate;
+	struct {
+		xtregs_opt_t opt;
+		xtregs_user_t user;
+#if XTENSA_HAVE_COPROCESSORS
+		xtregs_coprocessor_t cp;
+#endif
+	} xtregs;
 	unsigned char retcode[6];
 	unsigned int window[4];
 };
@@ -132,9 +136,10 @@ errout:
  */
 
 static int
-setup_sigcontext(struct sigcontext __user *sc, cp_state_t *cpstate,
-		 struct pt_regs *regs)
+setup_sigcontext(struct rt_sigframe __user *frame, struct pt_regs *regs)
 {
+	struct sigcontext __user *sc = &frame->uc.uc_mcontext;
+	struct thread_info *ti = current_thread_info();
 	int err = 0;
 
 #define COPY(x)	err |= __put_user(regs->x, &sc->sc_##x)
@@ -148,21 +153,32 @@ setup_sigcontext(struct sigcontext __user *sc, cp_state_t *cpstate,
 
 	err |= flush_window_regs_user(regs);
 	err |= __copy_to_user (sc->sc_a, regs->areg, 16 * 4);
+	err |= __put_user(0, &sc->sc_xtregs);
 
-	// err |= __copy_to_user (sc->sc_a, regs->areg, XCHAL_NUM_AREGS * 4)
+	if (err)
+		return err;
 
-#if XCHAL_HAVE_CP
-# error Coprocessors unsupported
-	err |= save_cpextra(cpstate);
-	err |= __put_user(err ? NULL : cpstate, &sc->sc_cpstate);
+#if XTENSA_HAVE_COPROCESSORS
+	coprocessor_flush_all(ti);
+	coprocessor_release_all(ti);
+	err |= __copy_to_user(&frame->xtregs.cp, &ti->xtregs_cp,
+			      sizeof (frame->xtregs.cp));
 #endif
+	err |= __copy_to_user(&frame->xtregs.opt, &regs->xtregs_opt,
+			      sizeof (xtregs_opt_t));
+	err |= __copy_to_user(&frame->xtregs.user, &ti->xtregs_user,
+			      sizeof (xtregs_user_t));
+
+	err |= __put_user(err ? NULL : &frame->xtregs, &sc->sc_xtregs);
 
 	return err;
 }
 
 static int
-restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc)
+restore_sigcontext(struct pt_regs *regs, struct rt_sigframe __user *frame)
 {
+	struct sigcontext __user *sc = &frame->uc.uc_mcontext;
+	struct thread_info *ti = current_thread_info();
 	unsigned int err = 0;
 	unsigned long ps;
 
@@ -180,6 +196,8 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc)
 	regs->windowbase = 0;
 	regs->windowstart = 1;
 
+	regs->syscall = -1;		/* disable syscall checks */
+
 	/* For PS, restore only PS.CALLINC.
 	 * Assume that all other bits are either the same as for the signal
 	 * handler, or the user mode value doesn't matter (e.g. PS.OWB).
@@ -195,8 +213,9 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc)
 
 	err |= __copy_from_user(regs->areg, sc->sc_a, 16 * 4);
 
-#if XCHAL_HAVE_CP
-# error Coprocessors unsupported
+	if (err)
+		return err;
+
  	/* The signal handler may have used coprocessors in which
 	 * case they are still enabled.  We disable them to force a
 	 * reloading of the original task's CP state by the lazy
@@ -204,20 +223,20 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc)
 	 * Also, we essentially discard any coprocessor state that the
 	 * signal handler created. */
 
-	if (!err) {
-	  struct task_struct *tsk = current;
-	  release_all_cp(tsk);
-	  err |= __copy_from_user(tsk->thread.cpextra, sc->sc_cpstate, 
-	      			  XTENSA_CP_EXTRA_SIZE);
-	}
+#if XTENSA_HAVE_COPROCESSORS
+	coprocessor_release_all(ti);
+	err |= __copy_from_user(&ti->xtregs_cp, &frame->xtregs.cp,
+				sizeof (frame->xtregs.cp));
 #endif
+	err |= __copy_from_user(&ti->xtregs_user, &frame->xtregs.user,
+				sizeof (xtregs_user_t));
+	err |= __copy_from_user(&regs->xtregs_opt, &frame->xtregs.opt,
+				sizeof (xtregs_opt_t));
 
-	regs->syscall = -1;		/* disable syscall checks */
 	return err;
 }
 
 
-
 /*
  * Do a signal return; undo the signal stack.
  */
@@ -246,7 +265,7 @@ asmlinkage long xtensa_rt_sigreturn(long a0, long a1, long a2, long a3,
 	recalc_sigpending();
 	spin_unlock_irq(&current->sighand->siglock);
 
