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.
111 static inline enum fault_type get_fault_type(struct pt_regs *regs)
113 unsigned long trans_exc_code;
115 trans_exc_code = regs->int_parm_long & 3;
116 if (likely(trans_exc_code == 0)) {
117 /* primary space exception */
118 if (IS_ENABLED(CONFIG_PGSTE) &&
119 test_pt_regs_flag(regs, PIF_GUEST_FAULT))
121 if (current->thread.mm_segment == USER_DS)
125 if (trans_exc_code == 2) {
126 /* secondary space exception */
127 if (current->thread.mm_segment & 1) {
128 if (current->thread.mm_segment == USER_DS_SACF)
134 if (trans_exc_code == 1) {
135 /* access register mode, not used in the kernel */
138 /* home space exception -> access via kernel ASCE */
142 static int bad_address(void *p)
146 return probe_kernel_address((unsigned long *)p, dummy);
149 static void dump_pagetable(unsigned long asce, unsigned long address)
151 unsigned long *table = __va(asce & _ASCE_ORIGIN);
153 pr_alert("AS:%016lx ", asce);
154 switch (asce & _ASCE_TYPE_MASK) {
155 case _ASCE_TYPE_REGION1:
156 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
157 if (bad_address(table))
159 pr_cont("R1:%016lx ", *table);
160 if (*table & _REGION_ENTRY_INVALID)
162 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
164 case _ASCE_TYPE_REGION2:
165 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
166 if (bad_address(table))
168 pr_cont("R2:%016lx ", *table);
169 if (*table & _REGION_ENTRY_INVALID)
171 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
173 case _ASCE_TYPE_REGION3:
174 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
175 if (bad_address(table))
177 pr_cont("R3:%016lx ", *table);
178 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
180 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
182 case _ASCE_TYPE_SEGMENT:
183 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
184 if (bad_address(table))
186 pr_cont("S:%016lx ", *table);
187 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
189 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
191 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
192 if (bad_address(table))
194 pr_cont("P:%016lx ", *table);
202 static void dump_fault_info(struct pt_regs *regs)
206 pr_alert("Failing address: %016lx TEID: %016lx\n",
207 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
208 pr_alert("Fault in ");
209 switch (regs->int_parm_long & 3) {
211 pr_cont("home space ");
214 pr_cont("secondary space ");
217 pr_cont("access register ");
220 pr_cont("primary space ");
223 pr_cont("mode while using ");
224 switch (get_fault_type(regs)) {
226 asce = S390_lowcore.user_asce;
230 asce = S390_lowcore.vdso_asce;
234 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
238 asce = S390_lowcore.kernel_asce;
243 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
246 int show_unhandled_signals = 1;
248 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
250 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
252 if (!unhandled_signal(current, signr))
254 if (!printk_ratelimit())
256 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
257 regs->int_code & 0xffff, regs->int_code >> 17);
258 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
259 printk(KERN_CONT "\n");
261 dump_fault_info(regs);
266 * Send SIGSEGV to task. This is an external routine
267 * to keep the stack usage of do_page_fault small.
269 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
271 report_user_fault(regs, SIGSEGV, 1);
272 force_sig_fault(SIGSEGV, si_code,
273 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK),
277 static noinline void do_no_context(struct pt_regs *regs)
279 const struct exception_table_entry *fixup;
281 /* Are we prepared to handle this kernel fault? */
282 fixup = search_exception_tables(regs->psw.addr);
284 regs->psw.addr = extable_fixup(fixup);
289 * Oops. The kernel tried to access some bad page. We'll have to
290 * terminate things with extreme prejudice.
