1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * S390 version
4 * Copyright IBM Corp. 1999
5 * Author(s): Hartmut Penner (hp@de.ibm.com)
6 * Ulrich Weigand (uweigand@de.ibm.com)
7 *
8 * Derived from "arch/i386/mm/fault.c"
9 * Copyright (C) 1995 Linus Torvalds
10 */
11
12 #include <linux/kernel_stat.h>
13 #include <linux/mmu_context.h>
14 #include <linux/cpufeature.h>
15 #include <linux/perf_event.h>
16 #include <linux/signal.h>
17 #include <linux/sched.h>
18 #include <linux/sched/debug.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>
25 #include <linux/mm.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 <linux/pagewalk.h>
38 #include <asm/asm-extable.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/ptrace.h>
41 #include <asm/fault.h>
42 #include <asm/diag.h>
43 #include <asm/gmap.h>
44 #include <asm/irq.h>
45 #include <asm/facility.h>
46 #include <asm/uv.h>
47 #include "../kernel/entry.h"
48
49 /*
50 * Find out which address space caused the exception.
51 */
is_kernel_fault(struct pt_regs * regs)52 static bool is_kernel_fault(struct pt_regs *regs)
53 {
54 union teid teid = { .val = regs->int_parm_long };
55
56 if (user_mode(regs))
57 return false;
58 if (teid.as == PSW_BITS_AS_SECONDARY)
59 return false;
60 return true;
61 }
62
get_fault_address(struct pt_regs * regs)63 static unsigned long get_fault_address(struct pt_regs *regs)
64 {
65 union teid teid = { .val = regs->int_parm_long };
66
67 return teid.addr * PAGE_SIZE;
68 }
69
fault_is_write(struct pt_regs * regs)70 static __always_inline bool fault_is_write(struct pt_regs *regs)
71 {
72 union teid teid = { .val = regs->int_parm_long };
73
74 if (test_facility(75))
75 return teid.fsi == TEID_FSI_STORE;
76 return false;
77 }
78
dump_pagetable(unsigned long asce,unsigned long address)79 static void dump_pagetable(unsigned long asce, unsigned long address)
80 {
81 unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
82
83 pr_alert("AS:%016lx ", asce);
84 switch (asce & _ASCE_TYPE_MASK) {
85 case _ASCE_TYPE_REGION1:
86 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
87 if (get_kernel_nofault(entry, table))
88 goto bad;
89 pr_cont("R1:%016lx ", entry);
90 if (entry & _REGION_ENTRY_INVALID)
91 goto out;
92 table = __va(entry & _REGION_ENTRY_ORIGIN);
93 fallthrough;
94 case _ASCE_TYPE_REGION2:
95 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
96 if (get_kernel_nofault(entry, table))
97 goto bad;
98 pr_cont("R2:%016lx ", entry);
99 if (entry & _REGION_ENTRY_INVALID)
100 goto out;
101 table = __va(entry & _REGION_ENTRY_ORIGIN);
102 fallthrough;
103 case _ASCE_TYPE_REGION3:
104 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
105 if (get_kernel_nofault(entry, table))
106 goto bad;
107 pr_cont("R3:%016lx ", entry);
108 if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
109 goto out;
110 table = __va(entry & _REGION_ENTRY_ORIGIN);
111 fallthrough;
112 case _ASCE_TYPE_SEGMENT:
113 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
114 if (get_kernel_nofault(entry, table))
115 goto bad;
116 pr_cont("S:%016lx ", entry);
117 if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
118 goto out;
119 table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
120 }
121 table += (address & _PAGE_INDEX) >> PAGE_SHIFT;
122 if (get_kernel_nofault(entry, table))
123 goto bad;
124 pr_cont("P:%016lx ", entry);
125 out:
126 pr_cont("\n");
127 return;
128 bad:
129 pr_cont("BAD\n");
130 }
131
dump_fault_info(struct pt_regs * regs)132 static void dump_fault_info(struct pt_regs *regs)
