linux/arch/s390/kernel/entry.S

1379 lines
37 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>
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
#include <asm/fpu-internal.h>
#include <asm/vx-insn.h>
__PT_R0 = __PT_GPRS
__PT_R1 = __PT_GPRS + 8
__PT_R2 = __PT_GPRS + 16
__PT_R3 = __PT_GPRS + 24
__PT_R4 = __PT_GPRS + 32
__PT_R5 = __PT_GPRS + 40
__PT_R6 = __PT_GPRS + 48
__PT_R7 = __PT_GPRS + 56
__PT_R8 = __PT_GPRS + 64
__PT_R9 = __PT_GPRS + 72
__PT_R10 = __PT_GPRS + 80
__PT_R11 = __PT_GPRS + 88
__PT_R12 = __PT_GPRS + 96
__PT_R13 = __PT_GPRS + 104
__PT_R14 = __PT_GPRS + 112
__PT_R15 = __PT_GPRS + 120
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_UPROBE)
_TIF_TRACE = (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | _TIF_SECCOMP | \
_TIF_SYSCALL_TRACEPOINT)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
_CIF_WORK = (_CIF_MCCK_PENDING | _CIF_ASCE | _CIF_FPU)
_PIF_WORK = (_PIF_PER_TRAP)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
#define BASED(name) name-cleanup_critical(%r13)
.macro TRACE_IRQS_ON
#ifdef CONFIG_TRACE_IRQFLAGS
basr %r2,%r0
brasl %r14,trace_hardirqs_on_caller
#endif
.endm
.macro TRACE_IRQS_OFF
#ifdef CONFIG_TRACE_IRQFLAGS
basr %r2,%r0
brasl %r14,trace_hardirqs_off_caller
#endif
.endm
.macro LOCKDEP_SYS_EXIT
#ifdef CONFIG_LOCKDEP
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jz .+10
brasl %r14,lockdep_sys_exit
#endif
.endm
.macro CHECK_STACK stacksize,savearea
#ifdef CONFIG_CHECK_STACK
tml %r15,\stacksize - CONFIG_STACK_GUARD
lghi %r14,\savearea
jz stack_overflow
#endif
.endm
.macro SWITCH_ASYNC savearea,timer
tmhh %r8,0x0001 # interrupting from user ?
jnz 1f
lgr %r14,%r9
slg %r14,BASED(.Lcritical_start)
clg %r14,BASED(.Lcritical_length)
jhe 0f
lghi %r11,\savearea # inside critical section, do cleanup
brasl %r14,cleanup_critical
tmhh %r8,0x0001 # 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
jnz 2f
CHECK_STACK 1<<STACK_SHIFT,\savearea
aghi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j 3f
1: LAST_BREAK %r14
UPDATE_VTIME %r14,%r15,\timer
2: lg %r15,__LC_ASYNC_STACK # load async stack
3: la %r11,STACK_FRAME_OVERHEAD(%r15)
.endm
.macro UPDATE_VTIME w1,w2,enter_timer
lg \w1,__LC_EXIT_TIMER
lg \w2,__LC_LAST_UPDATE_TIMER
slg \w1,\enter_timer
slg \w2,__LC_EXIT_TIMER
alg \w1,__LC_USER_TIMER
alg \w2,__LC_SYSTEM_TIMER
stg \w1,__LC_USER_TIMER
stg \w2,__LC_SYSTEM_TIMER
mvc __LC_LAST_UPDATE_TIMER(8),\enter_timer
.endm
.macro LAST_BREAK scratch
srag \scratch,%r10,23
jz .+10
stg %r10,__TI_last_break(%r12)
.endm
.macro REENABLE_IRQS
stg %r8,__LC_RETURN_PSW
ni __LC_RETURN_PSW,0xbf
ssm __LC_RETURN_PSW
.endm
.macro STCK savearea
#ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
.insn s,0xb27c0000,\savearea # store clock fast
#else
.insn s,0xb2050000,\savearea # store clock
#endif
.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)
stmg %r6,%r15,__SF_GPRS(%r15) # store gprs of prev task
lgr %r1,%r2
aghi %r1,__TASK_thread # thread_struct of prev task
lg %r4,__TASK_thread_info(%r2) # get thread_info of prev
lg %r5,__TASK_thread_info(%r3) # get thread_info of next
stg %r15,__THREAD_ksp(%r1) # store kernel stack of prev
lgr %r1,%r3
aghi %r1,__TASK_thread # thread_struct of next task
lgr %r15,%r5
aghi %r15,STACK_INIT # end of kernel stack of next
stg %r3,__LC_CURRENT # store task struct of next
stg %r5,__LC_THREAD_INFO # store thread info of next
stg %r15,__LC_KERNEL_STACK # store end of kernel stack
lg %r15,__THREAD_ksp(%r1) # load kernel stack of next
lctl %c4,%c4,__TASK_pid(%r3) # load pid to control reg. 4
mvc __LC_CURRENT_PID+4(4,%r0),__TASK_pid(%r3) # store pid of next
lmg %r6,%r15,__SF_GPRS(%r15) # load gprs of next task
br %r14
.L__critical_start:
#if IS_ENABLED(CONFIG_KVM)
/*
* sie64a calling convention:
* %r2 pointer to sie control block
* %r3 guest register save area
*/
ENTRY(sie64a)
stmg %r6,%r14,__SF_GPRS(%r15) # save kernel registers
stg %r2,__SF_EMPTY(%r15) # save control block pointer
stg %r3,__SF_EMPTY+8(%r15) # save guest register save area
xc __SF_EMPTY+16(16,%r15),__SF_EMPTY+16(%r15) # host id & reason
tm __LC_CPU_FLAGS+7,_CIF_FPU # load guest fp/vx registers ?
jno .Lsie_load_guest_gprs
brasl %r14,load_fpu_regs # load guest fp/vx regs
.Lsie_load_guest_gprs:
lmg %r0,%r13,0(%r3) # load guest gprs 0-13
lg %r14,__LC_GMAP # get gmap pointer
ltgr %r14,%r14
jz .Lsie_gmap
lctlg %c1,%c1,__GMAP_ASCE(%r14) # load primary asce
.Lsie_gmap:
lg %r14,__SF_EMPTY(%r15) # get control block pointer
oi __SIE_PROG0C+3(%r14),1 # we are going into SIE now
tm __SIE_PROG20+3(%r14),3 # last exit...
