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
54 static unsigned long store_indication;
58 if (test_facility(2) && test_facility(75))
59 store_indication = 0xc00;
62 static inline int notify_page_fault(struct pt_regs *regs)
66 /* kprobe_running() needs smp_processor_id() */
67 if (kprobes_built_in() && !user_mode(regs)) {
69 if (kprobe_running() && kprobe_fault_handler(regs, 14))
78 * Unlock any spinlocks which will prevent us from getting the
81 void bust_spinlocks(int yes)
86 int loglevel_save = console_loglevel;
90 * OK, the message is on the console. Now we call printk()
91 * without oops_in_progress set so that printk will give klogd
92 * a poke. Hold onto your hats...
94 console_loglevel = 15;
96 console_loglevel = loglevel_save;
101 * Returns the address space associated with the fault.
102 * Returns 0 for kernel space and 1 for user space.
104 static inline int user_space_fault(unsigned long trans_exc_code)
107 * The lowest two bits of the translation exception
108 * identification indicate which paging table was used.
111 if (trans_exc_code == 2)
112 /* Access via secondary space, set_fs setting decides */
113 return current->thread.mm_segment.ar4;
114 if (s390_user_mode == HOME_SPACE_MODE)
115 /* User space if the access has been done via home space. */
116 return trans_exc_code == 3;
118 * If the user space is not the home space the kernel runs in home
119 * space. Access via secondary space has already been covered,
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 = 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:
233 case VM_FAULT_SIGNAL:
234 if (!user_mode(regs))
237 default: /* fault & VM_FAULT_ERROR */
238 if (fault & VM_FAULT_OOM) {
239 if (!user_mode(regs))
242 pagefault_out_of_memory();
243 } else if (fault & VM_FAULT_SIGBUS) {
244 /* Kernel mode? Handle exceptions or die */
245 if (!user_mode(regs))
256 * This routine handles page faults. It determines the address,
257 * and the problem, and then passes it off to one of the appropriate
260 * interruption code (int_code):
261 * 04 Protection -> Write-Protection (suprression)
262 * 10 Segment translation -> Not present (nullification)
263 * 11 Page translation -> Not present (nullification)
264 * 3b Region third trans. -> Not present (nullification)
266 static inline int do_exception(struct pt_regs *regs, int access)
268 struct task_struct *tsk;
269 struct mm_struct *mm;
270 struct vm_area_struct *vma;
271 unsigned long trans_exc_code;
272 unsigned long address;
276 if (notify_page_fault(regs))
281 trans_exc_code = regs->int_parm_long;
284 * Verify that the fault happened in user space, that
285 * we are not in an interrupt and that there is a
288 fault = VM_FAULT_BADCONTEXT;
289 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
292 address = trans_exc_code & __FAIL_ADDR_MASK;
293 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
294 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
295 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
296 flags |= FAULT_FLAG_WRITE;
297 down_read(&mm->mmap_sem);
300 if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
301 address = __gmap_fault(address,
302 (struct gmap *) S390_lowcore.gmap);
303 if (address == -EFAULT) {
304 fault = VM_FAULT_BADMAP;
307 if (address == -ENOMEM) {
308 fault = VM_FAULT_OOM;
315 fault = VM_FAULT_BADMAP;
316 vma = find_vma(mm, address);
320 if (unlikely(vma->vm_start > address)) {
321 if (!(vma->vm_flags & VM_GROWSDOWN))
323 if (expand_stack(vma, address))
328 * Ok, we have a good vm_area for this memory access, so
331 fault = VM_FAULT_BADACCESS;
332 if (unlikely(!(vma->vm_flags & access)))
335 if (is_vm_hugetlb_page(vma))
336 address &= HPAGE_MASK;
338 * If for any reason at all we couldn't handle the fault,
339 * make sure we exit gracefully rather than endlessly redo
342 fault = handle_mm_fault(mm, vma, address, flags);
343 /* No reason to continue if interrupted by SIGKILL. */
344 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
345 fault = VM_FAULT_SIGNAL;
348 if (unlikely(fault & VM_FAULT_ERROR))
352 * Major/minor page fault accounting is only done on the
353 * initial attempt. If we go through a retry, it is extremely
354 * likely that the page will be found in page cache at that point.
356 if (flags & FAULT_FLAG_ALLOW_RETRY) {
357 if (fault & VM_FAULT_MAJOR) {
359 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
363 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
366 if (fault & VM_FAULT_RETRY) {
367 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
369 flags &= ~FAULT_FLAG_ALLOW_RETRY;
370 down_read(&mm->mmap_sem);
375 * The instruction that caused the program check will
376 * be repeated. Don't signal single step via SIGTRAP.
378 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
381 up_read(&mm->mmap_sem);
386 void __kprobes do_protection_exception(struct pt_regs *regs)
388 unsigned long trans_exc_code;
391 trans_exc_code = regs->int_parm_long;
392 /* Protection exception is suppressing, decrement psw address. */
393 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
395 * Check for low-address protection. This needs to be treated
396 * as a special case because the translation exception code
397 * field is not guaranteed to contain valid data in this case.
