linux_dsm_epyc7002/arch/s390/kernel/entry.S

967 lines
26 KiB
ArmAsm
Raw Normal View History

/*
* S390 low-level entry points.
*
* Copyright IBM Corp. 1999, 2012
* Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com),
* Hartmut Penner (hp@de.ibm.com),
* Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
* Heiko Carstens <heiko.carstens@de.ibm.com>
*/
#include <linux/init.h>
#include <linux/linkage.h>
#include <asm/processor.h>
#include <asm/cache.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>
#include <asm/sigp.h>
#include <asm/irq.h>
__PT_R0 = __PT_GPRS
__PT_R1 = __PT_GPRS + 4
__PT_R2 = __PT_GPRS + 8
__PT_R3 = __PT_GPRS + 12
__PT_R4 = __PT_GPRS + 16
__PT_R5 = __PT_GPRS + 20
__PT_R6 = __PT_GPRS + 24
__PT_R7 = __PT_GPRS + 28
__PT_R8 = __PT_GPRS + 32
__PT_R9 = __PT_GPRS + 36
__PT_R10 = __PT_GPRS + 40
__PT_R11 = __PT_GPRS + 44
__PT_R12 = __PT_GPRS + 48
__PT_R13 = __PT_GPRS + 524
__PT_R14 = __PT_GPRS + 56
__PT_R15 = __PT_GPRS + 60
STACK_SHIFT = PAGE_SHIFT + THREAD_ORDER
STACK_SIZE = 1 << STACK_SHIFT
STACK_INIT = STACK_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE
_TIF_WORK = (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_NEED_RESCHED)
_TIF_TRACE = (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | _TIF_SECCOMP | \
_TIF_SYSCALL_TRACEPOINT)
_CIF_WORK = (_CIF_MCCK_PENDING | _CIF_ASCE)
_PIF_WORK = (_PIF_PER_TRAP)
#define BASED(name) name-system_call(%r13)
.macro TRACE_IRQS_ON
#ifdef CONFIG_TRACE_IRQFLAGS
basr %r2,%r0
l %r1,BASED(.Lc_hardirqs_on)
basr %r14,%r1 # call trace_hardirqs_on_caller
#endif
.endm
.macro TRACE_IRQS_OFF
#ifdef CONFIG_TRACE_IRQFLAGS
basr %r2,%r0
l %r1,BASED(.Lc_hardirqs_off)
basr %r14,%r1 # call trace_hardirqs_off_caller
#endif
.endm
.macro LOCKDEP_SYS_EXIT
#ifdef CONFIG_LOCKDEP
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jz .+10
l %r1,BASED(.Lc_lockdep_sys_exit)
basr %r14,%r1 # call lockdep_sys_exit
#endif
.endm
.macro CHECK_STACK stacksize,savearea
#ifdef CONFIG_CHECK_STACK
tml %r15,\stacksize - CONFIG_STACK_GUARD
la %r14,\savearea
jz stack_overflow
#endif
.endm
.macro SWITCH_ASYNC savearea,stack,shift
tmh %r8,0x0001 # interrupting from user ?
jnz 1f
lr %r14,%r9
sl %r14,BASED(.Lc_critical_start)
cl %r14,BASED(.Lc_critical_length)
jhe 0f
la %r11,\savearea # inside critical section, do cleanup
bras %r14,cleanup_critical
tmh %r8,0x0001 # retest problem state after cleanup
jnz 1f
0: l %r14,\stack # are we already on the target stack?
slr %r14,%r15
sra %r14,\shift
jnz 1f
CHECK_STACK 1<<\shift,\savearea
ahi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j 2f
1: l %r15,\stack # load target stack
2: la %r11,STACK_FRAME_OVERHEAD(%r15)
.endm
.macro ADD64 high,low,timer
al \high,\timer
al \low,4+\timer
brc 12,.+8
ahi \high,1
.endm
.macro SUB64 high,low,timer
sl \high,\timer
sl \low,4+\timer
brc 3,.+8
ahi \high,-1
.endm
.macro UPDATE_VTIME high,low,enter_timer
lm \high,\low,__LC_EXIT_TIMER
SUB64 \high,\low,\enter_timer
ADD64 \high,\low,__LC_USER_TIMER
stm \high,\low,__LC_USER_TIMER
lm \high,\low,__LC_LAST_UPDATE_TIMER
SUB64 \high,\low,__LC_EXIT_TIMER
ADD64 \high,\low,__LC_SYSTEM_TIMER
stm \high,\low,__LC_SYSTEM_TIMER
mvc __LC_LAST_UPDATE_TIMER(8),\enter_timer
.endm
.macro REENABLE_IRQS
st %r8,__LC_RETURN_PSW
ni __LC_RETURN_PSW,0xbf
ssm __LC_RETURN_PSW
.endm
.section .kprobes.text, "ax"
/*
* Scheduler resume function, called by switch_to
* gpr2 = (task_struct *) prev
* gpr3 = (task_struct *) next
* Returns:
* gpr2 = prev
*/
ENTRY(__switch_to)
stm %r6,%r15,__SF_GPRS(%r15) # store gprs of prev task
st %r15,__THREAD_ksp(%r2) # store kernel stack of prev
l %r4,__THREAD_info(%r2) # get thread_info of prev
l %r5,__THREAD_info(%r3) # get thread_info of next
lr %r15,%r5
ahi %r15,STACK_INIT # end of kernel stack of next
st %r3,__LC_CURRENT # store task struct of next
st %r5,__LC_THREAD_INFO # store thread info of next
st %r15,__LC_KERNEL_STACK # store end of kernel stack
lctl %c4,%c4,__TASK_pid(%r3) # load pid to control reg. 4
mvc __LC_CURRENT_PID(4,%r0),__TASK_pid(%r3) # store pid of next
l %r15,__THREAD_ksp(%r3) # load kernel stack of next
lm %r6,%r15,__SF_GPRS(%r15) # load gprs of next task
br %r14
.L__critical_start:
/*
* SVC interrupt handler routine. System calls are synchronous events and
* are executed with interrupts enabled.