-	if (restore_sigcontext(regs, &frame->uc.uc_mcontext))
+	if (restore_sigcontext(regs, frame))
 		goto badframe;
 
 	ret = regs->areg[2];
@@ -359,7 +378,7 @@ static void setup_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
 	err |= __put_user(sas_ss_flags(regs->areg[1]),
 			  &frame->uc.uc_stack.ss_flags);
 	err |= __put_user(current->sas_ss_size, &frame->uc.uc_stack.ss_size);
-	err |= setup_sigcontext(&frame->uc.uc_mcontext, &frame->cpstate, regs);
+	err |= setup_sigcontext(frame, regs);
 	err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
 
 	/* Create sys_rt_sigreturn syscall in stack frame */
diff --git a/arch/xtensa/kernel/traps.c b/arch/xtensa/kernel/traps.c
index 6f722f9..c7a021d 100644
--- a/arch/xtensa/kernel/traps.c
+++ b/arch/xtensa/kernel/traps.c
@@ -118,28 +118,28 @@ static dispatch_init_table_t __initdata dispatch_init_table[] = {
 { EXCCAUSE_STORE_CACHE_ATTRIBUTE,	0,	   do_page_fault },
 { EXCCAUSE_LOAD_CACHE_ATTRIBUTE,	0,	   do_page_fault },
 /* XCCHAL_EXCCAUSE_FLOATING_POINT unhandled */
-#if (XCHAL_CP_MASK & 1)
+#if XTENSA_HAVE_COPROCESSOR(0)
 COPROCESSOR(0),
 #endif
-#if (XCHAL_CP_MASK & 2)
+#if XTENSA_HAVE_COPROCESSOR(1)
 COPROCESSOR(1),
 #endif
-#if (XCHAL_CP_MASK & 4)
+#if XTENSA_HAVE_COPROCESSOR(2)
 COPROCESSOR(2),
 #endif
-#if (XCHAL_CP_MASK & 8)
+#if XTENSA_HAVE_COPROCESSOR(3)
 COPROCESSOR(3),
 #endif
-#if (XCHAL_CP_MASK & 16)
+#if XTENSA_HAVE_COPROCESSOR(4)
 COPROCESSOR(4),
 #endif
-#if (XCHAL_CP_MASK & 32)
+#if XTENSA_HAVE_COPROCESSOR(5)
 COPROCESSOR(5),
 #endif
-#if (XCHAL_CP_MASK & 64)
+#if XTENSA_HAVE_COPROCESSOR(6)
 COPROCESSOR(6),
 #endif
-#if (XCHAL_CP_MASK & 128)
+#if XTENSA_HAVE_COPROCESSOR(7)
 COPROCESSOR(7),
 #endif
 { EXCCAUSE_MAPPED_DEBUG,		0,		do_debug },
diff --git a/include/asm-xtensa/coprocessor.h b/include/asm-xtensa/coprocessor.h
index aa21210..e5849bb 100644
--- a/include/asm-xtensa/coprocessor.h
+++ b/include/asm-xtensa/coprocessor.h
@@ -5,81 +5,168 @@
  * License.  See the file "COPYING" in the main directory of this archive
  * for more details.
  *
- * Copyright (C) 2003 - 2005 Tensilica Inc.
+ * Copyright (C) 2003 - 2007 Tensilica Inc.
  */
 
+
 #ifndef _XTENSA_COPROCESSOR_H
 #define _XTENSA_COPROCESSOR_H
 
-#include <asm/variant/core.h>
+#include <linux/stringify.h>
 #include <asm/variant/tie.h>
+#include <asm/types.h>
+
+#ifdef __ASSEMBLY__
+# include <asm/variant/tie-asm.h>
+
+.macro	xchal_sa_start  a b
+	.set .Lxchal_pofs_, 0
+	.set .Lxchal_ofs_, 0
+.endm
+
+.macro	xchal_sa_align  ptr minofs maxofs ofsalign totalign
+	.set	.Lxchal_ofs_, .Lxchal_ofs_ + .Lxchal_pofs_ + \totalign - 1
+	.set	.Lxchal_ofs_, (.Lxchal_ofs_ & -\totalign) - .Lxchal_pofs_
+.endm
+
+#define _SELECT	(  XTHAL_SAS_TIE | XTHAL_SAS_OPT \
+		 | XTHAL_SAS_CC \
+		 | XTHAL_SAS_CALR | XTHAL_SAS_CALE | XTHAL_SAS_GLOB )
+
+.macro save_xtregs_opt ptr clb at1 at2 at3 at4 offset
+	.if XTREGS_OPT_SIZE > 0
+		addi	\clb, \ptr, \offset
+		xchal_ncp_store \clb \at1 \at2 \at3 \at4 select=_SELECT
+	.endif
+.endm
+
+.macro load_xtregs_opt ptr clb at1 at2 at3 at4 offset
+	.if XTREGS_OPT_SIZE > 0
+		addi	\clb, \ptr, \offset
+		xchal_ncp_load \clb \at1 \at2 \at3 \at4 select=_SELECT
+	.endif
+.endm
+#undef _SELECT
+
+#define _SELECT	(  XTHAL_SAS_TIE | XTHAL_SAS_OPT \
+		 | XTHAL_SAS_NOCC \
+		 | XTHAL_SAS_CALR | XTHAL_SAS_CALE | XTHAL_SAS_GLOB )
+
+.macro save_xtregs_user ptr clb at1 at2 at3 at4 offset
+	.if XTREGS_USER_SIZE > 0
+		addi	\clb, \ptr, \offset
+		xchal_ncp_store \clb \at1 \at2 \at3 \at4 select=_SELECT
+	.endif
+.endm
+
+.macro load_xtregs_user ptr clb at1 at2 at3 at4 offset
+	.if XTREGS_USER_SIZE > 0
+		addi	\clb, \ptr, \offset
+		xchal_ncp_load \clb \at1 \at2 \at3 \at4 select=_SELECT
+	.endif
+.endm
+#undef _SELECT
+
+
+
+#endif	/* __ASSEMBLY__ */
 
