entry64.s

linux 内核源代码

/*
 *  arch/s390/kernel/entry64.S
 *    S390 low-level entry points.
 *
 *    Copyright (C) IBM Corp. 1999,2006
 *    Author(s): Martin Schwidefsky ([e-mail]),
 *		 Hartmut Penner ([e-mail]),
 *		 Denis Joseph Barrow ([e-mail],[e-mail]),
 *		 Heiko Carstens <[e-mail]>
 */

#include <linux/sys.h>
#include <linux/linkage.h>
#include <asm/cache.h>
#include <asm/lowcore.h>
#include <asm/errno.h>
#include <asm/ptrace.h>
#include <asm/thread_info.h>
#include <asm/asm-offsets.h>
#include <asm/unistd.h>
#include <asm/page.h>

/*
 * Stack layout for the system_call stack entry.
 * The first few entries are identical to the user_regs_struct.
 */
SP_PTREGS    =	STACK_FRAME_OVERHEAD
SP_ARGS      =	STACK_FRAME_OVERHEAD + __PT_ARGS
SP_PSW	     =	STACK_FRAME_OVERHEAD + __PT_PSW
SP_R0	     =	STACK_FRAME_OVERHEAD + __PT_GPRS
SP_R1	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 8
SP_R2	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 16
SP_R3	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 24
SP_R4	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 32
SP_R5	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 40
SP_R6	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 48
SP_R7	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 56
SP_R8	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 64
SP_R9	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 72
SP_R10	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 80
SP_R11	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 88
SP_R12	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 96
SP_R13	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 104
SP_R14	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 112
SP_R15	     =	STACK_FRAME_OVERHEAD + __PT_GPRS + 120
SP_ORIG_R2   =	STACK_FRAME_OVERHEAD + __PT_ORIG_GPR2
SP_ILC	     =	STACK_FRAME_OVERHEAD + __PT_ILC
SP_TRAP      =	STACK_FRAME_OVERHEAD + __PT_TRAP
SP_SIZE      =	STACK_FRAME_OVERHEAD + __PT_SIZE

STACK_SHIFT = PAGE_SHIFT + THREAD_ORDER
STACK_SIZE  = 1 << STACK_SHIFT

_TIF_WORK_SVC = (_TIF_SIGPENDING | _TIF_RESTORE_SIGMASK | _TIF_NEED_RESCHED | \
		 _TIF_MCCK_PENDING | _TIF_RESTART_SVC | _TIF_SINGLE_STEP )
_TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_RESTORE_SIGMASK | _TIF_NEED_RESCHED | \
		 _TIF_MCCK_PENDING)

#define BASED(name) name-system_call(%r13)

#ifdef CONFIG_TRACE_IRQFLAGS
	.macro	TRACE_IRQS_ON
	 brasl	%r14,trace_hardirqs_on
	.endm

	.macro	TRACE_IRQS_OFF
	 brasl	%r14,trace_hardirqs_off
	.endm

	.macro TRACE_IRQS_CHECK
	tm	SP_PSW(%r15),0x03	# irqs enabled?
	jz	0f
	brasl	%r14,trace_hardirqs_on
	j	1f
0:	brasl	%r14,trace_hardirqs_off
1:
	.endm
#else
#define TRACE_IRQS_ON
#define TRACE_IRQS_OFF
#define TRACE_IRQS_CHECK
#endif

#ifdef CONFIG_LOCKDEP
	.macro	LOCKDEP_SYS_EXIT
	tm	SP_PSW+1(%r15),0x01	# returning to user ?
	jz	0f
	brasl	%r14,lockdep_sys_exit
0:
	.endm
#else
#define LOCKDEP_SYS_EXIT
#endif

	.macro	STORE_TIMER lc_offset
#ifdef CONFIG_VIRT_CPU_ACCOUNTING
	stpt	\lc_offset
#endif
	.endm

#ifdef CONFIG_VIRT_CPU_ACCOUNTING
	.macro	UPDATE_VTIME lc_from,lc_to,lc_sum
	lg	%r10,\lc_from
	slg	%r10,\lc_to
	alg	%r10,\lc_sum
	stg	%r10,\lc_sum
	.endm
#endif

/*
 * Register usage in interrupt handlers:
 *    R9  - pointer to current task structure
 *    R13 - pointer to literal pool
 *    R14 - return register for function calls
 *    R15 - kernel stack pointer
 */

	.macro	SAVE_ALL_BASE savearea
	stmg	%r12,%r15,\savearea
	larl	%r13,system_call
	.endm