*/
ENTRY(system_call)
stpt __LC_SYNC_ENTER_TIMER
.Lsysc_stm:
stm %r8,%r15,__LC_SAVE_AREA_SYNC
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
lhi %r14,_PIF_SYSCALL
.Lsysc_per:
l %r15,__LC_KERNEL_STACK
la %r11,STACK_FRAME_OVERHEAD(%r15) # pointer to pt_regs
.Lsysc_vtime:
UPDATE_VTIME %r8,%r9,__LC_SYNC_ENTER_TIMER
stm %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(32,%r11),__LC_SAVE_AREA_SYNC
mvc __PT_PSW(8,%r11),__LC_SVC_OLD_PSW
mvc __PT_INT_CODE(4,%r11),__LC_SVC_ILC
st %r14,__PT_FLAGS(%r11)
.Lsysc_do_svc:
l %r10,__TI_sysc_table(%r12) # 31 bit system call table
lh %r8,__PT_INT_CODE+2(%r11)
sla %r8,2 # shift and test for svc0
jnz .Lsysc_nr_ok
# svc 0: system call number in %r1
cl %r1,BASED(.Lnr_syscalls)
jnl .Lsysc_nr_ok
sth %r1,__PT_INT_CODE+2(%r11)
lr %r8,%r1
sla %r8,2
.Lsysc_nr_ok:
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
st %r2,__PT_ORIG_GPR2(%r11)
st %r7,STACK_FRAME_OVERHEAD(%r15)
l %r9,0(%r8,%r10) # get system call addr.
tm __TI_flags+3(%r12),_TIF_TRACE
jnz .Lsysc_tracesys
basr %r14,%r9 # call sys_xxxx
st %r2,__PT_R2(%r11) # store return value
.Lsysc_return:
LOCKDEP_SYS_EXIT
.Lsysc_tif:
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jno .Lsysc_restore
tm __PT_FLAGS+3(%r11),_PIF_WORK
jnz .Lsysc_work
tm __TI_flags+3(%r12),_TIF_WORK
jnz .Lsysc_work # check for thread work
tm __LC_CPU_FLAGS+3,_CIF_WORK
jnz .Lsysc_work
.Lsysc_restore:
mvc __LC_RETURN_PSW(8),__PT_PSW(%r11)
stpt __LC_EXIT_TIMER
lm %r0,%r15,__PT_R0(%r11)
lpsw __LC_RETURN_PSW
.Lsysc_done:
#
# One of the work bits is on. Find out which one.
#
.Lsysc_work:
tm __LC_CPU_FLAGS+3,_CIF_MCCK_PENDING
jo .Lsysc_mcck_pending
tm __TI_flags+3(%r12),_TIF_NEED_RESCHED
jo .Lsysc_reschedule
tm __PT_FLAGS+3(%r11),_PIF_PER_TRAP
jo .Lsysc_singlestep
tm __TI_flags+3(%r12),_TIF_SIGPENDING
jo .Lsysc_sigpending
tm __TI_flags+3(%r12),_TIF_NOTIFY_RESUME
jo .Lsysc_notify_resume
tm __LC_CPU_FLAGS+3,_CIF_ASCE
jo .Lsysc_uaccess
j .Lsysc_return # beware of critical section cleanup
#
# _TIF_NEED_RESCHED is set, call schedule
#
.Lsysc_reschedule:
l %r1,BASED(.Lc_schedule)
la %r14,BASED(.Lsysc_return)
br %r1 # call schedule
#
# _CIF_MCCK_PENDING is set, call handler
#
.Lsysc_mcck_pending:
l %r1,BASED(.Lc_handle_mcck)
la %r14,BASED(.Lsysc_return)
br %r1 # TIF bit will be cleared by handler
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
# _CIF_ASCE is set, load user space asce
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
.Lsysc_uaccess:
ni __LC_CPU_FLAGS+3,255-_CIF_ASCE
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
lctl %c1,%c1,__LC_USER_ASCE # load primary asce
j .Lsysc_return
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
# _TIF_SIGPENDING is set, call do_signal
#
.Lsysc_sigpending:
lr %r2,%r11 # pass pointer to pt_regs
l %r1,BASED(.Lc_do_signal)
basr %r14,%r1 # call do_signal
tm __PT_FLAGS+3(%r11),_PIF_SYSCALL
jno .Lsysc_return
lm %r2,%r7,__PT_R2(%r11) # load svc arguments
l %r10,__TI_sysc_table(%r12) # 31 bit system call table
