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
53 static unsigned long store_indication __read_mostly;
55 static int __init fault_init(void)
57 if (test_facility(75))
58 store_indication = 0xc00;
61 early_initcall(fault_init);
63 static inline int notify_page_fault(struct pt_regs *regs)
67 /* kprobe_running() needs smp_processor_id() */
68 if (kprobes_built_in() && !user_mode(regs)) {
70 if (kprobe_running() && kprobe_fault_handler(regs, 14))
79 * Unlock any spinlocks which will prevent us from getting the
82 void bust_spinlocks(int yes)
87 int loglevel_save = console_loglevel;
91 * OK, the message is on the console. Now we call printk()
92 * without oops_in_progress set so that printk will give klogd
93 * a poke. Hold onto your hats...
95 console_loglevel = 15;
97 console_loglevel = loglevel_save;
102 * Returns the address space associated with the fault.
103 * Returns 0 for kernel space and 1 for user space.
105 static inline int user_space_fault(struct pt_regs *regs)
107 unsigned long trans_exc_code;
110 * The lowest two bits of the translation exception
111 * identification indicate which paging table was used.
113 trans_exc_code = regs->int_parm_long & 3;
114 if (trans_exc_code == 3) /* home space -> kernel */
118 if (trans_exc_code == 2) /* secondary space -> set_fs */
119 return current->thread.mm_segment.ar4;
120 if (current->flags & PF_VCPU)
125 static int bad_address(void *p)
129 return probe_kernel_address((unsigned long *)p, dummy);
132 static void dump_pagetable(unsigned long asce, unsigned long address)
134 unsigned long *table = __va(asce & _ASCE_ORIGIN);
136 pr_alert("AS:%016lx ", asce);
137 switch (asce & _ASCE_TYPE_MASK) {
138 case _ASCE_TYPE_REGION1:
139 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
140 if (bad_address(table))
142 pr_cont("R1:%016lx ", *table);
143 if (*table & _REGION_ENTRY_INVALID)
145 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
147 case _ASCE_TYPE_REGION2:
148 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
149 if (bad_address(table))
151 pr_cont("R2:%016lx ", *table);
152 if (*table & _REGION_ENTRY_INVALID)
154 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
156 case _ASCE_TYPE_REGION3:
157 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
158 if (bad_address(table))
160 pr_cont("R3:%016lx ", *table);
161 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
163 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
165 case _ASCE_TYPE_SEGMENT:
166 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
167 if (bad_address(table))
169 pr_cont("S:%016lx ", *table);
170 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
172 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
174 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
175 if (bad_address(table))
177 pr_cont("P:%016lx ", *table);
185 static void dump_fault_info(struct pt_regs *regs)
189 pr_alert("Failing address: %016lx TEID: %016lx\n",
190 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
191 pr_alert("Fault in ");
192 switch (regs->int_parm_long & 3) {
194 pr_cont("home space ");
197 pr_cont("secondary space ");
200 pr_cont("access register ");
203 pr_cont("primary space ");
206 pr_cont("mode while using ");
207 if (!user_space_fault(regs)) {
208 asce = S390_lowcore.kernel_asce;
212 else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
213 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
219 asce = S390_lowcore.user_asce;
223 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
226 int show_unhandled_signals = 1;
228 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
230 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
232 if (!unhandled_signal(current, signr))
234 if (!printk_ratelimit())
236 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
237 regs->int_code & 0xffff, regs->int_code >> 17);
238 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
239 printk(KERN_CONT "\n");
241 dump_fault_info(regs);
246 * Send SIGSEGV to task. This is an external routine
247 * to keep the stack usage of do_page_fault small.
249 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
253 report_user_fault(regs, SIGSEGV, 1);
254 si.si_signo = SIGSEGV;
256 si.si_code = si_code;
257 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
258 force_sig_info(SIGSEGV, &si, current);
261 static noinline void do_no_context(struct pt_regs *regs)
263 const struct exception_table_entry *fixup;
265 /* Are we prepared to handle this kernel fault? */
266 fixup = search_exception_tables(regs->psw.addr);
268 regs->psw.addr = extable_fixup(fixup);
273 * Oops. The kernel tried to access some bad page. We'll have to
274 * terminate things with extreme prejudice.