jnz .Lsie_skip
tm __LC_CPU_FLAGS+7,_CIF_FPU
jo .Lsie_skip # exit if fp/vx regs changed
tm __LC_MACHINE_FLAGS+6,0x20 # MACHINE_FLAG_LPP
jz .Lsie_enter
.insn s,0xb2800000,__LC_CURRENT_PID # set guest id to pid
.Lsie_enter:
sie 0(%r14)
tm __LC_MACHINE_FLAGS+6,0x20 # MACHINE_FLAG_LPP
jz .Lsie_skip
.insn s,0xb2800000,__SF_EMPTY+16(%r15)# set host id
.Lsie_skip:
ni __SIE_PROG0C+3(%r14),0xfe # no longer in SIE
lctlg %c1,%c1,__LC_USER_ASCE # load primary asce
.Lsie_done:
# some program checks are suppressing. C code (e.g. do_protection_exception)
# will rewind the PSW by the ILC, which is 4 bytes in case of SIE. Other
# instructions between sie64a and .Lsie_done should not cause program
# interrupts. So lets use a nop (47 00 00 00) as a landing pad.
# See also .Lcleanup_sie
.Lrewind_pad:
nop 0
.globl sie_exit
sie_exit:
lg %r14,__SF_EMPTY+8(%r15) # load guest register save area
stmg %r0,%r13,0(%r14) # save guest gprs 0-13
lmg %r6,%r14,__SF_GPRS(%r15) # restore kernel registers
lg %r2,__SF_EMPTY+24(%r15) # return exit reason code
br %r14
.Lsie_fault:
lghi %r14,-EFAULT
stg %r14,__SF_EMPTY+24(%r15) # set exit reason code
j sie_exit
EX_TABLE(.Lrewind_pad,.Lsie_fault)
EX_TABLE(sie_exit,.Lsie_fault)
#endif
/*
* SVC interrupt handler routine. System calls are synchronous events and
* are executed with interrupts enabled.
*/
ENTRY(system_call)
stpt __LC_SYNC_ENTER_TIMER
.Lsysc_stmg:
stmg %r8,%r15,__LC_SAVE_AREA_SYNC
lg %r10,__LC_LAST_BREAK
lg %r12,__LC_THREAD_INFO
lghi %r14,_PIF_SYSCALL
.Lsysc_per:
lg %r15,__LC_KERNEL_STACK
la %r11,STACK_FRAME_OVERHEAD(%r15) # pointer to pt_regs
LAST_BREAK %r13
.Lsysc_vtime:
UPDATE_VTIME %r10,%r13,__LC_SYNC_ENTER_TIMER
stmg %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(64,%r11),__LC_SAVE_AREA_SYNC
mvc __PT_PSW(16,%r11),__LC_SVC_OLD_PSW
mvc __PT_INT_CODE(4,%r11),__LC_SVC_ILC
stg %r14,__PT_FLAGS(%r11)
.Lsysc_do_svc:
lg %r10,__TI_sysc_table(%r12) # address of system call table
llgh %r8,__PT_INT_CODE+2(%r11)
slag %r8,%r8,2 # shift and test for svc 0
jnz .Lsysc_nr_ok
# svc 0: system call number in %r1
llgfr %r1,%r1 # clear high word in r1
cghi %r1,NR_syscalls
jnl .Lsysc_nr_ok
sth %r1,__PT_INT_CODE+2(%r11)
slag %r8,%r1,2
.Lsysc_nr_ok:
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
stg %r2,__PT_ORIG_GPR2(%r11)
stg %r7,STACK_FRAME_OVERHEAD(%r15)
lgf %r9,0(%r8,%r10) # get system call add.
tm __TI_flags+7(%r12),_TIF_TRACE
jnz .Lsysc_tracesys
basr %r14,%r9 # call sys_xxxx
stg %r2,__PT_R2(%r11) # store return value
.Lsysc_return:
LOCKDEP_SYS_EXIT
.Lsysc_tif:
tm __PT_FLAGS+7(%r11),_PIF_WORK
jnz .Lsysc_work
tm __TI_flags+7(%r12),_TIF_WORK
jnz .Lsysc_work # check for work
tm __LC_CPU_FLAGS+7,_CIF_WORK
jnz .Lsysc_work
.Lsysc_restore:
lg %r14,__LC_VDSO_PER_CPU
lmg %r0,%r10,__PT_R0(%r11)
mvc __LC_RETURN_PSW(16),__PT_PSW(%r11)
stpt __LC_EXIT_TIMER
mvc __VDSO_ECTG_BASE(16,%r14),__LC_EXIT_TIMER
lmg %r11,%r15,__PT_R11(%r11)
lpswe __LC_RETURN_PSW
.Lsysc_done:
#
# One of the work bits is on. Find out which one.
#
.Lsysc_work:
tm __LC_CPU_FLAGS+7,_CIF_MCCK_PENDING
jo .Lsysc_mcck_pending
tm __TI_flags+7(%r12),_TIF_NEED_RESCHED
jo .Lsysc_reschedule
#ifdef CONFIG_UPROBES
tm __TI_flags+7(%r12),_TIF_UPROBE
jo .Lsysc_uprobe_notify
#endif
tm __PT_FLAGS+7(%r11),_PIF_PER_TRAP
jo .Lsysc_singlestep
tm __TI_flags+7(%r12),_TIF_SIGPENDING
jo .Lsysc_sigpending
tm __TI_flags+7(%r12),_TIF_NOTIFY_RESUME
jo .Lsysc_notify_resume
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
tm __LC_CPU_FLAGS+7,_CIF_FPU
jo .Lsysc_vxrs
tm __LC_CPU_FLAGS+7,_CIF_ASCE
jo .Lsysc_uaccess
j .Lsysc_return # beware of critical section cleanup
#
# _TIF_NEED_RESCHED is set, call schedule
#
.Lsysc_reschedule:
larl %r14,.Lsysc_return
jg schedule
#
# _CIF_MCCK_PENDING is set, call handler
#
.Lsysc_mcck_pending:
larl %r14,.Lsysc_return
jg s390_handle_mcck # 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 09:42:25 +00:00
#
# _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 09:42:25 +00:00
#
.Lsysc_uaccess:
ni __LC_CPU_FLAGS+7,255-_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 09:42:25 +00:00
