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 */
51 #define VM_FAULT_BADCONTEXT 0x010000
52 #define VM_FAULT_BADMAP 0x020000
53 #define VM_FAULT_BADACCESS 0x040000
55 static unsigned long store_indication
;
59 if (test_facility(2) && test_facility(75))
60 store_indication
= 0xc00;
63 static inline int notify_page_fault(struct pt_regs
*regs
)
67 /* kprobe_running() needs smp_processor_id() */
68 if (kprobes_built_in() && !user_mode(regs
)) {
70 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
79 * Unlock any spinlocks which will prevent us from getting the
82 void bust_spinlocks(int yes
)
87 int loglevel_save
= console_loglevel
;
91 * OK, the message is on the console. Now we call printk()
92 * without oops_in_progress set so that printk will give klogd
93 * a poke. Hold onto your hats...
95 console_loglevel
= 15;
97 console_loglevel
= loglevel_save
;
102 * Returns the address space associated with the fault.
103 * Returns 0 for kernel space and 1 for user space.
105 static inline int user_space_fault(unsigned long trans_exc_code
)
108 * The lowest two bits of the translation exception
109 * identification indicate which paging table was used.
112 if (trans_exc_code
== 2)
113 /* Access via secondary space, set_fs setting decides */
114 return current
->thread
.mm_segment
.ar4
;
115 if (user_mode
== HOME_SPACE_MODE
)
116 /* User space if the access has been done via home space. */
117 return trans_exc_code
== 3;
119 * If the user space is not the home space the kernel runs in home
120 * space. Access via secondary space has already been covered,
121 * access via primary space or access register is from user space
122 * and access via home space is from the kernel.
124 return trans_exc_code
!= 3;
127 static inline void report_user_fault(struct pt_regs
*regs
, long int_code
,
128 int signr
, unsigned long address
)
130 if ((task_pid_nr(current
) > 1) && !show_unhandled_signals
)
132 if (!unhandled_signal(current
, signr
))
134 if (!printk_ratelimit())
136 printk("User process fault: interruption code 0x%lX ", int_code
);
137 print_vma_addr(KERN_CONT
"in ", regs
->psw
.addr
& PSW_ADDR_INSN
);
139 printk("failing address: %lX\n", address
);
144 * Send SIGSEGV to task. This is an external routine
145 * to keep the stack usage of do_page_fault small.
147 static noinline
void do_sigsegv(struct pt_regs
*regs
, long int_code
,
148 int si_code
, unsigned long trans_exc_code
)
151 unsigned long address
;
153 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
154 current
->thread
.prot_addr
= address
;
155 current
->thread
.trap_no
= int_code
;
156 report_user_fault(regs
, int_code
, SIGSEGV
, address
);
157 si
.si_signo
= SIGSEGV
;
158 si
.si_code
= si_code
;
159 si
.si_addr
= (void __user
*) address
;
160 force_sig_info(SIGSEGV
, &si
, current
);
163 static noinline
void do_no_context(struct pt_regs
*regs
, long int_code
,
164 unsigned long trans_exc_code
)
166 const struct exception_table_entry
*fixup
;
167 unsigned long address
;
169 /* Are we prepared to handle this kernel fault? */
170 fixup
= search_exception_tables(regs
->psw
.addr
& PSW_ADDR_INSN
);
172 regs
->psw
.addr
= fixup
->fixup
| PSW_ADDR_AMODE
;
177 * Oops. The kernel tried to access some bad page. We'll have to
178 * terminate things with extreme prejudice.
