head.s

linux-2.6.15.6

/*
 * Here is where the ball gets rolling as far as the kernel is concerned.
 * When control is transferred to _start, the bootload has already
 * loaded us to the correct address.  All that's left to do here is
 * to set up the kernel's global pointer and jump to the kernel
 * entry point.
 *
 * Copyright (C) 1998-2001, 2003, 2005 Hewlett-Packard Co
 *	David Mosberger-Tang <[e-mail]>
 *	Stephane Eranian <[e-mail]>
 * Copyright (C) 1999 VA Linux Systems
 * Copyright (C) 1999 Walt Drummond <[e-mail]>
 * Copyright (C) 1999 Intel Corp.
 * Copyright (C) 1999 Asit Mallick <[e-mail]>
 * Copyright (C) 1999 Don Dugger <[e-mail]>
 * Copyright (C) 2002 Fenghua Yu <[e-mail]>
 *   -Optimize __ia64_save_fpu() and __ia64_load_fpu() for Itanium 2.
 * Copyright (C) 2004 Ashok Raj <[e-mail]>
 *   Support for CPU Hotplug
 */

#include <linux/config.h>

#include <asm/asmmacro.h>
#include <asm/fpu.h>
#include <asm/kregs.h>
#include <asm/mmu_context.h>
#include <asm/asm-offsets.h>
#include <asm/pal.h>
#include <asm/pgtable.h>
#include <asm/processor.h>
#include <asm/ptrace.h>
#include <asm/system.h>
#include <asm/mca_asm.h>

#ifdef CONFIG_HOTPLUG_CPU
#define SAL_PSR_BITS_TO_SET				\
	(IA64_PSR_AC | IA64_PSR_BN | IA64_PSR_MFH | IA64_PSR_MFL)

#define SAVE_FROM_REG(src, ptr, dest)	\
	mov dest=src;;						\
	st8 [ptr]=dest,0x08

#define RESTORE_REG(reg, ptr, _tmp)		\
	ld8 _tmp=[ptr],0x08;;				\
	mov reg=_tmp

#define SAVE_BREAK_REGS(ptr, _idx, _breg, _dest)\
	mov ar.lc=IA64_NUM_DBG_REGS-1;; 			\
	mov _idx=0;; 								\
1: 												\
	SAVE_FROM_REG(_breg[_idx], ptr, _dest);;	\
	add _idx=1,_idx;;							\
	br.cloop.sptk.many 1b

#define RESTORE_BREAK_REGS(ptr, _idx, _breg, _tmp, _lbl)\
	mov ar.lc=IA64_NUM_DBG_REGS-1;;			\
	mov _idx=0;;							\
_lbl:  RESTORE_REG(_breg[_idx], ptr, _tmp);;	\
	add _idx=1, _idx;;						\
	br.cloop.sptk.many _lbl

#define SAVE_ONE_RR(num, _reg, _tmp) \
	movl _tmp=(num<<61);;	\
	mov _reg=rr[_tmp]

#define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
	SAVE_ONE_RR(0,_r0, _tmp);; \
	SAVE_ONE_RR(1,_r1, _tmp);; \
	SAVE_ONE_RR(2,_r2, _tmp);; \
	SAVE_ONE_RR(3,_r3, _tmp);; \
	SAVE_ONE_RR(4,_r4, _tmp);; \
	SAVE_ONE_RR(5,_r5, _tmp);; \
	SAVE_ONE_RR(6,_r6, _tmp);; \
	SAVE_ONE_RR(7,_r7, _tmp);;

#define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7) \
	st8 [ptr]=_r0, 8;; \
	st8 [ptr]=_r1, 8;; \
	st8 [ptr]=_r2, 8;; \
	st8 [ptr]=_r3, 8;; \
	st8 [ptr]=_r4, 8;; \
	st8 [ptr]=_r5, 8;; \
	st8 [ptr]=_r6, 8;; \
	st8 [ptr]=_r7, 8;;