lctlg %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 09:42:25 +00:00
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
#
# CIF_FPU is set, restore floating-point controls and floating-point registers.
#
.Lsysc_vxrs:
larl %r14,.Lsysc_return
jg load_fpu_regs
#
# _TIF_SIGPENDING is set, call do_signal
#
.Lsysc_sigpending:
lgr %r2,%r11 # pass pointer to pt_regs
brasl %r14,do_signal
tm __PT_FLAGS+7(%r11),_PIF_SYSCALL
jno .Lsysc_return
lmg %r2,%r7,__PT_R2(%r11) # load svc arguments
lg %r10,__TI_sysc_table(%r12) # address of system call table
lghi %r8,0 # svc 0 returns -ENOSYS
llgh %r1,__PT_INT_CODE+2(%r11) # load new svc number
cghi %r1,NR_syscalls
jnl .Lsysc_nr_ok # invalid svc number -> do svc 0
slag %r8,%r1,2
j .Lsysc_nr_ok # restart svc
#
# _TIF_NOTIFY_RESUME is set, call do_notify_resume
#
.Lsysc_notify_resume:
lgr %r2,%r11 # pass pointer to pt_regs
larl %r14,.Lsysc_return
jg do_notify_resume
#
# _TIF_UPROBE is set, call uprobe_notify_resume
#
#ifdef CONFIG_UPROBES
.Lsysc_uprobe_notify:
lgr %r2,%r11 # pass pointer to pt_regs
larl %r14,.Lsysc_return
jg uprobe_notify_resume
#endif
#
# _PIF_PER_TRAP is set, call do_per_trap
#
.Lsysc_singlestep:
ni __PT_FLAGS+7(%r11),255-_PIF_PER_TRAP
lgr %r2,%r11 # pass pointer to pt_regs
larl %r14,.Lsysc_return
jg do_per_trap
#
# call tracehook_report_syscall_entry/tracehook_report_syscall_exit before
# and after the system call
#
.Lsysc_tracesys:
lgr %r2,%r11 # pass pointer to pt_regs
la %r3,0
llgh %r0,__PT_INT_CODE+2(%r11)
stg %r0,__PT_R2(%r11)
brasl %r14,do_syscall_trace_enter
lghi %r0,NR_syscalls
clgr %r0,%r2
jnh .Lsysc_tracenogo
sllg %r8,%r2,2
lgf %r9,0(%r8,%r10)
.Lsysc_tracego:
lmg %r3,%r7,__PT_R3(%r11)
stg %r7,STACK_FRAME_OVERHEAD(%r15)
lg %r2,__PT_ORIG_GPR2(%r11)
basr %r14,%r9 # call sys_xxx
stg %r2,__PT_R2(%r11) # store return value
.Lsysc_tracenogo:
tm __TI_flags+7(%r12),_TIF_TRACE
jz .Lsysc_return
lgr %r2,%r11 # pass pointer to pt_regs
larl %r14,.Lsysc_return
jg do_syscall_trace_exit
#
# a new process exits the kernel with ret_from_fork
#
ENTRY(ret_from_fork)
la %r11,STACK_FRAME_OVERHEAD(%r15)
lg %r12,__LC_THREAD_INFO
brasl %r14,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
lmg %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
stmg %r8,%r15,__LC_SAVE_AREA_SYNC
lg %r10,__LC_LAST_BREAK
lg %r12,__LC_THREAD_INFO
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
larl %r13,cleanup_critical
lmg %r8,%r9,__LC_PGM_OLD_PSW
tmhh %r8,0x0001 # test problem state bit
jnz 2f # -> fault in user space
#if IS_ENABLED(CONFIG_KVM)
# cleanup critical section for sie64a
lgr %r14,%r9
slg %r14,BASED(.Lsie_critical_start)
clg %r14,BASED(.Lsie_critical_length)
jhe 0f
brasl %r14,.Lcleanup_sie
#endif
0: tmhh %r8,0x4000 # PER bit set in old PSW ?
jnz 1f # -> enabled, can't be a double fault
tm __LC_PGM_ILC+3,0x80 # check for per exception
jnz .Lpgm_svcper # -> single stepped svc
1: CHECK_STACK STACK_SIZE,__LC_SAVE_AREA_SYNC
aghi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j 3f
2: LAST_BREAK %r14
UPDATE_VTIME %r14,%r15,__LC_SYNC_ENTER_TIMER
lg %r15,__LC_KERNEL_STACK
lg %r14,__TI_task(%r12)
aghi %r14,__TASK_thread # pointer to thread_struct
lghi %r13,__LC_PGM_TDB
tm __LC_PGM_ILC+2,0x02 # check for transaction abort
jz 3f
mvc __THREAD_trap_tdb(256,%r14),0(%r13)
3: la %r11,STACK_FRAME_OVERHEAD(%r15)
stmg %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(64,%r11),__LC_SAVE_AREA_SYNC
stmg %r8,%r9,__PT_PSW(%r11)
mvc __PT_INT_CODE(4,%r11),__LC_PGM_ILC
mvc __PT_INT_PARM_LONG(8,%r11),__LC_TRANS_EXC_CODE
xc __PT_FLAGS(8,%r11),__PT_FLAGS(%r11)
stg %r10,__PT_ARGS(%r11)
tm __LC_PGM_ILC+3,0x80 # check for per exception
jz 4f
tmhh %r8,0x0001 # kernel per event ?
jz .Lpgm_kprobe
oi __PT_FLAGS+7(%r11),_PIF_PER_TRAP
mvc __THREAD_per_address(8,%r14),__LC_PER_ADDRESS
mvc __THREAD_per_cause(2,%r14),__LC_PER_CODE
mvc __THREAD_per_paid(1,%r14),__LC_PER_ACCESS_ID
4: REENABLE_IRQS
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
larl %r1,pgm_check_table
llgh %r10,__PT_INT_CODE+2(%r11)
nill %r10,0x007f
sll %r10,2
je .Lpgm_return
lgf %r1,0(%r10,%r1) # load address of handler routine
lgr %r2,%r11 # pass pointer to pt_regs
basr %r14,%r1 # branch to interrupt-handler
.Lpgm_return:
LOCKDEP_SYS_EXIT
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jno .Lsysc_restore
j .Lsysc_tif
#
# PER event in supervisor state, must be kprobes
#
.Lpgm_kprobe:
REENABLE_IRQS
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
lgr %r2,%r11 # pass pointer to pt_regs
brasl %r14,do_per_trap
j .Lpgm_return
#
# single stepped system call
#
.Lpgm_svcper:
mvc __LC_RETURN_PSW(8),__LC_SVC_NEW_PSW
larl %r14,.Lsysc_per
stg %r14,__LC_RETURN_PSW+8
lghi %r14,_PIF_SYSCALL | _PIF_PER_TRAP
lpswe __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
stmg %r8,%r15,__LC_SAVE_AREA_ASYNC
lg %r10,__LC_LAST_BREAK
lg %r12,__LC_THREAD_INFO
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
larl %r13,cleanup_critical
lmg %r8,%r9,__LC_IO_OLD_PSW
SWITCH_ASYNC __LC_SAVE_AREA_ASYNC,__LC_ASYNC_ENTER_TIMER
stmg %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(64,%r11),__LC_SAVE_AREA_ASYNC
stmg %r8,%r9,__PT_PSW(%r11)
mvc __PT_INT_CODE(12,%r11),__LC_SUBCHANNEL_ID
xc __PT_FLAGS(8,%r11),__PT_FLAGS(%r11)
TRACE_IRQS_OFF
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
.Lio_loop:
lgr %r2,%r11 # pass pointer to pt_regs
lghi %r3,IO_INTERRUPT
tm __PT_INT_CODE+8(%r11),0x80 # adapter interrupt ?
