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-offsets.h>
39 #include <asm/mmu_context.h>
40 #include <asm/facility.h>
42 #include "../kernel/entry.h"
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
48 #define VM_FAULT_BADCONTEXT ((__force vm_fault_t) 0x010000)
49 #define VM_FAULT_BADMAP ((__force vm_fault_t) 0x020000)
50 #define VM_FAULT_BADACCESS ((__force vm_fault_t) 0x040000)
51 #define VM_FAULT_SIGNAL ((__force vm_fault_t) 0x080000)
52 #define VM_FAULT_PFAULT ((__force vm_fault_t) 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);
71 * Find out which address space caused the exception.
73 static enum fault_type get_fault_type(struct pt_regs *regs)
75 unsigned long trans_exc_code;
77 trans_exc_code = regs->int_parm_long & 3;
78 if (likely(trans_exc_code == 0)) {
79 /* primary space exception */
82 if (!IS_ENABLED(CONFIG_PGSTE))
84 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
88 if (trans_exc_code == 2)
90 if (trans_exc_code == 1) {
91 /* access register mode, not used in the kernel */
94 /* home space exception -> access via kernel ASCE */
98 static int bad_address(void *p)
102 return get_kernel_nofault(dummy, (unsigned long *)p);
105 static void dump_pagetable(unsigned long asce, unsigned long address)
107 unsigned long *table = __va(asce & _ASCE_ORIGIN);
109 pr_alert("AS:%016lx ", asce);
110 switch (asce & _ASCE_TYPE_MASK) {
111 case _ASCE_TYPE_REGION1:
112 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
113 if (bad_address(table))
115 pr_cont("R1:%016lx ", *table);
116 if (*table & _REGION_ENTRY_INVALID)
118 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
120 case _ASCE_TYPE_REGION2:
121 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
122 if (bad_address(table))
124 pr_cont("R2:%016lx ", *table);
125 if (*table & _REGION_ENTRY_INVALID)
127 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
129 case _ASCE_TYPE_REGION3:
130 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
131 if (bad_address(table))
133 pr_cont("R3:%016lx ", *table);
134 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
136 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
138 case _ASCE_TYPE_SEGMENT:
139 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
140 if (bad_address(table))
142 pr_cont("S:%016lx ", *table);
143 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
145 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
147 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
148 if (bad_address(table))
150 pr_cont("P:%016lx ", *table);
158 static void dump_fault_info(struct pt_regs *regs)
162 pr_alert("Failing address: %016lx TEID: %016lx\n",
163 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
164 pr_alert("Fault in ");
165 switch (regs->int_parm_long & 3) {
167 pr_cont("home space ");
170 pr_cont("secondary space ");
173 pr_cont("access register ");
176 pr_cont("primary space ");
179 pr_cont("mode while using ");
180 switch (get_fault_type(regs)) {
182 asce = S390_lowcore.user_asce;
186 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
190 asce = S390_lowcore.kernel_asce;
197 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
200 int show_unhandled_signals = 1;
202 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
204 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
206 if (!unhandled_signal(current, signr))
208 if (!printk_ratelimit())
210 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
211 regs->int_code & 0xffff, regs->int_code >> 17);
212 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
213 printk(KERN_CONT "\n");
215 dump_fault_info(regs);
220 * Send SIGSEGV to task. This is an external routine
221 * to keep the stack usage of do_page_fault small.
223 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
225 report_user_fault(regs, SIGSEGV, 1);
226 force_sig_fault(SIGSEGV, si_code,
227 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
230 const struct exception_table_entry *s390_search_extables(unsigned long addr)
232 const struct exception_table_entry *fixup;
234 fixup = search_extable(__start_dma_ex_table,
235 __stop_dma_ex_table - __start_dma_ex_table,
238 fixup = search_exception_tables(addr);
242 static noinline void do_no_context(struct pt_regs *regs)
244 const struct exception_table_entry *fixup;
246 /* Are we prepared to handle this kernel fault? */
247 fixup = s390_search_extables(regs->psw.addr);
248 if (fixup && ex_handle(fixup, regs))
252 * Oops. The kernel tried to access some bad page. We'll have to
253 * terminate things with extreme prejudice.
