3 * Copyright IBM Corp. 1999
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/sched/debug.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/extable.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>
35 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
39 #include <asm/facility.h>
40 #include "../kernel/entry.h"
42 #define __FAIL_ADDR_MASK -4096L
43 #define __SUBCODE_MASK 0x0600
44 #define __PF_RES_FIELD 0x8000000000000000ULL
46 #define VM_FAULT_BADCONTEXT 0x010000
47 #define VM_FAULT_BADMAP 0x020000
48 #define VM_FAULT_BADACCESS 0x040000
49 #define VM_FAULT_SIGNAL 0x080000
50 #define VM_FAULT_PFAULT 0x100000
52 static unsigned long store_indication __read_mostly;
54 static int __init fault_init(void)
56 if (test_facility(75))
57 store_indication = 0xc00;
60 early_initcall(fault_init);
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(struct pt_regs *regs)
106 unsigned long trans_exc_code;
109 * The lowest two bits of the translation exception
110 * identification indicate which paging table was used.
112 trans_exc_code = regs->int_parm_long & 3;
113 if (trans_exc_code == 3) /* home space -> kernel */
117 if (trans_exc_code == 2) /* secondary space -> set_fs */
118 return current->thread.mm_segment.ar4;
119 if (current->flags & PF_VCPU)
124 static int bad_address(void *p)
128 return probe_kernel_address((unsigned long *)p, dummy);
131 static void dump_pagetable(unsigned long asce, unsigned long address)
133 unsigned long *table = __va(asce & _ASCE_ORIGIN);
135 pr_alert("AS:%016lx ", asce);
136 switch (asce & _ASCE_TYPE_MASK) {
137 case _ASCE_TYPE_REGION1:
138 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
139 if (bad_address(table))
141 pr_cont("R1:%016lx ", *table);
142 if (*table & _REGION_ENTRY_INVALID)
144 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
146 case _ASCE_TYPE_REGION2:
147 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
148 if (bad_address(table))
150 pr_cont("R2:%016lx ", *table);
151 if (*table & _REGION_ENTRY_INVALID)
153 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
155 case _ASCE_TYPE_REGION3:
156 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
157 if (bad_address(table))
159 pr_cont("R3:%016lx ", *table);
160 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
162 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
164 case _ASCE_TYPE_SEGMENT:
165 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
166 if (bad_address(table))
168 pr_cont("S:%016lx ", *table);
169 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
171 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
173 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
174 if (bad_address(table))
176 pr_cont("P:%016lx ", *table);
184 static void dump_fault_info(struct pt_regs *regs)
188 pr_alert("Failing address: %016lx TEID: %016lx\n",
189 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
190 pr_alert("Fault in ");
191 switch (regs->int_parm_long & 3) {
193 pr_cont("home space ");
196 pr_cont("secondary space ");
199 pr_cont("access register ");
202 pr_cont("primary space ");
205 pr_cont("mode while using ");
206 if (!user_space_fault(regs)) {
207 asce = S390_lowcore.kernel_asce;
211 else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
212 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
218 asce = S390_lowcore.user_asce;
222 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
225 int show_unhandled_signals = 1;
227 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
229 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
231 if (!unhandled_signal(current, signr))
233 if (!printk_ratelimit())
235 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
236 regs->int_code & 0xffff, regs->int_code >> 17);
237 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
238 printk(KERN_CONT "\n");
240 dump_fault_info(regs);
245 * Send SIGSEGV to task. This is an external routine
246 * to keep the stack usage of do_page_fault small.
248 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
252 report_user_fault(regs, SIGSEGV, 1);
253 si.si_signo = SIGSEGV;
255 si.si_code = si_code;
256 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
257 force_sig_info(SIGSEGV, &si, current);
260 static noinline void do_no_context(struct pt_regs *regs)
262 const struct exception_table_entry *fixup;
264 /* Are we prepared to handle this kernel fault? */
265 fixup = search_exception_tables(regs->psw.addr);
267 regs->psw.addr = extable_fixup(fixup);
272 * Oops. The kernel tried to access some bad page. We'll have to
273 * terminate things with extreme prejudice.