jz .Lio_call
lghi %r3,THIN_INTERRUPT
.Lio_call:
brasl %r14,do_IRQ
tm __LC_MACHINE_FLAGS+6,0x10 # 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+7(%r12),_TIF_WORK
jnz .Lio_work # there is work to do (signals etc.)
tm __LC_CPU_FLAGS+7,_CIF_WORK
jnz .Lio_work
.Lio_restore:
lg %r14,__LC_VDSO_PER_CPU
lmg %r0,%r10,__PT_R0(%r11)
mvc __LC_RETURN_PSW(16),__PT_PSW(%r11)
stpt __LC_EXIT_TIMER
mvc __VDSO_ECTG_BASE(16,%r14),__LC_EXIT_TIMER
lmg %r11,%r15,__PT_R11(%r11)
lpswe __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 kvm is enabled check if we need to
# modify the psw to leave SIE
# 3) 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 is disabled
tm __TI_flags+7(%r12),_TIF_NEED_RESCHED
jno .Lio_restore
# switch to kernel stack
lg %r1,__PT_R15(%r11)
aghi %r1,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(8,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r1)
lgr %r15,%r1
# TRACE_IRQS_ON already done at .Lio_return, call
# TRACE_IRQS_OFF to keep things symmetrical
TRACE_IRQS_OFF
brasl %r14,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:
lg %r1,__LC_KERNEL_STACK
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(8,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r1)
lgr %r15,%r1
#
# One of the work bits is on. Find out which one.
#
.Lio_work_tif:
tm __LC_CPU_FLAGS+7,_CIF_MCCK_PENDING
jo .Lio_mcck_pending
tm __TI_flags+7(%r12),_TIF_NEED_RESCHED
jo .Lio_reschedule
tm __TI_flags+7(%r12),_TIF_SIGPENDING
jo .Lio_sigpending
tm __TI_flags+7(%r12),_TIF_NOTIFY_RESUME
jo .Lio_notify_resume
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
tm __LC_CPU_FLAGS+7,_CIF_FPU
jo .Lio_vxrs
tm __LC_CPU_FLAGS+7,_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
brasl %r14,s390_handle_mcck # 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 09:42:25 +00:00
#
# _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 09:42:25 +00:00
#
.Lio_uaccess:
ni __LC_CPU_FLAGS+7,255-_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 09:42:25 +00:00
lctlg %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 09:42:25 +00:00
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
#
# CIF_FPU is set, restore floating-point controls and floating-point registers.
#
.Lio_vxrs:
larl %r14,.Lio_return
jg load_fpu_regs
#
# _TIF_NEED_RESCHED is set, call schedule
#
.Lio_reschedule:
# TRACE_IRQS_ON already done at .Lio_return
ssm __LC_SVC_NEW_PSW # reenable interrupts
brasl %r14,schedule # call scheduler
ssm __LC_PGM_NEW_PSW # disable I/O and ext. interrupts
TRACE_IRQS_OFF
j .Lio_return
#
# _TIF_SIGPENDING or is set, call do_signal
#
.Lio_sigpending:
# TRACE_IRQS_ON already done at .Lio_return
ssm __LC_SVC_NEW_PSW # reenable interrupts
lgr %r2,%r11 # pass pointer to pt_regs
brasl %r14,do_signal
ssm __LC_PGM_NEW_PSW # disable I/O and ext. interrupts
TRACE_IRQS_OFF
j .Lio_return
#
# _TIF_NOTIFY_RESUME or is set, call do_notify_resume
#
.Lio_notify_resume:
# TRACE_IRQS_ON already done at .Lio_return
ssm __LC_SVC_NEW_PSW # reenable interrupts
lgr %r2,%r11 # pass pointer to pt_regs
brasl %r14,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
stmg %r8,%r15,__LC_SAVE_AREA_ASYNC
lg %r10,__LC_LAST_BREAK
lg %r12,__LC_THREAD_INFO
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
larl %r13,cleanup_critical
lmg %r8,%r9,__LC_EXT_OLD_PSW
SWITCH_ASYNC __LC_SAVE_AREA_ASYNC,__LC_ASYNC_ENTER_TIMER
stmg %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(64,%r11),__LC_SAVE_AREA_ASYNC
stmg %r8,%r9,__PT_PSW(%r11)
lghi %r1,__LC_EXT_PARAMS2
mvc __PT_INT_CODE(4,%r11),__LC_EXT_CPU_ADDR
mvc __PT_INT_PARM(4,%r11),__LC_EXT_PARAMS
mvc __PT_INT_PARM_LONG(8,%r11),0(%r1)
xc __PT_FLAGS(8,%r11),__PT_FLAGS(%r11)
TRACE_IRQS_OFF
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
lgr %r2,%r11 # pass pointer to pt_regs
lghi %r3,EXT_INTERRUPT
brasl %r14,do_IRQ
j .Lio_return
/*
* Load idle PSW. The second "half" of this function is in .Lcleanup_idle.
*/
ENTRY(psw_idle)
stg %r3,__SF_EMPTY(%r15)
larl %r1,.Lpsw_idle_lpsw+4
stg %r1,__SF_EMPTY+8(%r15)
#ifdef CONFIG_SMP
larl %r1,smp_cpu_mtid
llgf %r1,0(%r1)
ltgr %r1,%r1
jz .Lpsw_idle_stcctm
.insn rsy,0xeb0000000017,%r1,5,__SF_EMPTY+16(%r15)
.Lpsw_idle_stcctm:
#endif
STCK __CLOCK_IDLE_ENTER(%r2)
stpt __TIMER_IDLE_ENTER(%r2)
.Lpsw_idle_lpsw:
lpswe __SF_EMPTY(%r15)
br %r14
.Lpsw_idle_end:
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
/* Store floating-point controls and floating-point or vector extension
* registers instead. A critical section cleanup assures that the registers
* are stored even if interrupted for some other work. The register %r2
* designates a struct fpu to store register contents. If the specified
* structure does not contain a register save area, the register store is
* omitted (see also comments in arch_dup_task_struct()).