180 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
181 if (!user_space_fault(trans_exc_code
))
182 printk(KERN_ALERT
"Unable to handle kernel pointer dereference"
183 " at virtual kernel address %p\n", (void *)address
);
185 printk(KERN_ALERT
"Unable to handle kernel paging request"
186 " at virtual user address %p\n", (void *)address
);
188 die("Oops", regs
, int_code
);
192 static noinline
void do_low_address(struct pt_regs
*regs
, long int_code
,
193 unsigned long trans_exc_code
)
195 /* Low-address protection hit in kernel mode means
196 NULL pointer write access in kernel mode. */
197 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
198 /* Low-address protection hit in user mode 'cannot happen'. */
199 die ("Low-address protection", regs
, int_code
);
203 do_no_context(regs
, int_code
, trans_exc_code
);
206 static noinline
void do_sigbus(struct pt_regs
*regs
, long int_code
,
207 unsigned long trans_exc_code
)
209 struct task_struct
*tsk
= current
;
210 unsigned long address
;
214 * Send a sigbus, regardless of whether we were in kernel
217 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
218 tsk
->thread
.prot_addr
= address
;
219 tsk
->thread
.trap_no
= int_code
;
220 si
.si_signo
= SIGBUS
;
222 si
.si_code
= BUS_ADRERR
;
223 si
.si_addr
= (void __user
*) address
;
224 force_sig_info(SIGBUS
, &si
, tsk
);
227 #ifdef CONFIG_S390_EXEC_PROTECT
228 static noinline
int signal_return(struct pt_regs
*regs
, long int_code
,
229 unsigned long trans_exc_code
)
234 rc
= __get_user(instruction
, (u16 __user
*) regs
->psw
.addr
);
236 if (!rc
&& instruction
== 0x0a77) {
237 clear_tsk_thread_flag(current
, TIF_SINGLE_STEP
);
238 if (is_compat_task())
242 } else if (!rc
&& instruction
== 0x0aad) {
243 clear_tsk_thread_flag(current
, TIF_SINGLE_STEP
);
244 if (is_compat_task())
245 sys32_rt_sigreturn();
249 do_sigsegv(regs
, int_code
, SEGV_MAPERR
, trans_exc_code
);
252 #endif /* CONFIG_S390_EXEC_PROTECT */
254 static noinline
void do_fault_error(struct pt_regs
*regs
, long int_code
,
255 unsigned long trans_exc_code
, int fault
)
260 case VM_FAULT_BADACCESS
:
261 #ifdef CONFIG_S390_EXEC_PROTECT
262 if ((regs
->psw
.mask
& PSW_MASK_ASC
) == PSW_ASC_SECONDARY
&&
263 (trans_exc_code
& 3) == 0) {
264 signal_return(regs
, int_code
, trans_exc_code
);
267 #endif /* CONFIG_S390_EXEC_PROTECT */
268 case VM_FAULT_BADMAP
:
269 /* Bad memory access. Check if it is kernel or user space. */
270 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
271 /* User mode accesses just cause a SIGSEGV */
272 si_code
= (fault
== VM_FAULT_BADMAP
) ?
273 SEGV_MAPERR
: SEGV_ACCERR
;
274 do_sigsegv(regs
, int_code
, si_code
, trans_exc_code
);
277 case VM_FAULT_BADCONTEXT
:
278 do_no_context(regs
, int_code
, trans_exc_code
);
280 default: /* fault & VM_FAULT_ERROR */
281 if (fault
& VM_FAULT_OOM
)
282 pagefault_out_of_memory();
283 else if (fault
& VM_FAULT_SIGBUS
) {
284 /* Kernel mode? Handle exceptions or die */
285 if (!(regs
->psw
.mask
& PSW_MASK_PSTATE
))
286 do_no_context(regs
, int_code
, trans_exc_code
);
288 do_sigbus(regs
, int_code
, trans_exc_code
);
296 * This routine handles page faults. It determines the address,
297 * and the problem, and then passes it off to one of the appropriate
300 * interruption code (int_code):
301 * 04 Protection -> Write-Protection (suprression)
302 * 10 Segment translation -> Not present (nullification)
303 * 11 Page translation -> Not present (nullification)
304 * 3b Region third trans. -> Not present (nullification)
306 static inline int do_exception(struct pt_regs
*regs
, int access
,
307 unsigned long trans_exc_code
)
309 struct task_struct
*tsk
;
310 struct mm_struct
*mm
;
311 struct vm_area_struct
*vma
;
312 unsigned long address
;
315 if (notify_page_fault(regs
))
322 * Verify that the fault happened in user space, that
323 * we are not in an interrupt and that there is a
326 fault
= VM_FAULT_BADCONTEXT
;
327 if (unlikely(!user_space_fault(trans_exc_code
) || in_atomic() || !mm
))
330 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
331 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, address
);
332 down_read(&mm
->mmap_sem
);
334 fault
= VM_FAULT_BADMAP
;
335 vma
= find_vma(mm
, address
);
339 if (unlikely(vma
->vm_start
> address
)) {
340 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
342 if (expand_stack(vma
, address
))
347 * Ok, we have a good vm_area for this memory access, so
350 fault
= VM_FAULT_BADACCESS
;
351 if (unlikely(!(vma
->vm_flags
& access
)))
354 if (is_vm_hugetlb_page(vma
))
355 address
&= HPAGE_MASK
;
357 * If for any reason at all we couldn't handle the fault,
358 * make sure we exit gracefully rather than endlessly redo
361 write
= (access
== VM_WRITE
||
362 (trans_exc_code
& store_indication
) == 0x400) ?