	.macro	SAVE_ALL_SVC psworg,savearea
	la	%r12,\psworg
	lg	%r15,__LC_KERNEL_STACK	# problem state -> load ksp
	.endm

	.macro	SAVE_ALL_SYNC psworg,savearea
	la	%r12,\psworg
	tm	\psworg+1,0x01		# test problem state bit
	jz	2f			# skip stack setup save
	lg	%r15,__LC_KERNEL_STACK	# problem state -> load ksp
#ifdef CONFIG_CHECK_STACK
	j	3f
2:	tml	%r15,STACK_SIZE - CONFIG_STACK_GUARD
	jz	stack_overflow
3:
#endif
2:
	.endm

	.macro	SAVE_ALL_ASYNC psworg,savearea
	la	%r12,\psworg
	tm	\psworg+1,0x01		# test problem state bit
	jnz	1f			# from user -> load kernel stack
	clc	\psworg+8(8),BASED(.Lcritical_end)
	jhe	0f
	clc	\psworg+8(8),BASED(.Lcritical_start)
	jl	0f
	brasl	%r14,cleanup_critical
	tm	1(%r12),0x01		# retest problem state after cleanup
	jnz	1f
0:	lg	%r14,__LC_ASYNC_STACK	# are we already on the async. stack ?
	slgr	%r14,%r15
	srag	%r14,%r14,STACK_SHIFT
	jz	2f
1:	lg	%r15,__LC_ASYNC_STACK	# load async stack
#ifdef CONFIG_CHECK_STACK
	j	3f
2:	tml	%r15,STACK_SIZE - CONFIG_STACK_GUARD
	jz	stack_overflow
3:
#endif
2:
	.endm

	.macro	CREATE_STACK_FRAME psworg,savearea
	aghi	%r15,-SP_SIZE		# make room for registers & psw
	mvc	SP_PSW(16,%r15),0(%r12)	# move user PSW to stack
	la	%r12,\psworg
	stg	%r2,SP_ORIG_R2(%r15)	# store original content of gpr 2
	icm	%r12,12,__LC_SVC_ILC
	stmg	%r0,%r11,SP_R0(%r15)	# store gprs %r0-%r11 to kernel stack
	st	%r12,SP_ILC(%r15)
	mvc	SP_R12(32,%r15),\savearea # move %r12-%r15 to stack
	la	%r12,0
	stg	%r12,__SF_BACKCHAIN(%r15)
	.endm

	.macro	RESTORE_ALL psworg,sync
	mvc	\psworg(16),SP_PSW(%r15) # move user PSW to lowcore
	.if !\sync
	ni	\psworg+1,0xfd		# clear wait state bit
	.endif
	lmg	%r0,%r15,SP_R0(%r15)	# load gprs 0-15 of user
	STORE_TIMER __LC_EXIT_TIMER
	lpswe	\psworg			# back to caller
	.endm

/*
 * Scheduler resume function, called by switch_to
 *  gpr2 = (task_struct *) prev
 *  gpr3 = (task_struct *) next
 * Returns:
 *  gpr2 = prev
 */
	.globl	__switch_to
__switch_to:
	tm	__THREAD_per+4(%r3),0xe8 # is the new process using per ?
	jz	__switch_to_noper		# if not we're fine
	stctg	%c9,%c11,__SF_EMPTY(%r15)# We are using per stuff
	clc	__THREAD_per(24,%r3),__SF_EMPTY(%r15)
	je	__switch_to_noper	     # we got away without bashing TLB's
	lctlg	%c9,%c11,__THREAD_per(%r3)	# Nope we didn't
__switch_to_noper:
	lg	%r4,__THREAD_info(%r2)		    # get thread_info of prev
	tm	__TI_flags+7(%r4),_TIF_MCCK_PENDING # machine check pending?
	jz	__switch_to_no_mcck
	ni	__TI_flags+7(%r4),255-_TIF_MCCK_PENDING # clear flag in prev
	lg	%r4,__THREAD_info(%r3)		    # get thread_info of next
	oi	__TI_flags+7(%r4),_TIF_MCCK_PENDING # set it in next
__switch_to_no_mcck:
	stmg	%r6,%r15,__SF_GPRS(%r15)# store __switch_to registers of prev task
	stg	%r15,__THREAD_ksp(%r2)	# store kernel stack to prev->tss.ksp
	lg	%r15,__THREAD_ksp(%r3)	# load kernel stack from next->tss.ksp
	lmg	%r6,%r15,__SF_GPRS(%r15)# load __switch_to registers of next task
	stg	%r3,__LC_CURRENT	# __LC_CURRENT = current task struct
	lctl	%c4,%c4,__TASK_pid(%r3) # load pid to control reg. 4
	lg	%r3,__THREAD_info(%r3)	# load thread_info from task struct
	stg	%r3,__LC_THREAD_INFO
	aghi	%r3,STACK_SIZE
	stg	%r3,__LC_KERNEL_STACK	# __LC_KERNEL_STACK = new kernel stack
	br	%r14