133 {
134 union teid teid = { .val = regs->int_parm_long };
135 unsigned long asce;
136
137 pr_alert("Failing address: %016lx TEID: %016lx\n",
138 get_fault_address(regs), teid.val);
139 pr_alert("Fault in ");
140 switch (teid.as) {
141 case PSW_BITS_AS_HOME:
142 pr_cont("home space ");
143 break;
144 case PSW_BITS_AS_SECONDARY:
145 pr_cont("secondary space ");
146 break;
147 case PSW_BITS_AS_ACCREG:
148 pr_cont("access register ");
149 break;
150 case PSW_BITS_AS_PRIMARY:
151 pr_cont("primary space ");
152 break;
153 }
154 pr_cont("mode while using ");
155 if (is_kernel_fault(regs)) {
156 asce = get_lowcore()->kernel_asce.val;
157 pr_cont("kernel ");
158 } else {
159 asce = get_lowcore()->user_asce.val;
160 pr_cont("user ");
161 }
162 pr_cont("ASCE.\n");
163 dump_pagetable(asce, get_fault_address(regs));
164 }
165
166 int show_unhandled_signals = 1;
167
168 static const struct ctl_table s390_fault_sysctl_table[] = {
169 {
170 .procname = "userprocess_debug",
171 .data = &show_unhandled_signals,
172 .maxlen = sizeof(int),
173 .mode = 0644,
174 .proc_handler = proc_dointvec,
175 },
176 };
177
init_s390_fault_sysctls(void)178 static int __init init_s390_fault_sysctls(void)
179 {
180 register_sysctl_init("kernel", s390_fault_sysctl_table);
181 return 0;
182 }
183 arch_initcall(init_s390_fault_sysctls);
184
report_user_fault(struct pt_regs * regs,long signr,int is_mm_fault)185 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
186 {
187 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
188
189 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
190 return;
191 if (!unhandled_signal(current, signr))
192 return;
193 if (!__ratelimit(&rs))
194 return;
195 pr_alert("User process fault: interruption code %04x ilc:%d ",
196 regs->int_code & 0xffff, regs->int_code >> 17);
197 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
198 pr_cont("\n");
199 if (is_mm_fault)
200 dump_fault_info(regs);
201 show_regs(regs);
202 }
203
do_sigsegv(struct pt_regs * regs,int si_code)204 static void do_sigsegv(struct pt_regs *regs, int si_code)
205 {
206 report_user_fault(regs, SIGSEGV, 1);
207 force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
208 }
209
handle_fault_error_nolock(struct pt_regs * regs,int si_code)210 static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
211 {
212 unsigned long address;
213 bool is_write;
214
215 if (user_mode(regs)) {
216 if (WARN_ON_ONCE(!si_code))
217 si_code = SEGV_MAPERR;
218 return do_sigsegv(regs, si_code);
219 }
220 if (fixup_exception(regs))
221 return;
222 if (is_kernel_fault(regs)) {
223 address = get_fault_address(regs);
224 is_write = fault_is_write(regs);
225 if (kfence_handle_page_fault(address, is_write, regs))
226 return;
227 pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
228 } else {
229 pr_alert("Unable to handle kernel paging request in virtual user address space\n");
230 }
231 dump_fault_info(regs);
232 die(regs, "Oops");
233 }
234
handle_fault_error(struct pt_regs * regs,int si_code)235 static void handle_fault_error(struct pt_regs *regs, int si_code)
236 {
237 struct mm_struct *mm = current->mm;
238
239 mmap_read_unlock(mm);
240 handle_fault_error_nolock(regs, si_code);
241 }
242
do_sigbus(struct pt_regs * regs)243 static void do_sigbus(struct pt_regs *regs)
244 {
245 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
246 }
247
248 /*
249 * This routine handles page faults. It determines the address,
250 * and the problem, and then passes it off to one of the appropriate
251 * routines.
