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/mmu_context.h>
14 #include <linux/perf_event.h>
15 #include <linux/signal.h>
16 #include <linux/sched.h>
17 #include <linux/sched/debug.h>
18 #include <linux/jump_label.h>
19 #include <linux/kernel.h>
20 #include <linux/errno.h>
21 #include <linux/string.h>
22 #include <linux/types.h>
23 #include <linux/ptrace.h>
24 #include <linux/mman.h>
26 #include <linux/compat.h>
27 #include <linux/smp.h>
28 #include <linux/kdebug.h>
29 #include <linux/init.h>
30 #include <linux/console.h>
31 #include <linux/extable.h>
32 #include <linux/hardirq.h>
33 #include <linux/kprobes.h>
34 #include <linux/uaccess.h>
35 #include <linux/hugetlb.h>
36 #include <linux/kfence.h>
37 #include <asm/asm-extable.h>
38 #include <asm/asm-offsets.h>
39 #include <asm/ptrace.h>
40 #include <asm/fault.h>
44 #include <asm/facility.h>
46 #include "../kernel/entry.h"
54 static DEFINE_STATIC_KEY_FALSE(have_store_indication);
56 static int __init fault_init(void)
58 if (test_facility(75))
59 static_branch_enable(&have_store_indication);
62 early_initcall(fault_init);
65 * Find out which address space caused the exception.
67 static enum fault_type get_fault_type(struct pt_regs *regs)
69 union teid teid = { .val = regs->int_parm_long };
71 if (likely(teid.as == PSW_BITS_AS_PRIMARY)) {
74 if (!IS_ENABLED(CONFIG_PGSTE))
76 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
80 if (teid.as == PSW_BITS_AS_SECONDARY)
82 /* Access register mode, not used in the kernel */
83 if (teid.as == PSW_BITS_AS_ACCREG)
85 /* Home space -> access via kernel ASCE */
89 static unsigned long get_fault_address(struct pt_regs *regs)
91 union teid teid = { .val = regs->int_parm_long };
93 return teid.addr * PAGE_SIZE;
96 static __always_inline bool fault_is_write(struct pt_regs *regs)
98 union teid teid = { .val = regs->int_parm_long };
100 if (static_branch_likely(&have_store_indication))
101 return teid.fsi == TEID_FSI_STORE;
105 static void dump_pagetable(unsigned long asce, unsigned long address)
107 unsigned long entry, *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 (get_kernel_nofault(entry, table))
115 pr_cont("R1:%016lx ", entry);
116 if (entry & _REGION_ENTRY_INVALID)
118 table = __va(entry & _REGION_ENTRY_ORIGIN);
120 case _ASCE_TYPE_REGION2:
121 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
122 if (get_kernel_nofault(entry, table))
124 pr_cont("R2:%016lx ", entry);
125 if (entry & _REGION_ENTRY_INVALID)
127 table = __va(entry & _REGION_ENTRY_ORIGIN);
129 case _ASCE_TYPE_REGION3:
130 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
131 if (get_kernel_nofault(entry, table))
133 pr_cont("R3:%016lx ", entry);
134 if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
136 table = __va(entry & _REGION_ENTRY_ORIGIN);
138 case _ASCE_TYPE_SEGMENT:
139 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
140 if (get_kernel_nofault(entry, table))
142 pr_cont("S:%016lx ", entry);
143 if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
145 table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
147 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
148 if (get_kernel_nofault(entry, table))
150 pr_cont("P:%016lx ", entry);
158 static void dump_fault_info(struct pt_regs *regs)
160 union teid teid = { .val = regs->int_parm_long };
163 pr_alert("Failing address: %016lx TEID: %016lx\n",
164 get_fault_address(regs), teid.val);
165 pr_alert("Fault in ");
167 case PSW_BITS_AS_HOME:
168 pr_cont("home space ");
170 case PSW_BITS_AS_SECONDARY:
171 pr_cont("secondary space ");
173 case PSW_BITS_AS_ACCREG:
174 pr_cont("access register ");
176 case PSW_BITS_AS_PRIMARY:
177 pr_cont("primary space ");
180 pr_cont("mode while using ");
181 switch (get_fault_type(regs)) {
183 asce = S390_lowcore.user_asce.val;
187 asce = ((struct gmap *)S390_lowcore.gmap)->asce;
191 asce = S390_lowcore.kernel_asce.val;
198 dump_pagetable(asce, get_fault_address(regs));
201 int show_unhandled_signals = 1;
203 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
205 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
207 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
209 if (!unhandled_signal(current, signr))
211 if (!__ratelimit(&rs))
213 pr_alert("User process fault: interruption code %04x ilc:%d ",
214 regs->int_code & 0xffff, regs->int_code >> 17);
215 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
218 dump_fault_info(regs);
222 static void do_sigsegv(struct pt_regs *regs, int si_code)
224 report_user_fault(regs, SIGSEGV, 1);
225 force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
228 static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
230 enum fault_type fault_type;
231 unsigned long address;
234 if (user_mode(regs)) {
235 if (WARN_ON_ONCE(!si_code))
236 si_code = SEGV_MAPERR;
237 return do_sigsegv(regs, si_code);
239 if (fixup_exception(regs))
241 fault_type = get_fault_type(regs);
242 if (fault_type == KERNEL_FAULT) {
243 address = get_fault_address(regs);