__critical_start:
/*
 * SVC interrupt handler routine. System calls are synchronous events and
 * are executed with interrupts enabled.
 */

	.globl	system_call
system_call:
	STORE_TIMER __LC_SYNC_ENTER_TIMER
sysc_saveall:
	SAVE_ALL_BASE __LC_SAVE_AREA
	SAVE_ALL_SVC __LC_SVC_OLD_PSW,__LC_SAVE_AREA
	CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
	llgh	%r7,__LC_SVC_INT_CODE	# get svc number from lowcore
#ifdef CONFIG_VIRT_CPU_ACCOUNTING
sysc_vtime:
	UPDATE_VTIME __LC_EXIT_TIMER,__LC_SYNC_ENTER_TIMER,__LC_USER_TIMER
sysc_stime:
	UPDATE_VTIME __LC_LAST_UPDATE_TIMER,__LC_EXIT_TIMER,__LC_SYSTEM_TIMER
sysc_update:
	mvc	__LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
#endif
sysc_do_svc:
	lg	%r9,__LC_THREAD_INFO	# load pointer to thread_info struct
	slag	%r7,%r7,2	# *4 and test for svc 0
	jnz	sysc_nr_ok
	# svc 0: system call number in %r1
	cl	%r1,BASED(.Lnr_syscalls)
	jnl	sysc_nr_ok
	lgfr	%r7,%r1 	# clear high word in r1
	slag	%r7,%r7,2	# svc 0: system call number in %r1
sysc_nr_ok:
	mvc	SP_ARGS(8,%r15),SP_R7(%r15)
sysc_do_restart:
	larl	%r10,sys_call_table
#ifdef CONFIG_COMPAT
	tm	__TI_flags+5(%r9),(_TIF_31BIT>>16)  # running in 31 bit mode ?
	jno	sysc_noemu
	larl	%r10,sys_call_table_emu  # use 31 bit emulation system calls
sysc_noemu:
#endif
	tm	__TI_flags+7(%r9),(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT)
	lgf	%r8,0(%r7,%r10) # load address of system call routine
	jnz	sysc_tracesys
	basr	%r14,%r8	# call sys_xxxx
	stg	%r2,SP_R2(%r15) # store return value (change R2 on stack)

sysc_return:
	tm	SP_PSW+1(%r15),0x01	# returning to user ?
	jno	sysc_restore
	tm	__TI_flags+7(%r9),_TIF_WORK_SVC
	jnz	sysc_work	# there is work to do (signals etc.)
sysc_restore:
#ifdef CONFIG_TRACE_IRQFLAGS
	larl	%r1,sysc_restore_trace_psw
	lpswe	0(%r1)
sysc_restore_trace:
	TRACE_IRQS_CHECK
	LOCKDEP_SYS_EXIT
#endif
sysc_leave:
	RESTORE_ALL __LC_RETURN_PSW,1
sysc_done:

#ifdef CONFIG_TRACE_IRQFLAGS
	.align	8
	.globl sysc_restore_trace_psw
sysc_restore_trace_psw:
	.quad	0, sysc_restore_trace
#endif

#
# recheck if there is more work to do
#
sysc_work_loop:
	tm	__TI_flags+7(%r9),_TIF_WORK_SVC
	jz	sysc_restore	  # there is no work to do
#
# One of the work bits is on. Find out which one.
#
sysc_work:
	tm	__TI_flags+7(%r9),_TIF_MCCK_PENDING
	jo	sysc_mcck_pending
	tm	__TI_flags+7(%r9),_TIF_NEED_RESCHED
	jo	sysc_reschedule
	tm	__TI_flags+7(%r9),(_TIF_SIGPENDING | _TIF_RESTORE_SIGMASK)
	jnz	sysc_sigpending
	tm	__TI_flags+7(%r9),_TIF_RESTART_SVC
	jo	sysc_restart
	tm	__TI_flags+7(%r9),_TIF_SINGLE_STEP
	jo	sysc_singlestep
	j	sysc_restore
sysc_work_done:

#
# _TIF_NEED_RESCHED is set, call schedule
#
sysc_reschedule:
	larl	%r14,sysc_work_loop
	jg	schedule	# return point is sysc_return

