3 * Copyright IBM Corp. 1999
4 * Author(s): Hartmut Penner (hp@de.ibm.com)
5 * Ulrich Weigand (uweigand@de.ibm.com)
7 * Derived from "arch/i386/mm/fault.c"
8 * Copyright (C) 1995 Linus Torvalds
11 #include <linux/kernel_stat.h>
12 #include <linux/perf_event.h>
13 #include <linux/signal.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/mman.h>
22 #include <linux/compat.h>
23 #include <linux/smp.h>
24 #include <linux/kdebug.h>
25 #include <linux/init.h>
26 #include <linux/console.h>
27 #include <linux/module.h>
28 #include <linux/hardirq.h>
29 #include <linux/kprobes.h>
30 #include <linux/uaccess.h>
31 #include <linux/hugetlb.h>
32 #include <asm/asm-offsets.h>
33 #include <asm/pgtable.h>
35 #include <asm/mmu_context.h>
36 #include <asm/facility.h>
37 #include "../kernel/entry.h"
40 #define __FAIL_ADDR_MASK 0x7ffff000
41 #define __SUBCODE_MASK 0x0200
42 #define __PF_RES_FIELD 0ULL
43 #else /* CONFIG_64BIT */
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #endif /* CONFIG_64BIT */
49 #define VM_FAULT_BADCONTEXT 0x010000
50 #define VM_FAULT_BADMAP 0x020000
51 #define VM_FAULT_BADACCESS 0x040000
52 #define VM_FAULT_SIGNAL 0x080000
53 #define VM_FAULT_PFAULT 0x100000
55 static unsigned long store_indication __read_mostly
;
58 static int __init
fault_init(void)
60 if (test_facility(75))
61 store_indication
= 0xc00;
64 early_initcall(fault_init
);
67 static inline int notify_page_fault(struct pt_regs
*regs
)
71 /* kprobe_running() needs smp_processor_id() */
72 if (kprobes_built_in() && !user_mode(regs
)) {
74 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
83 * Unlock any spinlocks which will prevent us from getting the
86 void bust_spinlocks(int yes
)
91 int loglevel_save
= console_loglevel
;
95 * OK, the message is on the console. Now we call printk()
96 * without oops_in_progress set so that printk will give klogd
97 * a poke. Hold onto your hats...
99 console_loglevel
= 15;
101 console_loglevel
= loglevel_save
;
106 * Returns the address space associated with the fault.
107 * Returns 0 for kernel space and 1 for user space.
109 static inline int user_space_fault(unsigned long trans_exc_code
)
112 * The lowest two bits of the translation exception
113 * identification indicate which paging table was used.
116 if (trans_exc_code
== 2)
117 /* Access via secondary space, set_fs setting decides */
118 return current
->thread
.mm_segment
.ar4
;
120 * Access via primary space or access register is from user space
121 * and access via home space is from the kernel.
123 return trans_exc_code
!= 3;
126 static inline void report_user_fault(struct pt_regs
*regs
, long signr
)
128 if ((task_pid_nr(current
) > 1) && !show_unhandled_signals
)
130 if (!unhandled_signal(current
, signr
))
132 if (!printk_ratelimit())
134 printk(KERN_ALERT
"User process fault: interruption code 0x%X ",
136 print_vma_addr(KERN_CONT
"in ", regs
->psw
.addr
& PSW_ADDR_INSN
);
137 printk(KERN_CONT
"\n");
138 printk(KERN_ALERT
"failing address: %lX\n",
139 regs
->int_parm_long
& __FAIL_ADDR_MASK
);
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
, int si_code
)
151 report_user_fault(regs
, SIGSEGV
);
152 si
.si_signo
= SIGSEGV
;
153 si
.si_code
= si_code
;
154 si
.si_addr
= (void __user
*)(regs
->int_parm_long
& __FAIL_ADDR_MASK
);
155 force_sig_info(SIGSEGV
, &si
, current
);
158 static noinline
void do_no_context(struct pt_regs
*regs
)
160 const struct exception_table_entry
*fixup
;
161 unsigned long address
;
163 /* Are we prepared to handle this kernel fault? */
164 fixup
= search_exception_tables(regs
->psw
.addr
& PSW_ADDR_INSN
);
166 regs
->psw
.addr
= extable_fixup(fixup
) | PSW_ADDR_AMODE
;
171 * Oops. The kernel tried to access some bad page. We'll have to
172 * terminate things with extreme prejudice.
