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 <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, int access,
277 case VM_FAULT_BADACCESS:
278 case VM_FAULT_BADMAP:
279 /* Bad memory access. Check if it is kernel or user space. */
280 if (user_mode(regs)) {
281 /* User mode accesses just cause a SIGSEGV */
282 si_code = (fault == VM_FAULT_BADMAP) ?
283 SEGV_MAPERR : SEGV_ACCERR;
284 do_sigsegv(regs, si_code);
288 case VM_FAULT_BADCONTEXT:
289 case VM_FAULT_PFAULT:
292 case VM_FAULT_SIGNAL:
293 if (!user_mode(regs))
296 default: /* fault & VM_FAULT_ERROR */
297 if (fault & VM_FAULT_OOM) {
298 if (!user_mode(regs))
301 pagefault_out_of_memory();
302 } else if (fault & VM_FAULT_SIGSEGV) {
303 /* Kernel mode? Handle exceptions or die */
304 if (!user_mode(regs))
307 do_sigsegv(regs, SEGV_MAPERR);
308 } else if (fault & VM_FAULT_SIGBUS) {
309 /* Kernel mode? Handle exceptions or die */
310 if (!user_mode(regs))
321 * This routine handles page faults. It determines the address,
322 * and the problem, and then passes it off to one of the appropriate
325 * interruption code (int_code):
326 * 04 Protection -> Write-Protection (suppression)
327 * 10 Segment translation -> Not present (nullification)
328 * 11 Page translation -> Not present (nullification)
329 * 3b Region third trans. -> Not present (nullification)
331 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
334 struct task_struct *tsk;
335 struct mm_struct *mm;
336 struct vm_area_struct *vma;
337 enum fault_type type;
338 unsigned long trans_exc_code;
339 unsigned long address;
346 * The instruction that caused the program check has
347 * been nullified. Don't signal single step via SIGTRAP.
349 clear_thread_flag(TIF_PER_TRAP);
351 if (kprobe_page_fault(regs, 14))
355 trans_exc_code = regs->int_parm_long;
356 address = trans_exc_code & __FAIL_ADDR_MASK;
357 is_write = (trans_exc_code & store_indication) == 0x400;
360 * Verify that the fault happened in user space, that
361 * we are not in an interrupt and that there is a
364 fault = VM_FAULT_BADCONTEXT;
365 type = get_fault_type(regs);
368 if (kfence_handle_page_fault(address, is_write, regs))
373 if (faulthandler_disabled() || !mm)
378 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
379 flags = FAULT_FLAG_DEFAULT;
381 flags |= FAULT_FLAG_USER;
382 if (access == VM_WRITE || is_write)
383 flags |= FAULT_FLAG_WRITE;
387 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
388 gmap = (struct gmap *) S390_lowcore.gmap;
389 current->thread.gmap_addr = address;
390 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
391 current->thread.gmap_int_code = regs->int_code & 0xffff;
392 address = __gmap_translate(gmap, address);
393 if (address == -EFAULT) {
394 fault = VM_FAULT_BADMAP;
397 if (gmap->pfault_enabled)
398 flags |= FAULT_FLAG_RETRY_NOWAIT;
402 fault = VM_FAULT_BADMAP;
403 vma = find_vma(mm, address);
407 if (unlikely(vma->vm_start > address)) {
408 if (!(vma->vm_flags & VM_GROWSDOWN))
410 if (expand_stack(vma, address))
415 * Ok, we have a good vm_area for this memory access, so
418 fault = VM_FAULT_BADACCESS;
419 if (unlikely(!(vma->vm_flags & access)))
422 if (is_vm_hugetlb_page(vma))
423 address &= HPAGE_MASK;
425 * If for any reason at all we couldn't handle the fault,
426 * make sure we exit gracefully rather than endlessly redo
429 fault = handle_mm_fault(vma, address, flags, regs);
430 if (fault_signal_pending(fault, regs)) {
431 fault = VM_FAULT_SIGNAL;
432 if (flags & FAULT_FLAG_RETRY_NOWAIT)
436 if (unlikely(fault & VM_FAULT_ERROR))
439 if (fault & VM_FAULT_RETRY) {
440 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
441 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
443 * FAULT_FLAG_RETRY_NOWAIT has been set, mmap_lock has
446 current->thread.gmap_pfault = 1;
447 fault = VM_FAULT_PFAULT;
450 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
451 flags |= FAULT_FLAG_TRIED;
455 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
456 address = __gmap_link(gmap, current->thread.gmap_addr,
458 if (address == -EFAULT) {
459 fault = VM_FAULT_BADMAP;
462 if (address == -ENOMEM) {
463 fault = VM_FAULT_OOM;
469 mmap_read_unlock(mm);
474 void do_protection_exception(struct pt_regs *regs)
476 unsigned long trans_exc_code;
480 trans_exc_code = regs->int_parm_long;
482 * Protection exceptions are suppressing, decrement psw address.