#
# _TIF_MCCK_PENDING is set, call handler
#
sysc_mcck_pending:
	larl	%r14,sysc_work_loop
	jg	s390_handle_mcck	# TIF bit will be cleared by handler

#
# _TIF_SIGPENDING or _TIF_RESTORE_SIGMASK is set, call do_signal
#
sysc_sigpending:
	ni	__TI_flags+7(%r9),255-_TIF_SINGLE_STEP # clear TIF_SINGLE_STEP
	la	%r2,SP_PTREGS(%r15)	# load pt_regs
	brasl	%r14,do_signal		# call do_signal
	tm	__TI_flags+7(%r9),_TIF_RESTART_SVC
	jo	sysc_restart
	tm	__TI_flags+7(%r9),_TIF_SINGLE_STEP
	jo	sysc_singlestep
	j	sysc_work_loop

#
# _TIF_RESTART_SVC is set, set up registers and restart svc
#
sysc_restart:
	ni	__TI_flags+7(%r9),255-_TIF_RESTART_SVC # clear TIF_RESTART_SVC
	lg	%r7,SP_R2(%r15)		# load new svc number
	slag	%r7,%r7,2		# *4
	mvc	SP_R2(8,%r15),SP_ORIG_R2(%r15) # restore first argument
	lmg	%r2,%r6,SP_R2(%r15)	# load svc arguments
	j	sysc_do_restart 	# restart svc

#
# _TIF_SINGLE_STEP is set, call do_single_step
#
sysc_singlestep:
	ni	__TI_flags+7(%r9),255-_TIF_SINGLE_STEP # clear TIF_SINGLE_STEP
	lhi	%r0,__LC_PGM_OLD_PSW
	sth	%r0,SP_TRAP(%r15)	# set trap indication to pgm check
	la	%r2,SP_PTREGS(%r15)	# address of register-save area
	larl	%r14,sysc_return	# load adr. of system return
	jg	do_single_step		# branch to do_sigtrap

#
# call syscall_trace before and after system call
# special linkage: %r12 contains the return address for trace_svc
#
sysc_tracesys:
	la	%r2,SP_PTREGS(%r15)	# load pt_regs
	la	%r3,0
	srl	%r7,2
	stg	%r7,SP_R2(%r15)
	brasl	%r14,syscall_trace
	lghi	%r0,NR_syscalls
	clg	%r0,SP_R2(%r15)
	jnh	sysc_tracenogo
	lg	%r7,SP_R2(%r15)		# strace might have changed the
	sll	%r7,2			# system call
	lgf	%r8,0(%r7,%r10)
sysc_tracego:
	lmg	%r3,%r6,SP_R3(%r15)
	lg	%r2,SP_ORIG_R2(%r15)
	basr	%r14,%r8		# call sys_xxx
	stg	%r2,SP_R2(%r15)		# store return value
sysc_tracenogo:
	tm	__TI_flags+7(%r9),(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT)
	jz	sysc_return
	la	%r2,SP_PTREGS(%r15)	# load pt_regs
	la	%r3,1
	larl	%r14,sysc_return	# return point is sysc_return
	jg	syscall_trace

#
# a new process exits the kernel with ret_from_fork
#
	.globl	ret_from_fork
ret_from_fork:
	lg	%r13,__LC_SVC_NEW_PSW+8
	lg	%r9,__LC_THREAD_INFO	# load pointer to thread_info struct
	tm	SP_PSW+1(%r15),0x01	# forking a kernel thread ?
	jo	0f
	stg	%r15,SP_R15(%r15)	# store stack pointer for new kthread
0:	brasl	%r14,schedule_tail
	TRACE_IRQS_ON
	stosm	24(%r15),0x03		# reenable interrupts
	j	sysc_return

#
# kernel_execve function needs to deal with pt_regs that is not
# at the usual place
#
	.globl	kernel_execve
kernel_execve:
	stmg	%r12,%r15,96(%r15)
	lgr	%r14,%r15
	aghi	%r15,-SP_SIZE
	stg	%r14,__SF_BACKCHAIN(%r15)
	la	%r12,SP_PTREGS(%r15)
	xc	0(__PT_SIZE,%r12),0(%r12)
	lgr	%r5,%r12
	brasl	%r14,do_execve
	ltgfr	%r2,%r2
	je	0f
	aghi	%r15,SP_SIZE
	lmg	%r12,%r15,96(%r15)
	br	%r14
	# execve succeeded.
0:	stnsm	__SF_EMPTY(%r15),0xfc	# disable interrupts
	lg	%r15,__LC_KERNEL_STACK	# load ksp
	aghi	%r15,-SP_SIZE		# make room for registers & psw
	lg	%r13,__LC_SVC_NEW_PSW+8
	lg	%r9,__LC_THREAD_INFO
	mvc	SP_PTREGS(__PT_SIZE,%r15),0(%r12)	# copy pt_regs
	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
	stosm	__SF_EMPTY(%r15),0x03	# reenable interrupts
	brasl	%r14,execve_tail
	j	sysc_return