275 if (!user_space_fault(regs))
276 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
277 " in virtual kernel address space\n");
279 printk(KERN_ALERT "Unable to handle kernel paging request"
280 " in virtual user address space\n");
281 dump_fault_info(regs);
286 static noinline void do_low_address(struct pt_regs *regs)
288 /* Low-address protection hit in kernel mode means
289 NULL pointer write access in kernel mode. */
290 if (regs->psw.mask & PSW_MASK_PSTATE) {
291 /* Low-address protection hit in user mode 'cannot happen'. */
292 die (regs, "Low-address protection");
299 static noinline void do_sigbus(struct pt_regs *regs)
301 struct task_struct *tsk = current;
305 * Send a sigbus, regardless of whether we were in kernel
308 si.si_signo = SIGBUS;
310 si.si_code = BUS_ADRERR;
311 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
312 force_sig_info(SIGBUS, &si, tsk);
315 static noinline int signal_return(struct pt_regs *regs)
320 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
323 if (instruction == 0x0a77) {
324 set_pt_regs_flag(regs, PIF_SYSCALL);
325 regs->int_code = 0x00040077;
327 } else if (instruction == 0x0aad) {
328 set_pt_regs_flag(regs, PIF_SYSCALL);
329 regs->int_code = 0x000400ad;
335 static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
340 case VM_FAULT_BADACCESS:
341 if (access == VM_EXEC && signal_return(regs) == 0)
343 case VM_FAULT_BADMAP:
344 /* Bad memory access. Check if it is kernel or user space. */
345 if (user_mode(regs)) {
346 /* User mode accesses just cause a SIGSEGV */
347 si_code = (fault == VM_FAULT_BADMAP) ?
348 SEGV_MAPERR : SEGV_ACCERR;
349 do_sigsegv(regs, si_code);
352 case VM_FAULT_BADCONTEXT:
353 case VM_FAULT_PFAULT:
356 case VM_FAULT_SIGNAL:
357 if (!user_mode(regs))
360 default: /* fault & VM_FAULT_ERROR */
361 if (fault & VM_FAULT_OOM) {
362 if (!user_mode(regs))
365 pagefault_out_of_memory();
366 } else if (fault & VM_FAULT_SIGSEGV) {
367 /* Kernel mode? Handle exceptions or die */
368 if (!user_mode(regs))
371 do_sigsegv(regs, SEGV_MAPERR);
372 } else if (fault & VM_FAULT_SIGBUS) {
373 /* Kernel mode? Handle exceptions or die */
374 if (!user_mode(regs))
385 * This routine handles page faults. It determines the address,
386 * and the problem, and then passes it off to one of the appropriate
389 * interruption code (int_code):
390 * 04 Protection -> Write-Protection (suprression)
391 * 10 Segment translation -> Not present (nullification)
392 * 11 Page translation -> Not present (nullification)
393 * 3b Region third trans. -> Not present (nullification)
395 static inline int do_exception(struct pt_regs *regs, int access)
400 struct task_struct *tsk;
401 struct mm_struct *mm;
402 struct vm_area_struct *vma;
403 unsigned long trans_exc_code;
404 unsigned long address;
410 * The instruction that caused the program check has
411 * been nullified. Don't signal single step via SIGTRAP.
413 clear_pt_regs_flag(regs, PIF_PER_TRAP);
415 if (notify_page_fault(regs))
419 trans_exc_code = regs->int_parm_long;
422 * Verify that the fault happened in user space, that
423 * we are not in an interrupt and that there is a
426 fault = VM_FAULT_BADCONTEXT;
427 if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
430 address = trans_exc_code & __FAIL_ADDR_MASK;
431 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
432 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
434 flags |= FAULT_FLAG_USER;
435 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
436 flags |= FAULT_FLAG_WRITE;
437 down_read(&mm->mmap_sem);
440 gmap = (current->flags & PF_VCPU) ?