-#if !XCHAL_HAVE_CP
-
-#define XTENSA_CP_EXTRA_OFFSET 	0
-#define XTENSA_CP_EXTRA_ALIGN	1	/* must be a power of 2 */
-#define XTENSA_CP_EXTRA_SIZE	0
-
-#else
-
-#define XTOFS(last_start,last_size,align) \
-	((last_start+last_size+align-1) & -align)
-
-#define XTENSA_CP_EXTRA_OFFSET	0
-#define XTENSA_CP_EXTRA_ALIGN	XCHAL_EXTRA_SA_ALIGN
-
-#define XTENSA_CPE_CP0_OFFSET	\
-	XTOFS(XTENSA_CP_EXTRA_OFFSET, XCHAL_EXTRA_SA_SIZE, XCHAL_CP0_SA_ALIGN)
-#define XTENSA_CPE_CP1_OFFSET	\
-	XTOFS(XTENSA_CPE_CP0_OFFSET, XCHAL_CP0_SA_SIZE, XCHAL_CP1_SA_ALIGN)
-#define XTENSA_CPE_CP2_OFFSET	\
-	XTOFS(XTENSA_CPE_CP1_OFFSET, XCHAL_CP1_SA_SIZE, XCHAL_CP2_SA_ALIGN)
-#define XTENSA_CPE_CP3_OFFSET	\
-	XTOFS(XTENSA_CPE_CP2_OFFSET, XCHAL_CP2_SA_SIZE, XCHAL_CP3_SA_ALIGN)
-#define XTENSA_CPE_CP4_OFFSET	\
-	XTOFS(XTENSA_CPE_CP3_OFFSET, XCHAL_CP3_SA_SIZE, XCHAL_CP4_SA_ALIGN)
-#define XTENSA_CPE_CP5_OFFSET	\
-	XTOFS(XTENSA_CPE_CP4_OFFSET, XCHAL_CP4_SA_SIZE, XCHAL_CP5_SA_ALIGN)
-#define XTENSA_CPE_CP6_OFFSET	\
-	XTOFS(XTENSA_CPE_CP5_OFFSET, XCHAL_CP5_SA_SIZE, XCHAL_CP6_SA_ALIGN)
-#define XTENSA_CPE_CP7_OFFSET	\
-	XTOFS(XTENSA_CPE_CP6_OFFSET, XCHAL_CP6_SA_SIZE, XCHAL_CP7_SA_ALIGN)
-#define XTENSA_CP_EXTRA_SIZE	\
-	XTOFS(XTENSA_CPE_CP7_OFFSET, XCHAL_CP7_SA_SIZE, 16)
-
-#if XCHAL_CP_NUM > 0
-# ifndef __ASSEMBLY__
 /*
- * Tasks that own contents of (last user) each coprocessor.
- * Entries are 0 for not-owned or non-existent coprocessors.
- * Note: The size of this structure is fixed to 8 bytes in entry.S
+ * XTENSA_HAVE_COPROCESSOR(x) returns 1 if coprocessor x is configured.
+ *
+ * XTENSA_HAVE_IO_PORT(x) returns 1 if io-port x is configured.
+ *
  */
-typedef struct {
-	struct task_struct *owner;	/* owner */
-	int offset;			/* offset in cpextra space. */
-} coprocessor_info_t;
-# else
-#  define COPROCESSOR_INFO_OWNER 0
-#  define COPROCESSOR_INFO_OFFSET 4
-#  define COPROCESSOR_INFO_SIZE 8
-# endif
-#endif
-#endif	/* XCHAL_HAVE_CP */
 
+#define XTENSA_HAVE_COPROCESSOR(x)					\
+	((XCHAL_CP_MASK ^ XCHAL_CP_PORT_MASK) & (1 << (x)))
+#define XTENSA_HAVE_COPROCESSORS					\
+	(XCHAL_CP_MASK ^ XCHAL_CP_PORT_MASK)
+#define XTENSA_HAVE_IO_PORT(x)						\
+	(XCHAL_CP_PORT_MASK & (1 << (x)))
+#define XTENSA_HAVE_IO_PORTS						\
+	XCHAL_CP_PORT_MASK
 