*
* The CIF_FPU flag is set in any case. The CIF_FPU triggers a lazy restore
* of the register contents at system call or io return.
*/
ENTRY(save_fpu_regs)
lg %r2,__LC_CURRENT
aghi %r2,__TASK_thread
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
tm __LC_CPU_FLAGS+7,_CIF_FPU
bor %r14
stfpc __THREAD_FPU_fpc(%r2)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
.Lsave_fpu_regs_fpc_end:
lg %r3,__THREAD_FPU_regs(%r2)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
ltgr %r3,%r3
jz .Lsave_fpu_regs_done # no save area -> set CIF_FPU
tm __THREAD_FPU_flags+3(%r2),FPU_USE_VX
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
jz .Lsave_fpu_regs_fp # no -> store FP regs
.Lsave_fpu_regs_vx_low:
VSTM %v0,%v15,0,%r3 # vstm 0,15,0(3)
.Lsave_fpu_regs_vx_high:
VSTM %v16,%v31,256,%r3 # vstm 16,31,256(3)
j .Lsave_fpu_regs_done # -> set CIF_FPU flag
.Lsave_fpu_regs_fp:
std 0,0(%r3)
std 1,8(%r3)
std 2,16(%r3)
std 3,24(%r3)
std 4,32(%r3)
std 5,40(%r3)
std 6,48(%r3)
std 7,56(%r3)
std 8,64(%r3)
std 9,72(%r3)
std 10,80(%r3)
std 11,88(%r3)
std 12,96(%r3)
std 13,104(%r3)
std 14,112(%r3)
std 15,120(%r3)
.Lsave_fpu_regs_done:
oi __LC_CPU_FLAGS+7,_CIF_FPU
br %r14
.Lsave_fpu_regs_end:
/* Load floating-point controls and floating-point or vector extension
* registers. A critical section cleanup assures that the register contents
* are loaded even if interrupted for some other work. Depending on the saved
* FP/VX state, the vector-enablement control, CR0.46, is either set or cleared.
*
* There are special calling conventions to fit into sysc and io return work:
* %r15: <kernel stack>
* The function requires:
* %r4 and __SF_EMPTY+32(%r15)
*/
load_fpu_regs:
lg %r4,__LC_CURRENT
aghi %r4,__TASK_thread
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
tm __LC_CPU_FLAGS+7,_CIF_FPU
bnor %r14
lfpc __THREAD_FPU_fpc(%r4)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
stctg %c0,%c0,__SF_EMPTY+32(%r15) # store CR0
tm __THREAD_FPU_flags+3(%r4),FPU_USE_VX # VX-enabled task ?
lg %r4,__THREAD_FPU_regs(%r4) # %r4 <- reg save area
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
jz .Lload_fpu_regs_fp_ctl # -> no VX, load FP regs
.Lload_fpu_regs_vx_ctl:
tm __SF_EMPTY+32+5(%r15),2 # test VX control
jo .Lload_fpu_regs_vx
oi __SF_EMPTY+32+5(%r15),2 # set VX control
lctlg %c0,%c0,__SF_EMPTY+32(%r15)
.Lload_fpu_regs_vx:
VLM %v0,%v15,0,%r4
.Lload_fpu_regs_vx_high:
VLM %v16,%v31,256,%r4
j .Lload_fpu_regs_done
.Lload_fpu_regs_fp_ctl:
tm __SF_EMPTY+32+5(%r15),2 # test VX control
jz .Lload_fpu_regs_fp
ni __SF_EMPTY+32+5(%r15),253 # clear VX control
lctlg %c0,%c0,__SF_EMPTY+32(%r15)
.Lload_fpu_regs_fp:
ld 0,0(%r4)
ld 1,8(%r4)
ld 2,16(%r4)
ld 3,24(%r4)
ld 4,32(%r4)
ld 5,40(%r4)
ld 6,48(%r4)
ld 7,56(%r4)
ld 8,64(%r4)
ld 9,72(%r4)
ld 10,80(%r4)
ld 11,88(%r4)
ld 12,96(%r4)
ld 13,104(%r4)
ld 14,112(%r4)
ld 15,120(%r4)
.Lload_fpu_regs_done:
ni __LC_CPU_FLAGS+7,255-_CIF_FPU
br %r14
.Lload_fpu_regs_end:
/* Test and set the vector enablement control in CR0.46 */
ENTRY(__ctl_set_vx)
stctg %c0,%c0,__SF_EMPTY(%r15)
tm __SF_EMPTY+5(%r15),2
bor %r14
oi __SF_EMPTY+5(%r15),2
lctlg %c0,%c0,__SF_EMPTY(%r15)
br %r14
.L__ctl_set_vx_end:
.L__critical_end:
/*
* Machine check handler routines
*/
ENTRY(mcck_int_handler)
STCK __LC_MCCK_CLOCK
la %r1,4095 # revalidate r1
spt __LC_CPU_TIMER_SAVE_AREA-4095(%r1) # revalidate cpu timer
lmg %r0,%r15,__LC_GPREGS_SAVE_AREA-4095(%r1)# revalidate gprs
lg %r10,__LC_LAST_BREAK
lg %r12,__LC_THREAD_INFO
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
larl %r13,cleanup_critical
lmg %r8,%r9,__LC_MCK_OLD_PSW
tm __LC_MCCK_CODE,0x80 # system damage?
jo .Lmcck_panic # yes -> rest of mcck code invalid
lghi %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
SWITCH_ASYNC __LC_GPREGS_SAVE_AREA+64,__LC_MCCK_ENTER_TIMER
.Lmcck_skip:
lghi %r14,__LC_GPREGS_SAVE_AREA+64
stmg %r0,%r7,__PT_R0(%r11)
mvc __PT_R8(64,%r11),0(%r14)
stmg %r8,%r9,__PT_PSW(%r11)
xc __PT_FLAGS(8,%r11),__PT_FLAGS(%r11)
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
lgr %r2,%r11 # pass pointer to pt_regs
brasl %r14,s390_do_machine_check
tm __PT_PSW+1(%r11),0x01 # returning to user ?
jno .Lmcck_return
lg %r1,__LC_KERNEL_STACK # switch to kernel stack
mvc STACK_FRAME_OVERHEAD(__PT_SIZE,%r1),0(%r11)
xc __SF_BACKCHAIN(8,%r1),__SF_BACKCHAIN(%r1)
la %r11,STACK_FRAME_OVERHEAD(%r1)
lgr %r15,%r1
ssm __LC_PGM_NEW_PSW # turn dat on, keep irqs off
tm __LC_CPU_FLAGS+7,_CIF_MCCK_PENDING
jno .Lmcck_return
TRACE_IRQS_OFF
brasl %r14,s390_handle_mcck
TRACE_IRQS_ON
.Lmcck_return:
lg %r14,__LC_VDSO_PER_CPU
lmg %r0,%r10,__PT_R0(%r11)
mvc __LC_RETURN_MCCK_PSW(16),__PT_PSW(%r11) # move return PSW
tm __LC_RETURN_MCCK_PSW+1,0x01 # returning to user ?