483 * The exception to this rule are aborted transactions, for these
484 * the PSW already points to the correct location.
486 if (!(regs->int_code & 0x200))
487 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
489 * Check for low-address protection. This needs to be treated
490 * as a special case because the translation exception code
491 * field is not guaranteed to contain valid data in this case.
493 if (unlikely(!(trans_exc_code & 4))) {
494 do_low_address(regs);
497 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
498 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
499 (regs->psw.addr & PAGE_MASK);
501 fault = VM_FAULT_BADACCESS;
504 fault = do_exception(regs, access);
507 do_fault_error(regs, access, fault);
509 NOKPROBE_SYMBOL(do_protection_exception);
511 void do_dat_exception(struct pt_regs *regs)
516 access = VM_ACCESS_FLAGS;
517 fault = do_exception(regs, access);
519 do_fault_error(regs, access, fault);
521 NOKPROBE_SYMBOL(do_dat_exception);
525 * 'pfault' pseudo page faults routines.
527 static int pfault_disable;
529 static int __init nopfault(char *str)
535 __setup("nopfault", nopfault);
537 struct pfault_refbk {
546 } __attribute__ ((packed, aligned(8)));
548 static struct pfault_refbk pfault_init_refbk = {
553 .refgaddr = __LC_LPP,
554 .refselmk = 1ULL << 48,
555 .refcmpmk = 1ULL << 48,
556 .reserved = __PF_RES_FIELD
559 int pfault_init(void)
565 diag_stat_inc(DIAG_STAT_X258);
567 " diag %1,%0,0x258\n"
573 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
577 static struct pfault_refbk pfault_fini_refbk = {
584 void pfault_fini(void)
589 diag_stat_inc(DIAG_STAT_X258);
594 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
597 static DEFINE_SPINLOCK(pfault_lock);
598 static LIST_HEAD(pfault_list);
600 #define PF_COMPLETE 0x0080
603 * The mechanism of our pfault code: if Linux is running as guest, runs a user
604 * space process and the user space process accesses a page that the host has
605 * paged out we get a pfault interrupt.
607 * This allows us, within the guest, to schedule a different process. Without
608 * this mechanism the host would have to suspend the whole virtual cpu until
609 * the page has been paged in.
611 * So when we get such an interrupt then we set the state of the current task
612 * to uninterruptible and also set the need_resched flag. Both happens within
613 * interrupt context(!). If we later on want to return to user space we
614 * recognize the need_resched flag and then call schedule(). It's not very
615 * obvious how this works...
617 * Of course we have a lot of additional fun with the completion interrupt (->
618 * host signals that a page of a process has been paged in and the process can
619 * continue to run). This interrupt can arrive on any cpu and, since we have
620 * virtual cpus, actually appear before the interrupt that signals that a page
623 static void pfault_interrupt(struct ext_code ext_code,
624 unsigned int param32, unsigned long param64)
626 struct task_struct *tsk;
631 * Get the external interruption subcode & pfault initial/completion
632 * signal bit. VM stores this in the 'cpu address' field associated
633 * with the external interrupt.
635 subcode = ext_code.subcode;
636 if ((subcode & 0xff00) != __SUBCODE_MASK)
638 inc_irq_stat(IRQEXT_PFL);
639 /* Get the token (= pid of the affected task). */
640 pid = param64 & LPP_PID_MASK;
642 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
644 get_task_struct(tsk);
648 spin_lock(&pfault_lock);
649 if (subcode & PF_COMPLETE) {
650 /* signal bit is set -> a page has been swapped in by VM */
651 if (tsk->thread.pfault_wait == 1) {
652 /* Initial interrupt was faster than the completion
653 * interrupt. pfault_wait is valid. Set pfault_wait
654 * back to zero and wake up the process. This can
655 * safely be done because the task is still sleeping
656 * and can't produce new pfaults. */
657 tsk->thread.pfault_wait = 0;
658 list_del(&tsk->thread.list);
659 wake_up_process(tsk);
660 put_task_struct(tsk);
662 /* Completion interrupt was faster than initial
663 * interrupt. Set pfault_wait to -1 so the initial
664 * interrupt doesn't put the task to sleep.