276 if (!user_space_fault(regs))
277 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
278 " in virtual kernel address space\n");
280 printk(KERN_ALERT "Unable to handle kernel paging request"
281 " in virtual user address space\n");
282 dump_fault_info(regs);
287 static noinline void do_low_address(struct pt_regs *regs)
289 /* Low-address protection hit in kernel mode means
290 NULL pointer write access in kernel mode. */
291 if (regs->psw.mask & PSW_MASK_PSTATE) {
292 /* Low-address protection hit in user mode 'cannot happen'. */
293 die (regs, "Low-address protection");
300 static noinline void do_sigbus(struct pt_regs *regs)
302 struct task_struct *tsk = current;
306 * Send a sigbus, regardless of whether we were in kernel
309 si.si_signo = SIGBUS;
311 si.si_code = BUS_ADRERR;
312 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
313 force_sig_info(SIGBUS, &si, tsk);
316 static noinline int signal_return(struct pt_regs *regs)
321 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
324 if (instruction == 0x0a77) {
325 set_pt_regs_flag(regs, PIF_SYSCALL);
326 regs->int_code = 0x00040077;
328 } else if (instruction == 0x0aad) {
329 set_pt_regs_flag(regs, PIF_SYSCALL);
330 regs->int_code = 0x000400ad;
336 static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
341 case VM_FAULT_BADACCESS:
342 if (access == VM_EXEC && signal_return(regs) == 0)
344 case VM_FAULT_BADMAP:
345 /* Bad memory access. Check if it is kernel or user space. */
346 if (user_mode(regs)) {
347 /* User mode accesses just cause a SIGSEGV */
348 si_code = (fault == VM_FAULT_BADMAP) ?
349 SEGV_MAPERR : SEGV_ACCERR;
350 do_sigsegv(regs, si_code);
353 case VM_FAULT_BADCONTEXT:
354 case VM_FAULT_PFAULT:
357 case VM_FAULT_SIGNAL:
358 if (!user_mode(regs))
361 default: /* fault & VM_FAULT_ERROR */
362 if (fault & VM_FAULT_OOM) {
363 if (!user_mode(regs))
366 pagefault_out_of_memory();
367 } else if (fault & VM_FAULT_SIGSEGV) {
368 /* Kernel mode? Handle exceptions or die */
369 if (!user_mode(regs))
372 do_sigsegv(regs, SEGV_MAPERR);
373 } else if (fault & VM_FAULT_SIGBUS) {
374 /* Kernel mode? Handle exceptions or die */
375 if (!user_mode(regs))
386 * This routine handles page faults. It determines the address,
387 * and the problem, and then passes it off to one of the appropriate
390 * interruption code (int_code):
391 * 04 Protection -> Write-Protection (suprression)
392 * 10 Segment translation -> Not present (nullification)
393 * 11 Page translation -> Not present (nullification)
394 * 3b Region third trans. -> Not present (nullification)
396 static inline int do_exception(struct pt_regs *regs, int access)
401 struct task_struct *tsk;
402 struct mm_struct *mm;
403 struct vm_area_struct *vma;
404 unsigned long trans_exc_code;
405 unsigned long address;
411 * The instruction that caused the program check has
412 * been nullified. Don't signal single step via SIGTRAP.
414 clear_pt_regs_flag(regs, PIF_PER_TRAP);
416 if (notify_page_fault(regs))
420 trans_exc_code = regs->int_parm_long;
423 * Verify that the fault happened in user space, that
424 * we are not in an interrupt and that there is a
427 fault = VM_FAULT_BADCONTEXT;
428 if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
431 address = trans_exc_code & __FAIL_ADDR_MASK;
432 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
433 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
435 flags |= FAULT_FLAG_USER;
436 if ((trans_exc_code & store_indication) == 0x400)
438 if (access == VM_WRITE)
439 flags |= FAULT_FLAG_WRITE;
440 down_read(&mm->mmap_sem);
443 gmap = (current->flags & PF_VCPU) ?
