5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Hartmut Penner (hp@de.ibm.com)
7 * Ulrich Weigand (uweigand@de.ibm.com)
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1995 Linus Torvalds
13 #include <linux/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/errno.h>
18 #include <linux/string.h>
19 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/mman.h>
23 #include <linux/compat.h>
24 #include <linux/smp.h>
25 #include <linux/kdebug.h>
26 #include <linux/init.h>
27 #include <linux/console.h>
28 #include <linux/module.h>
29 #include <linux/hardirq.h>
30 #include <linux/kprobes.h>
31 #include <linux/uaccess.h>
32 #include <linux/hugetlb.h>
33 #include <asm/asm-offsets.h>
34 #include <asm/system.h>
35 #include <asm/pgtable.h>
36 #include <asm/s390_ext.h>
37 #include <asm/mmu_context.h>
38 #include <asm/compat.h>
39 #include "../kernel/entry.h"
42 #define __FAIL_ADDR_MASK 0x7ffff000
43 #define __SUBCODE_MASK 0x0200
44 #define __PF_RES_FIELD 0ULL
45 #else /* CONFIG_64BIT */
46 #define __FAIL_ADDR_MASK -4096L
47 #define __SUBCODE_MASK 0x0600
48 #define __PF_RES_FIELD 0x8000000000000000ULL
49 #endif /* CONFIG_64BIT */
52 extern int sysctl_userprocess_debug
;
55 #define VM_FAULT_BADCONTEXT 0x010000
56 #define VM_FAULT_BADMAP 0x020000
57 #define VM_FAULT_BADACCESS 0x040000
59 static inline int notify_page_fault(struct pt_regs
*regs
)
63 /* kprobe_running() needs smp_processor_id() */
64 if (kprobes_built_in() && !user_mode(regs
)) {
66 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
75 * Unlock any spinlocks which will prevent us from getting the
78 void bust_spinlocks(int yes
)
83 int loglevel_save
= console_loglevel
;
87 * OK, the message is on the console. Now we call printk()
88 * without oops_in_progress set so that printk will give klogd
89 * a poke. Hold onto your hats...
91 console_loglevel
= 15;
93 console_loglevel
= loglevel_save
;
98 * Returns the address space associated with the fault.
99 * Returns 0 for kernel space and 1 for user space.
101 static inline int user_space_fault(unsigned long trans_exc_code
)
104 * The lowest two bits of the translation exception
105 * identification indicate which paging table was used.
108 if (trans_exc_code
== 2)
109 /* Access via secondary space, set_fs setting decides */
110 return current
->thread
.mm_segment
.ar4
;
111 if (user_mode
== HOME_SPACE_MODE
)
112 /* User space if the access has been done via home space. */
113 return trans_exc_code
== 3;
115 * If the user space is not the home space the kernel runs in home
116 * space. Access via secondary space has already been covered,
117 * access via primary space or access register is from user space
118 * and access via home space is from the kernel.
120 return trans_exc_code
!= 3;
124 * Send SIGSEGV to task. This is an external routine
125 * to keep the stack usage of do_page_fault small.
127 static noinline
void do_sigsegv(struct pt_regs
*regs
, long int_code
,
128 int si_code
, unsigned long trans_exc_code
)
131 unsigned long address
;
133 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
134 current
->thread
.prot_addr
= address
;
135 current
->thread
.trap_no
= int_code
;
136 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
137 #if defined(CONFIG_SYSCTL)
138 if (sysctl_userprocess_debug
)
141 printk("User process fault: interruption code 0x%lX\n",
143 printk("failing address: %lX\n", address
);
147 si
.si_signo
= SIGSEGV
;
148 si
.si_code
= si_code
;
149 si
.si_addr
= (void __user
*) address
;
150 force_sig_info(SIGSEGV
, &si
, current
);
153 static noinline
void do_no_context(struct pt_regs
*regs
, long int_code
,
154 unsigned long trans_exc_code
)
156 const struct exception_table_entry
*fixup
;
157 unsigned long address
;
159 /* Are we prepared to handle this kernel fault? */
160 fixup
= search_exception_tables(regs
->psw
.addr
& PSW_ADDR_INSN
);
162 regs
->psw
.addr
= fixup
->fixup
| PSW_ADDR_AMODE
;
167 * Oops. The kernel tried to access some bad page. We'll have to
168 * terminate things with extreme prejudice.