255 if (get_fault_type(regs) == KERNEL_FAULT)
256 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
257 " in virtual kernel address space\n");
259 printk(KERN_ALERT "Unable to handle kernel paging request"
260 " in virtual user address space\n");
261 dump_fault_info(regs);
266 static noinline void do_low_address(struct pt_regs *regs)
268 /* Low-address protection hit in kernel mode means
269 NULL pointer write access in kernel mode. */
270 if (regs->psw.mask & PSW_MASK_PSTATE) {
271 /* Low-address protection hit in user mode 'cannot happen'. */
272 die (regs, "Low-address protection");
279 static noinline void do_sigbus(struct pt_regs *regs)
282 * Send a sigbus, regardless of whether we were in kernel
285 force_sig_fault(SIGBUS, BUS_ADRERR,
286 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
289 static noinline void do_fault_error(struct pt_regs *regs, int access,
295 case VM_FAULT_BADACCESS:
296 case VM_FAULT_BADMAP:
297 /* Bad memory access. Check if it is kernel or user space. */
298 if (user_mode(regs)) {
299 /* User mode accesses just cause a SIGSEGV */
300 si_code = (fault == VM_FAULT_BADMAP) ?
301 SEGV_MAPERR : SEGV_ACCERR;
302 do_sigsegv(regs, si_code);
306 case VM_FAULT_BADCONTEXT:
307 case VM_FAULT_PFAULT:
310 case VM_FAULT_SIGNAL:
311 if (!user_mode(regs))
314 default: /* fault & VM_FAULT_ERROR */
315 if (fault & VM_FAULT_OOM) {
316 if (!user_mode(regs))
319 pagefault_out_of_memory();
320 } else if (fault & VM_FAULT_SIGSEGV) {
321 /* Kernel mode? Handle exceptions or die */
322 if (!user_mode(regs))
325 do_sigsegv(regs, SEGV_MAPERR);
326 } else if (fault & VM_FAULT_SIGBUS) {
327 /* Kernel mode? Handle exceptions or die */
328 if (!user_mode(regs))
339 * This routine handles page faults. It determines the address,
340 * and the problem, and then passes it off to one of the appropriate
343 * interruption code (int_code):
344 * 04 Protection -> Write-Protection (suppression)
345 * 10 Segment translation -> Not present (nullification)
346 * 11 Page translation -> Not present (nullification)
347 * 3b Region third trans. -> Not present (nullification)
349 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
352 struct task_struct *tsk;
353 struct mm_struct *mm;
354 struct vm_area_struct *vma;
355 enum fault_type type;
356 unsigned long trans_exc_code;
357 unsigned long address;
364 * The instruction that caused the program check has
365 * been nullified. Don't signal single step via SIGTRAP.
367 clear_thread_flag(TIF_PER_TRAP);
369 if (kprobe_page_fault(regs, 14))
373 trans_exc_code = regs->int_parm_long;
374 address = trans_exc_code & __FAIL_ADDR_MASK;
375 is_write = (trans_exc_code & store_indication) == 0x400;
378 * Verify that the fault happened in user space, that
379 * we are not in an interrupt and that there is a
382 fault = VM_FAULT_BADCONTEXT;
383 type = get_fault_type(regs);
386 if (kfence_handle_page_fault(address, is_write, regs))
391 if (faulthandler_disabled() || !mm)
396 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
397 flags = FAULT_FLAG_DEFAULT;
399 flags |= FAULT_FLAG_USER;
400 if (access == VM_WRITE || is_write)
401 flags |= FAULT_FLAG_WRITE;
405 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
406 gmap = (struct gmap *) S390_lowcore.gmap;
407 current->thread.gmap_addr = address;
408 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
409 current->thread.gmap_int_code = regs->int_code & 0xffff;
410 address = __gmap_translate(gmap, address);
411 if (address == -EFAULT) {
412 fault = VM_FAULT_BADMAP;
415 if (gmap->pfault_enabled)
416 flags |= FAULT_FLAG_RETRY_NOWAIT;
420 fault = VM_FAULT_BADMAP;
421 vma = find_vma(mm, address);
425 if (unlikely(vma->vm_start > address)) {
426 if (!(vma->vm_flags & VM_GROWSDOWN))
428 if (expand_stack(vma, address))
433 * Ok, we have a good vm_area for this memory access, so
436 fault = VM_FAULT_BADACCESS;
437 if (unlikely(!(vma->vm_flags & access)))
440 if (is_vm_hugetlb_page(vma))
441 address &= HPAGE_MASK;
443 * If for any reason at all we couldn't handle the fault,
444 * make sure we exit gracefully rather than endlessly redo
447 fault = handle_mm_fault(vma, address, flags, regs);
448 if (fault_signal_pending(fault, regs)) {
449 fault = VM_FAULT_SIGNAL;
450 if (flags & FAULT_FLAG_RETRY_NOWAIT)
454 if (unlikely(fault & VM_FAULT_ERROR))
457 if (flags & FAULT_FLAG_ALLOW_RETRY) {
458 if (fault & VM_FAULT_RETRY) {
459 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
460 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
461 /* FAULT_FLAG_RETRY_NOWAIT has been set,
462 * mmap_lock has not been released */
463 current->thread.gmap_pfault = 1;
464 fault = VM_FAULT_PFAULT;
467 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
468 flags |= FAULT_FLAG_TRIED;
473 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
474 address = __gmap_link(gmap, current->thread.gmap_addr,
476 if (address == -EFAULT) {
477 fault = VM_FAULT_BADMAP;
480 if (address == -ENOMEM) {
481 fault = VM_FAULT_OOM;
487 mmap_read_unlock(mm);
492 void do_protection_exception(struct pt_regs *regs)
494 unsigned long trans_exc_code;
498 trans_exc_code = regs->int_parm_long;
500 * Protection exceptions are suppressing, decrement psw address.