292 if (get_fault_type(regs) == KERNEL_FAULT)
293 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
294 " in virtual kernel address space\n");
296 printk(KERN_ALERT "Unable to handle kernel paging request"
297 " in virtual user address space\n");
298 dump_fault_info(regs);
303 static noinline void do_low_address(struct pt_regs *regs)
305 /* Low-address protection hit in kernel mode means
306 NULL pointer write access in kernel mode. */
307 if (regs->psw.mask & PSW_MASK_PSTATE) {
308 /* Low-address protection hit in user mode 'cannot happen'. */
309 die (regs, "Low-address protection");
316 static noinline void do_sigbus(struct pt_regs *regs)
319 * Send a sigbus, regardless of whether we were in kernel
322 force_sig_fault(SIGBUS, BUS_ADRERR,
323 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK),
327 static noinline int signal_return(struct pt_regs *regs)
332 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
335 if (instruction == 0x0a77) {
336 set_pt_regs_flag(regs, PIF_SYSCALL);
337 regs->int_code = 0x00040077;
339 } else if (instruction == 0x0aad) {
340 set_pt_regs_flag(regs, PIF_SYSCALL);
341 regs->int_code = 0x000400ad;
347 static noinline void do_fault_error(struct pt_regs *regs, int access,
353 case VM_FAULT_BADACCESS:
354 if (access == VM_EXEC && signal_return(regs) == 0)
356 case VM_FAULT_BADMAP:
357 /* Bad memory access. Check if it is kernel or user space. */
358 if (user_mode(regs)) {
359 /* User mode accesses just cause a SIGSEGV */
360 si_code = (fault == VM_FAULT_BADMAP) ?
361 SEGV_MAPERR : SEGV_ACCERR;
362 do_sigsegv(regs, si_code);
365 case VM_FAULT_BADCONTEXT:
366 case VM_FAULT_PFAULT:
369 case VM_FAULT_SIGNAL:
370 if (!user_mode(regs))
373 default: /* fault & VM_FAULT_ERROR */
374 if (fault & VM_FAULT_OOM) {
375 if (!user_mode(regs))
378 pagefault_out_of_memory();
379 } else if (fault & VM_FAULT_SIGSEGV) {
380 /* Kernel mode? Handle exceptions or die */
381 if (!user_mode(regs))
384 do_sigsegv(regs, SEGV_MAPERR);
385 } else if (fault & VM_FAULT_SIGBUS) {
386 /* Kernel mode? Handle exceptions or die */
387 if (!user_mode(regs))
398 * This routine handles page faults. It determines the address,
399 * and the problem, and then passes it off to one of the appropriate
402 * interruption code (int_code):
403 * 04 Protection -> Write-Protection (suprression)
404 * 10 Segment translation -> Not present (nullification)
405 * 11 Page translation -> Not present (nullification)
406 * 3b Region third trans. -> Not present (nullification)
408 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
411 struct task_struct *tsk;
412 struct mm_struct *mm;
413 struct vm_area_struct *vma;
414 enum fault_type type;
415 unsigned long trans_exc_code;
416 unsigned long address;
422 * The instruction that caused the program check has
423 * been nullified. Don't signal single step via SIGTRAP.
425 clear_pt_regs_flag(regs, PIF_PER_TRAP);
427 if (notify_page_fault(regs))
431 trans_exc_code = regs->int_parm_long;
434 * Verify that the fault happened in user space, that
435 * we are not in an interrupt and that there is a
438 fault = VM_FAULT_BADCONTEXT;
439 type = get_fault_type(regs);
444 fault = VM_FAULT_BADMAP;
448 if (faulthandler_disabled() || !mm)
453 address = trans_exc_code & __FAIL_ADDR_MASK;
454 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
455 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
457 flags |= FAULT_FLAG_USER;
458 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
459 flags |= FAULT_FLAG_WRITE;
460 down_read(&mm->mmap_sem);
463 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
464 gmap = (struct gmap *) S390_lowcore.gmap;
465 current->thread.gmap_addr = address;
466 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
467 current->thread.gmap_int_code = regs->int_code & 0xffff;
468 address = __gmap_translate(gmap, address);
469 if (address == -EFAULT) {
470 fault = VM_FAULT_BADMAP;
473 if (gmap->pfault_enabled)
474 flags |= FAULT_FLAG_RETRY_NOWAIT;
478 fault = VM_FAULT_BADMAP;
479 vma = find_vma(mm, address);
483 if (unlikely(vma->vm_start > address)) {
484 if (!(vma->vm_flags & VM_GROWSDOWN))
486 if (expand_stack(vma, address))
491 * Ok, we have a good vm_area for this memory access, so
494 fault = VM_FAULT_BADACCESS;
495 if (unlikely(!(vma->vm_flags & access)))
498 if (is_vm_hugetlb_page(vma))
499 address &= HPAGE_MASK;
501 * If for any reason at all we couldn't handle the fault,
502 * make sure we exit gracefully rather than endlessly redo
505 fault = handle_mm_fault(vma, address, flags);
506 /* No reason to continue if interrupted by SIGKILL. */
507 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
508 fault = VM_FAULT_SIGNAL;
509 if (flags & FAULT_FLAG_RETRY_NOWAIT)
513 if (unlikely(fault & VM_FAULT_ERROR))
517 * Major/minor page fault accounting is only done on the
518 * initial attempt. If we go through a retry, it is extremely
519 * likely that the page will be found in page cache at that point.