xr %r8,%r8 # svc 0 returns -ENOSYS
clc __PT_INT_CODE+2(2,%r11),BASED(.Lnr_syscalls+2)
jnl .Lsysc_nr_ok # invalid svc number -> do svc 0
lh %r8,__PT_INT_CODE+2(%r11) # load new svc number
sla %r8,2
j .Lsysc_nr_ok # restart svc
#
# _TIF_NOTIFY_RESUME is set, call do_notify_resume
#
.Lsysc_notify_resume:
lr %r2,%r11 # pass pointer to pt_regs
l %r1,BASED(.Lc_do_notify_resume)
la %r14,BASED(.Lsysc_return)
br %r1 # call do_notify_resume
#
# _PIF_PER_TRAP is set, call do_per_trap
#
.Lsysc_singlestep:
ni __PT_FLAGS+3(%r11),255-_PIF_PER_TRAP
lr %r2,%r11 # pass pointer to pt_regs
l %r1,BASED(.Lc_do_per_trap)
la %r14,BASED(.Lsysc_return)
br %r1 # call do_per_trap
#
# call tracehook_report_syscall_entry/tracehook_report_syscall_exit before
# and after the system call
#
.Lsysc_tracesys:
l %r1,BASED(.Lc_trace_enter)
lr %r2,%r11 # pass pointer to pt_regs
la %r3,0
xr %r0,%r0
icm %r0,3,__PT_INT_CODE+2(%r11)
st %r0,__PT_R2(%r11)
basr %r14,%r1 # call do_syscall_trace_enter
cl %r2,BASED(.Lnr_syscalls)
jnl .Lsysc_tracenogo
lr %r8,%r2
sll %r8,2
l %r9,0(%r8,%r10)
.Lsysc_tracego:
lm %r3,%r7,__PT_R3(%r11)
st %r7,STACK_FRAME_OVERHEAD(%r15)
l %r2,__PT_ORIG_GPR2(%r11)
basr %r14,%r9 # call sys_xxx
st %r2,__PT_R2(%r11) # store return value
.Lsysc_tracenogo:
tm __TI_flags+3(%r12),_TIF_TRACE
jz .Lsysc_return
l %r1,BASED(.Lc_trace_exit)
lr %r2,%r11 # pass pointer to pt_regs
la %r14,BASED(.Lsysc_return)
br %r1 # call do_syscall_trace_exit
#
# a new process exits the kernel with ret_from_fork
#
ENTRY(ret_from_fork)
la %r11,STACK_FRAME_OVERHEAD(%r15)
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
l %r1,BASED(.Lc_schedule_tail)
basr %r14,%r1 # call schedule_tail
TRACE_IRQS_ON
ssm __LC_SVC_NEW_PSW # reenable interrupts
tm __PT_PSW+1(%r11),0x01 # forking a kernel thread ?
jne .Lsysc_tracenogo
# it's a kernel thread
lm %r9,%r10,__PT_R9(%r11) # load gprs
ENTRY(kernel_thread_starter)
la %r2,0(%r10)
basr %r14,%r9
j .Lsysc_tracenogo
/*
* Program check handler routine
*/
ENTRY(pgm_check_handler)
stpt __LC_SYNC_ENTER_TIMER
stm %r8,%r15,__LC_SAVE_AREA_SYNC
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
lm %r8,%r9,__LC_PGM_OLD_PSW
tmh %r8,0x0001 # test problem state bit
jnz 1f # -> fault in user space
tmh %r8,0x4000 # PER bit set in old PSW ?
jnz 0f # -> enabled, can't be a double fault
tm __LC_PGM_ILC+3,0x80 # check for per exception
jnz .Lpgm_svcper # -> single stepped svc
0: CHECK_STACK STACK_SIZE,__LC_SAVE_AREA_SYNC
ahi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j 2f
1: UPDATE_VTIME %r14,%r15,__LC_SYNC_ENTER_TIMER
l %r15,__LC_KERNEL_STACK
2: la %r11,STACK_FRAME_OVERHEAD(%r15)
stm %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(32,%r11),__LC_SAVE_AREA_SYNC
stm %r8,%r9,__PT_PSW(%r11)
mvc __PT_INT_CODE(4,%r11),__LC_PGM_ILC
mvc __PT_INT_PARM_LONG(4,%r11),__LC_TRANS_EXC_CODE
xc __PT_FLAGS(4,%r11),__PT_FLAGS(%r11)
tm __LC_PGM_ILC+3,0x80 # check for per exception
jz 0f
l %r1,__TI_task(%r12)
tmh %r8,0x0001 # kernel per event ?
