1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright IBM Corp. 1999
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 <linux/kfence.h>
35 #include <asm/asm-extable.h>
36 #include <asm/asm-offsets.h>
40 #include <asm/mmu_context.h>
41 #include <asm/facility.h>
43 #include "../kernel/entry.h"
45 #define __FAIL_ADDR_MASK -4096L
46 #define __SUBCODE_MASK 0x0600
47 #define __PF_RES_FIELD 0x8000000000000000ULL
49 #define VM_FAULT_BADCONTEXT ((__force vm_fault_t) 0x010000)
50 #define VM_FAULT_BADMAP ((__force vm_fault_t) 0x020000)
51 #define VM_FAULT_BADACCESS ((__force vm_fault_t) 0x040000)
52 #define VM_FAULT_SIGNAL ((__force vm_fault_t) 0x080000)
53 #define VM_FAULT_PFAULT ((__force vm_fault_t) 0x100000)
61 static unsigned long store_indication __read_mostly;
63 static int __init fault_init(void)
65 if (test_facility(75))
66 store_indication = 0xc00;
69 early_initcall(fault_init);
72 * Find out which address space caused the exception.
74 static enum fault_type get_fault_type(struct pt_regs *regs)
76 unsigned long trans_exc_code;
78 trans_exc_code = regs->int_parm_long & 3;
79 if (likely(trans_exc_code == 0)) {
80 /* primary space exception */
83 if (!IS_ENABLED(CONFIG_PGSTE))
85 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
89 if (trans_exc_code == 2)
91 if (trans_exc_code == 1) {
92 /* access register mode, not used in the kernel */
95 /* home space exception -> access via kernel ASCE */
99 static int bad_address(void *p)
103 return get_kernel_nofault(dummy, (unsigned long *)p);
106 static void dump_pagetable(unsigned long asce, unsigned long address)
108 unsigned long *table = __va(asce & _ASCE_ORIGIN);
110 pr_alert("AS:%016lx ", asce);
111 switch (asce & _ASCE_TYPE_MASK) {
112 case _ASCE_TYPE_REGION1:
113 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
114 if (bad_address(table))
116 pr_cont("R1:%016lx ", *table);
117 if (*table & _REGION_ENTRY_INVALID)
119 table = __va(*table & _REGION_ENTRY_ORIGIN);
121 case _ASCE_TYPE_REGION2:
122 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
123 if (bad_address(table))
125 pr_cont("R2:%016lx ", *table);
126 if (*table & _REGION_ENTRY_INVALID)
128 table = __va(*table & _REGION_ENTRY_ORIGIN);
130 case _ASCE_TYPE_REGION3:
131 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
132 if (bad_address(table))
134 pr_cont("R3:%016lx ", *table);
135 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
137 table = __va(*table & _REGION_ENTRY_ORIGIN);
139 case _ASCE_TYPE_SEGMENT:
140 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
141 if (bad_address(table))
143 pr_cont("S:%016lx ", *table);
144 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
146 table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
148 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
149 if (bad_address(table))
151 pr_cont("P:%016lx ", *table);
159 static void dump_fault_info(struct pt_regs *regs)
163 pr_alert("Failing address: %016lx TEID: %016lx\n",
164 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
165 pr_alert("Fault in ");
166 switch (regs->int_parm_long & 3) {
168 pr_cont("home space ");
171 pr_cont("secondary space ");
174 pr_cont("access register ");
177 pr_cont("primary space ");
180 pr_cont("mode while using ");
181 switch (get_fault_type(regs)) {
183 asce = S390_lowcore.user_asce;
187 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
191 asce = S390_lowcore.kernel_asce;
198 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
201 int show_unhandled_signals = 1;
203 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
205 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
207 if (!unhandled_signal(current, signr))
209 if (!printk_ratelimit())
211 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
212 regs->int_code & 0xffff, regs->int_code >> 17);
213 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
214 printk(KERN_CONT "\n");
216 dump_fault_info(regs);
221 * Send SIGSEGV to task. This is an external routine
222 * to keep the stack usage of do_page_fault small.