399 if (unlikely(!(trans_exc_code & 4))) {
400 do_low_address(regs);
403 fault = do_exception(regs, VM_WRITE);
405 do_fault_error(regs, fault);
408 void __kprobes do_dat_exception(struct pt_regs *regs)
412 access = VM_READ | VM_EXEC | VM_WRITE;
413 fault = do_exception(regs, access);
415 do_fault_error(regs, fault);
419 void __kprobes do_asce_exception(struct pt_regs *regs)
421 struct mm_struct *mm = current->mm;
422 struct vm_area_struct *vma;
423 unsigned long trans_exc_code;
425 trans_exc_code = regs->int_parm_long;
426 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
429 down_read(&mm->mmap_sem);
430 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
431 up_read(&mm->mmap_sem);
434 update_mm(mm, current);
438 /* User mode accesses just cause a SIGSEGV */
439 if (user_mode(regs)) {
440 do_sigsegv(regs, SEGV_MAPERR);
449 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
454 /* Emulate a uaccess fault from kernel mode. */
455 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
456 if (!irqs_disabled())
457 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
458 regs.psw.addr = (unsigned long) __builtin_return_address(0);
459 regs.psw.addr |= PSW_ADDR_AMODE;
460 regs.int_code = pgm_int_code;
461 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
462 access = write ? VM_WRITE : VM_READ;
463 fault = do_exception(®s, access);
465 * Since the fault happened in kernel mode while performing a uaccess
466 * all we need to do now is emulating a fixup in case "fault" is not
468 * For the calling uaccess functions this results always in -EFAULT.
470 return fault ? -EFAULT : 0;
475 * 'pfault' pseudo page faults routines.
477 static int pfault_disable;
479 static int __init nopfault(char *str)
485 __setup("nopfault", nopfault);
487 struct pfault_refbk {
496 } __attribute__ ((packed, aligned(8)));
498 int pfault_init(void)
500 struct pfault_refbk refbk = {
505 .refgaddr = __LC_CURRENT_PID,
506 .refselmk = 1ULL << 48,
507 .refcmpmk = 1ULL << 48,
508 .reserved = __PF_RES_FIELD };
514 " diag %1,%0,0x258\n"
519 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
523 void pfault_fini(void)
525 struct pfault_refbk refbk = {
538 : : "a" (&refbk), "m" (refbk) : "cc");
541 static DEFINE_SPINLOCK(pfault_lock);
542 static LIST_HEAD(pfault_list);
544 static void pfault_interrupt(struct ext_code ext_code,
545 unsigned int param32, unsigned long param64)
547 struct task_struct *tsk;
552 * Get the external interruption subcode & pfault
553 * initial/completion signal bit. VM stores this
554 * in the 'cpu address' field associated with the
555 * external interrupt.
557 subcode = ext_code.subcode;
558 if ((subcode & 0xff00) != __SUBCODE_MASK)
560 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
561 /* Get the token (= pid of the affected task). */
562 pid = sizeof(void *) == 4 ? param32 : param64;
564 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
566 get_task_struct(tsk);
570 spin_lock(&pfault_lock);
571 if (subcode & 0x0080) {
572 /* signal bit is set -> a page has been swapped in by VM */
573 if (tsk->thread.pfault_wait == 1) {
574 /* Initial interrupt was faster than the completion
575 * interrupt. pfault_wait is valid. Set pfault_wait
576 * back to zero and wake up the process. This can
577 * safely be done because the task is still sleeping
578 * and can't produce new pfaults. */
579 tsk->thread.pfault_wait = 0;
580 list_del(&tsk->thread.list);
581 wake_up_process(tsk);
582 put_task_struct(tsk);
584 /* Completion interrupt was faster than initial
585 * interrupt. Set pfault_wait to -1 so the initial
586 * interrupt doesn't put the task to sleep.
587 * If the task is not running, ignore the completion
588 * interrupt since it must be a leftover of a PFAULT
589 * CANCEL operation which didn't remove all pending
590 * completion interrupts. */
591 if (tsk->state == TASK_RUNNING)
592 tsk->thread.pfault_wait = -1;
595 /* signal bit not set -> a real page is missing. */
596 if (WARN_ON_ONCE(tsk != current))
598 if (tsk->thread.pfault_wait == 1) {
599 /* Already on the list with a reference: put to sleep */
600 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
601 set_tsk_need_resched(tsk);
602 } else if (tsk->thread.pfault_wait == -1) {
603 /* Completion interrupt was faster than the initial
604 * interrupt (pfault_wait == -1). Set pfault_wait
605 * back to zero and exit. */
606 tsk->thread.pfault_wait = 0;
608 /* Initial interrupt arrived before completion
609 * interrupt. Let the task sleep.
610 * An extra task reference is needed since a different
611 * cpu may set the task state to TASK_RUNNING again
612 * before the scheduler is reached. */
613 get_task_struct(tsk);
614 tsk->thread.pfault_wait = 1;
615 list_add(&tsk->thread.list, &pfault_list);
616 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
617 set_tsk_need_resched(tsk);
621 spin_unlock(&pfault_lock);
622 put_task_struct(tsk);
625 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
626 unsigned long action, void *hcpu)
628 struct thread_struct *thread, *next;
629 struct task_struct *tsk;
631 switch (action & ~CPU_TASKS_FROZEN) {
633 spin_lock_irq(&pfault_lock);
634 list_for_each_entry_safe(thread, next, &pfault_list, list) {
635 thread->pfault_wait = 0;
636 list_del(&thread->list);
637 tsk = container_of(thread, struct task_struct, thread);
638 wake_up_process(tsk);
639 put_task_struct(tsk);
641 spin_unlock_irq(&pfault_lock);
649 static int __init pfault_irq_init(void)
653 rc = register_external_interrupt(0x2603, pfault_interrupt);
656 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
659 service_subclass_irq_register();
660 hotcpu_notifier(pfault_cpu_notify, 0);
664 unregister_external_interrupt(0x2603, pfault_interrupt);
669 early_initcall(pfault_irq_init);
671 #endif /* CONFIG_PFAULT */