174 address
= regs
->int_parm_long
& __FAIL_ADDR_MASK
;
175 if (!user_space_fault(regs
->int_parm_long
))
176 printk(KERN_ALERT
"Unable to handle kernel pointer dereference"
177 " at virtual kernel address %p\n", (void *)address
);
179 printk(KERN_ALERT
"Unable to handle kernel paging request"
180 " at virtual user address %p\n", (void *)address
);
186 static noinline
void do_low_address(struct pt_regs
*regs
)
188 /* Low-address protection hit in kernel mode means
189 NULL pointer write access in kernel mode. */
190 if (regs
->psw
.mask
& PSW_MASK_PSTATE
) {
191 /* Low-address protection hit in user mode 'cannot happen'. */
192 die (regs
, "Low-address protection");
199 static noinline
void do_sigbus(struct pt_regs
*regs
)
201 struct task_struct
*tsk
= current
;
205 * Send a sigbus, regardless of whether we were in kernel
208 si
.si_signo
= SIGBUS
;
210 si
.si_code
= BUS_ADRERR
;
211 si
.si_addr
= (void __user
*)(regs
->int_parm_long
& __FAIL_ADDR_MASK
);
212 force_sig_info(SIGBUS
, &si
, tsk
);
215 static noinline
void do_fault_error(struct pt_regs
*regs
, int fault
)
220 case VM_FAULT_BADACCESS
:
221 case VM_FAULT_BADMAP
:
222 /* Bad memory access. Check if it is kernel or user space. */
223 if (user_mode(regs
)) {
224 /* User mode accesses just cause a SIGSEGV */
225 si_code
= (fault
== VM_FAULT_BADMAP
) ?
226 SEGV_MAPERR
: SEGV_ACCERR
;
227 do_sigsegv(regs
, si_code
);
230 case VM_FAULT_BADCONTEXT
:
231 case VM_FAULT_PFAULT
:
234 case VM_FAULT_SIGNAL
:
235 if (!user_mode(regs
))
238 default: /* fault & VM_FAULT_ERROR */
239 if (fault
& VM_FAULT_OOM
) {
240 if (!user_mode(regs
))
243 pagefault_out_of_memory();
244 } else if (fault
& VM_FAULT_SIGBUS
) {
245 /* Kernel mode? Handle exceptions or die */
246 if (!user_mode(regs
))
257 * This routine handles page faults. It determines the address,
258 * and the problem, and then passes it off to one of the appropriate
261 * interruption code (int_code):
262 * 04 Protection -> Write-Protection (suprression)
263 * 10 Segment translation -> Not present (nullification)
264 * 11 Page translation -> Not present (nullification)
265 * 3b Region third trans. -> Not present (nullification)
267 static inline int do_exception(struct pt_regs
*regs
, int access
)
272 struct task_struct
*tsk
;
273 struct mm_struct
*mm
;
274 struct vm_area_struct
*vma
;
275 unsigned long trans_exc_code
;
276 unsigned long address
;
282 * The instruction that caused the program check has
283 * been nullified. Don't signal single step via SIGTRAP.
285 clear_tsk_thread_flag(tsk
, TIF_PER_TRAP
);
287 if (notify_page_fault(regs
))
291 trans_exc_code
= regs
->int_parm_long
;
294 * Verify that the fault happened in user space, that
295 * we are not in an interrupt and that there is a
298 fault
= VM_FAULT_BADCONTEXT
;
299 if (unlikely(!user_space_fault(trans_exc_code
) || in_atomic() || !mm
))
302 address
= trans_exc_code
& __FAIL_ADDR_MASK
;
303 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
304 flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
306 flags
|= FAULT_FLAG_USER
;
307 if (access
== VM_WRITE
|| (trans_exc_code
& store_indication
) == 0x400)
308 flags
|= FAULT_FLAG_WRITE
;
309 down_read(&mm
->mmap_sem
);
312 gmap
= (struct gmap
*)
313 ((current
->flags
& PF_VCPU
) ? S390_lowcore
.gmap
: 0);
315 address
= __gmap_fault(address
, gmap
);
316 if (address
== -EFAULT
) {
317 fault
= VM_FAULT_BADMAP
;
320 if (address
== -ENOMEM
) {
321 fault
= VM_FAULT_OOM
;
324 if (gmap
->pfault_enabled
)
325 flags
|= FAULT_FLAG_RETRY_NOWAIT
;
330 fault
= VM_FAULT_BADMAP
;
331 vma
= find_vma(mm
, address
);
335 if (unlikely(vma
->vm_start
> address
)) {
336 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
338 if (expand_stack(vma
, address
))
343 * Ok, we have a good vm_area for this memory access, so
346 fault
= VM_FAULT_BADACCESS
;
347 if (unlikely(!(vma
->vm_flags
& access
)))
350 if (is_vm_hugetlb_page(vma
))
351 address
&= HPAGE_MASK
;
353 * If for any reason at all we couldn't handle the fault,
354 * make sure we exit gracefully rather than endlessly redo
357 fault
= handle_mm_fault(mm
, vma
, address
, flags
);
358 /* No reason to continue if interrupted by SIGKILL. */
359 if ((fault
& VM_FAULT_RETRY
) && fatal_signal_pending(current
)) {
360 fault
= VM_FAULT_SIGNAL
;
363 if (unlikely(fault
& VM_FAULT_ERROR
))
367 * Major/minor page fault accounting is only done on the
368 * initial attempt. If we go through a retry, it is extremely
369 * likely that the page will be found in page cache at that point.
