/* Copyright (C) 1999-2018 Free Software Foundation, Inc.

   Contributed by David Mosberger-Tang <davidm@hpl.hp.com>.

   The GNU C Library is free software; you can redistribute it and/or

   modify it under the terms of the GNU Lesser General Public

   License as published by the Free Software Foundation; either

   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,

   but WITHOUT ANY WARRANTY; without even the implied warranty of

   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public

   License along with the GNU C Library; if not, see

   <http://www.gnu.org/licenses/>.

   Note that __sigsetjmp() did NOT flush the register stack.  Instead,

   we do it here since __longjmp() is usually much less frequently

   invoked than __sigsetjmp(). The only difficulty is that __sigsetjmp()

   didn't (and wouldn't be able to) save ar.rnat either.  This is a problem

   because if we're not careful, we could end up loading random NaT bits.

   There are two cases:

	(i)  ar.bsp < ia64_rse_rnat_addr(jmpbuf.ar_bsp)

		ar.rnat contains the desired bits---preserve ar.rnat

		across loadrs and write to ar.bspstore

	(ii) ar.bsp >= ia64_rse_rnat_addr(jmpbuf.ar_bsp)

		The desired ar.rnat is stored in

		ia64_rse_rnat_addr(jmpbuf.ar_bsp).  Load those

		bits into ar.rnat after setting ar.bspstore. */

#include <sysdep.h>

#include <features.h>

#	define	pPos	p6	/* is rotate count positive? */

#	define	pNeg	p7	/* is rotate count negative? */

	/* __longjmp(__jmp_buf buf, int val) */

LEAF(__longjmp)

#ifdef CHECK_RSP

	alloc r8=ar.pfs,2,1,3,0

	CHECK_RSP

#else

	alloc r8=ar.pfs,2,0,0,0

#endif

	mov r27=ar.rsc

	add r2=0x98,in0		// r2 <- &jmpbuf.orig_jmp_buf_addr

	;;

	ld8 r8=[r2],-16		// r8 <- orig_jmp_buf_addr

	mov r10=ar.bsp

	and r11=~0x3,r27	// clear ar.rsc.mode

	;;

	flushrs			// flush dirty regs to backing store (must be first in insn grp)

	ld8 r23=[r2],8		// r23 <- jmpbuf.ar_bsp

	sub r8=r8,in0		// r8 <- &orig_jmpbuf - &jmpbuf

	;;

	ld8 r25=[r2]		// r25 <- jmpbuf.ar_unat

	extr.u r8=r8,3,6	// r8 <- (&orig_jmpbuf - &jmpbuf)/8 & 0x3f

	;;

	cmp.lt pNeg,pPos=r8,r0

	mov r2=in0

	;;

(pPos)	mov r16=r8

(pNeg)	add r16=64,r8

(pPos)	sub r17=64,r8

(pNeg)	sub r17=r0,r8

	;;

	mov ar.rsc=r11		// put RSE in enforced lazy mode

	shr.u r8=r25,r16

	add r3=8,in0		// r3 <- &jmpbuf.r1

	shl r9=r25,r17

	;;

	ld8.fill.nta r28=[r2],16	// r28 <- jmpbuf.sp

	or r25=r8,r9

	;;

	mov r26=ar.rnat

	mov ar.unat=r25		// setup ar.unat (NaT bits for r1, r4-r7, and r12)

	;;

	ld8.fill.nta gp=[r3],32		// r1 (gp)

	dep r11=-1,r23,3,6	// r11 <- ia64_rse_rnat_addr(jmpbuf.ar_bsp)

	mov sp=r28		// r12 (sp)

	;;

	ld8.nta r16=[r2],16		// caller's unat

//	ld8.nta r17=[r3],16		// fpsr

	;;

	ld8.fill.nta r4=[r2],16	// r4

	ld8.fill.nta r5=[r3],16		// r5 (gp)

	cmp.geu p8,p0=r10,r11	// p8 <- (ar.bsp >= jmpbuf.ar_bsp)

	;;

	ld8.fill.nta r6=[r2],16	// r6

	ld8.fill.nta r7=[r3],16		// r7

	;;

	mov ar.unat=r16			// restore caller's unat

//	mov ar.fpsr=r17			// restore fpsr

	;;

	ld8.nta r16=[r2],16		// b0

	ld8.nta r17=[r3],16		// b1

	;;

(p8)	ld8 r26=[r11]		// r26 <- *ia64_rse_rnat_addr(jmpbuf.ar_bsp)

	mov ar.bspstore=r23	// restore ar.bspstore

	;;

	ld8.nta r18=[r2],16		// b2

	ld8.nta r19=[r3],16		// b3

	;;

#ifdef PTR_DEMANGLE

	PTR_DEMANGLE (r16, r24)

#endif

	ld8.nta r20=[r2],16		// b4

	ld8.nta r21=[r3],16		// b5

	;;

	ld8.nta r11=[r2],16		// ar.pfs

	ld8.nta r22=[r3],56		// ar.lc

	;;

	ld8.nta r24=[r2],32		// pr

	mov b0=r16

	;;

	ldf.fill.nta f2=[r2],32

	ldf.fill.nta f3=[r3],32

	mov b1=r17

	;;

	ldf.fill.nta f4=[r2],32

	ldf.fill.nta f5=[r3],32

	mov b2=r18

	;;

	ldf.fill.nta f16=[r2],32

	ldf.fill.nta f17=[r3],32

	mov b3=r19

	;;

	ldf.fill.nta f18=[r2],32

	ldf.fill.nta f19=[r3],32

	mov b4=r20

	;;

	ldf.fill.nta f20=[r2],32

	ldf.fill.nta f21=[r3],32

	mov b5=r21

	;;

	ldf.fill.nta f22=[r2],32

	ldf.fill.nta f23=[r3],32

	mov ar.lc=r22

	;;

	ldf.fill.nta f24=[r2],32

	ldf.fill.nta f25=[r3],32

	cmp.eq p8,p9=0,in1

	;;

	ldf.fill.nta f26=[r2],32

	ldf.fill.nta f27=[r3],32

	mov ar.pfs=r11

	;;

	ldf.fill.nta f28=[r2],32

	ldf.fill.nta f29=[r3],32

	;;

	ldf.fill.nta f30=[r2]

	ldf.fill.nta f31=[r3]

(p8)	mov r8=1

	mov ar.rnat=r26		// restore ar.rnat

	;;

	mov ar.rsc=r27		// restore ar.rsc

(p9)	mov r8=in1

	invala			// virt. -> phys. regnum mapping may change

	mov pr=r24,-1

	ret

END(__longjmp)