244 is_write = fault_is_write(regs);
245 if (kfence_handle_page_fault(address, is_write, regs))
248 if (fault_type == KERNEL_FAULT)
249 pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
251 pr_alert("Unable to handle kernel paging request in virtual user address space\n");
252 dump_fault_info(regs);
256 static void handle_fault_error(struct pt_regs *regs, int si_code)
258 struct mm_struct *mm = current->mm;
260 mmap_read_unlock(mm);
261 handle_fault_error_nolock(regs, si_code);
264 static void do_sigbus(struct pt_regs *regs)
266 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
270 * This routine handles page faults. It determines the address,
271 * and the problem, and then passes it off to one of the appropriate
274 * interruption code (int_code):
275 * 04 Protection -> Write-Protection (suppression)
276 * 10 Segment translation -> Not present (nullification)
277 * 11 Page translation -> Not present (nullification)
278 * 3b Region third trans. -> Not present (nullification)
280 static void do_exception(struct pt_regs *regs, int access)
282 struct vm_area_struct *vma;
283 struct task_struct *tsk;
284 unsigned long address;
285 struct mm_struct *mm;
286 enum fault_type type;
294 * The instruction that caused the program check has
295 * been nullified. Don't signal single step via SIGTRAP.
297 clear_thread_flag(TIF_PER_TRAP);
298 if (kprobe_page_fault(regs, 14))
301 address = get_fault_address(regs);
302 is_write = fault_is_write(regs);
303 type = get_fault_type(regs);
306 return handle_fault_error_nolock(regs, 0);
309 if (faulthandler_disabled() || !mm)
310 return handle_fault_error_nolock(regs, 0);
313 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
314 flags = FAULT_FLAG_DEFAULT;
316 flags |= FAULT_FLAG_USER;
319 if (access == VM_WRITE)
320 flags |= FAULT_FLAG_WRITE;
321 if (!(flags & FAULT_FLAG_USER))
323 vma = lock_vma_under_rcu(mm, address);
326 if (!(vma->vm_flags & access)) {
330 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
331 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
333 if (!(fault & VM_FAULT_RETRY)) {
334 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
335 if (unlikely(fault & VM_FAULT_ERROR))
339 count_vm_vma_lock_event(VMA_LOCK_RETRY);
340 if (fault & VM_FAULT_MAJOR)
341 flags |= FAULT_FLAG_TRIED;
343 /* Quick path to respond to signals */
344 if (fault_signal_pending(fault, regs)) {
345 if (!user_mode(regs))
346 handle_fault_error_nolock(regs, 0);
352 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
353 gmap = (struct gmap *)S390_lowcore.gmap;
354 current->thread.gmap_addr = address;
355 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
356 current->thread.gmap_int_code = regs->int_code & 0xffff;
357 address = __gmap_translate(gmap, address);
358 if (address == -EFAULT)
359 return handle_fault_error(regs, SEGV_MAPERR);
360 if (gmap->pfault_enabled)
361 flags |= FAULT_FLAG_RETRY_NOWAIT;
364 vma = find_vma(mm, address);
366 return handle_fault_error(regs, SEGV_MAPERR);
367 if (unlikely(vma->vm_start > address)) {
368 if (!(vma->vm_flags & VM_GROWSDOWN))
369 return handle_fault_error(regs, SEGV_MAPERR);
370 vma = expand_stack(mm, address);
372 return handle_fault_error_nolock(regs, SEGV_MAPERR);
374 if (unlikely(!(vma->vm_flags & access)))
375 return handle_fault_error(regs, SEGV_ACCERR);
376 fault = handle_mm_fault(vma, address, flags, regs);
377 if (fault_signal_pending(fault, regs)) {
378 if (flags & FAULT_FLAG_RETRY_NOWAIT)
379 mmap_read_unlock(mm);
380 if (!user_mode(regs))
381 handle_fault_error_nolock(regs, 0);
384 /* The fault is fully completed (including releasing mmap lock) */
385 if (fault & VM_FAULT_COMPLETED) {
392 if (unlikely(fault & VM_FAULT_ERROR)) {
393 mmap_read_unlock(mm);
396 if (fault & VM_FAULT_RETRY) {
397 if (IS_ENABLED(CONFIG_PGSTE) && gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
399 * FAULT_FLAG_RETRY_NOWAIT has been set,
400 * mmap_lock has not been released
402 current->thread.gmap_pfault = 1;
403 return handle_fault_error(regs, 0);
405 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
406 flags |= FAULT_FLAG_TRIED;
411 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
412 address = __gmap_link(gmap, current->thread.gmap_addr,
414 if (address == -EFAULT)
415 return handle_fault_error(regs, SEGV_MAPERR);
416 if (address == -ENOMEM) {
417 fault = VM_FAULT_OOM;
418 mmap_read_unlock(mm);
422 mmap_read_unlock(mm);
425 if (fault & VM_FAULT_OOM) {
426 if (!user_mode(regs))
427 handle_fault_error_nolock(regs, 0);
429 pagefault_out_of_memory();
430 } else if (fault & VM_FAULT_SIGSEGV) {
431 if (!user_mode(regs))
432 handle_fault_error_nolock(regs, 0);
434 do_sigsegv(regs, SEGV_MAPERR);
435 } else if (fault & VM_FAULT_SIGBUS) {
436 if (!user_mode(regs))
437 handle_fault_error_nolock(regs, 0);
445 void do_protection_exception(struct pt_regs *regs)
447 union teid teid = { .val = regs->int_parm_long };
450 * Protection exceptions are suppressing, decrement psw address.