#define RESTORE_REGION_REGS(ptr, _idx1, _idx2, _tmp) \
	mov		ar.lc=0x08-1;;						\
	movl	_idx1=0x00;;						\
RestRR:											\
	dep.z	_idx2=_idx1,61,3;;					\
	ld8		_tmp=[ptr],8;;						\
	mov		rr[_idx2]=_tmp;;					\
	srlz.d;;									\
	add		_idx1=1,_idx1;;						\
	br.cloop.sptk.few	RestRR

#define SET_AREA_FOR_BOOTING_CPU(reg1, reg2) \
	movl reg1=sal_state_for_booting_cpu;;	\
	ld8 reg2=[reg1];;

/*
 * Adjust region registers saved before starting to save
 * break regs and rest of the states that need to be preserved.
 */
#define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(_reg1,_reg2,_pred)  \
	SAVE_FROM_REG(b0,_reg1,_reg2);;						\
	SAVE_FROM_REG(b1,_reg1,_reg2);;						\
	SAVE_FROM_REG(b2,_reg1,_reg2);;						\
	SAVE_FROM_REG(b3,_reg1,_reg2);;						\
	SAVE_FROM_REG(b4,_reg1,_reg2);;						\
	SAVE_FROM_REG(b5,_reg1,_reg2);;						\
	st8 [_reg1]=r1,0x08;;								\
	st8 [_reg1]=r12,0x08;;								\
	st8 [_reg1]=r13,0x08;;								\
	SAVE_FROM_REG(ar.fpsr,_reg1,_reg2);;				\
	SAVE_FROM_REG(ar.pfs,_reg1,_reg2);;					\
	SAVE_FROM_REG(ar.rnat,_reg1,_reg2);;				\
	SAVE_FROM_REG(ar.unat,_reg1,_reg2);;				\
	SAVE_FROM_REG(ar.bspstore,_reg1,_reg2);;			\
	SAVE_FROM_REG(cr.dcr,_reg1,_reg2);;					\
	SAVE_FROM_REG(cr.iva,_reg1,_reg2);;					\
	SAVE_FROM_REG(cr.pta,_reg1,_reg2);;					\
	SAVE_FROM_REG(cr.itv,_reg1,_reg2);;					\
	SAVE_FROM_REG(cr.pmv,_reg1,_reg2);;					\
	SAVE_FROM_REG(cr.cmcv,_reg1,_reg2);;				\
	SAVE_FROM_REG(cr.lrr0,_reg1,_reg2);;				\
	SAVE_FROM_REG(cr.lrr1,_reg1,_reg2);;				\
	st8 [_reg1]=r4,0x08;;								\
	st8 [_reg1]=r5,0x08;;								\
	st8 [_reg1]=r6,0x08;;								\
	st8 [_reg1]=r7,0x08;;								\
	st8 [_reg1]=_pred,0x08;;							\
	SAVE_FROM_REG(ar.lc, _reg1, _reg2);;				\
	stf.spill.nta [_reg1]=f2,16;;						\
	stf.spill.nta [_reg1]=f3,16;;						\
	stf.spill.nta [_reg1]=f4,16;;						\
	stf.spill.nta [_reg1]=f5,16;;						\
	stf.spill.nta [_reg1]=f16,16;;						\
	stf.spill.nta [_reg1]=f17,16;;						\
	stf.spill.nta [_reg1]=f18,16;;						\
	stf.spill.nta [_reg1]=f19,16;;						\
	stf.spill.nta [_reg1]=f20,16;;						\
	stf.spill.nta [_reg1]=f21,16;;						\
	stf.spill.nta [_reg1]=f22,16;;						\
	stf.spill.nta [_reg1]=f23,16;;						\
	stf.spill.nta [_reg1]=f24,16;;						\
	stf.spill.nta [_reg1]=f25,16;;						\
	stf.spill.nta [_reg1]=f26,16;;						\
	stf.spill.nta [_reg1]=f27,16;;						\
	stf.spill.nta [_reg1]=f28,16;;						\
	stf.spill.nta [_reg1]=f29,16;;						\
	stf.spill.nta [_reg1]=f30,16;;						\
	stf.spill.nta [_reg1]=f31,16;;