jno 0f
stpt __LC_EXIT_TIMER
mvc __VDSO_ECTG_BASE(16,%r14),__LC_EXIT_TIMER
0: lmg %r11,%r15,__PT_R11(%r11)
lpswe __LC_RETURN_MCCK_PSW
.Lmcck_panic:
lg %r15,__LC_PANIC_STACK
aghi %r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
j .Lmcck_skip
#
# PSW restart interrupt handler
#
ENTRY(restart_int_handler)
stg %r15,__LC_SAVE_AREA_RESTART
lg %r15,__LC_RESTART_STACK
aghi %r15,-__PT_SIZE # create pt_regs on stack
xc 0(__PT_SIZE,%r15),0(%r15)
stmg %r0,%r14,__PT_R0(%r15)
mvc __PT_R15(8,%r15),__LC_SAVE_AREA_RESTART
mvc __PT_PSW(16,%r15),__LC_RST_OLD_PSW # store restart old psw
aghi %r15,-STACK_FRAME_OVERHEAD # create stack frame on stack
xc 0(STACK_FRAME_OVERHEAD,%r15),0(%r15)
lg %r1,__LC_RESTART_FN # load fn, parm & source cpu
lg %r2,__LC_RESTART_DATA
lg %r3,__LC_RESTART_SOURCE
ltgr %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
llgh %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:
lg %r15,__LC_PANIC_STACK # change to panic stack
la %r11,STACK_FRAME_OVERHEAD(%r15)
stmg %r0,%r7,__PT_R0(%r11)
stmg %r8,%r9,__PT_PSW(%r11)
mvc __PT_R8(64,%r11),0(%r14)
stg %r10,__PT_ORIG_GPR2(%r11) # store last break to orig_gpr2
xc __SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
lgr %r2,%r11 # pass pointer to pt_regs
jg kernel_stack_overflow
#endif
cleanup_critical:
#if IS_ENABLED(CONFIG_KVM)
clg %r9,BASED(.Lcleanup_table_sie) # .Lsie_gmap
jl 0f
clg %r9,BASED(.Lcleanup_table_sie+8)# .Lsie_done
jl .Lcleanup_sie
#endif
clg %r9,BASED(.Lcleanup_table) # system_call
jl 0f
clg %r9,BASED(.Lcleanup_table+8) # .Lsysc_do_svc
jl .Lcleanup_system_call
clg %r9,BASED(.Lcleanup_table+16) # .Lsysc_tif
jl 0f
clg %r9,BASED(.Lcleanup_table+24) # .Lsysc_restore
jl .Lcleanup_sysc_tif
clg %r9,BASED(.Lcleanup_table+32) # .Lsysc_done
jl .Lcleanup_sysc_restore
clg %r9,BASED(.Lcleanup_table+40) # .Lio_tif
jl 0f
clg %r9,BASED(.Lcleanup_table+48) # .Lio_restore
jl .Lcleanup_io_tif
clg %r9,BASED(.Lcleanup_table+56) # .Lio_done
jl .Lcleanup_io_restore
clg %r9,BASED(.Lcleanup_table+64) # psw_idle
jl 0f
clg %r9,BASED(.Lcleanup_table+72) # .Lpsw_idle_end
jl .Lcleanup_idle
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
clg %r9,BASED(.Lcleanup_table+80) # save_fpu_regs
jl 0f
clg %r9,BASED(.Lcleanup_table+88) # .Lsave_fpu_regs_end
jl .Lcleanup_save_fpu_regs
clg %r9,BASED(.Lcleanup_table+96) # load_fpu_regs
jl 0f
clg %r9,BASED(.Lcleanup_table+104) # .Lload_fpu_regs_end
jl .Lcleanup_load_fpu_regs
clg %r9,BASED(.Lcleanup_table+112) # __ctl_set_vx
jl 0f
clg %r9,BASED(.Lcleanup_table+120) # .L__ctl_set_vx_end
jl .Lcleanup___ctl_set_vx
0: br %r14
.align 8
.Lcleanup_table:
.quad system_call
.quad .Lsysc_do_svc
.quad .Lsysc_tif
.quad .Lsysc_restore
.quad .Lsysc_done
.quad .Lio_tif
.quad .Lio_restore
.quad .Lio_done
.quad psw_idle
.quad .Lpsw_idle_end
.quad save_fpu_regs
.quad .Lsave_fpu_regs_end
.quad load_fpu_regs
.quad .Lload_fpu_regs_end
.quad __ctl_set_vx
.quad .L__ctl_set_vx_end
#if IS_ENABLED(CONFIG_KVM)
.Lcleanup_table_sie:
.quad .Lsie_gmap
.quad .Lsie_done
.Lcleanup_sie:
lg %r9,__SF_EMPTY(%r15) # get control block pointer
tm __LC_MACHINE_FLAGS+6,0x20 # MACHINE_FLAG_LPP
jz 0f
.insn s,0xb2800000,__SF_EMPTY+16(%r15)# set host id
0: ni __SIE_PROG0C+3(%r9),0xfe # no longer in SIE
lctlg %c1,%c1,__LC_USER_ASCE # load primary asce
larl %r9,sie_exit # skip forward to sie_exit
br %r14
#endif
.Lcleanup_system_call:
# check if stpt has been executed
clg %r9,BASED(.Lcleanup_system_call_insn)
jh 0f
mvc __LC_SYNC_ENTER_TIMER(8),__LC_ASYNC_ENTER_TIMER
cghi %r11,__LC_SAVE_AREA_ASYNC
je 0f
mvc __LC_SYNC_ENTER_TIMER(8),__LC_MCCK_ENTER_TIMER
0: # check if stmg has been executed
clg %r9,BASED(.Lcleanup_system_call_insn+8)
jh 0f
mvc __LC_SAVE_AREA_SYNC(64),0(%r11)
0: # check if base register setup + TIF bit load has been done
clg %r9,BASED(.Lcleanup_system_call_insn+16)
jhe 0f
# set up saved registers r10 and r12
stg %r10,16(%r11) # r10 last break
stg %r12,32(%r11) # r12 thread-info pointer
0: # check if the user time update has been done
clg %r9,BASED(.Lcleanup_system_call_insn+24)
jh 0f
lg %r15,__LC_EXIT_TIMER
slg %r15,__LC_SYNC_ENTER_TIMER
alg %r15,__LC_USER_TIMER
stg %r15,__LC_USER_TIMER
0: # check if the system time update has been done
clg %r9,BASED(.Lcleanup_system_call_insn+32)
jh 0f
lg %r15,__LC_LAST_UPDATE_TIMER
slg %r15,__LC_EXIT_TIMER
alg %r15,__LC_SYSTEM_TIMER
stg %r15,__LC_SYSTEM_TIMER
0: # update accounting time stamp
mvc __LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
# do LAST_BREAK
lg %r9,16(%r11)
srag %r9,%r9,23
jz 0f
mvc __TI_last_break(8,%r12),16(%r11)
0: # set up saved register r11
lg %r15,__LC_KERNEL_STACK