jz .Lpgm_kprobe
oi __PT_FLAGS+3(%r11),_PIF_PER_TRAP
mvc __THREAD_per_address(4,%r1),__LC_PER_ADDRESS
mvc __THREAD_per_cause(2,%r1),__LC_PER_CODE
mvc __THREAD_per_paid(1,%r1),__LC_PER_ACCESS_ID
0: REENABLE_IRQS
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
l %r1,BASED(.Lc_jump_table)
la %r10,0x7f
n %r10,__PT_INT_CODE(%r11)
je .Lsysc_return
sll %r10,2
l %r1,0(%r10,%r1) # load address of handler routine
lr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # branch to interrupt-handler
j .Lsysc_return
#
# PER event in supervisor state, must be kprobes
#
.Lpgm_kprobe:
REENABLE_IRQS
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
l %r1,BASED(.Lc_do_per_trap)
lr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # call do_per_trap
j .Lsysc_return
#
# single stepped system call
#
.Lpgm_svcper:
mvc __LC_RETURN_PSW(4),__LC_SVC_NEW_PSW
mvc __LC_RETURN_PSW+4(4),BASED(.Lc_sysc_per)
lhi %r14,_PIF_SYSCALL | _PIF_PER_TRAP
lpsw __LC_RETURN_PSW # branch to .Lsysc_per and enable irqs
/*
* IO interrupt handler routine
*/
ENTRY(io_int_handler)
stck __LC_INT_CLOCK
stpt __LC_ASYNC_ENTER_TIMER
stm %r8,%r15,__LC_SAVE_AREA_ASYNC
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
lm %r8,%r9,__LC_IO_OLD_PSW
tmh %r8,0x0001 # interrupting from user ?
jz .Lio_skip
UPDATE_VTIME %r14,%r15,__LC_ASYNC_ENTER_TIMER
.Lio_skip:
SWITCH_ASYNC __LC_SAVE_AREA_ASYNC,__LC_ASYNC_STACK,STACK_SHIFT
stm %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(32,%r11),__LC_SAVE_AREA_ASYNC
stm %r8,%r9,__PT_PSW(%r11)
mvc __PT_INT_CODE(12,%r11),__LC_SUBCHANNEL_ID
xc __PT_FLAGS(4,%r11),__PT_FLAGS(%r11)
TRACE_IRQS_OFF
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
.Lio_loop:
l %r1,BASED(.Lc_do_IRQ)
lr %r2,%r11 # pass pointer to pt_regs
lhi %r3,IO_INTERRUPT
tm __PT_INT_CODE+8(%r11),0x80 # adapter interrupt ?
jz .Lio_call
lhi %r3,THIN_INTERRUPT
.Lio_call:
basr %r14,%r1 # call do_IRQ
tm __LC_MACHINE_FLAGS+2,0x10 # MACHINE_FLAG_LPAR
jz .Lio_return
tpi 0
jz .Lio_return
mvc __PT_INT_CODE(12,%r11),__LC_SUBCHANNEL_ID
j .Lio_loop
.Lio_return:
LOCKDEP_SYS_EXIT
TRACE_IRQS_ON
.Lio_tif:
tm __TI_flags+3(%r12),_TIF_WORK
jnz .Lio_work # there is work to do (signals etc.)
tm __LC_CPU_FLAGS+3,_CIF_WORK
jnz .Lio_work
.Lio_restore:
mvc __LC_RETURN_PSW(8),__PT_PSW(%r11)
stpt __LC_EXIT_TIMER
lm %r0,%r15,__PT_R0(%r11)
lpsw __LC_RETURN_PSW
.Lio_done:
#
# There is work todo, find out in which context we have been interrupted:
# 1) if we return to user space we can do all _TIF_WORK work
# 2) if we return to kernel code and preemptive scheduling is enabled check
# the preemption counter and if it is zero call preempt_schedule_irq
# Before any work can be done, a switch to the kernel stack is required.
#
.Lio_work:
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jo .Lio_work_user # yes -> do resched & signal
#ifdef CONFIG_PREEMPT
# check for preemptive scheduling
icm %r0,15,__TI_precount(%r12)
jnz .Lio_restore # preemption disabled
tm __TI_flags+3(%r12),_TIF_NEED_RESCHED
jno .Lio_restore
# switch to kernel stack
l %r1,__PT_R15(%r11)
ahi %r1,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(4,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r1)
lr %r15,%r1
# TRACE_IRQS_ON already done at .Lio_return, call
# TRACE_IRQS_OFF to keep things symmetrical
TRACE_IRQS_OFF
l %r1,BASED(.Lc_preempt_irq)
basr %r14,%r1 # call preempt_schedule_irq
j .Lio_return
#else
j .Lio_restore
#endif
#
# Need to do work before returning to userspace, switch to kernel stack
#
.Lio_work_user:
l %r1,__LC_KERNEL_STACK
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(4,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r1)