224 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
226 report_user_fault(regs, SIGSEGV, 1);
227 force_sig_fault(SIGSEGV, si_code,
228 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
231 static noinline void do_no_context(struct pt_regs *regs)
233 if (fixup_exception(regs))
236 * Oops. The kernel tried to access some bad page. We'll have to
237 * terminate things with extreme prejudice.
239 if (get_fault_type(regs) == KERNEL_FAULT)
240 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
241 " in virtual kernel address space\n");
243 printk(KERN_ALERT "Unable to handle kernel paging request"
244 " in virtual user address space\n");
245 dump_fault_info(regs);
249 static noinline void do_low_address(struct pt_regs *regs)
251 /* Low-address protection hit in kernel mode means
252 NULL pointer write access in kernel mode. */
253 if (regs->psw.mask & PSW_MASK_PSTATE) {
254 /* Low-address protection hit in user mode 'cannot happen'. */
255 die (regs, "Low-address protection");
261 static noinline void do_sigbus(struct pt_regs *regs)
264 * Send a sigbus, regardless of whether we were in kernel
267 force_sig_fault(SIGBUS, BUS_ADRERR,
268 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
271 static noinline void do_fault_error(struct pt_regs *regs, vm_fault_t fault)
276 case VM_FAULT_BADACCESS:
277 case VM_FAULT_BADMAP:
278 /* Bad memory access. Check if it is kernel or user space. */
279 if (user_mode(regs)) {
280 /* User mode accesses just cause a SIGSEGV */
281 si_code = (fault == VM_FAULT_BADMAP) ?
282 SEGV_MAPERR : SEGV_ACCERR;
283 do_sigsegv(regs, si_code);
287 case VM_FAULT_BADCONTEXT:
288 case VM_FAULT_PFAULT:
291 case VM_FAULT_SIGNAL:
292 if (!user_mode(regs))
295 default: /* fault & VM_FAULT_ERROR */
296 if (fault & VM_FAULT_OOM) {
297 if (!user_mode(regs))
300 pagefault_out_of_memory();
301 } else if (fault & VM_FAULT_SIGSEGV) {
302 /* Kernel mode? Handle exceptions or die */
303 if (!user_mode(regs))
306 do_sigsegv(regs, SEGV_MAPERR);
307 } else if (fault & VM_FAULT_SIGBUS) {
308 /* Kernel mode? Handle exceptions or die */
309 if (!user_mode(regs))
320 * This routine handles page faults. It determines the address,
321 * and the problem, and then passes it off to one of the appropriate
324 * interruption code (int_code):
325 * 04 Protection -> Write-Protection (suppression)
326 * 10 Segment translation -> Not present (nullification)
327 * 11 Page translation -> Not present (nullification)
328 * 3b Region third trans. -> Not present (nullification)
330 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
333 struct task_struct *tsk;
334 struct mm_struct *mm;
335 struct vm_area_struct *vma;
336 enum fault_type type;
337 unsigned long trans_exc_code;
338 unsigned long address;
345 * The instruction that caused the program check has
346 * been nullified. Don't signal single step via SIGTRAP.