521 if (flags & FAULT_FLAG_ALLOW_RETRY) {
522 if (fault & VM_FAULT_MAJOR) {
524 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
528 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
531 if (fault & VM_FAULT_RETRY) {
532 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
533 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
534 /* FAULT_FLAG_RETRY_NOWAIT has been set,
535 * mmap_sem has not been released */
536 current->thread.gmap_pfault = 1;
537 fault = VM_FAULT_PFAULT;
540 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
542 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
543 FAULT_FLAG_RETRY_NOWAIT);
544 flags |= FAULT_FLAG_TRIED;
545 down_read(&mm->mmap_sem);
549 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
550 address = __gmap_link(gmap, current->thread.gmap_addr,
552 if (address == -EFAULT) {
553 fault = VM_FAULT_BADMAP;
556 if (address == -ENOMEM) {
557 fault = VM_FAULT_OOM;
563 up_read(&mm->mmap_sem);
568 void do_protection_exception(struct pt_regs *regs)
570 unsigned long trans_exc_code;
574 trans_exc_code = regs->int_parm_long;
576 * Protection exceptions are suppressing, decrement psw address.
577 * The exception to this rule are aborted transactions, for these
578 * the PSW already points to the correct location.
580 if (!(regs->int_code & 0x200))
581 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
583 * Check for low-address protection. This needs to be treated
584 * as a special case because the translation exception code
585 * field is not guaranteed to contain valid data in this case.
587 if (unlikely(!(trans_exc_code & 4))) {
588 do_low_address(regs);
591 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
592 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
593 (regs->psw.addr & PAGE_MASK);
595 fault = VM_FAULT_BADACCESS;
598 fault = do_exception(regs, access);
601 do_fault_error(regs, access, fault);
603 NOKPROBE_SYMBOL(do_protection_exception);
605 void do_dat_exception(struct pt_regs *regs)
610 access = VM_READ | VM_EXEC | VM_WRITE;
611 fault = do_exception(regs, access);
613 do_fault_error(regs, access, fault);
615 NOKPROBE_SYMBOL(do_dat_exception);
619 * 'pfault' pseudo page faults routines.
621 static int pfault_disable;
623 static int __init nopfault(char *str)
629 __setup("nopfault", nopfault);
631 struct pfault_refbk {
640 } __attribute__ ((packed, aligned(8)));
642 int pfault_init(void)
644 struct pfault_refbk refbk = {
649 .refgaddr = __LC_LPP,
650 .refselmk = 1ULL << 48,
651 .refcmpmk = 1ULL << 48,
652 .reserved = __PF_RES_FIELD };
657 diag_stat_inc(DIAG_STAT_X258);
659 " diag %1,%0,0x258\n"
664 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
668 void pfault_fini(void)
670 struct pfault_refbk refbk = {
679 diag_stat_inc(DIAG_STAT_X258);
684 : : "a" (&refbk), "m" (refbk) : "cc");
687 static DEFINE_SPINLOCK(pfault_lock);
688 static LIST_HEAD(pfault_list);
690 #define PF_COMPLETE 0x0080
693 * The mechanism of our pfault code: if Linux is running as guest, runs a user
694 * space process and the user space process accesses a page that the host has
695 * paged out we get a pfault interrupt.
697 * This allows us, within the guest, to schedule a different process. Without
698 * this mechanism the host would have to suspend the whole virtual cpu until
699 * the page has been paged in.