444 (struct gmap *) S390_lowcore.gmap : NULL;
446 current->thread.gmap_addr = address;
447 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
448 current->thread.gmap_int_code = regs->int_code & 0xffff;
449 address = __gmap_translate(gmap, address);
450 if (address == -EFAULT) {
451 fault = VM_FAULT_BADMAP;
454 if (gmap->pfault_enabled)
455 flags |= FAULT_FLAG_RETRY_NOWAIT;
460 fault = VM_FAULT_BADMAP;
461 vma = find_vma(mm, address);
465 if (unlikely(vma->vm_start > address)) {
466 if (!(vma->vm_flags & VM_GROWSDOWN))
468 if (expand_stack(vma, address))
473 * Ok, we have a good vm_area for this memory access, so
476 fault = VM_FAULT_BADACCESS;
477 if (unlikely(!(vma->vm_flags & access)))
480 if (is_vm_hugetlb_page(vma))
481 address &= HPAGE_MASK;
483 * If for any reason at all we couldn't handle the fault,
484 * make sure we exit gracefully rather than endlessly redo
487 fault = handle_mm_fault(vma, address, flags);
488 /* No reason to continue if interrupted by SIGKILL. */
489 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
490 fault = VM_FAULT_SIGNAL;
491 if (flags & FAULT_FLAG_RETRY_NOWAIT)
495 if (unlikely(fault & VM_FAULT_ERROR))
499 * Major/minor page fault accounting is only done on the
500 * initial attempt. If we go through a retry, it is extremely
501 * likely that the page will be found in page cache at that point.
503 if (flags & FAULT_FLAG_ALLOW_RETRY) {
504 if (fault & VM_FAULT_MAJOR) {
506 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
510 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
513 if (fault & VM_FAULT_RETRY) {
515 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
516 /* FAULT_FLAG_RETRY_NOWAIT has been set,
517 * mmap_sem has not been released */
518 current->thread.gmap_pfault = 1;
519 fault = VM_FAULT_PFAULT;
523 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
525 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
526 FAULT_FLAG_RETRY_NOWAIT);
527 flags |= FAULT_FLAG_TRIED;
528 down_read(&mm->mmap_sem);
534 address = __gmap_link(gmap, current->thread.gmap_addr,
536 if (address == -EFAULT) {
537 fault = VM_FAULT_BADMAP;
540 if (address == -ENOMEM) {
541 fault = VM_FAULT_OOM;
548 up_read(&mm->mmap_sem);
553 void do_protection_exception(struct pt_regs *regs)
555 unsigned long trans_exc_code;
558 trans_exc_code = regs->int_parm_long;
560 * Protection exceptions are suppressing, decrement psw address.
561 * The exception to this rule are aborted transactions, for these
562 * the PSW already points to the correct location.
564 if (!(regs->int_code & 0x200))
565 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
567 * Check for low-address protection. This needs to be treated
568 * as a special case because the translation exception code
569 * field is not guaranteed to contain valid data in this case.
571 if (unlikely(!(trans_exc_code & 4))) {
572 do_low_address(regs);
575 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
576 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
577 (regs->psw.addr & PAGE_MASK);
579 fault = VM_FAULT_BADACCESS;
582 fault = do_exception(regs, access);
585 do_fault_error(regs, access, fault);
587 NOKPROBE_SYMBOL(do_protection_exception);
589 void do_dat_exception(struct pt_regs *regs)
593 access = VM_READ | VM_EXEC | VM_WRITE;
594 fault = do_exception(regs, access);
596 do_fault_error(regs, access, fault);
598 NOKPROBE_SYMBOL(do_dat_exception);
602 * 'pfault' pseudo page faults routines.
604 static int pfault_disable;
606 static int __init nopfault(char *str)
612 __setup("nopfault", nopfault);
614 struct pfault_refbk {
623 } __attribute__ ((packed, aligned(8)));
625 int pfault_init(void)
627 struct pfault_refbk refbk = {
632 .refgaddr = __LC_LPP,
633 .refselmk = 1ULL << 48,
634 .refcmpmk = 1ULL << 48,
635 .reserved = __PF_RES_FIELD };
640 diag_stat_inc(DIAG_STAT_X258);
642 " diag %1,%0,0x258\n"
647 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
651 void pfault_fini(void)
653 struct pfault_refbk refbk = {
662 diag_stat_inc(DIAG_STAT_X258);
667 : : "a" (&refbk), "m" (refbk) : "cc");
670 static DEFINE_SPINLOCK(pfault_lock);
671 static LIST_HEAD(pfault_list);
673 #define PF_COMPLETE 0x0080
676 * The mechanism of our pfault code: if Linux is running as guest, runs a user
677 * space process and the user space process accesses a page that the host has
678 * paged out we get a pfault interrupt.