363 FAULT_FLAG_WRITE
: 0;
364 fault
= handle_mm_fault(mm
, vma
, address
, write
);
365 if (unlikely(fault
& VM_FAULT_ERROR
))
368 if (fault
& VM_FAULT_MAJOR
) {
370 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0,
374 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0,
378 * The instruction that caused the program check will
379 * be repeated. Don't signal single step via SIGTRAP.
381 clear_tsk_thread_flag(tsk
, TIF_SINGLE_STEP
);
384 up_read(&mm
->mmap_sem
);
389 void __kprobes
do_protection_exception(struct pt_regs
*regs
, long pgm_int_code
,
390 unsigned long trans_exc_code
)
394 /* Protection exception is supressing, decrement psw address. */
395 regs
->psw
.addr
-= (pgm_int_code
>> 16);
397 * Check for low-address protection. This needs to be treated
398 * as a special case because the translation exception code
399 * field is not guaranteed to contain valid data in this case.
401 if (unlikely(!(trans_exc_code
& 4))) {
402 do_low_address(regs
, pgm_int_code
, trans_exc_code
);
405 fault
= do_exception(regs
, VM_WRITE
, trans_exc_code
);
407 do_fault_error(regs
, 4, trans_exc_code
, fault
);
410 void __kprobes
do_dat_exception(struct pt_regs
*regs
, long pgm_int_code
,
411 unsigned long trans_exc_code
)
415 access
= VM_READ
| VM_EXEC
| VM_WRITE
;
416 #ifdef CONFIG_S390_EXEC_PROTECT
417 if ((regs
->psw
.mask
& PSW_MASK_ASC
) == PSW_ASC_SECONDARY
&&
418 (trans_exc_code
& 3) == 0)
421 fault
= do_exception(regs
, access
, trans_exc_code
);
423 do_fault_error(regs
, pgm_int_code
& 255, trans_exc_code
, fault
);
427 void __kprobes
do_asce_exception(struct pt_regs
*regs
, long pgm_int_code
,
428 unsigned long trans_exc_code
)
430 struct mm_struct
*mm
= current
->mm
;
431 struct vm_area_struct
*vma
;
433 if (unlikely(!user_space_fault(trans_exc_code
) || in_atomic() || !mm
))
436 down_read(&mm
->mmap_sem
);
437 vma
= find_vma(mm
, trans_exc_code
& __FAIL_ADDR_MASK
);
438 up_read(&mm
->mmap_sem
);
441 update_mm(mm
, current
);
445 /* User mode accesses just cause a SIGSEGV */
446 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
447 do_sigsegv(regs
, pgm_int_code
, SEGV_MAPERR
, trans_exc_code
);
452 do_no_context(regs
, pgm_int_code
, trans_exc_code
);
456 int __handle_fault(unsigned long uaddr
, unsigned long pgm_int_code
, int write
)
461 regs
.psw
.mask
= psw_kernel_bits
;
462 if (!irqs_disabled())
463 regs
.psw
.mask
|= PSW_MASK_IO
| PSW_MASK_EXT
;
464 regs
.psw
.addr
= (unsigned long) __builtin_return_address(0);
465 regs
.psw
.addr
|= PSW_ADDR_AMODE
;
467 access
= write
? VM_WRITE
: VM_READ
;
468 fault
= do_exception(®s
, access
, uaddr
| 2);
469 if (unlikely(fault
)) {
470 if (fault
& VM_FAULT_OOM
) {
471 pagefault_out_of_memory();
473 } else if (fault
& VM_FAULT_SIGBUS
)
474 do_sigbus(®s
, pgm_int_code
, uaddr
);
476 return fault
? -EFAULT
: 0;
481 * 'pfault' pseudo page faults routines.