lr %r15,%r1
#
# One of the work bits is on. Find out which one.
#
.Lio_work_tif:
tm __LC_CPU_FLAGS+3(%r12),_CIF_MCCK_PENDING
jo .Lio_mcck_pending
tm __TI_flags+3(%r12),_TIF_NEED_RESCHED
jo .Lio_reschedule
tm __TI_flags+3(%r12),_TIF_SIGPENDING
jo .Lio_sigpending
tm __TI_flags+3(%r12),_TIF_NOTIFY_RESUME
jo .Lio_notify_resume
tm __LC_CPU_FLAGS+3,_CIF_ASCE
jo .Lio_uaccess
j .Lio_return # beware of critical section cleanup
#
# _CIF_MCCK_PENDING is set, call handler
#
.Lio_mcck_pending:
# TRACE_IRQS_ON already done at .Lio_return
l %r1,BASED(.Lc_handle_mcck)
basr %r14,%r1 # TIF bit will be cleared by handler
TRACE_IRQS_OFF
j .Lio_return
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
# _CIF_ASCE is set, load user space asce
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
.Lio_uaccess:
ni __LC_CPU_FLAGS+3,255-_CIF_ASCE
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
lctl %c1,%c1,__LC_USER_ASCE # load primary asce
j .Lio_return
s390/uaccess: rework uaccess code - fix locking issues The current uaccess code uses a page table walk in some circumstances, e.g. in case of the in atomic futex operations or if running on old hardware which doesn't support the mvcos instruction. However it turned out that the page table walk code does not correctly lock page tables when accessing page table entries. In other words: a different cpu may invalidate a page table entry while the current cpu inspects the pte. This may lead to random data corruption. Adding correct locking however isn't trivial for all uaccess operations. Especially copy_in_user() is problematic since that requires to hold at least two locks, but must be protected against ABBA deadlock when a different cpu also performs a copy_in_user() operation. So the solution is a different approach where we change address spaces: User space runs in primary address mode, or access register mode within vdso code, like it currently already does. The kernel usually also runs in home space mode, however when accessing user space the kernel switches to primary or secondary address mode if the mvcos instruction is not available or if a compare-and-swap (futex) instruction on a user space address is performed. KVM however is special, since that requires the kernel to run in home address space while implicitly accessing user space with the sie instruction. So we end up with: User space: - runs in primary or access register mode - cr1 contains the user asce - cr7 contains the user asce - cr13 contains the kernel asce Kernel space: - runs in home space mode - cr1 contains the user or kernel asce -> the kernel asce is loaded when a uaccess requires primary or secondary address mode - cr7 contains the user or kernel asce, (changed with set_fs()) - cr13 contains the kernel asce In case of uaccess the kernel changes to: - primary space mode in case of a uaccess (copy_to_user) and uses e.g. the mvcp instruction to access user space. However the kernel will stay in home space mode if the mvcos instruction is available - secondary space mode in case of futex atomic operations, so that the instructions come from primary address space and data from secondary space In case of kvm the kernel runs in home space mode, but cr1 gets switched to contain the gmap asce before the sie instruction gets executed. When the sie instruction is finished cr1 will be switched back to contain the user asce. A context switch between two processes will always load the kernel asce for the next process in cr1. So the first exit to user space is a bit more expensive (one extra load control register instruction) than before, however keeps the code rather simple. In sum this means there is no need to perform any error prone page table walks anymore when accessing user space. The patch seems to be rather large, however it mainly removes the the page table walk code and restores the previously deleted "standard" uaccess code, with a couple of changes. The uaccess without mvcos mode can be enforced with the "uaccess_primary" kernel parameter. Reported-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2014-03-21 16:42:25 +07:00
#
# _TIF_NEED_RESCHED is set, call schedule
#
.Lio_reschedule:
# TRACE_IRQS_ON already done at .Lio_return
l %r1,BASED(.Lc_schedule)
ssm __LC_SVC_NEW_PSW # reenable interrupts
basr %r14,%r1 # call scheduler
ssm __LC_PGM_NEW_PSW # disable I/O and ext. interrupts
TRACE_IRQS_OFF
j .Lio_return
#
# _TIF_SIGPENDING is set, call do_signal
#
.Lio_sigpending:
# TRACE_IRQS_ON already done at .Lio_return
l %r1,BASED(.Lc_do_signal)
ssm __LC_SVC_NEW_PSW # reenable interrupts
lr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # call do_signal
ssm __LC_PGM_NEW_PSW # disable I/O and ext. interrupts
TRACE_IRQS_OFF
j .Lio_return
#
# _TIF_SIGPENDING is set, call do_signal
#
.Lio_notify_resume:
# TRACE_IRQS_ON already done at .Lio_return
l %r1,BASED(.Lc_do_notify_resume)
ssm __LC_SVC_NEW_PSW # reenable interrupts
lr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # call do_notify_resume
ssm __LC_PGM_NEW_PSW # disable I/O and ext. interrupts
TRACE_IRQS_OFF
j .Lio_return
/*
* External interrupt handler routine
*/
ENTRY(ext_int_handler)
stck __LC_INT_CLOCK
stpt __LC_ASYNC_ENTER_TIMER
stm %r8,%r15,__LC_SAVE_AREA_ASYNC
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
lm %r8,%r9,__LC_EXT_OLD_PSW
tmh %r8,0x0001 # interrupting from user ?
jz .Lext_skip
UPDATE_VTIME %r14,%r15,__LC_ASYNC_ENTER_TIMER
.Lext_skip:
SWITCH_ASYNC __LC_SAVE_AREA_ASYNC,__LC_ASYNC_STACK,STACK_SHIFT
stm %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(32,%r11),__LC_SAVE_AREA_ASYNC
stm %r8,%r9,__PT_PSW(%r11)
mvc __PT_INT_CODE(4,%r11),__LC_EXT_CPU_ADDR
mvc __PT_INT_PARM(4,%r11),__LC_EXT_PARAMS
xc __PT_FLAGS(4,%r11),__PT_FLAGS(%r11)
TRACE_IRQS_OFF
l %r1,BASED(.Lc_do_IRQ)
lr %r2,%r11 # pass pointer to pt_regs
lhi %r3,EXT_INTERRUPT
basr %r14,%r1 # call do_IRQ
j .Lio_return
/*
* Load idle PSW. The second "half" of this function is in .Lcleanup_idle.