441 (struct gmap *) S390_lowcore.gmap : NULL;
443 current->thread.gmap_addr = address;
444 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
445 current->thread.gmap_int_code = regs->int_code & 0xffff;
446 address = __gmap_translate(gmap, address);
447 if (address == -EFAULT) {
448 fault = VM_FAULT_BADMAP;
451 if (gmap->pfault_enabled)
452 flags |= FAULT_FLAG_RETRY_NOWAIT;
457 fault = VM_FAULT_BADMAP;
458 vma = find_vma(mm, address);
462 if (unlikely(vma->vm_start > address)) {
463 if (!(vma->vm_flags & VM_GROWSDOWN))
465 if (expand_stack(vma, address))
470 * Ok, we have a good vm_area for this memory access, so
473 fault = VM_FAULT_BADACCESS;
474 if (unlikely(!(vma->vm_flags & access)))
477 if (is_vm_hugetlb_page(vma))
478 address &= HPAGE_MASK;
480 * If for any reason at all we couldn't handle the fault,
481 * make sure we exit gracefully rather than endlessly redo
484 fault = handle_mm_fault(vma, address, flags);
485 /* No reason to continue if interrupted by SIGKILL. */
486 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
487 fault = VM_FAULT_SIGNAL;
490 if (unlikely(fault & VM_FAULT_ERROR))
494 * Major/minor page fault accounting is only done on the
495 * initial attempt. If we go through a retry, it is extremely
496 * likely that the page will be found in page cache at that point.
498 if (flags & FAULT_FLAG_ALLOW_RETRY) {
499 if (fault & VM_FAULT_MAJOR) {
501 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
505 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
508 if (fault & VM_FAULT_RETRY) {
510 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
511 /* FAULT_FLAG_RETRY_NOWAIT has been set,
512 * mmap_sem has not been released */
513 current->thread.gmap_pfault = 1;
514 fault = VM_FAULT_PFAULT;
518 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
520 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
521 FAULT_FLAG_RETRY_NOWAIT);
522 flags |= FAULT_FLAG_TRIED;
523 down_read(&mm->mmap_sem);
529 address = __gmap_link(gmap, current->thread.gmap_addr,
531 if (address == -EFAULT) {
532 fault = VM_FAULT_BADMAP;
535 if (address == -ENOMEM) {
536 fault = VM_FAULT_OOM;
543 up_read(&mm->mmap_sem);
548 void do_protection_exception(struct pt_regs *regs)
550 unsigned long trans_exc_code;
553 trans_exc_code = regs->int_parm_long;
555 * Protection exceptions are suppressing, decrement psw address.
556 * The exception to this rule are aborted transactions, for these
557 * the PSW already points to the correct location.
559 if (!(regs->int_code & 0x200))
560 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
562 * Check for low-address protection. This needs to be treated
563 * as a special case because the translation exception code
564 * field is not guaranteed to contain valid data in this case.
566 if (unlikely(!(trans_exc_code & 4))) {
567 do_low_address(regs);
570 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
571 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
572 (regs->psw.addr & PAGE_MASK);
574 fault = VM_FAULT_BADACCESS;
577 fault = do_exception(regs, access);
580 do_fault_error(regs, access, fault);
582 NOKPROBE_SYMBOL(do_protection_exception);
584 void do_dat_exception(struct pt_regs *regs)
588 access = VM_READ | VM_EXEC | VM_WRITE;
589 fault = do_exception(regs, access);
591 do_fault_error(regs, access, fault);
593 NOKPROBE_SYMBOL(do_dat_exception);
597 * 'pfault' pseudo page faults routines.
599 static int pfault_disable;
601 static int __init nopfault(char *str)
607 __setup("nopfault", nopfault);
609 struct pfault_refbk {
618 } __attribute__ ((packed, aligned(8)));
620 int pfault_init(void)
622 struct pfault_refbk refbk = {
627 .refgaddr = __LC_LPP,
628 .refselmk = 1ULL << 48,
629 .refcmpmk = 1ULL << 48,
630 .reserved = __PF_RES_FIELD };
635 diag_stat_inc(DIAG_STAT_X258);
637 " diag %1,%0,0x258\n"
642 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
646 void pfault_fini(void)
648 struct pfault_refbk refbk = {
657 diag_stat_inc(DIAG_STAT_X258);
662 : : "a" (&refbk), "m" (refbk) : "cc");
665 static DEFINE_SPINLOCK(pfault_lock);
666 static LIST_HEAD(pfault_list);
668 #define PF_COMPLETE 0x0080
671 * The mechanism of our pfault code: if Linux is running as guest, runs a user
672 * space process and the user space process accesses a page that the host has
673 * paged out we get a pfault interrupt.
675 * This allows us, within the guest, to schedule a different process. Without
676 * this mechanism the host would have to suspend the whole virtual cpu until
677 * the page has been paged in.