la %r9,STACK_FRAME_OVERHEAD(%r15)
stg %r9,24(%r11) # r11 pt_regs pointer
# fill pt_regs
mvc __PT_R8(64,%r9),__LC_SAVE_AREA_SYNC
stmg %r0,%r7,__PT_R0(%r9)
mvc __PT_PSW(16,%r9),__LC_SVC_OLD_PSW
mvc __PT_INT_CODE(4,%r9),__LC_SVC_ILC
xc __PT_FLAGS(8,%r9),__PT_FLAGS(%r9)
mvi __PT_FLAGS+7(%r9),_PIF_SYSCALL
# setup saved register r15
stg %r15,56(%r11) # r15 stack pointer
# set new psw address and exit
larl %r9,.Lsysc_do_svc
br %r14
.Lcleanup_system_call_insn:
.quad system_call
.quad .Lsysc_stmg
.quad .Lsysc_per
.quad .Lsysc_vtime+36
.quad .Lsysc_vtime+42
.Lcleanup_sysc_tif:
larl %r9,.Lsysc_tif
br %r14
.Lcleanup_sysc_restore:
clg %r9,BASED(.Lcleanup_sysc_restore_insn)
je 0f
lg %r9,24(%r11) # get saved pointer to pt_regs
mvc __LC_RETURN_PSW(16),__PT_PSW(%r9)
mvc 0(64,%r11),__PT_R8(%r9)
lmg %r0,%r7,__PT_R0(%r9)
0: lmg %r8,%r9,__LC_RETURN_PSW
br %r14
.Lcleanup_sysc_restore_insn:
.quad .Lsysc_done - 4
.Lcleanup_io_tif:
larl %r9,.Lio_tif
br %r14
.Lcleanup_io_restore:
clg %r9,BASED(.Lcleanup_io_restore_insn)
je 0f
lg %r9,24(%r11) # get saved r11 pointer to pt_regs
mvc __LC_RETURN_PSW(16),__PT_PSW(%r9)
mvc 0(64,%r11),__PT_R8(%r9)
lmg %r0,%r7,__PT_R0(%r9)
0: lmg %r8,%r9,__LC_RETURN_PSW
br %r14
.Lcleanup_io_restore_insn:
.quad .Lio_done - 4
.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
cghi %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 & stpt have been executed
clg %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(%r2)
1: # calculate idle cycles
#ifdef CONFIG_SMP
clg %r9,BASED(.Lcleanup_idle_insn)
jl 3f
larl %r1,smp_cpu_mtid
llgf %r1,0(%r1)
ltgr %r1,%r1
jz 3f
.insn rsy,0xeb0000000017,%r1,5,__SF_EMPTY+80(%r15)
larl %r3,mt_cycles
ag %r3,__LC_PERCPU_OFFSET
la %r4,__SF_EMPTY+16(%r15)
2: lg %r0,0(%r3)
slg %r0,0(%r4)
alg %r0,64(%r4)
stg %r0,0(%r3)
la %r3,8(%r3)
la %r4,8(%r4)
brct %r1,2b
#endif
3: # account system time going idle
lg %r9,__LC_STEAL_TIMER
alg %r9,__CLOCK_IDLE_ENTER(%r2)
slg %r9,__LC_LAST_UPDATE_CLOCK
stg %r9,__LC_STEAL_TIMER
mvc __LC_LAST_UPDATE_CLOCK(8),__CLOCK_IDLE_EXIT(%r2)
lg %r9,__LC_SYSTEM_TIMER
alg %r9,__LC_LAST_UPDATE_TIMER
slg %r9,__TIMER_IDLE_ENTER(%r2)
stg %r9,__LC_SYSTEM_TIMER
mvc __LC_LAST_UPDATE_TIMER(8),__TIMER_IDLE_EXIT(%r2)
# prepare return psw
nihh %r8,0xfcfd # clear irq & wait state bits
lg %r9,48(%r11) # return from psw_idle
br %r14
.Lcleanup_idle_insn:
.quad .Lpsw_idle_lpsw
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
.Lcleanup_save_fpu_regs:
tm __LC_CPU_FLAGS+7,_CIF_FPU
bor %r14
clg %r9,BASED(.Lcleanup_save_fpu_regs_done)
jhe 5f
clg %r9,BASED(.Lcleanup_save_fpu_regs_fp)
jhe 4f
clg %r9,BASED(.Lcleanup_save_fpu_regs_vx_high)
jhe 3f
clg %r9,BASED(.Lcleanup_save_fpu_regs_vx_low)
jhe 2f
clg %r9,BASED(.Lcleanup_save_fpu_fpc_end)
jhe 1f
lg %r2,__LC_CURRENT
aghi %r2,__TASK_thread
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
0: # Store floating-point controls
stfpc __THREAD_FPU_fpc(%r2)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
1: # Load register save area and check if VX is active
lg %r3,__THREAD_FPU_regs(%r2)
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
ltgr %r3,%r3
jz 5f # no save area -> set CIF_FPU
tm __THREAD_FPU_flags+3(%r2),FPU_USE_VX
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
jz 4f # no VX -> store FP regs
2: # Store vector registers (V0-V15)
VSTM %v0,%v15,0,%r3 # vstm 0,15,0(3)
3: # Store vector registers (V16-V31)
VSTM %v16,%v31,256,%r3 # vstm 16,31,256(3)
j 5f # -> done, set CIF_FPU flag
4: # Store floating-point registers
std 0,0(%r3)
std 1,8(%r3)
std 2,16(%r3)
std 3,24(%r3)
std 4,32(%r3)
std 5,40(%r3)
std 6,48(%r3)
std 7,56(%r3)
std 8,64(%r3)
std 9,72(%r3)
std 10,80(%r3)
std 11,88(%r3)
std 12,96(%r3)
std 13,104(%r3)
std 14,112(%r3)
std 15,120(%r3)
5: # Set CIF_FPU flag
oi __LC_CPU_FLAGS+7,_CIF_FPU
lg %r9,48(%r11) # return from save_fpu_regs
br %r14
.Lcleanup_save_fpu_fpc_end:
.quad .Lsave_fpu_regs_fpc_end
.Lcleanup_save_fpu_regs_vx_low:
.quad .Lsave_fpu_regs_vx_low
.Lcleanup_save_fpu_regs_vx_high:
.quad .Lsave_fpu_regs_vx_high
.Lcleanup_save_fpu_regs_fp:
.quad .Lsave_fpu_regs_fp
.Lcleanup_save_fpu_regs_done:
.quad .Lsave_fpu_regs_done
.Lcleanup_load_fpu_regs:
tm __LC_CPU_FLAGS+7,_CIF_FPU
bnor %r14
clg %r9,BASED(.Lcleanup_load_fpu_regs_done)
jhe 1f
clg %r9,BASED(.Lcleanup_load_fpu_regs_fp)
jhe 2f
clg %r9,BASED(.Lcleanup_load_fpu_regs_fp_ctl)
jhe 3f
clg %r9,BASED(.Lcleanup_load_fpu_regs_vx_high)
jhe 4f
clg %r9,BASED(.Lcleanup_load_fpu_regs_vx)
jhe 5f
clg %r9,BASED(.Lcleanup_load_fpu_regs_vx_ctl)
jhe 6f
lg %r4,__LC_CURRENT
aghi %r4,__TASK_thread
lfpc __THREAD_FPU_fpc(%r4)
tm __THREAD_FPU_flags+3(%r4),FPU_USE_VX # VX-enabled task ?