501 * The exception to this rule are aborted transactions, for these
502 * the PSW already points to the correct location.
504 if (!(regs->int_code & 0x200))
505 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
507 * Check for low-address protection. This needs to be treated
508 * as a special case because the translation exception code
509 * field is not guaranteed to contain valid data in this case.
511 if (unlikely(!(trans_exc_code & 4))) {
512 do_low_address(regs);
515 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
516 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
517 (regs->psw.addr & PAGE_MASK);
519 fault = VM_FAULT_BADACCESS;
522 fault = do_exception(regs, access);
525 do_fault_error(regs, access, fault);
527 NOKPROBE_SYMBOL(do_protection_exception);
529 void do_dat_exception(struct pt_regs *regs)
534 access = VM_ACCESS_FLAGS;
535 fault = do_exception(regs, access);
537 do_fault_error(regs, access, fault);
539 NOKPROBE_SYMBOL(do_dat_exception);
543 * 'pfault' pseudo page faults routines.
545 static int pfault_disable;
547 static int __init nopfault(char *str)
553 __setup("nopfault", nopfault);
555 struct pfault_refbk {
564 } __attribute__ ((packed, aligned(8)));
566 static struct pfault_refbk pfault_init_refbk = {
571 .refgaddr = __LC_LPP,
572 .refselmk = 1ULL << 48,
573 .refcmpmk = 1ULL << 48,
574 .reserved = __PF_RES_FIELD
577 int pfault_init(void)
583 diag_stat_inc(DIAG_STAT_X258);
585 " diag %1,%0,0x258\n"
591 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
595 static struct pfault_refbk pfault_fini_refbk = {
602 void pfault_fini(void)
607 diag_stat_inc(DIAG_STAT_X258);
612 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
615 static DEFINE_SPINLOCK(pfault_lock);
616 static LIST_HEAD(pfault_list);
618 #define PF_COMPLETE 0x0080
621 * The mechanism of our pfault code: if Linux is running as guest, runs a user
622 * space process and the user space process accesses a page that the host has
623 * paged out we get a pfault interrupt.
625 * This allows us, within the guest, to schedule a different process. Without
626 * this mechanism the host would have to suspend the whole virtual cpu until
627 * the page has been paged in.
629 * So when we get such an interrupt then we set the state of the current task
630 * to uninterruptible and also set the need_resched flag. Both happens within
631 * interrupt context(!). If we later on want to return to user space we
632 * recognize the need_resched flag and then call schedule(). It's not very
633 * obvious how this works...
635 * Of course we have a lot of additional fun with the completion interrupt (->
636 * host signals that a page of a process has been paged in and the process can
637 * continue to run). This interrupt can arrive on any cpu and, since we have
638 * virtual cpus, actually appear before the interrupt that signals that a page
641 static void pfault_interrupt(struct ext_code ext_code,
642 unsigned int param32, unsigned long param64)
644 struct task_struct *tsk;
649 * Get the external interruption subcode & pfault initial/completion
650 * signal bit. VM stores this in the 'cpu address' field associated
651 * with the external interrupt.
653 subcode = ext_code.subcode;
654 if ((subcode & 0xff00) != __SUBCODE_MASK)
656 inc_irq_stat(IRQEXT_PFL);
657 /* Get the token (= pid of the affected task). */
658 pid = param64 & LPP_PID_MASK;
660 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
662 get_task_struct(tsk);
666 spin_lock(&pfault_lock);
667 if (subcode & PF_COMPLETE) {
668 /* signal bit is set -> a page has been swapped in by VM */
669 if (tsk->thread.pfault_wait == 1) {
670 /* Initial interrupt was faster than the completion
671 * interrupt. pfault_wait is valid. Set pfault_wait
672 * back to zero and wake up the process. This can
673 * safely be done because the task is still sleeping
674 * and can't produce new pfaults. */
675 tsk->thread.pfault_wait = 0;
676 list_del(&tsk->thread.list);
677 wake_up_process(tsk);
678 put_task_struct(tsk);
680 /* Completion interrupt was faster than initial
681 * interrupt. Set pfault_wait to -1 so the initial
682 * interrupt doesn't put the task to sleep.