/*
 * Program check handler routine
 */

	.globl	pgm_check_handler
pgm_check_handler:
/*
 * First we need to check for a special case:
 * Single stepping an instruction that disables the PER event mask will
 * cause a PER event AFTER the mask has been set. Example: SVC or LPSW.
 * For a single stepped SVC the program check handler gets control after
 * the SVC new PSW has been loaded. But we want to execute the SVC first and
 * then handle the PER event. Therefore we update the SVC old PSW to point
 * to the pgm_check_handler and branch to the SVC handler after we checked
 * if we have to load the kernel stack register.
 * For every other possible cause for PER event without the PER mask set
 * we just ignore the PER event (FIXME: is there anything we have to do
 * for LPSW?).
 */
	STORE_TIMER __LC_SYNC_ENTER_TIMER
	SAVE_ALL_BASE __LC_SAVE_AREA
	tm	__LC_PGM_INT_CODE+1,0x80 # check whether we got a per exception
	jnz	pgm_per 		 # got per exception -> special case
	SAVE_ALL_SYNC __LC_PGM_OLD_PSW,__LC_SAVE_AREA
	CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
#ifdef CONFIG_VIRT_CPU_ACCOUNTING
	tm	SP_PSW+1(%r15),0x01	# interrupting from user ?
	jz	pgm_no_vtime
	UPDATE_VTIME __LC_EXIT_TIMER,__LC_SYNC_ENTER_TIMER,__LC_USER_TIMER
	UPDATE_VTIME __LC_LAST_UPDATE_TIMER,__LC_EXIT_TIMER,__LC_SYSTEM_TIMER
	mvc	__LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
pgm_no_vtime:
#endif
	lg	%r9,__LC_THREAD_INFO	# load pointer to thread_info struct
	TRACE_IRQS_OFF
	lgf	%r3,__LC_PGM_ILC	# load program interruption code
	lghi	%r8,0x7f
	ngr	%r8,%r3
pgm_do_call:
	sll	%r8,3
	larl	%r1,pgm_check_table
	lg	%r1,0(%r8,%r1)		# load address of handler routine
	la	%r2,SP_PTREGS(%r15)	# address of register-save area
	larl	%r14,sysc_return
	br	%r1			# branch to interrupt-handler

#
# handle per exception
#
pgm_per:
	tm	__LC_PGM_OLD_PSW,0x40	# test if per event recording is on
	jnz	pgm_per_std		# ok, normal per event from user space
# ok its one of the special cases, now we need to find out which one
	clc	__LC_PGM_OLD_PSW(16),__LC_SVC_NEW_PSW
	je	pgm_svcper
# no interesting special case, ignore PER event
	lmg	%r12,%r15,__LC_SAVE_AREA
	lpswe	__LC_PGM_OLD_PSW

#
# Normal per exception
#
pgm_per_std:
	SAVE_ALL_SYNC __LC_PGM_OLD_PSW,__LC_SAVE_AREA
	CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
#ifdef CONFIG_VIRT_CPU_ACCOUNTING
	tm	SP_PSW+1(%r15),0x01	# interrupting from user ?
	jz	pgm_no_vtime2
	UPDATE_VTIME __LC_EXIT_TIMER,__LC_SYNC_ENTER_TIMER,__LC_USER_TIMER
	UPDATE_VTIME __LC_LAST_UPDATE_TIMER,__LC_EXIT_TIMER,__LC_SYSTEM_TIMER
	mvc	__LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
pgm_no_vtime2:
#endif
	lg	%r9,__LC_THREAD_INFO	# load pointer to thread_info struct
	TRACE_IRQS_OFF
	lg	%r1,__TI_task(%r9)
	tm	SP_PSW+1(%r15),0x01	# kernel per event ?
	jz	kernel_per
	mvc	__THREAD_per+__PER_atmid(2,%r1),__LC_PER_ATMID
	mvc	__THREAD_per+__PER_address(8,%r1),__LC_PER_ADDRESS
	mvc	__THREAD_per+__PER_access_id(1,%r1),__LC_PER_ACCESS_ID
	oi	__TI_flags+7(%r9),_TIF_SINGLE_STEP # set TIF_SINGLE_STEP
	lgf	%r3,__LC_PGM_ILC	# load program interruption code