1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright IBM Corp. 1999
5 * Author(s): Hartmut Penner (hp@de.ibm.com)
6 * Ulrich Weigand (uweigand@de.ibm.com)
8 * Derived from "arch/i386/mm/fault.c"
9 * Copyright (C) 1995 Linus Torvalds
12 #include <linux/kernel_stat.h>
13 #include <linux/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
24 #include <linux/compat.h>
25 #include <linux/smp.h>
26 #include <linux/kdebug.h>
27 #include <linux/init.h>
28 #include <linux/console.h>
29 #include <linux/extable.h>
30 #include <linux/hardirq.h>
31 #include <linux/kprobes.h>
32 #include <linux/uaccess.h>
33 #include <linux/hugetlb.h>
34 #include <asm/asm-offsets.h>
36 #include <asm/pgtable.h>
39 #include <asm/mmu_context.h>
40 #include <asm/facility.h>
41 #include "../kernel/entry.h"
43 #define __FAIL_ADDR_MASK -4096L
44 #define __SUBCODE_MASK 0x0600
45 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #define VM_FAULT_BADCONTEXT 0x010000
48 #define VM_FAULT_BADMAP 0x020000
49 #define VM_FAULT_BADACCESS 0x040000
50 #define VM_FAULT_SIGNAL 0x080000
51 #define VM_FAULT_PFAULT 0x100000
60 static unsigned long store_indication __read_mostly;
62 static int __init fault_init(void)
64 if (test_facility(75))
65 store_indication = 0xc00;
68 early_initcall(fault_init);
70 static inline int notify_page_fault(struct pt_regs *regs)
74 /* kprobe_running() needs smp_processor_id() */
75 if (kprobes_built_in() && !user_mode(regs)) {
77 if (kprobe_running() && kprobe_fault_handler(regs, 14))
86 * Unlock any spinlocks which will prevent us from getting the
89 void bust_spinlocks(int yes)
94 int loglevel_save = console_loglevel;
98 * OK, the message is on the console. Now we call printk()
99 * without oops_in_progress set so that printk will give klogd
100 * a poke. Hold onto your hats...
102 console_loglevel = 15;
104 console_loglevel = loglevel_save;
109 * Find out which address space caused the exception.
110 * Access register mode is impossible, ignore space == 3.
112 static inline enum fault_type get_fault_type(struct pt_regs *regs)
114 unsigned long trans_exc_code;
116 trans_exc_code = regs->int_parm_long & 3;
117 if (likely(trans_exc_code == 0)) {
118 /* primary space exception */
119 if (IS_ENABLED(CONFIG_PGSTE) &&
120 test_pt_regs_flag(regs, PIF_GUEST_FAULT))
122 if (current->thread.mm_segment == USER_DS)
126 if (trans_exc_code == 2) {
127 /* secondary space exception */
128 if (current->thread.mm_segment & 1) {
129 if (current->thread.mm_segment == USER_DS_SACF)
135 /* home space exception -> access via kernel ASCE */
139 static int bad_address(void *p)
143 return probe_kernel_address((unsigned long *)p, dummy);
146 static void dump_pagetable(unsigned long asce, unsigned long address)
148 unsigned long *table = __va(asce & _ASCE_ORIGIN);
150 pr_alert("AS:%016lx ", asce);
151 switch (asce & _ASCE_TYPE_MASK) {
152 case _ASCE_TYPE_REGION1:
153 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
154 if (bad_address(table))
156 pr_cont("R1:%016lx ", *table);
157 if (*table & _REGION_ENTRY_INVALID)
159 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
161 case _ASCE_TYPE_REGION2:
162 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
163 if (bad_address(table))
165 pr_cont("R2:%016lx ", *table);
166 if (*table & _REGION_ENTRY_INVALID)
168 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
170 case _ASCE_TYPE_REGION3:
171 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
172 if (bad_address(table))
174 pr_cont("R3:%016lx ", *table);
175 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
177 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
179 case _ASCE_TYPE_SEGMENT:
180 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
181 if (bad_address(table))
183 pr_cont("S:%016lx ", *table);
184 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
186 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
188 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
189 if (bad_address(table))
191 pr_cont("P:%016lx ", *table);
199 static void dump_fault_info(struct pt_regs *regs)
203 pr_alert("Failing address: %016lx TEID: %016lx\n",
204 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
205 pr_alert("Fault in ");
206 switch (regs->int_parm_long & 3) {
208 pr_cont("home space ");
211 pr_cont("secondary space ");
214 pr_cont("access register ");
217 pr_cont("primary space ");
220 pr_cont("mode while using ");
221 switch (get_fault_type(regs)) {
223 asce = S390_lowcore.user_asce;
227 asce = S390_lowcore.vdso_asce;
231 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
235 asce = S390_lowcore.kernel_asce;
240 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
243 int show_unhandled_signals = 1;
245 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
247 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
249 if (!unhandled_signal(current, signr))
251 if (!printk_ratelimit())
253 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
254 regs->int_code & 0xffff, regs->int_code >> 17);
255 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
256 printk(KERN_CONT "\n");
258 dump_fault_info(regs);
263 * Send SIGSEGV to task. This is an external routine
264 * to keep the stack usage of do_page_fault small.