665 * If the task is not running, ignore the completion
666 * interrupt since it must be a leftover of a PFAULT
667 * CANCEL operation which didn't remove all pending
668 * completion interrupts. */
669 if (task_is_running(tsk))
670 tsk->thread.pfault_wait = -1;
673 /* signal bit not set -> a real page is missing. */
674 if (WARN_ON_ONCE(tsk != current))
676 if (tsk->thread.pfault_wait == 1) {
677 /* Already on the list with a reference: put to sleep */
679 } else if (tsk->thread.pfault_wait == -1) {
680 /* Completion interrupt was faster than the initial
681 * interrupt (pfault_wait == -1). Set pfault_wait
682 * back to zero and exit. */
683 tsk->thread.pfault_wait = 0;
685 /* Initial interrupt arrived before completion
686 * interrupt. Let the task sleep.
687 * An extra task reference is needed since a different
688 * cpu may set the task state to TASK_RUNNING again
689 * before the scheduler is reached. */
690 get_task_struct(tsk);
691 tsk->thread.pfault_wait = 1;
692 list_add(&tsk->thread.list, &pfault_list);
694 /* Since this must be a userspace fault, there
695 * is no kernel task state to trample. Rely on the
696 * return to userspace schedule() to block. */
697 __set_current_state(TASK_UNINTERRUPTIBLE);
698 set_tsk_need_resched(tsk);
699 set_preempt_need_resched();
703 spin_unlock(&pfault_lock);
704 put_task_struct(tsk);
707 static int pfault_cpu_dead(unsigned int cpu)
709 struct thread_struct *thread, *next;
710 struct task_struct *tsk;
712 spin_lock_irq(&pfault_lock);
713 list_for_each_entry_safe(thread, next, &pfault_list, list) {
714 thread->pfault_wait = 0;
715 list_del(&thread->list);
716 tsk = container_of(thread, struct task_struct, thread);
717 wake_up_process(tsk);
718 put_task_struct(tsk);
720 spin_unlock_irq(&pfault_lock);
724 static int __init pfault_irq_init(void)
728 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
731 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
734 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
735 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
736 NULL, pfault_cpu_dead);
740 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
745 early_initcall(pfault_irq_init);
747 #endif /* CONFIG_PFAULT */
749 #if IS_ENABLED(CONFIG_PGSTE)
751 void do_secure_storage_access(struct pt_regs *regs)
753 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
754 struct vm_area_struct *vma;
755 struct mm_struct *mm;
760 * bit 61 tells us if the address is valid, if it's not we
761 * have a major problem and should stop the kernel or send a
762 * SIGSEGV to the process. Unfortunately bit 61 is not
763 * reliable without the misc UV feature so we need to check
766 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
767 !test_bit_inv(61, ®s->int_parm_long)) {
769 * When this happens, userspace did something that it
770 * was not supposed to do, e.g. branching into secure
771 * memory. Trigger a segmentation fault.
773 if (user_mode(regs)) {
774 send_sig(SIGSEGV, current, 0);
779 * The kernel should never run into this case and we
780 * have no way out of this situation.
782 panic("Unexpected PGM 0x3d with TEID bit 61=0");
785 switch (get_fault_type(regs)) {
789 vma = find_vma(mm, addr);
791 mmap_read_unlock(mm);
792 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
795 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
796 if (IS_ERR_OR_NULL(page)) {
797 mmap_read_unlock(mm);
800 if (arch_make_page_accessible(page))
801 send_sig(SIGSEGV, current, 0);
803 mmap_read_unlock(mm);
806 page = phys_to_page(addr);
807 if (unlikely(!try_get_page(page)))
809 rc = arch_make_page_accessible(page);
816 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
820 NOKPROBE_SYMBOL(do_secure_storage_access);
822 void do_non_secure_storage_access(struct pt_regs *regs)
824 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
825 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
827 if (get_fault_type(regs) != GMAP_FAULT) {
828 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
833 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
834 send_sig(SIGSEGV, current, 0);
836 NOKPROBE_SYMBOL(do_non_secure_storage_access);
838 void do_secure_storage_violation(struct pt_regs *regs)
841 * Either KVM messed up the secure guest mapping or the same
842 * page is mapped into multiple secure guests.
844 * This exception is only triggered when a guest 2 is running
845 * and can therefore never occur in kernel context.
847 printk_ratelimited(KERN_WARNING
848 "Secure storage violation in task: %s, pid %d\n",
849 current->comm, current->pid);
850 send_sig(SIGSEGV, current, 0);
853 #endif /* CONFIG_PGSTE */