 #ifndef __ASSEMBLY__
-# if XCHAL_CP_NUM > 0
-struct task_struct;
-extern void release_coprocessors (struct task_struct*);
-extern void save_coprocessor_registers(void*, int);
-# else
-#  define release_coprocessors(task)
-# endif
 
-typedef unsigned char cp_state_t[XTENSA_CP_EXTRA_SIZE]
-	__attribute__ ((aligned (XTENSA_CP_EXTRA_ALIGN)));
 
-#endif	/* !__ASSEMBLY__ */
+#if XCHAL_HAVE_CP
 
+#define RSR_CPENABLE(x)	do {						  \
+	__asm__ __volatile__("rsr %0," __stringify(CPENABLE) : "=a" (x)); \
+	} while(0);
+#define WSR_CPENABLE(x)	do {						  \
+  	__asm__ __volatile__("wsr %0," __stringify(CPENABLE) "; rsync" 	  \
+	    		     :: "a" (x));				  \
+	} while(0);
 
+#endif /* XCHAL_HAVE_CP */
+
+
+/*
+ * Additional registers.
+ * We define three types of additional registers:
+ *  ext: extra registers that are used by the compiler
+ *  cpn: optional registers that can be used by a user application
+ *  cpX: coprocessor registers that can only be used if the corresponding
+ *       CPENABLE bit is set.
+ */
+
+#define XCHAL_SA_REG(list,compiler,x,type,y,name,z,align,size,...)	\
+	__REG ## list (compiler, type, name, size, align)
+
+#define __REG0(compiler,t,name,s,a)	__REG0_ ## compiler (name)
+#define __REG1(compiler,t,name,s,a)	__REG1_ ## compiler (name)
+#define __REG2(c,type,...)		__REG2_ ## type (__VA_ARGS__)
+
+#define __REG0_0(name)
+#define __REG0_1(name)	__u32 name;
+#define __REG1_0(name)	__u32 name;
+#define __REG1_1(name)
+#define __REG2_0(n,s,a)	__u32 name;
+#define __REG2_1(n,s,a)	unsigned char n[s] __attribute__ ((aligned(a)));
+#define __REG2_2(n,s,a) unsigned char n[s] __attribute__ ((aligned(a)));
+
+typedef struct { XCHAL_NCP_SA_LIST(0) } xtregs_opt_t
+	__attribute__ ((aligned (XCHAL_NCP_SA_ALIGN)));
+typedef struct { XCHAL_NCP_SA_LIST(1) } xtregs_user_t
+	__attribute__ ((aligned (XCHAL_NCP_SA_ALIGN)));
+
+#if XTENSA_HAVE_COPROCESSORS
+
+typedef struct { XCHAL_CP0_SA_LIST(2) } xtregs_cp0_t
+	__attribute__ ((aligned (XCHAL_CP0_SA_ALIGN)));
+typedef struct { XCHAL_CP1_SA_LIST(2) } xtregs_cp1_t
+	__attribute__ ((aligned (XCHAL_CP1_SA_ALIGN)));
+typedef struct { XCHAL_CP2_SA_LIST(2) } xtregs_cp2_t
+	__attribute__ ((aligned (XCHAL_CP2_SA_ALIGN)));
+typedef struct { XCHAL_CP3_SA_LIST(2) } xtregs_cp3_t
+	__attribute__ ((aligned (XCHAL_CP3_SA_ALIGN)));
+typedef struct { XCHAL_CP4_SA_LIST(2) } xtregs_cp4_t
+	__attribute__ ((aligned (XCHAL_CP4_SA_ALIGN)));
+typedef struct { XCHAL_CP5_SA_LIST(2) } xtregs_cp5_t
+	__attribute__ ((aligned (XCHAL_CP5_SA_ALIGN)));
+typedef struct { XCHAL_CP6_SA_LIST(2) } xtregs_cp6_t
+	__attribute__ ((aligned (XCHAL_CP6_SA_ALIGN)));
+typedef struct { XCHAL_CP7_SA_LIST(2) } xtregs_cp7_t
+	__attribute__ ((aligned (XCHAL_CP7_SA_ALIGN)));
+
+extern struct thread_info* coprocessor_owner[XCHAL_CP_MAX];
+extern void coprocessor_save(void*, int);
+extern void coprocessor_load(void*, int);
+extern void coprocessor_flush(struct thread_info*, int);
+extern void coprocessor_restore(struct thread_info*, int);
+
+extern void coprocessor_release_all(struct thread_info*);
+extern void coprocessor_flush_all(struct thread_info*);
+
+static inline void coprocessor_clear_cpenable(void)
+{
+	unsigned long i = 0;
+	WSR_CPENABLE(i);
+}
+
+#endif	/* XTENSA_HAVE_COPROCESSORS */
+
+#endif	/* !__ASSEMBLY__ */
 #endif	/* _XTENSA_COPROCESSOR_H */
diff --git a/include/asm-xtensa/elf.h b/include/asm-xtensa/elf.h
index 86479b8..11103e0 100644
--- a/include/asm-xtensa/elf.h
+++ b/include/asm-xtensa/elf.h
@@ -173,6 +173,21 @@ extern void xtensa_elf_core_copy_regs (xtensa_gregset_t *, struct pt_regs *);
        _r->areg[12]=0; _r->areg[13]=0;   _r->areg[14]=0; _r->areg[15]=0; \
   } while (0)
 
