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>
34 #include <asm/pgtable.h>
37 #include <asm/mmu_context.h>
38 #include <asm/facility.h>
39 #include "../kernel/entry.h"
41 #define __FAIL_ADDR_MASK -4096L
42 #define __SUBCODE_MASK 0x0600
43 #define __PF_RES_FIELD 0x8000000000000000ULL
45 #define VM_FAULT_BADCONTEXT 0x010000
46 #define VM_FAULT_BADMAP 0x020000
47 #define VM_FAULT_BADACCESS 0x040000
48 #define VM_FAULT_SIGNAL 0x080000
49 #define VM_FAULT_PFAULT 0x100000
51 static unsigned long store_indication __read_mostly;
53 static int __init fault_init(void)
55 if (test_facility(75))
56 store_indication = 0xc00;
59 early_initcall(fault_init);
61 static inline int notify_page_fault(struct pt_regs *regs)
65 /* kprobe_running() needs smp_processor_id() */
66 if (kprobes_built_in() && !user_mode(regs)) {
68 if (kprobe_running() && kprobe_fault_handler(regs, 14))
77 * Unlock any spinlocks which will prevent us from getting the
80 void bust_spinlocks(int yes)
85 int loglevel_save = console_loglevel;
89 * OK, the message is on the console. Now we call printk()
90 * without oops_in_progress set so that printk will give klogd
91 * a poke. Hold onto your hats...
93 console_loglevel = 15;
95 console_loglevel = loglevel_save;
100 * Returns the address space associated with the fault.
101 * Returns 0 for kernel space and 1 for user space.
103 static inline int user_space_fault(struct pt_regs *regs)
105 unsigned long trans_exc_code;
108 * The lowest two bits of the translation exception
109 * identification indicate which paging table was used.
111 trans_exc_code = regs->int_parm_long & 3;
112 if (trans_exc_code == 3) /* home space -> kernel */
116 if (trans_exc_code == 2) /* secondary space -> set_fs */
117 return current->thread.mm_segment.ar4;
118 if (current->flags & PF_VCPU)
123 static int bad_address(void *p)
127 return probe_kernel_address((unsigned long *)p, dummy);
130 static void dump_pagetable(unsigned long asce, unsigned long address)
132 unsigned long *table = __va(asce & PAGE_MASK);
134 pr_alert("AS:%016lx ", asce);
135 switch (asce & _ASCE_TYPE_MASK) {
136 case _ASCE_TYPE_REGION1:
137 table = table + ((address >> 53) & 0x7ff);
138 if (bad_address(table))
140 pr_cont("R1:%016lx ", *table);
141 if (*table & _REGION_ENTRY_INVALID)
143 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
145 case _ASCE_TYPE_REGION2:
146 table = table + ((address >> 42) & 0x7ff);
147 if (bad_address(table))
149 pr_cont("R2:%016lx ", *table);
150 if (*table & _REGION_ENTRY_INVALID)
152 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
154 case _ASCE_TYPE_REGION3:
155 table = table + ((address >> 31) & 0x7ff);
156 if (bad_address(table))
158 pr_cont("R3:%016lx ", *table);
159 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
161 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
163 case _ASCE_TYPE_SEGMENT:
164 table = table + ((address >> 20) & 0x7ff);
165 if (bad_address(table))
167 pr_cont("S:%016lx ", *table);
168 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
170 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
172 table = table + ((address >> 12) & 0xff);
173 if (bad_address(table))
175 pr_cont("P:%016lx ", *table);
183 static void dump_fault_info(struct pt_regs *regs)
187 pr_alert("Failing address: %016lx TEID: %016lx\n",
188 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
189 pr_alert("Fault in ");
190 switch (regs->int_parm_long & 3) {
192 pr_cont("home space ");
195 pr_cont("secondary space ");
198 pr_cont("access register ");
201 pr_cont("primary space ");
204 pr_cont("mode while using ");
205 if (!user_space_fault(regs)) {
206 asce = S390_lowcore.kernel_asce;
210 else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
211 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
217 asce = S390_lowcore.user_asce;
221 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
224 int show_unhandled_signals = 1;
226 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
228 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
230 if (!unhandled_signal(current, signr))
232 if (!printk_ratelimit())
234 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
235 regs->int_code & 0xffff, regs->int_code >> 17);
236 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
237 printk(KERN_CONT "\n");
239 dump_fault_info(regs);
244 * Send SIGSEGV to task. This is an external routine
245 * to keep the stack usage of do_page_fault small.