679 * So when we get such an interrupt then we set the state of the current task
680 * to uninterruptible and also set the need_resched flag. Both happens within
681 * interrupt context(!). If we later on want to return to user space we
682 * recognize the need_resched flag and then call schedule(). It's not very
683 * obvious how this works...
685 * Of course we have a lot of additional fun with the completion interrupt (->
686 * host signals that a page of a process has been paged in and the process can
687 * continue to run). This interrupt can arrive on any cpu and, since we have
688 * virtual cpus, actually appear before the interrupt that signals that a page
691 static void pfault_interrupt(struct ext_code ext_code,
692 unsigned int param32, unsigned long param64)
694 struct task_struct *tsk;
699 * Get the external interruption subcode & pfault initial/completion
700 * signal bit. VM stores this in the 'cpu address' field associated
701 * with the external interrupt.
703 subcode = ext_code.subcode;
704 if ((subcode & 0xff00) != __SUBCODE_MASK)
706 inc_irq_stat(IRQEXT_PFL);
707 /* Get the token (= pid of the affected task). */
708 pid = param64 & LPP_PFAULT_PID_MASK;
710 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
712 get_task_struct(tsk);
716 spin_lock(&pfault_lock);
717 if (subcode & PF_COMPLETE) {
718 /* signal bit is set -> a page has been swapped in by VM */
719 if (tsk->thread.pfault_wait == 1) {
720 /* Initial interrupt was faster than the completion
721 * interrupt. pfault_wait is valid. Set pfault_wait
722 * back to zero and wake up the process. This can
723 * safely be done because the task is still sleeping
724 * and can't produce new pfaults. */
725 tsk->thread.pfault_wait = 0;
726 list_del(&tsk->thread.list);
727 wake_up_process(tsk);
728 put_task_struct(tsk);
730 /* Completion interrupt was faster than initial
731 * interrupt. Set pfault_wait to -1 so the initial
732 * interrupt doesn't put the task to sleep.
733 * If the task is not running, ignore the completion
734 * interrupt since it must be a leftover of a PFAULT
735 * CANCEL operation which didn't remove all pending
736 * completion interrupts. */
737 if (tsk->state == TASK_RUNNING)
738 tsk->thread.pfault_wait = -1;
741 /* signal bit not set -> a real page is missing. */
742 if (WARN_ON_ONCE(tsk != current))
744 if (tsk->thread.pfault_wait == 1) {
745 /* Already on the list with a reference: put to sleep */
747 } else if (tsk->thread.pfault_wait == -1) {
748 /* Completion interrupt was faster than the initial
749 * interrupt (pfault_wait == -1). Set pfault_wait
750 * back to zero and exit. */
751 tsk->thread.pfault_wait = 0;
753 /* Initial interrupt arrived before completion
754 * interrupt. Let the task sleep.
755 * An extra task reference is needed since a different
756 * cpu may set the task state to TASK_RUNNING again
757 * before the scheduler is reached. */
758 get_task_struct(tsk);
759 tsk->thread.pfault_wait = 1;
760 list_add(&tsk->thread.list, &pfault_list);
762 /* Since this must be a userspace fault, there
763 * is no kernel task state to trample. Rely on the
764 * return to userspace schedule() to block. */
765 __set_current_state(TASK_UNINTERRUPTIBLE);
766 set_tsk_need_resched(tsk);
767 set_preempt_need_resched();
771 spin_unlock(&pfault_lock);
772 put_task_struct(tsk);
775 static int pfault_cpu_dead(unsigned int cpu)
777 struct thread_struct *thread, *next;
778 struct task_struct *tsk;
780 spin_lock_irq(&pfault_lock);
781 list_for_each_entry_safe(thread, next, &pfault_list, list) {
782 thread->pfault_wait = 0;
783 list_del(&thread->list);
784 tsk = container_of(thread, struct task_struct, thread);
785 wake_up_process(tsk);
786 put_task_struct(tsk);
788 spin_unlock_irq(&pfault_lock);
792 static int __init pfault_irq_init(void)
796 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
799 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
802 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
803 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
804 NULL, pfault_cpu_dead);
808 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
813 early_initcall(pfault_irq_init);
815 #endif /* CONFIG_PFAULT */