451 * The exception to this rule are aborted transactions, for these
452 * the PSW already points to the correct location.
454 if (!(regs->int_code & 0x200))
455 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
457 * Check for low-address protection. This needs to be treated
458 * as a special case because the translation exception code
459 * field is not guaranteed to contain valid data in this case.
461 if (unlikely(!teid.b61)) {
462 if (user_mode(regs)) {
463 /* Low-address protection in user mode: cannot happen */
464 die(regs, "Low-address protection");
467 * Low-address protection in kernel mode means
468 * NULL pointer write access in kernel mode.
470 return handle_fault_error_nolock(regs, 0);
472 if (unlikely(MACHINE_HAS_NX && teid.b56)) {
473 regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
474 return handle_fault_error_nolock(regs, SEGV_ACCERR);
476 do_exception(regs, VM_WRITE);
478 NOKPROBE_SYMBOL(do_protection_exception);
480 void do_dat_exception(struct pt_regs *regs)
482 do_exception(regs, VM_ACCESS_FLAGS);
484 NOKPROBE_SYMBOL(do_dat_exception);
486 #if IS_ENABLED(CONFIG_PGSTE)
488 void do_secure_storage_access(struct pt_regs *regs)
490 union teid teid = { .val = regs->int_parm_long };
491 unsigned long addr = get_fault_address(regs);
492 struct vm_area_struct *vma;
493 struct mm_struct *mm;
499 * Bit 61 indicates if the address is valid, if it is not the
500 * kernel should be stopped or SIGSEGV should be sent to the
501 * process. Bit 61 is not reliable without the misc UV feature,
502 * therefore this needs to be checked too.
504 if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
506 * When this happens, userspace did something that it
507 * was not supposed to do, e.g. branching into secure
508 * memory. Trigger a segmentation fault.
510 if (user_mode(regs)) {
511 send_sig(SIGSEGV, current, 0);
515 * The kernel should never run into this case and
516 * there is no way out of this situation.
518 panic("Unexpected PGM 0x3d with TEID bit 61=0");
520 switch (get_fault_type(regs)) {
523 gmap = (struct gmap *)S390_lowcore.gmap;
525 addr = __gmap_translate(gmap, addr);
526 mmap_read_unlock(mm);
527 if (IS_ERR_VALUE(addr))
528 return handle_fault_error_nolock(regs, SEGV_MAPERR);
533 vma = find_vma(mm, addr);
535 return handle_fault_error(regs, SEGV_MAPERR);
536 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
537 if (IS_ERR_OR_NULL(page)) {
538 mmap_read_unlock(mm);
541 if (arch_make_page_accessible(page))
542 send_sig(SIGSEGV, current, 0);
544 mmap_read_unlock(mm);
547 page = phys_to_page(addr);
548 if (unlikely(!try_get_page(page)))
550 rc = arch_make_page_accessible(page);
559 NOKPROBE_SYMBOL(do_secure_storage_access);
561 void do_non_secure_storage_access(struct pt_regs *regs)
563 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
564 unsigned long gaddr = get_fault_address(regs);
566 if (WARN_ON_ONCE(get_fault_type(regs) != GMAP_FAULT))
567 return handle_fault_error_nolock(regs, SEGV_MAPERR);
568 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
569 send_sig(SIGSEGV, current, 0);
571 NOKPROBE_SYMBOL(do_non_secure_storage_access);
573 void do_secure_storage_violation(struct pt_regs *regs)
575 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
576 unsigned long gaddr = get_fault_address(regs);
579 * If the VM has been rebooted, its address space might still contain
580 * secure pages from the previous boot.
581 * Clear the page so it can be reused.
583 if (!gmap_destroy_page(gmap, gaddr))
586 * Either KVM messed up the secure guest mapping or the same
587 * page is mapped into multiple secure guests.
589 * This exception is only triggered when a guest 2 is running
590 * and can therefore never occur in kernel context.
592 pr_warn_ratelimited("Secure storage violation in task: %s, pid %d\n",
593 current->comm, current->pid);
594 send_sig(SIGSEGV, current, 0);
597 #endif /* CONFIG_PGSTE */