#else
#define SET_AREA_FOR_BOOTING_CPU(a1, a2)
#define SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(a1,a2, a3)
#define SAVE_REGION_REGS(_tmp, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
#define STORE_REGION_REGS(ptr, _r0, _r1, _r2, _r3, _r4, _r5, _r6, _r7)
#endif

#define SET_ONE_RR(num, pgsize, _tmp1, _tmp2, vhpt) \
	movl _tmp1=(num << 61);;	\
	mov _tmp2=((ia64_rid(IA64_REGION_ID_KERNEL, (num<<61)) << 8) | (pgsize << 2) | vhpt);; \
	mov rr[_tmp1]=_tmp2

	.section __special_page_section,"ax"

	.global empty_zero_page
empty_zero_page:
	.skip PAGE_SIZE

	.global swapper_pg_dir
swapper_pg_dir:
	.skip PAGE_SIZE

	.rodata
halt_msg:
	stringz "Halting kernel\n"

	.text

	.global start_ap

	/*
	 * Start the kernel.  When the bootloader passes control to _start(), r28
	 * points to the address of the boot parameter area.  Execution reaches
	 * here in physical mode.
	 */
GLOBAL_ENTRY(_start)
start_ap:
	.prologue
	.save rp, r0		// terminate unwind chain with a NULL rp
	.body

	rsm psr.i | psr.ic
	;;
	srlz.i
	;;
	/*
	 * Save the region registers, predicate before they get clobbered
	 */
	SAVE_REGION_REGS(r2, r8,r9,r10,r11,r12,r13,r14,r15);
	mov r25=pr;;

	/*
	 * Initialize kernel region registers:
	 *	rr[0]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[1]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[2]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[3]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[4]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[5]: VHPT enabled, page size = PAGE_SHIFT
	 *	rr[6]: VHPT disabled, page size = IA64_GRANULE_SHIFT
	 *	rr[7]: VHPT disabled, page size = IA64_GRANULE_SHIFT
	 * We initialize all of them to prevent inadvertently assuming
	 * something about the state of address translation early in boot.
	 */
	SET_ONE_RR(0, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(1, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(2, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(3, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(4, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(5, PAGE_SHIFT, r2, r16, 1);;
	SET_ONE_RR(6, IA64_GRANULE_SHIFT, r2, r16, 0);;
	SET_ONE_RR(7, IA64_GRANULE_SHIFT, r2, r16, 0);;
	/*
	 * Now pin mappings into the TLB for kernel text and data
	 */
	mov r18=KERNEL_TR_PAGE_SHIFT<<2
	movl r17=KERNEL_START
	;;
	mov cr.itir=r18
	mov cr.ifa=r17
	mov r16=IA64_TR_KERNEL
	mov r3=ip
	movl r18=PAGE_KERNEL
	;;
	dep r2=0,r3,0,KERNEL_TR_PAGE_SHIFT
	;;
	or r18=r2,r18
	;;
	srlz.i
	;;
	itr.i itr[r16]=r18
	;;
	itr.d dtr[r16]=r18
	;;
	srlz.i

	/*
	 * Switch into virtual mode:
	 */
	movl r16=(IA64_PSR_IT|IA64_PSR_IC|IA64_PSR_DT|IA64_PSR_RT|IA64_PSR_DFH|IA64_PSR_BN \
		  |IA64_PSR_DI)
	;;
	mov cr.ipsr=r16
	movl r17=1f
	;;
	mov cr.iip=r17
	mov cr.ifs=r0
	;;
	rfi
	;;
1:	// now we are in virtual mode

	SET_AREA_FOR_BOOTING_CPU(r2, r16);

	STORE_REGION_REGS(r16, r8,r9,r10,r11,r12,r13,r14,r15);
	SAL_TO_OS_BOOT_HANDOFF_STATE_SAVE(r16,r17,r25)
	;;

	// set IVT entry point---can't access I/O ports without it
	movl r3=ia64_ivt
	;;
	mov cr.iva=r3
	movl r2=FPSR_DEFAULT
	;;
	srlz.i
	movl gp=__gp

	mov ar.fpsr=r2
	;;

#define isAP	p2	// are we an Application Processor?
#define isBP	p3	// are we the Bootstrap Processor?