483 static ext_int_info_t ext_int_pfault
;
484 static int pfault_disable
= 0;
486 static int __init
nopfault(char *str
)
492 __setup("nopfault", nopfault
);
503 } __attribute__ ((packed
, aligned(8))) pfault_refbk_t
;
505 int pfault_init(void)
507 pfault_refbk_t refbk
=
508 { 0x258, 0, 5, 2, __LC_CURRENT
, 1ULL << 48, 1ULL << 48,
512 if (!MACHINE_IS_VM
|| pfault_disable
)
515 " diag %1,%0,0x258\n"
520 : "=d" (rc
) : "a" (&refbk
), "m" (refbk
) : "cc");
525 void pfault_fini(void)
527 pfault_refbk_t refbk
=
528 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
530 if (!MACHINE_IS_VM
|| pfault_disable
)
532 __ctl_clear_bit(0,9);
537 : : "a" (&refbk
), "m" (refbk
) : "cc");
540 static void pfault_interrupt(unsigned int ext_int_code
,
541 unsigned int param32
, unsigned long param64
)
543 struct task_struct
*tsk
;
547 * Get the external interruption subcode & pfault
548 * initial/completion signal bit. VM stores this
549 * in the 'cpu address' field associated with the
550 * external interrupt.
552 subcode
= ext_int_code
>> 16;
553 if ((subcode
& 0xff00) != __SUBCODE_MASK
)
557 * Get the token (= address of the task structure of the affected task).
560 tsk
= *(struct task_struct
**) param64
;
562 tsk
= *(struct task_struct
**) param32
;
565 if (subcode
& 0x0080) {
566 /* signal bit is set -> a page has been swapped in by VM */
567 if (xchg(&tsk
->thread
.pfault_wait
, -1) != 0) {
568 /* Initial interrupt was faster than the completion
569 * interrupt. pfault_wait is valid. Set pfault_wait
570 * back to zero and wake up the process. This can
571 * safely be done because the task is still sleeping
572 * and can't produce new pfaults. */
573 tsk
->thread
.pfault_wait
= 0;
574 wake_up_process(tsk
);
575 put_task_struct(tsk
);
578 /* signal bit not set -> a real page is missing. */
579 get_task_struct(tsk
);
580 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
581 if (xchg(&tsk
->thread
.pfault_wait
, 1) != 0) {
582 /* Completion interrupt was faster than the initial
583 * interrupt (swapped in a -1 for pfault_wait). Set
584 * pfault_wait back to zero and exit. This can be
585 * done safely because tsk is running in kernel
586 * mode and can't produce new pfaults. */
587 tsk
->thread
.pfault_wait
= 0;
588 set_task_state(tsk
, TASK_RUNNING
);
589 put_task_struct(tsk
);
591 set_tsk_need_resched(tsk
);
595 void __init
pfault_irq_init(void)
601 * Try to get pfault pseudo page faults going.
603 if (register_early_external_interrupt(0x2603, pfault_interrupt
,
604 &ext_int_pfault
) != 0)
605 panic("Couldn't request external interrupt 0x2603");
607 if (pfault_init() == 0)
610 /* Tough luck, no pfault. */
612 unregister_early_external_interrupt(0x2603, pfault_interrupt
,