170 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
171 if (!user_space_fault(trans_exc_code
))
172 printk(KERN_ALERT
"Unable to handle kernel pointer dereference"
173 " at virtual kernel address %p\n", (void *)address
);
175 printk(KERN_ALERT
"Unable to handle kernel paging request"
176 " at virtual user address %p\n", (void *)address
);
178 die("Oops", regs
, int_code
);
182 static noinline
void do_low_address(struct pt_regs
*regs
, long int_code
,
183 unsigned long trans_exc_code
)
185 /* Low-address protection hit in kernel mode means
186 NULL pointer write access in kernel mode. */
187 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
188 /* Low-address protection hit in user mode 'cannot happen'. */
189 die ("Low-address protection", regs
, int_code
);
193 do_no_context(regs
, int_code
, trans_exc_code
);
196 static noinline
void do_sigbus(struct pt_regs
*regs
, long int_code
,
197 unsigned long trans_exc_code
)
199 struct task_struct
*tsk
= current
;
202 * Send a sigbus, regardless of whether we were in kernel
205 tsk
->thread
.prot_addr
= trans_exc_code
& __FAIL_ADDR_MASK
;
206 tsk
->thread
.trap_no
= int_code
;
207 force_sig(SIGBUS
, tsk
);
210 #ifdef CONFIG_S390_EXEC_PROTECT
211 static noinline
int signal_return(struct pt_regs
*regs
, long int_code
,
212 unsigned long trans_exc_code
)
217 rc
= __get_user(instruction
, (u16 __user
*) regs
->psw
.addr
);
219 if (!rc
&& instruction
== 0x0a77) {
220 clear_tsk_thread_flag(current
, TIF_SINGLE_STEP
);
221 if (is_compat_task())
225 } else if (!rc
&& instruction
== 0x0aad) {
226 clear_tsk_thread_flag(current
, TIF_SINGLE_STEP
);
227 if (is_compat_task())
228 sys32_rt_sigreturn();
232 do_sigsegv(regs
, int_code
, SEGV_MAPERR
, trans_exc_code
);
235 #endif /* CONFIG_S390_EXEC_PROTECT */
237 static noinline
void do_fault_error(struct pt_regs
*regs
, long int_code
,
238 unsigned long trans_exc_code
, int fault
)
243 case VM_FAULT_BADACCESS
:
244 #ifdef CONFIG_S390_EXEC_PROTECT
245 if ((regs
->psw
.mask
& PSW_MASK_ASC
) == PSW_ASC_SECONDARY
&&
246 (trans_exc_code
& 3) == 0) {
247 signal_return(regs
, int_code
, trans_exc_code
);
250 #endif /* CONFIG_S390_EXEC_PROTECT */
251 case VM_FAULT_BADMAP
:
252 /* Bad memory access. Check if it is kernel or user space. */
253 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
254 /* User mode accesses just cause a SIGSEGV */
255 si_code
= (fault
== VM_FAULT_BADMAP
) ?