247 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
251 report_user_fault(regs, SIGSEGV, 1);
252 si.si_signo = SIGSEGV;
253 si.si_code = si_code;
254 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
255 force_sig_info(SIGSEGV, &si, current);
258 static noinline void do_no_context(struct pt_regs *regs)
260 const struct exception_table_entry *fixup;
262 /* Are we prepared to handle this kernel fault? */
263 fixup = search_exception_tables(regs->psw.addr);
265 regs->psw.addr = extable_fixup(fixup);
270 * Oops. The kernel tried to access some bad page. We'll have to
271 * terminate things with extreme prejudice.
273 if (!user_space_fault(regs))
274 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
275 " in virtual kernel address space\n");
277 printk(KERN_ALERT "Unable to handle kernel paging request"
278 " in virtual user address space\n");
279 dump_fault_info(regs);
284 static noinline void do_low_address(struct pt_regs *regs)
286 /* Low-address protection hit in kernel mode means
287 NULL pointer write access in kernel mode. */
288 if (regs->psw.mask & PSW_MASK_PSTATE) {
289 /* Low-address protection hit in user mode 'cannot happen'. */
290 die (regs, "Low-address protection");
297 static noinline void do_sigbus(struct pt_regs *regs)
299 struct task_struct *tsk = current;
303 * Send a sigbus, regardless of whether we were in kernel
306 si.si_signo = SIGBUS;
308 si.si_code = BUS_ADRERR;
309 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
310 force_sig_info(SIGBUS, &si, tsk);
313 static noinline void do_fault_error(struct pt_regs *regs, int fault)
318 case VM_FAULT_BADACCESS:
319 case VM_FAULT_BADMAP:
320 /* Bad memory access. Check if it is kernel or user space. */
321 if (user_mode(regs)) {
322 /* User mode accesses just cause a SIGSEGV */
323 si_code = (fault == VM_FAULT_BADMAP) ?
324 SEGV_MAPERR : SEGV_ACCERR;
325 do_sigsegv(regs, si_code);
328 case VM_FAULT_BADCONTEXT:
329 case VM_FAULT_PFAULT:
332 case VM_FAULT_SIGNAL:
333 if (!user_mode(regs))
336 default: /* fault & VM_FAULT_ERROR */
337 if (fault & VM_FAULT_OOM) {
338 if (!user_mode(regs))
341 pagefault_out_of_memory();
342 } else if (fault & VM_FAULT_SIGSEGV) {
343 /* Kernel mode? Handle exceptions or die */
344 if (!user_mode(regs))
347 do_sigsegv(regs, SEGV_MAPERR);
348 } else if (fault & VM_FAULT_SIGBUS) {
349 /* Kernel mode? Handle exceptions or die */
350 if (!user_mode(regs))
361 * This routine handles page faults. It determines the address,
362 * and the problem, and then passes it off to one of the appropriate
365 * interruption code (int_code):
366 * 04 Protection -> Write-Protection (suprression)
367 * 10 Segment translation -> Not present (nullification)
368 * 11 Page translation -> Not present (nullification)
369 * 3b Region third trans. -> Not present (nullification)
371 static inline int do_exception(struct pt_regs *regs, int access)
376 struct task_struct *tsk;
377 struct mm_struct *mm;
378 struct vm_area_struct *vma;
379 unsigned long trans_exc_code;
380 unsigned long address;
386 * The instruction that caused the program check has
387 * been nullified. Don't signal single step via SIGTRAP.
389 clear_pt_regs_flag(regs, PIF_PER_TRAP);
391 if (notify_page_fault(regs))
395 trans_exc_code = regs->int_parm_long;
398 * Verify that the fault happened in user space, that
399 * we are not in an interrupt and that there is a
402 fault = VM_FAULT_BADCONTEXT;
403 if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
406 address = trans_exc_code & __FAIL_ADDR_MASK;
407 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
408 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
410 flags |= FAULT_FLAG_USER;
411 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
412 flags |= FAULT_FLAG_WRITE;
413 down_read(&mm->mmap_sem);
416 gmap = (current->flags & PF_VCPU) ?