+typedef struct {
+	xtregs_opt_t	opt;
+	xtregs_user_t	user;
+#if XTENSA_HAVE_COPROCESSORS
+	xtregs_cp0_t	cp0;
+	xtregs_cp1_t	cp1;
+	xtregs_cp2_t	cp2;
+	xtregs_cp3_t	cp3;
+	xtregs_cp4_t	cp4;
+	xtregs_cp5_t	cp5;
+	xtregs_cp6_t	cp6;
+	xtregs_cp7_t	cp7;
+#endif
+} elf_xtregs_t;
+
 #define SET_PERSONALITY(ex, ibcs2) set_personality(PER_LINUX_32BIT)
 
 struct task_struct;
diff --git a/include/asm-xtensa/processor.h b/include/asm-xtensa/processor.h
index 96408f4..4918a4e 100644
--- a/include/asm-xtensa/processor.h
+++ b/include/asm-xtensa/processor.h
@@ -103,10 +103,6 @@ struct thread_struct {
 	unsigned long dbreaka[XCHAL_NUM_DBREAK];
 	unsigned long dbreakc[XCHAL_NUM_DBREAK];
 
-	/* Allocate storage for extra state and coprocessor state. */
-	unsigned char cp_save[XTENSA_CP_EXTRA_SIZE]
-		__attribute__ ((aligned(XTENSA_CP_EXTRA_ALIGN)));
-
 	/* Make structure 16 bytes aligned. */
 	int align[0] __attribute__ ((aligned(16)));
 };
@@ -162,21 +158,16 @@ struct thread_struct {
 struct task_struct;
 struct mm_struct;
 
-// FIXME: do we need release_thread for CP??
 /* Free all resources held by a thread. */
 #define release_thread(thread) do { } while(0)
 
-// FIXME: do we need prepare_to_copy (lazy status) for CP??
 /* Prepare to copy thread state - unlazy all lazy status */
-#define prepare_to_copy(tsk)	do { } while (0)
+extern void prepare_to_copy(struct task_struct*);
 
-/*
- * create a kernel thread without removing it from tasklists
- */
+/* Create a kernel thread without removing it from tasklists */
 extern int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
 
 /* Copy and release all segment info associated with a VM */
-
 #define copy_segments(p, mm)	do { } while(0)
 #define release_segments(mm)	do { } while(0)
 #define forget_segments()	do { } while (0)
diff --git a/include/asm-xtensa/ptrace.h b/include/asm-xtensa/ptrace.h
index 77ff02d..422c73e 100644
--- a/include/asm-xtensa/ptrace.h
+++ b/include/asm-xtensa/ptrace.h
@@ -53,33 +53,30 @@
 
 /* Registers used by strace */
 
-#define REG_A_BASE	0xfc000000
-#define REG_AR_BASE	0x04000000
-#define REG_PC		0x14000000
-#define REG_PS		0x080000e6
-#define REG_WB		0x08000048
-#define REG_WS		0x08000049
-#define REG_LBEG	0x08000000
-#define REG_LEND	0x08000001
-#define REG_LCOUNT	0x08000002
-#define REG_SAR		0x08000003
-#define REG_DEPC	0x080000c0
-#define	REG_EXCCAUSE	0x080000e8
-#define REG_EXCVADDR	0x080000ee
-#define SYSCALL_NR	0x1
-
-#define AR_REGNO_TO_A_REGNO(ar, wb) (ar - wb*4) & ~(XCHAL_NUM_AREGS - 1)
+#define REG_A_BASE	0x0000
+#define REG_AR_BASE	0x0100
+#define REG_PC		0x0020
+#define REG_PS		0x02e6
+#define REG_WB		0x0248
+#define REG_WS		0x0249
+#define REG_LBEG	0x0200
+#define REG_LEND	0x0201
+#define REG_LCOUNT	0x0202
+#define REG_SAR		0x0203
+
+#define SYSCALL_NR	0x00ff
 
 /* Other PTRACE_ values defined in <linux/ptrace.h> using values 0-9,16,17,24 */
 
-#define PTRACE_GETREGS            12
-#define PTRACE_SETREGS            13
-#define PTRACE_GETFPREGS          14
-#define PTRACE_SETFPREGS          15
-#define PTRACE_GETFPREGSIZE       18
+#define PTRACE_GETREGS		12
+#define PTRACE_SETREGS		13
+#define PTRACE_GETXTREGS	18
+#define PTRACE_SETXTREGS	19
 
 #ifndef __ASSEMBLY__
 
+#ifdef __KERNEL__
+
 /*
  * This struct defines the way the registers are stored on the
  * kernel stack during a system call or other kernel entry.
@@ -102,6 +99,9 @@ struct pt_regs {
 	unsigned long icountlevel;	/*  60 */
 	int reserved[1];		/*  64 */
 