lg %r4,__THREAD_FPU_regs(%r4) # %r4 <- reg save area
s390/kernel: lazy restore fpu registers Improve the save and restore behavior of FPU register contents to use the vector extension within the kernel. The kernel does not use floating-point or vector registers and, therefore, saving and restoring the FPU register contents are performed for handling signals or switching processes only. To prepare for using vector instructions and vector registers within the kernel, enhance the save behavior and implement a lazy restore at return to user space from a system call or interrupt. To implement the lazy restore, the save_fpu_regs() sets a CPU information flag, CIF_FPU, to indicate that the FPU registers must be restored. Saving and setting CIF_FPU is performed in an atomic fashion to be interrupt-safe. When the kernel wants to use the vector extension or wants to change the FPU register state for a task during signal handling, the save_fpu_regs() must be called first. The CIF_FPU flag is also set at process switch. At return to user space, the FPU state is restored. In particular, the FPU state includes the floating-point or vector register contents, as well as, vector-enablement and floating-point control. The FPU state restore and clearing CIF_FPU is also performed in an atomic fashion. For KVM, the restore of the FPU register state is performed when restoring the general-purpose guest registers before the SIE instructions is started. Because the path towards the SIE instruction is interruptible, the CIF_FPU flag must be checked again right before going into SIE. If set, the guest registers must be reloaded again by re-entering the outer SIE loop. This is the same behavior as if the SIE critical section is interrupted. Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-06-10 10:53:42 +00:00
jz 3f # -> no VX, load FP regs
6: # Set VX-enablement control
stctg %c0,%c0,__SF_EMPTY+32(%r15) # store CR0
tm __SF_EMPTY+32+5(%r15),2 # test VX control
jo 5f
oi __SF_EMPTY+32+5(%r15),2 # set VX control
lctlg %c0,%c0,__SF_EMPTY+32(%r15)
5: # Load V0 ..V15 registers
VLM %v0,%v15,0,%r4
4: # Load V16..V31 registers
VLM %v16,%v31,256,%r4
j 1f
3: # Clear VX-enablement control for FP
stctg %c0,%c0,__SF_EMPTY+32(%r15) # store CR0
tm __SF_EMPTY+32+5(%r15),2 # test VX control
jz 2f
ni __SF_EMPTY+32+5(%r15),253 # clear VX control
lctlg %c0,%c0,__SF_EMPTY+32(%r15)
2: # Load floating-point registers
ld 0,0(%r4)
ld 1,8(%r4)
ld 2,16(%r4)
ld 3,24(%r4)
ld 4,32(%r4)
ld 5,40(%r4)
ld 6,48(%r4)
ld 7,56(%r4)
ld 8,64(%r4)
ld 9,72(%r4)
ld 10,80(%r4)
ld 11,88(%r4)
ld 12,96(%r4)
ld 13,104(%r4)
ld 14,112(%r4)
ld 15,120(%r4)
1: # Clear CIF_FPU bit
ni __LC_CPU_FLAGS+7,255-_CIF_FPU
lg %r9,48(%r11) # return from load_fpu_regs
br %r14
.Lcleanup_load_fpu_regs_vx_ctl:
.quad .Lload_fpu_regs_vx_ctl
.Lcleanup_load_fpu_regs_vx:
.quad .Lload_fpu_regs_vx
.Lcleanup_load_fpu_regs_vx_high:
.quad .Lload_fpu_regs_vx_high
.Lcleanup_load_fpu_regs_fp_ctl:
.quad .Lload_fpu_regs_fp_ctl
.Lcleanup_load_fpu_regs_fp:
.quad .Lload_fpu_regs_fp
.Lcleanup_load_fpu_regs_done:
.quad .Lload_fpu_regs_done
.Lcleanup___ctl_set_vx:
stctg %c0,%c0,__SF_EMPTY(%r15)
tm __SF_EMPTY+5(%r15),2
bor %r14
oi __SF_EMPTY+5(%r15),2
lctlg %c0,%c0,__SF_EMPTY(%r15)
lg %r9,48(%r11) # return from __ctl_set_vx
br %r14
/*
* Integer constants
*/
.align 8
.Lcritical_start:
.quad .L__critical_start
.Lcritical_length:
.quad .L__critical_end - .L__critical_start
#if IS_ENABLED(CONFIG_KVM)
.Lsie_critical_start:
.quad .Lsie_gmap
.Lsie_critical_length:
.quad .Lsie_done - .Lsie_gmap
#endif
.section .rodata, "a"
#define SYSCALL(esame,emu) .long esame
.globl sys_call_table
sys_call_table:
#include "syscalls.S"
#undef SYSCALL
#ifdef CONFIG_COMPAT
#define SYSCALL(esame,emu) .long emu
.globl sys_call_table_emu
sys_call_table_emu:
#include "syscalls.S"
#undef SYSCALL
#endif