348 clear_thread_flag(TIF_PER_TRAP);
350 if (kprobe_page_fault(regs, 14))
354 trans_exc_code = regs->int_parm_long;
355 address = trans_exc_code & __FAIL_ADDR_MASK;
356 is_write = (trans_exc_code & store_indication) == 0x400;
359 * Verify that the fault happened in user space, that
360 * we are not in an interrupt and that there is a
363 fault = VM_FAULT_BADCONTEXT;
364 type = get_fault_type(regs);
367 if (kfence_handle_page_fault(address, is_write, regs))
372 if (faulthandler_disabled() || !mm)
377 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
378 flags = FAULT_FLAG_DEFAULT;
380 flags |= FAULT_FLAG_USER;
383 if (access == VM_WRITE)
384 flags |= FAULT_FLAG_WRITE;
388 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
389 gmap = (struct gmap *) S390_lowcore.gmap;
390 current->thread.gmap_addr = address;
391 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
392 current->thread.gmap_int_code = regs->int_code & 0xffff;
393 address = __gmap_translate(gmap, address);
394 if (address == -EFAULT) {
395 fault = VM_FAULT_BADMAP;
398 if (gmap->pfault_enabled)
399 flags |= FAULT_FLAG_RETRY_NOWAIT;
403 fault = VM_FAULT_BADMAP;
404 vma = find_vma(mm, address);
408 if (unlikely(vma->vm_start > address)) {
409 if (!(vma->vm_flags & VM_GROWSDOWN))
411 if (expand_stack(vma, address))
416 * Ok, we have a good vm_area for this memory access, so
419 fault = VM_FAULT_BADACCESS;
420 if (unlikely(!(vma->vm_flags & access)))
424 * If for any reason at all we couldn't handle the fault,
425 * make sure we exit gracefully rather than endlessly redo
428 fault = handle_mm_fault(vma, address, flags, regs);
429 if (fault_signal_pending(fault, regs)) {
430 fault = VM_FAULT_SIGNAL;
431 if (flags & FAULT_FLAG_RETRY_NOWAIT)
436 /* The fault is fully completed (including releasing mmap lock) */
437 if (fault & VM_FAULT_COMPLETED) {
446 if (unlikely(fault & VM_FAULT_ERROR))
449 if (fault & VM_FAULT_RETRY) {
450 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
451 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
453 * FAULT_FLAG_RETRY_NOWAIT has been set, mmap_lock has
456 current->thread.gmap_pfault = 1;
457 fault = VM_FAULT_PFAULT;
460 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
461 flags |= FAULT_FLAG_TRIED;
466 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
467 address = __gmap_link(gmap, current->thread.gmap_addr,
469 if (address == -EFAULT) {
470 fault = VM_FAULT_BADMAP;
473 if (address == -ENOMEM) {
474 fault = VM_FAULT_OOM;
480 mmap_read_unlock(mm);
485 void do_protection_exception(struct pt_regs *regs)
487 unsigned long trans_exc_code;
491 trans_exc_code = regs->int_parm_long;
493 * Protection exceptions are suppressing, decrement psw address.
494 * The exception to this rule are aborted transactions, for these
495 * the PSW already points to the correct location.
497 if (!(regs->int_code & 0x200))
498 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
500 * Check for low-address protection. This needs to be treated
501 * as a special case because the translation exception code
502 * field is not guaranteed to contain valid data in this case.
504 if (unlikely(!(trans_exc_code & 4))) {
505 do_low_address(regs);
508 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
509 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
510 (regs->psw.addr & PAGE_MASK);
512 fault = VM_FAULT_BADACCESS;
515 fault = do_exception(regs, access);
518 do_fault_error(regs, fault);
520 NOKPROBE_SYMBOL(do_protection_exception);
522 void do_dat_exception(struct pt_regs *regs)
527 access = VM_ACCESS_FLAGS;
528 fault = do_exception(regs, access);
530 do_fault_error(regs, fault);
532 NOKPROBE_SYMBOL(do_dat_exception);
536 * 'pfault' pseudo page faults routines.
538 static int pfault_disable;
540 static int __init nopfault(char *str)
546 __setup("nopfault", nopfault);
548 struct pfault_refbk {
557 } __attribute__ ((packed, aligned(8)));
559 static struct pfault_refbk pfault_init_refbk = {
564 .refgaddr = __LC_LPP,
565 .refselmk = 1ULL << 48,
566 .refcmpmk = 1ULL << 48,
567 .reserved = __PF_RES_FIELD
570 int pfault_init(void)
576 diag_stat_inc(DIAG_STAT_X258);
578 " diag %1,%0,0x258\n"
584 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
588 static struct pfault_refbk pfault_fini_refbk = {
595 void pfault_fini(void)
600 diag_stat_inc(DIAG_STAT_X258);
605 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
608 static DEFINE_SPINLOCK(pfault_lock);
609 static LIST_HEAD(pfault_list);
611 #define PF_COMPLETE 0x0080
614 * The mechanism of our pfault code: if Linux is running as guest, runs a user
615 * space process and the user space process accesses a page that the host has
616 * paged out we get a pfault interrupt.