+	/* Additional configurable registers that are used by the compiler. */
+	xtregs_opt_t xtregs_opt;
+
 	/* Make sure the areg field is 16 bytes aligned. */
 	int align[0] __attribute__ ((aligned(16)));
 
@@ -111,8 +111,6 @@ struct pt_regs {
 	unsigned long areg[16];		/* 128 (64) */
 };
 
-#ifdef __KERNEL__
-
 #include <asm/variant/core.h>
 
 # define task_pt_regs(tsk) ((struct pt_regs*) \
diff --git a/include/asm-xtensa/regs.h b/include/asm-xtensa/regs.h
index c913d25..d4baed2 100644
--- a/include/asm-xtensa/regs.h
+++ b/include/asm-xtensa/regs.h
@@ -100,7 +100,14 @@
 #define EXCCAUSE_DTLB_SIZE_RESTRICTION		27
 #define EXCCAUSE_LOAD_CACHE_ATTRIBUTE		28
 #define EXCCAUSE_STORE_CACHE_ATTRIBUTE		29
-#define EXCCAUSE_FLOATING_POINT			40
+#define EXCCAUSE_COPROCESSOR0_DISABLED		32
+#define EXCCAUSE_COPROCESSOR1_DISABLED		33
+#define EXCCAUSE_COPROCESSOR2_DISABLED		34
+#define EXCCAUSE_COPROCESSOR3_DISABLED		35
+#define EXCCAUSE_COPROCESSOR4_DISABLED		36
+#define EXCCAUSE_COPROCESSOR5_DISABLED		37
+#define EXCCAUSE_COPROCESSOR6_DISABLED		38
+#define EXCCAUSE_COPROCESSOR7_DISABLED		39
 
 /*  PS register fields.  */
 
diff --git a/include/asm-xtensa/sigcontext.h b/include/asm-xtensa/sigcontext.h
index dff3c54..03383af 100644
--- a/include/asm-xtensa/sigcontext.h
+++ b/include/asm-xtensa/sigcontext.h
@@ -22,6 +22,7 @@ struct sigcontext {
 	unsigned long sc_acclo;
 	unsigned long sc_acchi;
 	unsigned long sc_a[16];
+	void *sc_xtregs;
 };
 
 #endif /* _XTENSA_SIGCONTEXT_H */
diff --git a/include/asm-xtensa/system.h b/include/asm-xtensa/system.h
index e0cb911..62b1e8f 100644
--- a/include/asm-xtensa/system.h
+++ b/include/asm-xtensa/system.h
@@ -46,42 +46,6 @@ static inline int irqs_disabled(void)
 	return flags & 0xf;
 }
 
-#define RSR_CPENABLE(x)	do {						  \
-	__asm__ __volatile__("rsr %0," __stringify(CPENABLE) : "=a" (x)); \
-	} while(0);
-#define WSR_CPENABLE(x)	do {						  \
-  	__asm__ __volatile__("wsr %0," __stringify(CPENABLE)";rsync" 	  \
-	    		     :: "a" (x));} while(0);
-
-#define clear_cpenable() __clear_cpenable()
-
-static inline void __clear_cpenable(void)
-{
-#if XCHAL_HAVE_CP
-	unsigned long i = 0;
-	WSR_CPENABLE(i);
-#endif
-}
-
-static inline void enable_coprocessor(int i)
-{
-#if XCHAL_HAVE_CP
-	int cp;
-	RSR_CPENABLE(cp);
-	cp |= 1 << i;
-	WSR_CPENABLE(cp);
-#endif
-}
-
-static inline void disable_coprocessor(int i)
-{
-#if XCHAL_HAVE_CP
-	int cp;
-	RSR_CPENABLE(cp);
-	cp &= ~(1 << i);
-	WSR_CPENABLE(cp);
-#endif
-}
 
 #define smp_read_barrier_depends() do { } while(0)
 #define read_barrier_depends() do { } while(0)
@@ -111,7 +75,6 @@ extern void *_switch_to(void *last, void *next);
 
 #define switch_to(prev,next,last)		\
 do {						\
-	clear_cpenable();			\
 	(last) = _switch_to(prev, next);	\
 } while(0)
 
@@ -244,7 +207,7 @@ static inline void spill_registers(void)
 		"wsr	a13," __stringify(SAR) "\n\t"
 		"wsr	a14," __stringify(PS) "\n\t"
 		:: "a" (&a0), "a" (&ps)
-		: "a2", "a3", "a12", "a13", "a14", "a15", "memory");
+		: "a2", "a3", "a4", "a7", "a11", "a12", "a13", "a14", "a15", "memory");
 }
 
 #define arch_align_stack(x) (x)
diff --git a/include/asm-xtensa/thread_info.h b/include/asm-xtensa/thread_info.h
index 52c9582..a2c6406 100644
--- a/include/asm-xtensa/thread_info.h
+++ b/include/asm-xtensa/thread_info.h
@@ -27,6 +27,21 @@
 