417 (struct gmap *) S390_lowcore.gmap : NULL;
419 current->thread.gmap_addr = address;
420 address = __gmap_translate(gmap, address);
421 if (address == -EFAULT) {
422 fault = VM_FAULT_BADMAP;
425 if (gmap->pfault_enabled)
426 flags |= FAULT_FLAG_RETRY_NOWAIT;
431 fault = VM_FAULT_BADMAP;
432 vma = find_vma(mm, address);
436 if (unlikely(vma->vm_start > address)) {
437 if (!(vma->vm_flags & VM_GROWSDOWN))
439 if (expand_stack(vma, address))
444 * Ok, we have a good vm_area for this memory access, so
447 fault = VM_FAULT_BADACCESS;
448 if (unlikely(!(vma->vm_flags & access)))
451 if (is_vm_hugetlb_page(vma))
452 address &= HPAGE_MASK;
454 * If for any reason at all we couldn't handle the fault,
455 * make sure we exit gracefully rather than endlessly redo
458 fault = handle_mm_fault(mm, vma, address, flags);
459 /* No reason to continue if interrupted by SIGKILL. */
460 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
461 fault = VM_FAULT_SIGNAL;
464 if (unlikely(fault & VM_FAULT_ERROR))
468 * Major/minor page fault accounting is only done on the
469 * initial attempt. If we go through a retry, it is extremely
470 * likely that the page will be found in page cache at that point.
472 if (flags & FAULT_FLAG_ALLOW_RETRY) {
473 if (fault & VM_FAULT_MAJOR) {
475 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
479 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
482 if (fault & VM_FAULT_RETRY) {
484 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
485 /* FAULT_FLAG_RETRY_NOWAIT has been set,
486 * mmap_sem has not been released */
487 current->thread.gmap_pfault = 1;
488 fault = VM_FAULT_PFAULT;
492 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
494 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
495 FAULT_FLAG_RETRY_NOWAIT);
496 flags |= FAULT_FLAG_TRIED;
497 down_read(&mm->mmap_sem);
503 address = __gmap_link(gmap, current->thread.gmap_addr,
505 if (address == -EFAULT) {
506 fault = VM_FAULT_BADMAP;
509 if (address == -ENOMEM) {
510 fault = VM_FAULT_OOM;
517 up_read(&mm->mmap_sem);
522 void do_protection_exception(struct pt_regs *regs)
524 unsigned long trans_exc_code;
527 trans_exc_code = regs->int_parm_long;
529 * Protection exceptions are suppressing, decrement psw address.
530 * The exception to this rule are aborted transactions, for these
531 * the PSW already points to the correct location.
533 if (!(regs->int_code & 0x200))
534 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
536 * Check for low-address protection. This needs to be treated
537 * as a special case because the translation exception code
538 * field is not guaranteed to contain valid data in this case.
540 if (unlikely(!(trans_exc_code & 4))) {
541 do_low_address(regs);
544 fault = do_exception(regs, VM_WRITE);
546 do_fault_error(regs, fault);
548 NOKPROBE_SYMBOL(do_protection_exception);
550 void do_dat_exception(struct pt_regs *regs)
554 access = VM_READ | VM_EXEC | VM_WRITE;
555 fault = do_exception(regs, access);
557 do_fault_error(regs, fault);
559 NOKPROBE_SYMBOL(do_dat_exception);
563 * 'pfault' pseudo page faults routines.
565 static int pfault_disable;
567 static int __init nopfault(char *str)
573 __setup("nopfault", nopfault);
575 struct pfault_refbk {
584 } __attribute__ ((packed, aligned(8)));
586 int pfault_init(void)
588 struct pfault_refbk refbk = {
593 .refgaddr = __LC_LPP,
594 .refselmk = 1ULL << 48,
595 .refcmpmk = 1ULL << 48,
596 .reserved = __PF_RES_FIELD };
601 diag_stat_inc(DIAG_STAT_X258);
603 " diag %1,%0,0x258\n"
608 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
612 void pfault_fini(void)
614 struct pfault_refbk refbk = {
623 diag_stat_inc(DIAG_STAT_X258);
628 : : "a" (&refbk), "m" (refbk) : "cc");
631 static DEFINE_SPINLOCK(pfault_lock);
632 static LIST_HEAD(pfault_list);
634 #define PF_COMPLETE 0x0080
637 * The mechanism of our pfault code: if Linux is running as guest, runs a user
638 * space process and the user space process accesses a page that the host has
639 * paged out we get a pfault interrupt.