266 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
270 report_user_fault(regs, SIGSEGV, 1);
271 si.si_signo = SIGSEGV;
273 si.si_code = si_code;
274 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
275 force_sig_info(SIGSEGV, &si, current);
278 static noinline void do_no_context(struct pt_regs *regs)
280 const struct exception_table_entry *fixup;
282 /* Are we prepared to handle this kernel fault? */
283 fixup = search_exception_tables(regs->psw.addr);
285 regs->psw.addr = extable_fixup(fixup);
290 * Oops. The kernel tried to access some bad page. We'll have to
291 * terminate things with extreme prejudice.
293 if (get_fault_type(regs) == KERNEL_FAULT)
294 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
295 " in virtual kernel address space\n");
297 printk(KERN_ALERT "Unable to handle kernel paging request"
298 " in virtual user address space\n");
299 dump_fault_info(regs);
304 static noinline void do_low_address(struct pt_regs *regs)
306 /* Low-address protection hit in kernel mode means
307 NULL pointer write access in kernel mode. */
308 if (regs->psw.mask & PSW_MASK_PSTATE) {
309 /* Low-address protection hit in user mode 'cannot happen'. */
310 die (regs, "Low-address protection");
317 static noinline void do_sigbus(struct pt_regs *regs)
319 struct task_struct *tsk = current;
323 * Send a sigbus, regardless of whether we were in kernel
326 si.si_signo = SIGBUS;
328 si.si_code = BUS_ADRERR;
329 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
330 force_sig_info(SIGBUS, &si, tsk);
333 static noinline int signal_return(struct pt_regs *regs)
338 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
341 if (instruction == 0x0a77) {
342 set_pt_regs_flag(regs, PIF_SYSCALL);
343 regs->int_code = 0x00040077;
345 } else if (instruction == 0x0aad) {
346 set_pt_regs_flag(regs, PIF_SYSCALL);
347 regs->int_code = 0x000400ad;
353 static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
358 case VM_FAULT_BADACCESS:
359 if (access == VM_EXEC && signal_return(regs) == 0)
361 case VM_FAULT_BADMAP:
362 /* Bad memory access. Check if it is kernel or user space. */
363 if (user_mode(regs)) {
364 /* User mode accesses just cause a SIGSEGV */
365 si_code = (fault == VM_FAULT_BADMAP) ?
366 SEGV_MAPERR : SEGV_ACCERR;
367 do_sigsegv(regs, si_code);
370 case VM_FAULT_BADCONTEXT:
371 case VM_FAULT_PFAULT:
374 case VM_FAULT_SIGNAL:
375 if (!user_mode(regs))
378 default: /* fault & VM_FAULT_ERROR */
379 if (fault & VM_FAULT_OOM) {
380 if (!user_mode(regs))
383 pagefault_out_of_memory();
384 } else if (fault & VM_FAULT_SIGSEGV) {
385 /* Kernel mode? Handle exceptions or die */
386 if (!user_mode(regs))
389 do_sigsegv(regs, SEGV_MAPERR);
390 } else if (fault & VM_FAULT_SIGBUS) {
391 /* Kernel mode? Handle exceptions or die */
392 if (!user_mode(regs))
403 * This routine handles page faults. It determines the address,
404 * and the problem, and then passes it off to one of the appropriate
407 * interruption code (int_code):
408 * 04 Protection -> Write-Protection (suprression)
409 * 10 Segment translation -> Not present (nullification)
410 * 11 Page translation -> Not present (nullification)
411 * 3b Region third trans. -> Not present (nullification)
413 static inline int do_exception(struct pt_regs *regs, int access)
416 struct task_struct *tsk;
417 struct mm_struct *mm;
418 struct vm_area_struct *vma;
419 enum fault_type type;
420 unsigned long trans_exc_code;
421 unsigned long address;
427 * The instruction that caused the program check has
428 * been nullified. Don't signal single step via SIGTRAP.