 #ifndef __ASSEMBLY__
 
+#if XTENSA_HAVE_COPROCESSORS
+
+typedef struct xtregs_coprocessor {
+	xtregs_cp0_t cp0;
+	xtregs_cp1_t cp1;
+	xtregs_cp2_t cp2;
+	xtregs_cp3_t cp3;
+	xtregs_cp4_t cp4;
+	xtregs_cp5_t cp5;
+	xtregs_cp6_t cp6;
+	xtregs_cp7_t cp7;
+} xtregs_coprocessor_t;
+
+#endif
+
 struct thread_info {
 	struct task_struct	*task;		/* main task structure */
 	struct exec_domain	*exec_domain;	/* execution domain */
@@ -38,7 +53,13 @@ struct thread_info {
 	mm_segment_t		addr_limit;	/* thread address space */
 	struct restart_block    restart_block;
 
+	unsigned long		cpenable;
 
+	/* Allocate storage for extra user states and coprocessor states. */
+#if XTENSA_HAVE_COPROCESSORS
+	xtregs_coprocessor_t	xtregs_cp;
+#endif
+	xtregs_user_t		xtregs_user;
 };
 
 #else /* !__ASSEMBLY__ */
diff --git a/include/asm-xtensa/variant-fsf/tie-asm.h b/include/asm-xtensa/variant-fsf/tie-asm.h
new file mode 100644
index 0000000..68a73bf
--- /dev/null
+++ b/include/asm-xtensa/variant-fsf/tie-asm.h
@@ -0,0 +1,70 @@
+/*
+ * This header file contains assembly-language definitions (assembly
+ * macros, etc.) for this specific Xtensa processor's TIE extensions
+ * and options.  It is customized to this Xtensa processor configuration.
+ *
+ * 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) 1999-2008 Tensilica Inc.
+ */
+
+#ifndef _XTENSA_CORE_TIE_ASM_H
+#define _XTENSA_CORE_TIE_ASM_H
+
+/*  Selection parameter values for save-area save/restore macros:  */
+/*  Option vs. TIE:  */
+#define XTHAL_SAS_TIE	0x0001	/* custom extension or coprocessor */
+#define XTHAL_SAS_OPT	0x0002	/* optional (and not a coprocessor) */
+/*  Whether used automatically by compiler:  */
+#define XTHAL_SAS_NOCC	0x0004	/* not used by compiler w/o special opts/code */
+#define XTHAL_SAS_CC	0x0008	/* used by compiler without special opts/code */
+/*  ABI handling across function calls:  */
+#define XTHAL_SAS_CALR	0x0010	/* caller-saved */
+#define XTHAL_SAS_CALE	0x0020	/* callee-saved */
+#define XTHAL_SAS_GLOB	0x0040	/* global across function calls (in thread) */
+/*  Misc  */
+#define XTHAL_SAS_ALL	0xFFFF	/* include all default NCP contents */
+
+
+
+/* Macro to save all non-coprocessor (extra) custom TIE and optional state
+ * (not including zero-overhead loop registers).
+ * Save area ptr (clobbered):  ptr  (1 byte aligned)
+ * Scratch regs  (clobbered):  at1..at4  (only first XCHAL_NCP_NUM_ATMPS needed)
+ */
+	.macro xchal_ncp_store  ptr at1 at2 at3 at4  continue=0 ofs=-1 select=XTHAL_SAS_ALL
+	xchal_sa_start	\continue, \ofs
+	.ifeq (XTHAL_SAS_OPT | XTHAL_SAS_CC | XTHAL_SAS_GLOB) & ~\select
+	xchal_sa_align	\ptr, 0, 1024-4, 4, 4
+	rur	\at1, THREADPTR		// threadptr option
+	s32i	\at1, \ptr, .Lxchal_ofs_ + 0
+	.set	.Lxchal_ofs_, .Lxchal_ofs_ + 4
+	.endif
+	.endm	// xchal_ncp_store
+
+/* Macro to save all non-coprocessor (extra) custom TIE and optional state
+ * (not including zero-overhead loop registers).
+ * Save area ptr (clobbered):  ptr  (1 byte aligned)
+ * Scratch regs  (clobbered):  at1..at4  (only first XCHAL_NCP_NUM_ATMPS needed)
+ */
+	.macro xchal_ncp_load  ptr at1 at2 at3 at4  continue=0 ofs=-1 select=XTHAL_SAS_ALL
+	xchal_sa_start	\continue, \ofs
+	.ifeq (XTHAL_SAS_OPT | XTHAL_SAS_CC | XTHAL_SAS_GLOB) & ~\select
+	xchal_sa_align	\ptr, 0, 1024-4, 4, 4
+	l32i	\at1, \ptr, .Lxchal_ofs_ + 0
+	wur	\at1, THREADPTR		// threadptr option
+	.set	.Lxchal_ofs_, .Lxchal_ofs_ + 4
+	.endif
+	.endm	// xchal_ncp_load
+
+
+
+#define XCHAL_NCP_NUM_ATMPS	1
+
+
+#define XCHAL_SA_NUM_ATMPS	1
+
+#endif /*_XTENSA_CORE_TIE_ASM_H*/
+
diff --git a/include/asm-xtensa/variant-fsf/tie.h b/include/asm-xtensa/variant-fsf/tie.h
index a73c716..bf40201 100644
--- a/include/asm-xtensa/variant-fsf/tie.h
+++ b/include/asm-xtensa/variant-fsf/tie.h
@@ -1,22 +1,77 @@
 /*
- * Xtensa processor core configuration information.
+ * This header file describes this specific Xtensa processor's TIE extensions
+ * that extend basic Xtensa core functionality.  It is customized to this
+ * Xtensa processor configuration.
  *
  * 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) 1999-2006 Tensilica Inc.
+ * Copyright (C) 1999-2007 Tensilica Inc.
  */
 