256 SEGV_MAPERR
: SEGV_ACCERR
;
257 do_sigsegv(regs
, int_code
, si_code
, trans_exc_code
);
260 case VM_FAULT_BADCONTEXT
:
261 do_no_context(regs
, int_code
, trans_exc_code
);
263 default: /* fault & VM_FAULT_ERROR */
264 if (fault
& VM_FAULT_OOM
)
265 pagefault_out_of_memory();
266 else if (fault
& VM_FAULT_SIGBUS
) {
267 do_sigbus(regs
, int_code
, trans_exc_code
);
268 /* Kernel mode? Handle exceptions or die */
269 if (!(regs
->psw
.mask
& PSW_MASK_PSTATE
))
270 do_no_context(regs
, int_code
, trans_exc_code
);
278 * This routine handles page faults. It determines the address,
279 * and the problem, and then passes it off to one of the appropriate
282 * interruption code (int_code):
283 * 04 Protection -> Write-Protection (suprression)
284 * 10 Segment translation -> Not present (nullification)
285 * 11 Page translation -> Not present (nullification)
286 * 3b Region third trans. -> Not present (nullification)
288 static inline int do_exception(struct pt_regs
*regs
, int access
,
289 unsigned long trans_exc_code
)
291 struct task_struct
*tsk
;
292 struct mm_struct
*mm
;
293 struct vm_area_struct
*vma
;
294 unsigned long address
;
297 if (notify_page_fault(regs
))
304 * Verify that the fault happened in user space, that
305 * we are not in an interrupt and that there is a
308 fault
= VM_FAULT_BADCONTEXT
;
309 if (unlikely(!user_space_fault(trans_exc_code
) || in_atomic() || !mm
))
312 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
314 * When we get here, the fault happened in the current
315 * task's user address space, so we can switch on the
316 * interrupts again and then search the VMAs
319 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, address
);
320 down_read(&mm
->mmap_sem
);
322 fault
= VM_FAULT_BADMAP
;
323 vma
= find_vma(mm
, address
);
327 if (unlikely(vma
->vm_start
> address
)) {
328 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
330 if (expand_stack(vma
, address
))
335 * Ok, we have a good vm_area for this memory access, so
338 fault
= VM_FAULT_BADACCESS
;
339 if (unlikely(!(vma
->vm_flags
& access
)))
342 if (is_vm_hugetlb_page(vma
))
343 address
&= HPAGE_MASK
;
345 * If for any reason at all we couldn't handle the fault,
346 * make sure we exit gracefully rather than endlessly redo
349 fault
= handle_mm_fault(mm
, vma
, address
,
350 (access
== VM_WRITE
) ? FAULT_FLAG_WRITE
: 0);
351 if (unlikely(fault
& VM_FAULT_ERROR
))
354 if (fault
& VM_FAULT_MAJOR
) {
356 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0,
360 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0,
364 * The instruction that caused the program check will
365 * be repeated. Don't signal single step via SIGTRAP.
367 clear_tsk_thread_flag(tsk
, TIF_SINGLE_STEP
);
370 up_read(&mm
->mmap_sem
);
375 void __kprobes
do_protection_exception(struct pt_regs
*regs
, long int_code
)
377 unsigned long trans_exc_code
= S390_lowcore
.trans_exc_code
;
380 /* Protection exception is supressing, decrement psw address. */
381 regs
->psw
.addr
-= (int_code
>> 16);
383 * Check for low-address protection. This needs to be treated
384 * as a special case because the translation exception code
385 * field is not guaranteed to contain valid data in this case.
387 if (unlikely(!(trans_exc_code
& 4))) {
388 do_low_address(regs
, int_code
, trans_exc_code
);
391 fault
= do_exception(regs
, VM_WRITE
, trans_exc_code
);
393 do_fault_error(regs
, 4, trans_exc_code
, fault
);
396 void __kprobes
do_dat_exception(struct pt_regs
*regs
, long int_code
)
398 unsigned long trans_exc_code
= S390_lowcore
.trans_exc_code
;
401 access
= VM_READ
| VM_EXEC
| VM_WRITE
;
402 #ifdef CONFIG_S390_EXEC_PROTECT
403 if ((regs
->psw
.mask
& PSW_MASK_ASC
) == PSW_ASC_SECONDARY
&&
404 (trans_exc_code
& 3) == 0)
407 fault
= do_exception(regs
, access
, trans_exc_code
);
409 do_fault_error(regs
, int_code
& 255, trans_exc_code
, fault
);
413 void __kprobes
do_asce_exception(struct pt_regs
*regs
, long int_code
)
415 unsigned long trans_exc_code
= S390_lowcore
.trans_exc_code
;
416 struct mm_struct
*mm
= current
->mm
;
417 struct vm_area_struct
*vma
;
419 if (unlikely(!user_space_fault(trans_exc_code
) || in_atomic() || !mm
))
424 down_read(&mm
->mmap_sem
);
425 vma
= find_vma(mm
, trans_exc_code
& __FAIL_ADDR_MASK
);
426 up_read(&mm
->mmap_sem
);
429 update_mm(mm
, current
);
433 /* User mode accesses just cause a SIGSEGV */
434 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
435 do_sigsegv(regs
, int_code
, SEGV_MAPERR
, trans_exc_code
);
440 do_no_context(regs
, int_code
, trans_exc_code
);
444 int __handle_fault(unsigned long uaddr
, unsigned long int_code
, int write_user
)
449 regs
.psw
.mask
= psw_kernel_bits
;
450 if (!irqs_disabled())
451 regs
.psw
.mask
|= PSW_MASK_IO
| PSW_MASK_EXT
;
452 regs
.psw
.addr
= (unsigned long) __builtin_return_address(0);
453 regs
.psw
.addr
|= PSW_ADDR_AMODE
;
455 access
= write_user
? VM_WRITE
: VM_READ
;
456 fault
= do_exception(®s
, access
, uaddr
| 2);
457 if (unlikely(fault
)) {
458 if (fault
& VM_FAULT_OOM
) {
459 pagefault_out_of_memory();
461 } else if (fault
& VM_FAULT_SIGBUS
)
462 do_sigbus(®s
, int_code
, uaddr
);
464 return fault
? -EFAULT
: 0;
469 * 'pfault' pseudo page faults routines.