430 clear_pt_regs_flag(regs, PIF_PER_TRAP);
432 if (notify_page_fault(regs))
436 trans_exc_code = regs->int_parm_long;
439 * Verify that the fault happened in user space, that
440 * we are not in an interrupt and that there is a
443 fault = VM_FAULT_BADCONTEXT;
444 type = get_fault_type(regs);
449 fault = VM_FAULT_BADMAP;
453 if (faulthandler_disabled() || !mm)
458 address = trans_exc_code & __FAIL_ADDR_MASK;
459 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
460 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
462 flags |= FAULT_FLAG_USER;
463 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
464 flags |= FAULT_FLAG_WRITE;
465 down_read(&mm->mmap_sem);
468 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
469 gmap = (struct gmap *) S390_lowcore.gmap;
470 current->thread.gmap_addr = address;
471 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
472 current->thread.gmap_int_code = regs->int_code & 0xffff;
473 address = __gmap_translate(gmap, address);
474 if (address == -EFAULT) {
475 fault = VM_FAULT_BADMAP;
478 if (gmap->pfault_enabled)
479 flags |= FAULT_FLAG_RETRY_NOWAIT;
483 fault = VM_FAULT_BADMAP;
484 vma = find_vma(mm, address);
488 if (unlikely(vma->vm_start > address)) {
489 if (!(vma->vm_flags & VM_GROWSDOWN))
491 if (expand_stack(vma, address))
496 * Ok, we have a good vm_area for this memory access, so
499 fault = VM_FAULT_BADACCESS;
500 if (unlikely(!(vma->vm_flags & access)))
503 if (is_vm_hugetlb_page(vma))
504 address &= HPAGE_MASK;
506 * If for any reason at all we couldn't handle the fault,
507 * make sure we exit gracefully rather than endlessly redo
510 fault = handle_mm_fault(vma, address, flags);
511 /* No reason to continue if interrupted by SIGKILL. */
512 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
513 fault = VM_FAULT_SIGNAL;
516 if (unlikely(fault & VM_FAULT_ERROR))
520 * Major/minor page fault accounting is only done on the
521 * initial attempt. If we go through a retry, it is extremely
522 * likely that the page will be found in page cache at that point.
524 if (flags & FAULT_FLAG_ALLOW_RETRY) {
525 if (fault & VM_FAULT_MAJOR) {
527 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
531 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
534 if (fault & VM_FAULT_RETRY) {
535 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
536 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
537 /* FAULT_FLAG_RETRY_NOWAIT has been set,
538 * mmap_sem has not been released */
539 current->thread.gmap_pfault = 1;
540 fault = VM_FAULT_PFAULT;
543 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
545 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
546 FAULT_FLAG_RETRY_NOWAIT);
547 flags |= FAULT_FLAG_TRIED;
548 down_read(&mm->mmap_sem);
552 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
553 address = __gmap_link(gmap, current->thread.gmap_addr,
555 if (address == -EFAULT) {
556 fault = VM_FAULT_BADMAP;
559 if (address == -ENOMEM) {
560 fault = VM_FAULT_OOM;
566 up_read(&mm->mmap_sem);
571 void do_protection_exception(struct pt_regs *regs)
573 unsigned long trans_exc_code;
576 trans_exc_code = regs->int_parm_long;
578 * Protection exceptions are suppressing, decrement psw address.
579 * The exception to this rule are aborted transactions, for these
580 * the PSW already points to the correct location.
582 if (!(regs->int_code & 0x200))
583 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
585 * Check for low-address protection. This needs to be treated
586 * as a special case because the translation exception code
587 * field is not guaranteed to contain valid data in this case.
589 if (unlikely(!(trans_exc_code & 4))) {
590 do_low_address(regs);
593 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
594 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
595 (regs->psw.addr & PAGE_MASK);
597 fault = VM_FAULT_BADACCESS;
600 fault = do_exception(regs, access);
603 do_fault_error(regs, access, fault);
605 NOKPROBE_SYMBOL(do_protection_exception);
607 void do_dat_exception(struct pt_regs *regs)
611 access = VM_READ | VM_EXEC | VM_WRITE;
612 fault = do_exception(regs, access);
614 do_fault_error(regs, access, fault);
616 NOKPROBE_SYMBOL(do_dat_exception);
620 * 'pfault' pseudo page faults routines.
622 static int pfault_disable;
624 static int __init nopfault(char *str)
630 __setup("nopfault", nopfault);
632 struct pfault_refbk {
641 } __attribute__ ((packed, aligned(8)));
643 int pfault_init(void)
645 struct pfault_refbk refbk = {
650 .refgaddr = __LC_LPP,
651 .refselmk = 1ULL << 48,
652 .refcmpmk = 1ULL << 48,
653 .reserved = __PF_RES_FIELD };
658 diag_stat_inc(DIAG_STAT_X258);
660 " diag %1,%0,0x258\n"
665 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
669 void pfault_fini(void)
671 struct pfault_refbk refbk = {
680 diag_stat_inc(DIAG_STAT_X258);
685 : : "a" (&refbk), "m" (refbk) : "cc");
688 static DEFINE_SPINLOCK(pfault_lock);
689 static LIST_HEAD(pfault_list);
691 #define PF_COMPLETE 0x0080
694 * The mechanism of our pfault code: if Linux is running as guest, runs a user
695 * space process and the user space process accesses a page that the host has
696 * paged out we get a pfault interrupt.