#ifdef CONFIG_SMP
	/*
	 * Find the init_task for the currently booting CPU.  At poweron, and in
	 * UP mode, task_for_booting_cpu is NULL.
	 */
	movl r3=task_for_booting_cpu
 	;;
	ld8 r3=[r3]
	movl r2=init_task
	;;
	cmp.eq isBP,isAP=r3,r0
	;;
(isAP)	mov r2=r3
#else
	movl r2=init_task
	cmp.eq isBP,isAP=r0,r0
#endif
	;;
	tpa r3=r2		// r3 == phys addr of task struct
	mov r16=-1
(isBP)	br.cond.dpnt .load_current // BP stack is on region 5 --- no need to map it

	// load mapping for stack (virtaddr in r2, physaddr in r3)
	rsm psr.ic
	movl r17=PAGE_KERNEL
	;;
	srlz.d
	dep r18=0,r3,0,12
	;;
	or r18=r17,r18
	dep r2=-1,r3,61,3	// IMVA of task
	;;
	mov r17=rr[r2]
	shr.u r16=r3,IA64_GRANULE_SHIFT
	;;
	dep r17=0,r17,8,24
	;;
	mov cr.itir=r17
	mov cr.ifa=r2

	mov r19=IA64_TR_CURRENT_STACK
	;;
	itr.d dtr[r19]=r18
	;;
	ssm psr.ic
	srlz.d
  	;;

.load_current:
	// load the "current" pointer (r13) and ar.k6 with the current task
	mov IA64_KR(CURRENT)=r2		// virtual address
	mov IA64_KR(CURRENT_STACK)=r16
	mov r13=r2
	/*
	 * Reserve space at the top of the stack for "struct pt_regs".  Kernel
	 * threads don't store interesting values in that structure, but the space
	 * still needs to be there because time-critical stuff such as the context
	 * switching can be implemented more efficiently (for example, __switch_to()
	 * always sets the psr.dfh bit of the task it is switching to).
	 */

	addl r12=IA64_STK_OFFSET-IA64_PT_REGS_SIZE-16,r2
	addl r2=IA64_RBS_OFFSET,r2	// initialize the RSE
	mov ar.rsc=0		// place RSE in enforced lazy mode
	;;
	loadrs			// clear the dirty partition
	;;
	mov ar.bspstore=r2	// establish the new RSE stack
	;;
	mov ar.rsc=0x3		// place RSE in eager mode

(isBP)	dep r28=-1,r28,61,3	// make address virtual
(isBP)	movl r2=ia64_boot_param
	;;
(isBP)	st8 [r2]=r28		// save the address of the boot param area passed by the bootloader

#ifdef CONFIG_SMP
(isAP)	br.call.sptk.many rp=start_secondary
.ret0:
(isAP)	br.cond.sptk self
#endif

	// This is executed by the bootstrap processor (bsp) only:

#ifdef CONFIG_IA64_FW_EMU
	// initialize PAL & SAL emulator:
	br.call.sptk.many rp=sys_fw_init
.ret1:
#endif
	br.call.sptk.many rp=start_kernel
.ret2:	addl r3=@ltoff(halt_msg),gp
	;;
	alloc r2=ar.pfs,8,0,2,0
	;;
	ld8 out0=[r3]
	br.call.sptk.many b0=console_print

self:	hint @pause
	br.sptk.many self		// endless loop
END(_start)