371 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
372 if (fault
& VM_FAULT_MAJOR
) {
374 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1,
378 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1,
381 if (fault
& VM_FAULT_RETRY
) {
383 if (gmap
&& (flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
384 /* FAULT_FLAG_RETRY_NOWAIT has been set,
385 * mmap_sem has not been released */
386 current
->thread
.gmap_pfault
= 1;
387 fault
= VM_FAULT_PFAULT
;
391 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
393 flags
&= ~(FAULT_FLAG_ALLOW_RETRY
|
394 FAULT_FLAG_RETRY_NOWAIT
);
395 flags
|= FAULT_FLAG_TRIED
;
396 down_read(&mm
->mmap_sem
);
402 up_read(&mm
->mmap_sem
);
407 void __kprobes
do_protection_exception(struct pt_regs
*regs
)
409 unsigned long trans_exc_code
;
412 trans_exc_code
= regs
->int_parm_long
;
414 * Protection exceptions are suppressing, decrement psw address.
415 * The exception to this rule are aborted transactions, for these
416 * the PSW already points to the correct location.
418 if (!(regs
->int_code
& 0x200))
419 regs
->psw
.addr
= __rewind_psw(regs
->psw
, regs
->int_code
>> 16);
421 * Check for low-address protection. This needs to be treated
422 * as a special case because the translation exception code
423 * field is not guaranteed to contain valid data in this case.
425 if (unlikely(!(trans_exc_code
& 4))) {
426 do_low_address(regs
);
429 fault
= do_exception(regs
, VM_WRITE
);
431 do_fault_error(regs
, fault
);
434 void __kprobes
do_dat_exception(struct pt_regs
*regs
)
438 access
= VM_READ
| VM_EXEC
| VM_WRITE
;
439 fault
= do_exception(regs
, access
);
441 do_fault_error(regs
, fault
);
444 int __handle_fault(unsigned long uaddr
, unsigned long pgm_int_code
, int write
)
449 /* Emulate a uaccess fault from kernel mode. */
450 regs
.psw
.mask
= PSW_KERNEL_BITS
| PSW_MASK_DAT
| PSW_MASK_MCHECK
;
451 if (!irqs_disabled())
452 regs
.psw
.mask
|= PSW_MASK_IO
| PSW_MASK_EXT
;
453 regs
.psw
.addr
= (unsigned long) __builtin_return_address(0);
454 regs
.psw
.addr
|= PSW_ADDR_AMODE
;
455 regs
.int_code
= pgm_int_code
;
456 regs
.int_parm_long
= (uaddr
& PAGE_MASK
) | 2;
457 access
= write
? VM_WRITE
: VM_READ
;
458 fault
= do_exception(®s
, access
);
460 * Since the fault happened in kernel mode while performing a uaccess
461 * all we need to do now is emulating a fixup in case "fault" is not
463 * For the calling uaccess functions this results always in -EFAULT.
465 return fault
? -EFAULT
: 0;
470 * 'pfault' pseudo page faults routines.