698 * This allows us, within the guest, to schedule a different process. Without
699 * this mechanism the host would have to suspend the whole virtual cpu until
700 * the page has been paged in.
702 * So when we get such an interrupt then we set the state of the current task
703 * to uninterruptible and also set the need_resched flag. Both happens within
704 * interrupt context(!). If we later on want to return to user space we
705 * recognize the need_resched flag and then call schedule(). It's not very
706 * obvious how this works...
708 * Of course we have a lot of additional fun with the completion interrupt (->
709 * host signals that a page of a process has been paged in and the process can
710 * continue to run). This interrupt can arrive on any cpu and, since we have
711 * virtual cpus, actually appear before the interrupt that signals that a page
714 static void pfault_interrupt(struct ext_code ext_code,
715 unsigned int param32, unsigned long param64)
717 struct task_struct *tsk;
722 * Get the external interruption subcode & pfault initial/completion
723 * signal bit. VM stores this in the 'cpu address' field associated
724 * with the external interrupt.
726 subcode = ext_code.subcode;
727 if ((subcode & 0xff00) != __SUBCODE_MASK)
729 inc_irq_stat(IRQEXT_PFL);
730 /* Get the token (= pid of the affected task). */
731 pid = param64 & LPP_PID_MASK;
733 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
735 get_task_struct(tsk);
739 spin_lock(&pfault_lock);
740 if (subcode & PF_COMPLETE) {
741 /* signal bit is set -> a page has been swapped in by VM */
742 if (tsk->thread.pfault_wait == 1) {
743 /* Initial interrupt was faster than the completion
744 * interrupt. pfault_wait is valid. Set pfault_wait
745 * back to zero and wake up the process. This can
746 * safely be done because the task is still sleeping
747 * and can't produce new pfaults. */
748 tsk->thread.pfault_wait = 0;
749 list_del(&tsk->thread.list);
750 wake_up_process(tsk);
751 put_task_struct(tsk);
753 /* Completion interrupt was faster than initial
754 * interrupt. Set pfault_wait to -1 so the initial
755 * interrupt doesn't put the task to sleep.
756 * If the task is not running, ignore the completion
757 * interrupt since it must be a leftover of a PFAULT
758 * CANCEL operation which didn't remove all pending
759 * completion interrupts. */
760 if (tsk->state == TASK_RUNNING)
761 tsk->thread.pfault_wait = -1;
764 /* signal bit not set -> a real page is missing. */
765 if (WARN_ON_ONCE(tsk != current))
767 if (tsk->thread.pfault_wait == 1) {
768 /* Already on the list with a reference: put to sleep */
770 } else if (tsk->thread.pfault_wait == -1) {
771 /* Completion interrupt was faster than the initial
772 * interrupt (pfault_wait == -1). Set pfault_wait
773 * back to zero and exit. */
774 tsk->thread.pfault_wait = 0;
776 /* Initial interrupt arrived before completion
777 * interrupt. Let the task sleep.
778 * An extra task reference is needed since a different
779 * cpu may set the task state to TASK_RUNNING again
780 * before the scheduler is reached. */
781 get_task_struct(tsk);
782 tsk->thread.pfault_wait = 1;
783 list_add(&tsk->thread.list, &pfault_list);
785 /* Since this must be a userspace fault, there
786 * is no kernel task state to trample. Rely on the
787 * return to userspace schedule() to block. */
788 __set_current_state(TASK_UNINTERRUPTIBLE);
789 set_tsk_need_resched(tsk);
790 set_preempt_need_resched();
794 spin_unlock(&pfault_lock);
795 put_task_struct(tsk);
798 static int pfault_cpu_dead(unsigned int cpu)
800 struct thread_struct *thread, *next;
801 struct task_struct *tsk;
803 spin_lock_irq(&pfault_lock);
804 list_for_each_entry_safe(thread, next, &pfault_list, list) {
805 thread->pfault_wait = 0;
806 list_del(&thread->list);
807 tsk = container_of(thread, struct task_struct, thread);
808 wake_up_process(tsk);
809 put_task_struct(tsk);
811 spin_unlock_irq(&pfault_lock);
815 static int __init pfault_irq_init(void)
819 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
822 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
825 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
826 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
827 NULL, pfault_cpu_dead);
831 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
836 early_initcall(pfault_irq_init);
838 #endif /* CONFIG_PFAULT */