GLOBAL_ENTRY(ia64_save_debug_regs)
	alloc r16=ar.pfs,1,0,0,0
	mov r20=ar.lc			// preserve ar.lc
	mov ar.lc=IA64_NUM_DBG_REGS-1
	mov r18=0
	add r19=IA64_NUM_DBG_REGS*8,in0
	;;
1:	mov r16=dbr[r18]
#ifdef CONFIG_ITANIUM
	;;
	srlz.d
#endif
	mov r17=ibr[r18]
	add r18=1,r18
	;;
	st8.nta [in0]=r16,8
	st8.nta [r19]=r17,8
	br.cloop.sptk.many 1b
	;;
	mov ar.lc=r20			// restore ar.lc
	br.ret.sptk.many rp
END(ia64_save_debug_regs)

GLOBAL_ENTRY(ia64_load_debug_regs)
	alloc r16=ar.pfs,1,0,0,0
	lfetch.nta [in0]
	mov r20=ar.lc			// preserve ar.lc
	add r19=IA64_NUM_DBG_REGS*8,in0
	mov ar.lc=IA64_NUM_DBG_REGS-1
	mov r18=-1
	;;
1:	ld8.nta r16=[in0],8
	ld8.nta r17=[r19],8
	add r18=1,r18
	;;
	mov dbr[r18]=r16
#ifdef CONFIG_ITANIUM
	;;
	srlz.d				// Errata 132 (NoFix status)
#endif
	mov ibr[r18]=r17
	br.cloop.sptk.many 1b
	;;
	mov ar.lc=r20			// restore ar.lc
	br.ret.sptk.many rp
END(ia64_load_debug_regs)