472 static int pfault_disable
;
474 static int __init
nopfault(char *str
)
480 __setup("nopfault", nopfault
);
482 struct pfault_refbk
{
491 } __attribute__ ((packed
, aligned(8)));
493 int pfault_init(void)
495 struct pfault_refbk refbk
= {
500 .refgaddr
= __LC_CURRENT_PID
,
501 .refselmk
= 1ULL << 48,
502 .refcmpmk
= 1ULL << 48,
503 .reserved
= __PF_RES_FIELD
};
509 " diag %1,%0,0x258\n"
514 : "=d" (rc
) : "a" (&refbk
), "m" (refbk
) : "cc");
518 void pfault_fini(void)
520 struct pfault_refbk refbk
= {
533 : : "a" (&refbk
), "m" (refbk
) : "cc");
536 static DEFINE_SPINLOCK(pfault_lock
);
537 static LIST_HEAD(pfault_list
);
539 static void pfault_interrupt(struct ext_code ext_code
,
540 unsigned int param32
, unsigned long param64
)
542 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_code
.subcode
;
553 if ((subcode
& 0xff00) != __SUBCODE_MASK
)
555 inc_irq_stat(IRQEXT_PFL
);
556 /* Get the token (= pid of the affected task). */
557 pid
= sizeof(void *) == 4 ? param32
: param64
;
559 tsk
= find_task_by_pid_ns(pid
, &init_pid_ns
);
561 get_task_struct(tsk
);
565 spin_lock(&pfault_lock
);
566 if (subcode
& 0x0080) {
567 /* signal bit is set -> a page has been swapped in by VM */
568 if (tsk
->thread
.pfault_wait
== 1) {
569 /* Initial interrupt was faster than the completion
570 * interrupt. pfault_wait is valid. Set pfault_wait
571 * back to zero and wake up the process. This can
572 * safely be done because the task is still sleeping
573 * and can't produce new pfaults. */
574 tsk
->thread
.pfault_wait
= 0;
575 list_del(&tsk
->thread
.list
);
576 wake_up_process(tsk
);
577 put_task_struct(tsk
);
579 /* Completion interrupt was faster than initial
580 * interrupt. Set pfault_wait to -1 so the initial
581 * interrupt doesn't put the task to sleep.
582 * If the task is not running, ignore the completion
583 * interrupt since it must be a leftover of a PFAULT
584 * CANCEL operation which didn't remove all pending
585 * completion interrupts. */
586 if (tsk
->state
== TASK_RUNNING
)
587 tsk
->thread
.pfault_wait
= -1;
590 /* signal bit not set -> a real page is missing. */
591 if (WARN_ON_ONCE(tsk
!= current
))
593 if (tsk
->thread
.pfault_wait
== 1) {
594 /* Already on the list with a reference: put to sleep */
595 __set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
596 set_tsk_need_resched(tsk
);
597 } else if (tsk
->thread
.pfault_wait
== -1) {
598 /* Completion interrupt was faster than the initial
599 * interrupt (pfault_wait == -1). Set pfault_wait
600 * back to zero and exit. */
601 tsk
->thread
.pfault_wait
= 0;
603 /* Initial interrupt arrived before completion
604 * interrupt. Let the task sleep.
605 * An extra task reference is needed since a different
606 * cpu may set the task state to TASK_RUNNING again
607 * before the scheduler is reached. */
608 get_task_struct(tsk
);
609 tsk
->thread
.pfault_wait
= 1;
610 list_add(&tsk
->thread
.list
, &pfault_list
);
611 __set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
612 set_tsk_need_resched(tsk
);
616 spin_unlock(&pfault_lock
);
617 put_task_struct(tsk
);
620 static int pfault_cpu_notify(struct notifier_block
*self
, unsigned long action
,
623 struct thread_struct
*thread
, *next
;
624 struct task_struct
*tsk
;
626 switch (action
& ~CPU_TASKS_FROZEN
) {
628 spin_lock_irq(&pfault_lock
);
629 list_for_each_entry_safe(thread
, next
, &pfault_list
, list
) {
630 thread
->pfault_wait
= 0;
631 list_del(&thread
->list
);
632 tsk
= container_of(thread
, struct task_struct
, thread
);
633 wake_up_process(tsk
);
634 put_task_struct(tsk
);
636 spin_unlock_irq(&pfault_lock
);
644 static int __init
pfault_irq_init(void)
648 rc
= register_external_interrupt(0x2603, pfault_interrupt
);
651 rc
= pfault_init() == 0 ? 0 : -EOPNOTSUPP
;
654 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL
);
655 hotcpu_notifier(pfault_cpu_notify
, 0);
659 unregister_external_interrupt(0x2603, pfault_interrupt
);
664 early_initcall(pfault_irq_init
);
666 #endif /* CONFIG_PFAULT */