683 * If the task is not running, ignore the completion
684 * interrupt since it must be a leftover of a PFAULT
685 * CANCEL operation which didn't remove all pending
686 * completion interrupts. */
687 if (task_is_running(tsk))
688 tsk->thread.pfault_wait = -1;
691 /* signal bit not set -> a real page is missing. */
692 if (WARN_ON_ONCE(tsk != current))
694 if (tsk->thread.pfault_wait == 1) {
695 /* Already on the list with a reference: put to sleep */
697 } else if (tsk->thread.pfault_wait == -1) {
698 /* Completion interrupt was faster than the initial
699 * interrupt (pfault_wait == -1). Set pfault_wait
700 * back to zero and exit. */
701 tsk->thread.pfault_wait = 0;
703 /* Initial interrupt arrived before completion
704 * interrupt. Let the task sleep.
705 * An extra task reference is needed since a different
706 * cpu may set the task state to TASK_RUNNING again
707 * before the scheduler is reached. */
708 get_task_struct(tsk);
709 tsk->thread.pfault_wait = 1;
710 list_add(&tsk->thread.list, &pfault_list);
712 /* Since this must be a userspace fault, there
713 * is no kernel task state to trample. Rely on the
714 * return to userspace schedule() to block. */
715 __set_current_state(TASK_UNINTERRUPTIBLE);
716 set_tsk_need_resched(tsk);
717 set_preempt_need_resched();
721 spin_unlock(&pfault_lock);
722 put_task_struct(tsk);
725 static int pfault_cpu_dead(unsigned int cpu)
727 struct thread_struct *thread, *next;
728 struct task_struct *tsk;
730 spin_lock_irq(&pfault_lock);
731 list_for_each_entry_safe(thread, next, &pfault_list, list) {
732 thread->pfault_wait = 0;
733 list_del(&thread->list);
734 tsk = container_of(thread, struct task_struct, thread);
735 wake_up_process(tsk);
736 put_task_struct(tsk);
738 spin_unlock_irq(&pfault_lock);
742 static int __init pfault_irq_init(void)
746 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
749 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
752 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
753 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
754 NULL, pfault_cpu_dead);
758 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
763 early_initcall(pfault_irq_init);
765 #endif /* CONFIG_PFAULT */
767 #if IS_ENABLED(CONFIG_PGSTE)
769 void do_secure_storage_access(struct pt_regs *regs)
771 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
772 struct vm_area_struct *vma;
773 struct mm_struct *mm;
778 * bit 61 tells us if the address is valid, if it's not we
779 * have a major problem and should stop the kernel or send a
780 * SIGSEGV to the process. Unfortunately bit 61 is not
781 * reliable without the misc UV feature so we need to check
784 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
785 !test_bit_inv(61, ®s->int_parm_long)) {
787 * When this happens, userspace did something that it
788 * was not supposed to do, e.g. branching into secure
789 * memory. Trigger a segmentation fault.
791 if (user_mode(regs)) {
792 send_sig(SIGSEGV, current, 0);
797 * The kernel should never run into this case and we
798 * have no way out of this situation.
800 panic("Unexpected PGM 0x3d with TEID bit 61=0");
803 switch (get_fault_type(regs)) {
807 vma = find_vma(mm, addr);
809 mmap_read_unlock(mm);
810 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
813 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
814 if (IS_ERR_OR_NULL(page)) {
815 mmap_read_unlock(mm);
818 if (arch_make_page_accessible(page))
819 send_sig(SIGSEGV, current, 0);
821 mmap_read_unlock(mm);
824 page = phys_to_page(addr);
825 if (unlikely(!try_get_page(page)))
827 rc = arch_make_page_accessible(page);
834 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
838 NOKPROBE_SYMBOL(do_secure_storage_access);
840 void do_non_secure_storage_access(struct pt_regs *regs)
842 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
843 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
845 if (get_fault_type(regs) != GMAP_FAULT) {
846 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
851 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
852 send_sig(SIGSEGV, current, 0);
854 NOKPROBE_SYMBOL(do_non_secure_storage_access);
856 void do_secure_storage_violation(struct pt_regs *regs)
859 * Either KVM messed up the secure guest mapping or the same
860 * page is mapped into multiple secure guests.
862 * This exception is only triggered when a guest 2 is running
863 * and can therefore never occur in kernel context.
865 printk_ratelimited(KERN_WARNING
866 "Secure storage violation in task: %s, pid %d\n",
867 current->comm, current->pid);
868 send_sig(SIGSEGV, current, 0);
871 #endif /* CONFIG_PGSTE */