252 *
253 * interruption code (int_code):
254 * 04 Protection -> Write-Protection (suppression)
255 * 10 Segment translation -> Not present (nullification)
256 * 11 Page translation -> Not present (nullification)
257 * 3b Region third trans. -> Not present (nullification)
258 */
do_exception(struct pt_regs * regs,int access)259 static void do_exception(struct pt_regs *regs, int access)
260 {
261 struct vm_area_struct *vma;
262 unsigned long address;
263 struct mm_struct *mm;
264 unsigned int flags;
265 vm_fault_t fault;
266 bool is_write;
267
268 /*
269 * The instruction that caused the program check has
270 * been nullified. Don't signal single step via SIGTRAP.
271 */
272 clear_thread_flag(TIF_PER_TRAP);
273 if (kprobe_page_fault(regs, 14))
274 return;
275 mm = current->mm;
276 address = get_fault_address(regs);
277 is_write = fault_is_write(regs);
278 if (is_kernel_fault(regs) || faulthandler_disabled() || !mm)
279 return handle_fault_error_nolock(regs, 0);
280 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
281 flags = FAULT_FLAG_DEFAULT;
282 if (user_mode(regs))
283 flags |= FAULT_FLAG_USER;
284 if (is_write)
285 access = VM_WRITE;
286 if (access == VM_WRITE)
287 flags |= FAULT_FLAG_WRITE;
288 if (!(flags & FAULT_FLAG_USER))
289 goto lock_mmap;
290 vma = lock_vma_under_rcu(mm, address);
291 if (!vma)
292 goto lock_mmap;
293 if (!(vma->vm_flags & access)) {
294 vma_end_read(vma);
295 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
296 return handle_fault_error_nolock(regs, SEGV_ACCERR);
297 }
298 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
299 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
300 vma_end_read(vma);
301 if (!(fault & VM_FAULT_RETRY)) {
302 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
303 goto done;
304 }
305 count_vm_vma_lock_event(VMA_LOCK_RETRY);
306 if (fault & VM_FAULT_MAJOR)
307 flags |= FAULT_FLAG_TRIED;
308 /* Quick path to respond to signals */
309 if (fault_signal_pending(fault, regs)) {
310 if (!user_mode(regs))
311 handle_fault_error_nolock(regs, 0);
312 return;
313 }
314 lock_mmap:
315 retry:
316 vma = lock_mm_and_find_vma(mm, address, regs);
317 if (!vma)
318 return handle_fault_error_nolock(regs, SEGV_MAPERR);
319 if (unlikely(!(vma->vm_flags & access)))
320 return handle_fault_error(regs, SEGV_ACCERR);
321 fault = handle_mm_fault(vma, address, flags, regs);
322 if (fault_signal_pending(fault, regs)) {
323 if (!user_mode(regs))
324 handle_fault_error_nolock(regs, 0);
325 return;
326 }
327 /* The fault is fully completed (including releasing mmap lock) */
328 if (fault & VM_FAULT_COMPLETED)
329 return;
330 if (fault & VM_FAULT_RETRY) {
331 flags |= FAULT_FLAG_TRIED;
332 goto retry;
333 }
334 mmap_read_unlock(mm);
335 done:
336 if (!(fault & VM_FAULT_ERROR))
337 return;
338 if (fault & VM_FAULT_OOM) {
339 if (!user_mode(regs))
340 handle_fault_error_nolock(regs, 0);
341 else
342 pagefault_out_of_memory();
343 } else if (fault & VM_FAULT_SIGSEGV) {
344 if (!user_mode(regs))
345 handle_fault_error_nolock(regs, 0);
346 else
347 do_sigsegv(regs, SEGV_MAPERR);
348 } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
349 VM_FAULT_HWPOISON_LARGE)) {
350 if (!user_mode(regs))
351 handle_fault_error_nolock(regs, 0);
352 else
353 do_sigbus(regs);
354 } else {
355 pr_emerg("Unexpected fault flags: %08x\n", fault);
356 BUG();
357 }
358 }
359
do_protection_exception(struct pt_regs * regs)360 void do_protection_exception(struct pt_regs *regs)
361 {
362 union teid teid = { .val = regs->int_parm_long };
363
364 /*
365 * Protection exceptions are suppressing, decrement psw address.