*/
ENTRY(psw_idle)
st %r3,__SF_EMPTY(%r15)
basr %r1,0
la %r1,.Lpsw_idle_lpsw+4-.(%r1)
st %r1,__SF_EMPTY+4(%r15)
oi __SF_EMPTY+4(%r15),0x80
stck __CLOCK_IDLE_ENTER(%r2)
stpt __TIMER_IDLE_ENTER(%r2)
.Lpsw_idle_lpsw:
lpsw __SF_EMPTY(%r15)
br %r14
.Lpsw_idle_end:
.L__critical_end:
/*
* Machine check handler routines
*/
ENTRY(mcck_int_handler)
stck __LC_MCCK_CLOCK
spt __LC_CPU_TIMER_SAVE_AREA # revalidate cpu timer
lm %r0,%r15,__LC_GPREGS_SAVE_AREA # revalidate gprs
l %r12,__LC_THREAD_INFO
l %r13,__LC_SVC_NEW_PSW+4
lm %r8,%r9,__LC_MCK_OLD_PSW
tm __LC_MCCK_CODE,0x80 # system damage?
jo .Lmcck_panic # yes -> rest of mcck code invalid
la %r14,__LC_CPU_TIMER_SAVE_AREA
mvc __LC_MCCK_ENTER_TIMER(8),0(%r14)
tm __LC_MCCK_CODE+5,0x02 # stored cpu timer value valid?
jo 3f
la %r14,__LC_SYNC_ENTER_TIMER
clc 0(8,%r14),__LC_ASYNC_ENTER_TIMER
jl 0f
la %r14,__LC_ASYNC_ENTER_TIMER
0: clc 0(8,%r14),__LC_EXIT_TIMER
jl 1f
la %r14,__LC_EXIT_TIMER
1: clc 0(8,%r14),__LC_LAST_UPDATE_TIMER
jl 2f
la %r14,__LC_LAST_UPDATE_TIMER
2: spt 0(%r14)
mvc __LC_MCCK_ENTER_TIMER(8),0(%r14)
3: tm __LC_MCCK_CODE+2,0x09 # mwp + ia of old psw valid?
jno .Lmcck_panic # no -> skip cleanup critical
tm %r8,0x0001 # interrupting from user ?
jz .Lmcck_skip
UPDATE_VTIME %r14,%r15,__LC_MCCK_ENTER_TIMER
.Lmcck_skip:
SWITCH_ASYNC __LC_GPREGS_SAVE_AREA+32,__LC_PANIC_STACK,PAGE_SHIFT
stm %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(32,%r11),__LC_GPREGS_SAVE_AREA+32
stm %r8,%r9,__PT_PSW(%r11)
xc __PT_FLAGS(4,%r11),__PT_FLAGS(%r11)
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
l %r1,BASED(.Lc_do_machine_check)
lr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # call s390_do_machine_check
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jno .Lmcck_return
l %r1,__LC_KERNEL_STACK # switch to kernel stack
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(4,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r15)
lr %r15,%r1
ssm __LC_PGM_NEW_PSW # turn dat on, keep irqs off
tm __LC_CPU_FLAGS+3,_CIF_MCCK_PENDING
jno .Lmcck_return
TRACE_IRQS_OFF
l %r1,BASED(.Lc_handle_mcck)
basr %r14,%r1 # call s390_handle_mcck
TRACE_IRQS_ON
.Lmcck_return:
mvc __LC_RETURN_MCCK_PSW(8),__PT_PSW(%r11) # move return PSW
tm __LC_RETURN_MCCK_PSW+1,0x01 # returning to user ?
jno 0f
lm %r0,%r15,__PT_R0(%r11)
stpt __LC_EXIT_TIMER
lpsw __LC_RETURN_MCCK_PSW
0: lm %r0,%r15,__PT_R0(%r11)
lpsw __LC_RETURN_MCCK_PSW
.Lmcck_panic:
l %r14,__LC_PANIC_STACK
slr %r14,%r15
sra %r14,PAGE_SHIFT
jz 0f
l %r15,__LC_PANIC_STACK
j .Lmcck_skip
0: ahi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j .Lmcck_skip
#
# PSW restart interrupt handler
#
ENTRY(restart_int_handler)
st %r15,__LC_SAVE_AREA_RESTART
l %r15,__LC_RESTART_STACK
ahi %r15,-__PT_SIZE # create pt_regs on stack
xc 0(__PT_SIZE,%r15),0(%r15)
stm %r0,%r14,__PT_R0(%r15)
mvc __PT_R15(4,%r15),__LC_SAVE_AREA_RESTART
mvc __PT_PSW(8,%r15),__LC_RST_OLD_PSW # store restart old psw
ahi %r15,-STACK_FRAME_OVERHEAD # create stack frame on stack
xc 0(STACK_FRAME_OVERHEAD,%r15),0(%r15)
l %r1,__LC_RESTART_FN # load fn, parm & source cpu
l %r2,__LC_RESTART_DATA
l %r3,__LC_RESTART_SOURCE
ltr %r3,%r3 # test source cpu address
jm 1f # negative -> skip source stop
0: sigp %r4,%r3,SIGP_SENSE # sigp sense to source cpu
brc 10,0b # wait for status stored
1: basr %r14,%r1 # call function
stap __SF_EMPTY(%r15) # store cpu address
lh %r3,__SF_EMPTY(%r15)
2: sigp %r4,%r3,SIGP_STOP # sigp stop to current cpu
brc 2,2b
3: j 3b
.section .kprobes.text, "ax"
#ifdef CONFIG_CHECK_STACK
/*
* The synchronous or the asynchronous stack overflowed. We are dead.