-#ifndef XTENSA_TIE_H
-#define XTENSA_TIE_H
-
-/*----------------------------------------------------------------------
-			COPROCESSORS and EXTRA STATE
-  ----------------------------------------------------------------------*/
+#ifndef _XTENSA_CORE_TIE_H
+#define _XTENSA_CORE_TIE_H
 
 #define XCHAL_CP_NUM			0	/* number of coprocessors */
-#define XCHAL_CP_MASK			0x00
+#define XCHAL_CP_MAX			0	/* max CP ID + 1 (0 if none) */
+#define XCHAL_CP_MASK			0x00	/* bitmask of all CPs by ID */
+#define XCHAL_CP_PORT_MASK		0x00	/* bitmask of only port CPs */
+
+/*  Basic parameters of each coprocessor:  */
+#define XCHAL_CP7_NAME			"XTIOP"
+#define XCHAL_CP7_IDENT			XTIOP
+#define XCHAL_CP7_SA_SIZE		0	/* size of state save area */
+#define XCHAL_CP7_SA_ALIGN		1	/* min alignment of save area */
+#define XCHAL_CP_ID_XTIOP		7	/* coprocessor ID (0..7) */
+
+/*  Filler info for unassigned coprocessors, to simplify arrays etc:  */
+#define XCHAL_NCP_SA_SIZE		0
+#define XCHAL_NCP_SA_ALIGN		1
+#define XCHAL_CP0_SA_SIZE		0
+#define XCHAL_CP0_SA_ALIGN		1
+#define XCHAL_CP1_SA_SIZE		0
+#define XCHAL_CP1_SA_ALIGN		1
+#define XCHAL_CP2_SA_SIZE		0
+#define XCHAL_CP2_SA_ALIGN		1
+#define XCHAL_CP3_SA_SIZE		0
+#define XCHAL_CP3_SA_ALIGN		1
+#define XCHAL_CP4_SA_SIZE		0
+#define XCHAL_CP4_SA_ALIGN		1
+#define XCHAL_CP5_SA_SIZE		0
+#define XCHAL_CP5_SA_ALIGN		1
+#define XCHAL_CP6_SA_SIZE		0
+#define XCHAL_CP6_SA_ALIGN		1
+
+/*  Save area for non-coprocessor optional and custom (TIE) state:  */
+#define XCHAL_NCP_SA_SIZE		0
+#define XCHAL_NCP_SA_ALIGN		1
+
+/*  Total save area for optional and custom state (NCP + CPn):  */
+#define XCHAL_TOTAL_SA_SIZE		0	/* with 16-byte align padding */
+#define XCHAL_TOTAL_SA_ALIGN		1	/* actual minimum alignment */
+
+#define XCHAL_NCP_SA_NUM	0
+#define XCHAL_NCP_SA_LIST(s)
+#define XCHAL_CP0_SA_NUM	0
+#define XCHAL_CP0_SA_LIST(s)
+#define XCHAL_CP1_SA_NUM	0
+#define XCHAL_CP1_SA_LIST(s)
+#define XCHAL_CP2_SA_NUM	0
+#define XCHAL_CP2_SA_LIST(s)
+#define XCHAL_CP3_SA_NUM	0
+#define XCHAL_CP3_SA_LIST(s)
+#define XCHAL_CP4_SA_NUM	0
+#define XCHAL_CP4_SA_LIST(s)
+#define XCHAL_CP5_SA_NUM	0
+#define XCHAL_CP5_SA_LIST(s)
+#define XCHAL_CP6_SA_NUM	0
+#define XCHAL_CP6_SA_LIST(s)
+#define XCHAL_CP7_SA_NUM	0
+#define XCHAL_CP7_SA_LIST(s)
+
+/* Byte length of instruction from its first nibble (op0 field), per FLIX.  */
+#define XCHAL_OP0_FORMAT_LENGTHS	3,3,3,3,3,3,3,3,2,2,2,2,2,2,3,3
 
-#endif /*XTENSA_CONFIG_TIE_H*/
+#endif /*_XTENSA_CORE_TIE_H*/