618 * This allows us, within the guest, to schedule a different process. Without
619 * this mechanism the host would have to suspend the whole virtual cpu until
620 * the page has been paged in.
622 * So when we get such an interrupt then we set the state of the current task
623 * to uninterruptible and also set the need_resched flag. Both happens within
624 * interrupt context(!). If we later on want to return to user space we
625 * recognize the need_resched flag and then call schedule(). It's not very
626 * obvious how this works...
628 * Of course we have a lot of additional fun with the completion interrupt (->
629 * host signals that a page of a process has been paged in and the process can
630 * continue to run). This interrupt can arrive on any cpu and, since we have
631 * virtual cpus, actually appear before the interrupt that signals that a page
634 static void pfault_interrupt(struct ext_code ext_code,
635 unsigned int param32, unsigned long param64)
637 struct task_struct *tsk;
642 * Get the external interruption subcode & pfault initial/completion
643 * signal bit. VM stores this in the 'cpu address' field associated
644 * with the external interrupt.
646 subcode = ext_code.subcode;
647 if ((subcode & 0xff00) != __SUBCODE_MASK)
649 inc_irq_stat(IRQEXT_PFL);
650 /* Get the token (= pid of the affected task). */
651 pid = param64 & LPP_PID_MASK;
653 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
655 get_task_struct(tsk);
659 spin_lock(&pfault_lock);
660 if (subcode & PF_COMPLETE) {
661 /* signal bit is set -> a page has been swapped in by VM */
662 if (tsk->thread.pfault_wait == 1) {
663 /* Initial interrupt was faster than the completion
664 * interrupt. pfault_wait is valid. Set pfault_wait
665 * back to zero and wake up the process. This can
666 * safely be done because the task is still sleeping
667 * and can't produce new pfaults. */
668 tsk->thread.pfault_wait = 0;
669 list_del(&tsk->thread.list);
670 wake_up_process(tsk);
671 put_task_struct(tsk);
673 /* Completion interrupt was faster than initial
674 * interrupt. Set pfault_wait to -1 so the initial
675 * interrupt doesn't put the task to sleep.
676 * If the task is not running, ignore the completion
677 * interrupt since it must be a leftover of a PFAULT
678 * CANCEL operation which didn't remove all pending
679 * completion interrupts. */
680 if (task_is_running(tsk))
681 tsk->thread.pfault_wait = -1;
684 /* signal bit not set -> a real page is missing. */
685 if (WARN_ON_ONCE(tsk != current))
687 if (tsk->thread.pfault_wait == 1) {
688 /* Already on the list with a reference: put to sleep */
690 } else if (tsk->thread.pfault_wait == -1) {
691 /* Completion interrupt was faster than the initial
692 * interrupt (pfault_wait == -1). Set pfault_wait
693 * back to zero and exit. */
694 tsk->thread.pfault_wait = 0;
696 /* Initial interrupt arrived before completion
697 * interrupt. Let the task sleep.