366 * The exception to this rule are aborted transactions, for these
367 * the PSW already points to the correct location.
368 */
369 if (!(regs->int_code & 0x200))
370 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
371 /*
372 * Check for low-address protection. This needs to be treated
373 * as a special case because the translation exception code
374 * field is not guaranteed to contain valid data in this case.
375 */
376 if (unlikely(!teid.b61)) {
377 if (user_mode(regs)) {
378 /* Low-address protection in user mode: cannot happen */
379 dump_fault_info(regs);
380 die(regs, "Low-address protection");
381 }
382 /*
383 * Low-address protection in kernel mode means
384 * NULL pointer write access in kernel mode.
385 */
386 return handle_fault_error_nolock(regs, 0);
387 }
388 if (unlikely(cpu_has_nx() && teid.b56)) {
389 regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
390 return handle_fault_error_nolock(regs, SEGV_ACCERR);
391 }
392 do_exception(regs, VM_WRITE);
393 }
394 NOKPROBE_SYMBOL(do_protection_exception);
395
do_dat_exception(struct pt_regs * regs)396 void do_dat_exception(struct pt_regs *regs)
397 {
398 do_exception(regs, VM_ACCESS_FLAGS);
399 }
400 NOKPROBE_SYMBOL(do_dat_exception);
401
402 #if IS_ENABLED(CONFIG_PGSTE)
403
do_secure_storage_access(struct pt_regs * regs)404 void do_secure_storage_access(struct pt_regs *regs)
405 {
406 union teid teid = { .val = regs->int_parm_long };
407 unsigned long addr = get_fault_address(regs);
408 struct vm_area_struct *vma;
409 struct folio_walk fw;
410 struct mm_struct *mm;
411 struct folio *folio;
412 int rc;
413
414 /*
415 * Bit 61 indicates if the address is valid, if it is not the
416 * kernel should be stopped or SIGSEGV should be sent to the
417 * process. Bit 61 is not reliable without the misc UV feature,
418 * therefore this needs to be checked too.
419 */
420 if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
421 /*
422 * When this happens, userspace did something that it
423 * was not supposed to do, e.g. branching into secure
424 * memory. Trigger a segmentation fault.
425 */
426 if (user_mode(regs)) {
427 send_sig(SIGSEGV, current, 0);
428 return;
429 }
430 /*
431 * The kernel should never run into this case and
432 * there is no way out of this situation.
433 */
434 panic("Unexpected PGM 0x3d with TEID bit 61=0");
435 }
436 if (is_kernel_fault(regs)) {
437 folio = phys_to_folio(addr);
438 if (unlikely(!folio_try_get(folio)))
439 return;
440 rc = arch_make_folio_accessible(folio);
441 folio_put(folio);
442 if (rc)
443 BUG();
444 } else {
445 mm = current->mm;
446 mmap_read_lock(mm);
447 vma = find_vma(mm, addr);
448 if (!vma)
449 return handle_fault_error(regs, SEGV_MAPERR);
450 folio = folio_walk_start(&fw, vma, addr, 0);
451 if (!folio) {
452 mmap_read_unlock(mm);
453 return;
454 }
455 /* arch_make_folio_accessible() needs a raised refcount. */
456 folio_get(folio);
457 rc = arch_make_folio_accessible(folio);
458 folio_put(folio);
459 folio_walk_end(&fw, vma);
460 if (rc)
461 send_sig(SIGSEGV, current, 0);
462 mmap_read_unlock(mm);
463 }
464 }
465 NOKPROBE_SYMBOL(do_secure_storage_access);
466
467 #endif /* CONFIG_PGSTE */
468