GLOBAL_ENTRY(__ia64_save_fpu)
	alloc r2=ar.pfs,1,4,0,0
	adds loc0=96*16-16,in0
	adds loc1=96*16-16-128,in0
	;;
	stf.spill.nta [loc0]=f127,-256
	stf.spill.nta [loc1]=f119,-256
	;;
	stf.spill.nta [loc0]=f111,-256
	stf.spill.nta [loc1]=f103,-256
	;;
	stf.spill.nta [loc0]=f95,-256
	stf.spill.nta [loc1]=f87,-256
	;;
	stf.spill.nta [loc0]=f79,-256
	stf.spill.nta [loc1]=f71,-256
	;;
	stf.spill.nta [loc0]=f63,-256
	stf.spill.nta [loc1]=f55,-256
	adds loc2=96*16-32,in0
	;;
	stf.spill.nta [loc0]=f47,-256
	stf.spill.nta [loc1]=f39,-256
	adds loc3=96*16-32-128,in0
	;;
	stf.spill.nta [loc2]=f126,-256
	stf.spill.nta [loc3]=f118,-256
	;;
	stf.spill.nta [loc2]=f110,-256
	stf.spill.nta [loc3]=f102,-256
	;;
	stf.spill.nta [loc2]=f94,-256
	stf.spill.nta [loc3]=f86,-256
	;;
	stf.spill.nta [loc2]=f78,-256
	stf.spill.nta [loc3]=f70,-256
	;;
	stf.spill.nta [loc2]=f62,-256
	stf.spill.nta [loc3]=f54,-256
	adds loc0=96*16-48,in0
	;;
	stf.spill.nta [loc2]=f46,-256
	stf.spill.nta [loc3]=f38,-256
	adds loc1=96*16-48-128,in0
	;;
	stf.spill.nta [loc0]=f125,-256
	stf.spill.nta [loc1]=f117,-256
	;;
	stf.spill.nta [loc0]=f109,-256
	stf.spill.nta [loc1]=f101,-256
	;;
	stf.spill.nta [loc0]=f93,-256
	stf.spill.nta [loc1]=f85,-256
	;;
	stf.spill.nta [loc0]=f77,-256
	stf.spill.nta [loc1]=f69,-256
	;;
	stf.spill.nta [loc0]=f61,-256
	stf.spill.nta [loc1]=f53,-256
	adds loc2=96*16-64,in0
	;;
	stf.spill.nta [loc0]=f45,-256
	stf.spill.nta [loc1]=f37,-256
	adds loc3=96*16-64-128,in0
	;;
	stf.spill.nta [loc2]=f124,-256
	stf.spill.nta [loc3]=f116,-256
	;;
	stf.spill.nta [loc2]=f108,-256
	stf.spill.nta [loc3]=f100,-256
	;;
	stf.spill.nta [loc2]=f92,-256
	stf.spill.nta [loc3]=f84,-256
	;;
	stf.spill.nta [loc2]=f76,-256
	stf.spill.nta [loc3]=f68,-256
	;;
	stf.spill.nta [loc2]=f60,-256
	stf.spill.nta [loc3]=f52,-256
	adds loc0=96*16-80,in0
	;;
	stf.spill.nta [loc2]=f44,-256
	stf.spill.nta [loc3]=f36,-256
	adds loc1=96*16-80-128,in0
	;;
	stf.spill.nta [loc0]=f123,-256
	stf.spill.nta [loc1]=f115,-256
	;;
	stf.spill.nta [loc0]=f107,-256
	stf.spill.nta [loc1]=f99,-256
	;;
	stf.spill.nta [loc0]=f91,-256
	stf.spill.nta [loc1]=f83,-256
	;;
	stf.spill.nta [loc0]=f75,-256
	stf.spill.nta [loc1]=f67,-256
	;;
	stf.spill.nta [loc0]=f59,-256
	stf.spill.nta [loc1]=f51,-256
	adds loc2=96*16-96,in0
	;;
	stf.spill.nta [loc0]=f43,-256
	stf.spill.nta [loc1]=f35,-256
	adds loc3=96*16-96-128,in0
	;;
	stf.spill.nta [loc2]=f122,-256
	stf.spill.nta [loc3]=f114,-256
	;;
	stf.spill.nta [loc2]=f106,-256
	stf.spill.nta [loc3]=f98,-256
	;;
	stf.spill.nta [loc2]=f90,-256
	stf.spill.nta [loc3]=f82,-256
	;;
	stf.spill.nta [loc2]=f74,-256
	stf.spill.nta [loc3]=f66,-256
	;;
	stf.spill.nta [loc2]=f58,-256
	stf.spill.nta [loc3]=f50,-256
	adds loc0=96*16-112,in0
	;;
	stf.spill.nta [loc2]=f42,-256
	stf.spill.nta [loc3]=f34,-256
	adds loc1=96*16-112-128,in0
	;;
	stf.spill.nta [loc0]=f121,-256
	stf.spill.nta [loc1]=f113,-256
	;;
	stf.spill.nta [loc0]=f105,-256
	stf.spill.nta [loc1]=f97,-256
	;;
	stf.spill.nta [loc0]=f89,-256
	stf.spill.nta [loc1]=f81,-256
	;;
	stf.spill.nta [loc0]=f73,-256
	stf.spill.nta [loc1]=f65,-256
	;;
	stf.spill.nta [loc0]=f57,-256
	stf.spill.nta [loc1]=f49,-256
	adds loc2=96*16-128,in0
	;;
	stf.spill.nta [loc0]=f41,-256
	stf.spill.nta [loc1]=f33,-256
	adds loc3=96*16-128-128,in0
	;;
	stf.spill.nta [loc2]=f120,-256
	stf.spill.nta [loc3]=f112,-256
	;;
	stf.spill.nta [loc2]=f104,-256
	stf.spill.nta [loc3]=f96,-256
	;;
	stf.spill.nta [loc2]=f88,-256
	stf.spill.nta [loc3]=f80,-256
	;;
	stf.spill.nta [loc2]=f72,-256
	stf.spill.nta [loc3]=f64,-256
	;;
	stf.spill.nta [loc2]=f56,-256
	stf.spill.nta [loc3]=f48,-256
	;;
	stf.spill.nta [loc2]=f40
	stf.spill.nta [loc3]=f32
	br.ret.sptk.many rp
END(__ia64_save_fpu)

GLOBAL_ENTRY(__ia64_load_fpu)
	alloc r2=ar.pfs,1,2,0,0
	adds r3=128,in0
	adds r14=256,in0
	adds r15=384,in0
	mov loc0=512
	mov loc1=-1024+16
	;;
	ldf.fill.nta f32=[in0],loc0
	ldf.fill.nta f40=[ r3],loc0