680 * This allows us, within the guest, to schedule a different process. Without
681 * this mechanism the host would have to suspend the whole virtual cpu until
682 * the page has been paged in.
684 * So when we get such an interrupt then we set the state of the current task
685 * to uninterruptible and also set the need_resched flag. Both happens within
686 * interrupt context(!). If we later on want to return to user space we
687 * recognize the need_resched flag and then call schedule(). It's not very
688 * obvious how this works...
690 * Of course we have a lot of additional fun with the completion interrupt (->
691 * host signals that a page of a process has been paged in and the process can
692 * continue to run). This interrupt can arrive on any cpu and, since we have
693 * virtual cpus, actually appear before the interrupt that signals that a page
696 static void pfault_interrupt(struct ext_code ext_code,
697 unsigned int param32, unsigned long param64)
699 struct task_struct *tsk;
704 * Get the external interruption subcode & pfault initial/completion
705 * signal bit. VM stores this in the 'cpu address' field associated
706 * with the external interrupt.
708 subcode = ext_code.subcode;
709 if ((subcode & 0xff00) != __SUBCODE_MASK)
711 inc_irq_stat(IRQEXT_PFL);
712 /* Get the token (= pid of the affected task). */
713 pid = param64 & LPP_PFAULT_PID_MASK;
715 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
717 get_task_struct(tsk);
721 spin_lock(&pfault_lock);
722 if (subcode & PF_COMPLETE) {
723 /* signal bit is set -> a page has been swapped in by VM */
724 if (tsk->thread.pfault_wait == 1) {
725 /* Initial interrupt was faster than the completion
726 * interrupt. pfault_wait is valid. Set pfault_wait
727 * back to zero and wake up the process. This can
728 * safely be done because the task is still sleeping
729 * and can't produce new pfaults. */
730 tsk->thread.pfault_wait = 0;
731 list_del(&tsk->thread.list);
732 wake_up_process(tsk);
733 put_task_struct(tsk);
735 /* Completion interrupt was faster than initial
736 * interrupt. Set pfault_wait to -1 so the initial
737 * interrupt doesn't put the task to sleep.
738 * If the task is not running, ignore the completion
739 * interrupt since it must be a leftover of a PFAULT
740 * CANCEL operation which didn't remove all pending
741 * completion interrupts. */
742 if (tsk->state == TASK_RUNNING)
743 tsk->thread.pfault_wait = -1;
746 /* signal bit not set -> a real page is missing. */
747 if (WARN_ON_ONCE(tsk != current))
749 if (tsk->thread.pfault_wait == 1) {
750 /* Already on the list with a reference: put to sleep */
752 } else if (tsk->thread.pfault_wait == -1) {
753 /* Completion interrupt was faster than the initial
754 * interrupt (pfault_wait == -1). Set pfault_wait
755 * back to zero and exit. */
756 tsk->thread.pfault_wait = 0;
758 /* Initial interrupt arrived before completion
759 * interrupt. Let the task sleep.
760 * An extra task reference is needed since a different
761 * cpu may set the task state to TASK_RUNNING again
762 * before the scheduler is reached. */
763 get_task_struct(tsk);
764 tsk->thread.pfault_wait = 1;
765 list_add(&tsk->thread.list, &pfault_list);
767 /* Since this must be a userspace fault, there
768 * is no kernel task state to trample. Rely on the
769 * return to userspace schedule() to block. */
770 __set_current_state(TASK_UNINTERRUPTIBLE);
771 set_tsk_need_resched(tsk);
772 set_preempt_need_resched();
776 spin_unlock(&pfault_lock);
777 put_task_struct(tsk);
780 static int pfault_cpu_dead(unsigned int cpu)
782 struct thread_struct *thread, *next;
783 struct task_struct *tsk;
785 spin_lock_irq(&pfault_lock);
786 list_for_each_entry_safe(thread, next, &pfault_list, list) {
787 thread->pfault_wait = 0;
788 list_del(&thread->list);
789 tsk = container_of(thread, struct task_struct, thread);
790 wake_up_process(tsk);
791 put_task_struct(tsk);
793 spin_unlock_irq(&pfault_lock);
797 static int __init pfault_irq_init(void)
801 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
804 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
807 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
808 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
809 NULL, pfault_cpu_dead);
813 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
818 early_initcall(pfault_irq_init);
820 #endif /* CONFIG_PFAULT */