471 static ext_int_info_t ext_int_pfault
;
472 static int pfault_disable
= 0;
474 static int __init
nopfault(char *str
)
480 __setup("nopfault", nopfault
);
491 } __attribute__ ((packed
, aligned(8))) pfault_refbk_t
;
493 int pfault_init(void)
495 pfault_refbk_t refbk
=
496 { 0x258, 0, 5, 2, __LC_CURRENT
, 1ULL << 48, 1ULL << 48,
500 if (!MACHINE_IS_VM
|| pfault_disable
)
503 " diag %1,%0,0x258\n"
508 : "=d" (rc
) : "a" (&refbk
), "m" (refbk
) : "cc");
513 void pfault_fini(void)
515 pfault_refbk_t refbk
=
516 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
518 if (!MACHINE_IS_VM
|| pfault_disable
)
520 __ctl_clear_bit(0,9);
525 : : "a" (&refbk
), "m" (refbk
) : "cc");
528 static void pfault_interrupt(__u16 int_code
)
530 struct task_struct
*tsk
;
534 * Get the external interruption subcode & pfault
535 * initial/completion signal bit. VM stores this
536 * in the 'cpu address' field associated with the
537 * external interrupt.
539 subcode
= S390_lowcore
.cpu_addr
;
540 if ((subcode
& 0xff00) != __SUBCODE_MASK
)
544 * Get the token (= address of the task structure of the affected task).
546 tsk
= *(struct task_struct
**) __LC_PFAULT_INTPARM
;
548 if (subcode
& 0x0080) {
549 /* signal bit is set -> a page has been swapped in by VM */
550 if (xchg(&tsk
->thread
.pfault_wait
, -1) != 0) {
551 /* Initial interrupt was faster than the completion
552 * interrupt. pfault_wait is valid. Set pfault_wait
553 * back to zero and wake up the process. This can
554 * safely be done because the task is still sleeping
555 * and can't produce new pfaults. */
556 tsk
->thread
.pfault_wait
= 0;
557 wake_up_process(tsk
);
558 put_task_struct(tsk
);
561 /* signal bit not set -> a real page is missing. */
562 get_task_struct(tsk
);
563 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
564 if (xchg(&tsk
->thread
.pfault_wait
, 1) != 0) {
565 /* Completion interrupt was faster than the initial
566 * interrupt (swapped in a -1 for pfault_wait). Set
567 * pfault_wait back to zero and exit. This can be
568 * done safely because tsk is running in kernel
569 * mode and can't produce new pfaults. */
570 tsk
->thread
.pfault_wait
= 0;
571 set_task_state(tsk
, TASK_RUNNING
);
572 put_task_struct(tsk
);
574 set_tsk_need_resched(tsk
);
578 void __init
pfault_irq_init(void)
584 * Try to get pfault pseudo page faults going.
586 if (register_early_external_interrupt(0x2603, pfault_interrupt
,
587 &ext_int_pfault
) != 0)
588 panic("Couldn't request external interrupt 0x2603");
590 if (pfault_init() == 0)
593 /* Tough luck, no pfault. */
595 unregister_early_external_interrupt(0x2603, pfault_interrupt
,