641 * This allows us, within the guest, to schedule a different process. Without
642 * this mechanism the host would have to suspend the whole virtual cpu until
643 * the page has been paged in.
645 * So when we get such an interrupt then we set the state of the current task
646 * to uninterruptible and also set the need_resched flag. Both happens within
647 * interrupt context(!). If we later on want to return to user space we
648 * recognize the need_resched flag and then call schedule(). It's not very
649 * obvious how this works...
651 * Of course we have a lot of additional fun with the completion interrupt (->
652 * host signals that a page of a process has been paged in and the process can
653 * continue to run). This interrupt can arrive on any cpu and, since we have
654 * virtual cpus, actually appear before the interrupt that signals that a page
657 static void pfault_interrupt(struct ext_code ext_code,
658 unsigned int param32, unsigned long param64)
660 struct task_struct *tsk;
665 * Get the external interruption subcode & pfault initial/completion
666 * signal bit. VM stores this in the 'cpu address' field associated
667 * with the external interrupt.
669 subcode = ext_code.subcode;
670 if ((subcode & 0xff00) != __SUBCODE_MASK)
672 inc_irq_stat(IRQEXT_PFL);
673 /* Get the token (= pid of the affected task). */
674 pid = param64 & LPP_PFAULT_PID_MASK;
676 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
678 get_task_struct(tsk);
682 spin_lock(&pfault_lock);
683 if (subcode & PF_COMPLETE) {
684 /* signal bit is set -> a page has been swapped in by VM */
685 if (tsk->thread.pfault_wait == 1) {
686 /* Initial interrupt was faster than the completion
687 * interrupt. pfault_wait is valid. Set pfault_wait
688 * back to zero and wake up the process. This can
689 * safely be done because the task is still sleeping
690 * and can't produce new pfaults. */
691 tsk->thread.pfault_wait = 0;
692 list_del(&tsk->thread.list);
693 wake_up_process(tsk);
694 put_task_struct(tsk);
696 /* Completion interrupt was faster than initial
697 * interrupt. Set pfault_wait to -1 so the initial
698 * interrupt doesn't put the task to sleep.
699 * If the task is not running, ignore the completion
700 * interrupt since it must be a leftover of a PFAULT
701 * CANCEL operation which didn't remove all pending
702 * completion interrupts. */
703 if (tsk->state == TASK_RUNNING)
704 tsk->thread.pfault_wait = -1;
707 /* signal bit not set -> a real page is missing. */
708 if (WARN_ON_ONCE(tsk != current))
710 if (tsk->thread.pfault_wait == 1) {
711 /* Already on the list with a reference: put to sleep */
713 } else if (tsk->thread.pfault_wait == -1) {
714 /* Completion interrupt was faster than the initial
715 * interrupt (pfault_wait == -1). Set pfault_wait
716 * back to zero and exit. */
717 tsk->thread.pfault_wait = 0;
719 /* Initial interrupt arrived before completion
720 * interrupt. Let the task sleep.
721 * An extra task reference is needed since a different
722 * cpu may set the task state to TASK_RUNNING again
723 * before the scheduler is reached. */
724 get_task_struct(tsk);
725 tsk->thread.pfault_wait = 1;
726 list_add(&tsk->thread.list, &pfault_list);
728 /* Since this must be a userspace fault, there
729 * is no kernel task state to trample. Rely on the
730 * return to userspace schedule() to block. */
731 __set_current_state(TASK_UNINTERRUPTIBLE);
732 set_tsk_need_resched(tsk);
736 spin_unlock(&pfault_lock);
737 put_task_struct(tsk);
740 static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
743 struct thread_struct *thread, *next;
744 struct task_struct *tsk;
746 switch (action & ~CPU_TASKS_FROZEN) {
748 spin_lock_irq(&pfault_lock);
749 list_for_each_entry_safe(thread, next, &pfault_list, list) {
750 thread->pfault_wait = 0;
751 list_del(&thread->list);
752 tsk = container_of(thread, struct task_struct, thread);
753 wake_up_process(tsk);
754 put_task_struct(tsk);
756 spin_unlock_irq(&pfault_lock);
764 static int __init pfault_irq_init(void)
768 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
771 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
774 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
775 hotcpu_notifier(pfault_cpu_notify, 0);
779 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
784 early_initcall(pfault_irq_init);
786 #endif /* CONFIG_PFAULT */