701 * So when we get such an interrupt then we set the state of the current task
702 * to uninterruptible and also set the need_resched flag. Both happens within
703 * interrupt context(!). If we later on want to return to user space we
704 * recognize the need_resched flag and then call schedule(). It's not very
705 * obvious how this works...
707 * Of course we have a lot of additional fun with the completion interrupt (->
708 * host signals that a page of a process has been paged in and the process can
709 * continue to run). This interrupt can arrive on any cpu and, since we have
710 * virtual cpus, actually appear before the interrupt that signals that a page
713 static void pfault_interrupt(struct ext_code ext_code,
714 unsigned int param32, unsigned long param64)
716 struct task_struct *tsk;
721 * Get the external interruption subcode & pfault initial/completion
722 * signal bit. VM stores this in the 'cpu address' field associated
723 * with the external interrupt.
725 subcode = ext_code.subcode;
726 if ((subcode & 0xff00) != __SUBCODE_MASK)
728 inc_irq_stat(IRQEXT_PFL);
729 /* Get the token (= pid of the affected task). */
730 pid = param64 & LPP_PID_MASK;
732 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
734 get_task_struct(tsk);
738 spin_lock(&pfault_lock);
739 if (subcode & PF_COMPLETE) {
740 /* signal bit is set -> a page has been swapped in by VM */
741 if (tsk->thread.pfault_wait == 1) {
742 /* Initial interrupt was faster than the completion
743 * interrupt. pfault_wait is valid. Set pfault_wait
744 * back to zero and wake up the process. This can
745 * safely be done because the task is still sleeping
746 * and can't produce new pfaults. */
747 tsk->thread.pfault_wait = 0;
748 list_del(&tsk->thread.list);
749 wake_up_process(tsk);
750 put_task_struct(tsk);
752 /* Completion interrupt was faster than initial
753 * interrupt. Set pfault_wait to -1 so the initial
754 * interrupt doesn't put the task to sleep.
755 * If the task is not running, ignore the completion
756 * interrupt since it must be a leftover of a PFAULT
757 * CANCEL operation which didn't remove all pending
758 * completion interrupts. */
759 if (tsk->state == TASK_RUNNING)
760 tsk->thread.pfault_wait = -1;
763 /* signal bit not set -> a real page is missing. */
764 if (WARN_ON_ONCE(tsk != current))
766 if (tsk->thread.pfault_wait == 1) {
767 /* Already on the list with a reference: put to sleep */
769 } else if (tsk->thread.pfault_wait == -1) {
770 /* Completion interrupt was faster than the initial
771 * interrupt (pfault_wait == -1). Set pfault_wait
772 * back to zero and exit. */
773 tsk->thread.pfault_wait = 0;
775 /* Initial interrupt arrived before completion
776 * interrupt. Let the task sleep.
777 * An extra task reference is needed since a different
778 * cpu may set the task state to TASK_RUNNING again
779 * before the scheduler is reached. */
780 get_task_struct(tsk);
781 tsk->thread.pfault_wait = 1;
782 list_add(&tsk->thread.list, &pfault_list);
784 /* Since this must be a userspace fault, there
785 * is no kernel task state to trample. Rely on the
786 * return to userspace schedule() to block. */
787 __set_current_state(TASK_UNINTERRUPTIBLE);
788 set_tsk_need_resched(tsk);
789 set_preempt_need_resched();
793 spin_unlock(&pfault_lock);
794 put_task_struct(tsk);
797 static int pfault_cpu_dead(unsigned int cpu)
799 struct thread_struct *thread, *next;
800 struct task_struct *tsk;
802 spin_lock_irq(&pfault_lock);
803 list_for_each_entry_safe(thread, next, &pfault_list, list) {
804 thread->pfault_wait = 0;
805 list_del(&thread->list);
806 tsk = container_of(thread, struct task_struct, thread);
807 wake_up_process(tsk);
808 put_task_struct(tsk);
810 spin_unlock_irq(&pfault_lock);
814 static int __init pfault_irq_init(void)
818 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
821 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
824 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
825 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
826 NULL, pfault_cpu_dead);
830 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
835 early_initcall(pfault_irq_init);
837 #endif /* CONFIG_PFAULT */