* No need to properly save the registers, we are going to panic anyway.
* Setup a pt_regs so that show_trace can provide a good call trace.
*/
stack_overflow:
l %r15,__LC_PANIC_STACK # change to panic stack
la %r11,STACK_FRAME_OVERHEAD(%r15)
stm %r0,%r7,__PT_R0(%r11)
stm %r8,%r9,__PT_PSW(%r11)
mvc __PT_R8(32,%r11),0(%r14)
l %r1,BASED(1f)
xc __SF_BACKCHAIN(4,%r15),__SF_BACKCHAIN(%r15)
lr %r2,%r11 # pass pointer to pt_regs
br %r1 # branch to kernel_stack_overflow
1: .long kernel_stack_overflow
#endif
.Lcleanup_table:
.long system_call + 0x80000000
.long .Lsysc_do_svc + 0x80000000
.long .Lsysc_tif + 0x80000000
.long .Lsysc_restore + 0x80000000
.long .Lsysc_done + 0x80000000
.long .Lio_tif + 0x80000000
.long .Lio_restore + 0x80000000
.long .Lio_done + 0x80000000
.long psw_idle + 0x80000000
.long .Lpsw_idle_end + 0x80000000
cleanup_critical:
cl %r9,BASED(.Lcleanup_table) # system_call
jl 0f
cl %r9,BASED(.Lcleanup_table+4) # .Lsysc_do_svc
jl .Lcleanup_system_call
cl %r9,BASED(.Lcleanup_table+8) # .Lsysc_tif
jl 0f
cl %r9,BASED(.Lcleanup_table+12) # .Lsysc_restore
jl .Lcleanup_sysc_tif
cl %r9,BASED(.Lcleanup_table+16) # .Lsysc_done
jl .Lcleanup_sysc_restore
cl %r9,BASED(.Lcleanup_table+20) # .Lio_tif
jl 0f
cl %r9,BASED(.Lcleanup_table+24) # .Lio_restore
jl .Lcleanup_io_tif
cl %r9,BASED(.Lcleanup_table+28) # .Lio_done
jl .Lcleanup_io_restore
cl %r9,BASED(.Lcleanup_table+32) # psw_idle
jl 0f
cl %r9,BASED(.Lcleanup_table+36) # .Lpsw_idle_end
jl .Lcleanup_idle
0: br %r14
.Lcleanup_system_call:
# check if stpt has been executed
cl %r9,BASED(.Lcleanup_system_call_insn)
jh 0f
mvc __LC_SYNC_ENTER_TIMER(8),__LC_ASYNC_ENTER_TIMER
chi %r11,__LC_SAVE_AREA_ASYNC
je 0f
mvc __LC_SYNC_ENTER_TIMER(8),__LC_MCCK_ENTER_TIMER
0: # check if stm has been executed
cl %r9,BASED(.Lcleanup_system_call_insn+4)
jh 0f
mvc __LC_SAVE_AREA_SYNC(32),0(%r11)
0: # set up saved registers r12, and r13
st %r12,16(%r11) # r12 thread-info pointer
st %r13,20(%r11) # r13 literal-pool pointer
# check if the user time calculation has been done
cl %r9,BASED(.Lcleanup_system_call_insn+8)
jh 0f
l %r10,__LC_EXIT_TIMER
l %r15,__LC_EXIT_TIMER+4
SUB64 %r10,%r15,__LC_SYNC_ENTER_TIMER
ADD64 %r10,%r15,__LC_USER_TIMER
st %r10,__LC_USER_TIMER
st %r15,__LC_USER_TIMER+4
0: # check if the system time calculation has been done
cl %r9,BASED(.Lcleanup_system_call_insn+12)
jh 0f
l %r10,__LC_LAST_UPDATE_TIMER
l %r15,__LC_LAST_UPDATE_TIMER+4
SUB64 %r10,%r15,__LC_EXIT_TIMER
ADD64 %r10,%r15,__LC_SYSTEM_TIMER
st %r10,__LC_SYSTEM_TIMER
st %r15,__LC_SYSTEM_TIMER+4
0: # update accounting time stamp
mvc __LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
# set up saved register 11
l %r15,__LC_KERNEL_STACK
la %r9,STACK_FRAME_OVERHEAD(%r15)
st %r9,12(%r11) # r11 pt_regs pointer
# fill pt_regs
mvc __PT_R8(32,%r9),__LC_SAVE_AREA_SYNC
stm %r0,%r7,__PT_R0(%r9)
mvc __PT_PSW(8,%r9),__LC_SVC_OLD_PSW
mvc __PT_INT_CODE(4,%r9),__LC_SVC_ILC
xc __PT_FLAGS(4,%r9),__PT_FLAGS(%r9)
mvi __PT_FLAGS+3(%r9),_PIF_SYSCALL
# setup saved register 15
st %r15,28(%r11) # r15 stack pointer
# set new psw address and exit
l %r9,BASED(.Lcleanup_table+4) # .Lsysc_do_svc + 0x80000000
br %r14
.Lcleanup_system_call_insn:
.long system_call + 0x80000000
.long .Lsysc_stm + 0x80000000
.long .Lsysc_vtime + 0x80000000 + 36
.long .Lsysc_vtime + 0x80000000 + 76
.Lcleanup_sysc_tif:
l %r9,BASED(.Lcleanup_table+8) # .Lsysc_tif + 0x80000000
br %r14
.Lcleanup_sysc_restore:
cl %r9,BASED(.Lcleanup_sysc_restore_insn)
jhe 0f
l %r9,12(%r11) # get saved pointer to pt_regs
mvc __LC_RETURN_PSW(8),__PT_PSW(%r9)
mvc 0(32,%r11),__PT_R8(%r9)
lm %r0,%r7,__PT_R0(%r9)
0: lm %r8,%r9,__LC_RETURN_PSW
br %r14
.Lcleanup_sysc_restore_insn:
.long .Lsysc_done - 4 + 0x80000000
.Lcleanup_io_tif:
l %r9,BASED(.Lcleanup_table+20) # .Lio_tif + 0x80000000
br %r14
.Lcleanup_io_restore:
cl %r9,BASED(.Lcleanup_io_restore_insn)
jhe 0f
l %r9,12(%r11) # get saved r11 pointer to pt_regs
mvc __LC_RETURN_PSW(8),__PT_PSW(%r9)
mvc 0(32,%r11),__PT_R8(%r9)
lm %r0,%r7,__PT_R0(%r9)
0: lm %r8,%r9,__LC_RETURN_PSW
br %r14
.Lcleanup_io_restore_insn:
.long .Lio_done - 4 + 0x80000000
.Lcleanup_idle:
# copy interrupt clock & cpu timer
mvc __CLOCK_IDLE_EXIT(8,%r2),__LC_INT_CLOCK
mvc __TIMER_IDLE_EXIT(8,%r2),__LC_ASYNC_ENTER_TIMER
chi %r11,__LC_SAVE_AREA_ASYNC
je 0f
mvc __CLOCK_IDLE_EXIT(8,%r2),__LC_MCCK_CLOCK
mvc __TIMER_IDLE_EXIT(8,%r2),__LC_MCCK_ENTER_TIMER
0: # check if stck has been executed
cl %r9,BASED(.Lcleanup_idle_insn)
jhe 1f
mvc __CLOCK_IDLE_ENTER(8,%r2),__CLOCK_IDLE_EXIT(%r2)
mvc __TIMER_IDLE_ENTER(8,%r2),__TIMER_IDLE_EXIT(%r3)
1: # account system time going idle
lm %r9,%r10,__LC_STEAL_TIMER
ADD64 %r9,%r10,__CLOCK_IDLE_ENTER(%r2)
SUB64 %r9,%r10,__LC_LAST_UPDATE_CLOCK
stm %r9,%r10,__LC_STEAL_TIMER
mvc __LC_LAST_UPDATE_CLOCK(8),__CLOCK_IDLE_EXIT(%r2)
lm %r9,%r10,__LC_SYSTEM_TIMER
ADD64 %r9,%r10,__LC_LAST_UPDATE_TIMER
SUB64 %r9,%r10,__TIMER_IDLE_ENTER(%r2)
stm %r9,%r10,__LC_SYSTEM_TIMER
mvc __LC_LAST_UPDATE_TIMER(8),__TIMER_IDLE_EXIT(%r2)
# prepare return psw
n %r8,BASED(.Lcleanup_idle_wait) # clear irq & wait state bits
l %r9,24(%r11) # return from psw_idle
br %r14
.Lcleanup_idle_insn:
.long .Lpsw_idle_lpsw + 0x80000000
.Lcleanup_idle_wait:
.long 0xfcfdffff
/*
* Integer constants
*/
.align 4
.Lnr_syscalls:
.long NR_syscalls
.Lvtimer_max:
.quad 0x7fffffffffffffff
/*
* Symbol constants
*/
.Lc_do_machine_check: .long s390_do_machine_check
.Lc_handle_mcck: .long s390_handle_mcck
.Lc_do_IRQ: .long do_IRQ
.Lc_do_signal: .long do_signal
.Lc_do_notify_resume: .long do_notify_resume
.Lc_do_per_trap: .long do_per_trap
.Lc_jump_table: .long pgm_check_table
.Lc_schedule: .long schedule
#ifdef CONFIG_PREEMPT
.Lc_preempt_irq: .long preempt_schedule_irq
#endif
.Lc_trace_enter: .long do_syscall_trace_enter
.Lc_trace_exit: .long do_syscall_trace_exit
.Lc_schedule_tail: .long schedule_tail
.Lc_sysc_per: .long .Lsysc_per + 0x80000000
#ifdef CONFIG_TRACE_IRQFLAGS
.Lc_hardirqs_on: .long trace_hardirqs_on_caller
.Lc_hardirqs_off: .long trace_hardirqs_off_caller
#endif
#ifdef CONFIG_LOCKDEP
.Lc_lockdep_sys_exit: .long lockdep_sys_exit
#endif
.Lc_critical_start: .long .L__critical_start + 0x80000000
.Lc_critical_length: .long .L__critical_end - .L__critical_start
.section .rodata, "a"
#define SYSCALL(esa,esame,emu) .long esa
.globl sys_call_table
sys_call_table:
#include "syscalls.S"
#undef SYSCALL