698 * An extra task reference is needed since a different
699 * cpu may set the task state to TASK_RUNNING again
700 * before the scheduler is reached. */
701 get_task_struct(tsk);
702 tsk->thread.pfault_wait = 1;
703 list_add(&tsk->thread.list, &pfault_list);
705 /* Since this must be a userspace fault, there
706 * is no kernel task state to trample. Rely on the
707 * return to userspace schedule() to block. */
708 __set_current_state(TASK_UNINTERRUPTIBLE);
709 set_tsk_need_resched(tsk);
710 set_preempt_need_resched();
714 spin_unlock(&pfault_lock);
715 put_task_struct(tsk);
718 static int pfault_cpu_dead(unsigned int cpu)
720 struct thread_struct *thread, *next;
721 struct task_struct *tsk;
723 spin_lock_irq(&pfault_lock);
724 list_for_each_entry_safe(thread, next, &pfault_list, list) {
725 thread->pfault_wait = 0;
726 list_del(&thread->list);
727 tsk = container_of(thread, struct task_struct, thread);
728 wake_up_process(tsk);
729 put_task_struct(tsk);
731 spin_unlock_irq(&pfault_lock);
735 static int __init pfault_irq_init(void)
739 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
742 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
745 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
746 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
747 NULL, pfault_cpu_dead);
751 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
756 early_initcall(pfault_irq_init);
758 #endif /* CONFIG_PFAULT */
760 #if IS_ENABLED(CONFIG_PGSTE)
762 void do_secure_storage_access(struct pt_regs *regs)
764 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
765 struct vm_area_struct *vma;
766 struct mm_struct *mm;
772 * bit 61 tells us if the address is valid, if it's not we
773 * have a major problem and should stop the kernel or send a
774 * SIGSEGV to the process. Unfortunately bit 61 is not
775 * reliable without the misc UV feature so we need to check
778 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
779 !test_bit_inv(61, ®s->int_parm_long)) {
781 * When this happens, userspace did something that it
782 * was not supposed to do, e.g. branching into secure
783 * memory. Trigger a segmentation fault.
785 if (user_mode(regs)) {
786 send_sig(SIGSEGV, current, 0);
791 * The kernel should never run into this case and we
792 * have no way out of this situation.
794 panic("Unexpected PGM 0x3d with TEID bit 61=0");
797 switch (get_fault_type(regs)) {
800 gmap = (struct gmap *)S390_lowcore.gmap;
802 addr = __gmap_translate(gmap, addr);
803 mmap_read_unlock(mm);
804 if (IS_ERR_VALUE(addr)) {
805 do_fault_error(regs, VM_FAULT_BADMAP);
812 vma = find_vma(mm, addr);
814 mmap_read_unlock(mm);
815 do_fault_error(regs, VM_FAULT_BADMAP);
818 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
819 if (IS_ERR_OR_NULL(page)) {
820 mmap_read_unlock(mm);
823 if (arch_make_page_accessible(page))
824 send_sig(SIGSEGV, current, 0);
826 mmap_read_unlock(mm);
829 page = phys_to_page(addr);
830 if (unlikely(!try_get_page(page)))
832 rc = arch_make_page_accessible(page);
838 do_fault_error(regs, VM_FAULT_BADMAP);
842 NOKPROBE_SYMBOL(do_secure_storage_access);
844 void do_non_secure_storage_access(struct pt_regs *regs)
846 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
847 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
849 if (get_fault_type(regs) != GMAP_FAULT) {
850 do_fault_error(regs, VM_FAULT_BADMAP);
855 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
856 send_sig(SIGSEGV, current, 0);
858 NOKPROBE_SYMBOL(do_non_secure_storage_access);
860 void do_secure_storage_violation(struct pt_regs *regs)
862 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
863 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
866 * If the VM has been rebooted, its address space might still contain
867 * secure pages from the previous boot.
868 * Clear the page so it can be reused.
870 if (!gmap_destroy_page(gmap, gaddr))
873 * Either KVM messed up the secure guest mapping or the same
874 * page is mapped into multiple secure guests.
876 * This exception is only triggered when a guest 2 is running
877 * and can therefore never occur in kernel context.
879 printk_ratelimited(KERN_WARNING
880 "Secure storage violation in task: %s, pid %d\n",
881 current->comm, current->pid);
882 send_sig(SIGSEGV, current, 0);
885 #endif /* CONFIG_PGSTE */