fault.c

  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/smp.h>
 27#include <linux/kdebug.h>
 28#include <linux/init.h>
 29#include <linux/console.h>
 30#include <linux/extable.h>
 31#include <linux/hardirq.h>
 32#include <linux/kprobes.h>
 33#include <linux/uaccess.h>
 34#include <linux/hugetlb.h>
 35#include <linux/kfence.h>
 36#include <linux/pagewalk.h>
 37#include <asm/asm-extable.h>
 38#include <asm/asm-offsets.h>
 39#include <asm/ptrace.h>
 40#include <asm/fault.h>
 41#include <asm/diag.h>
 42#include <asm/irq.h>
 43#include <asm/facility.h>
 44#include <asm/uv.h>
 45#include "../kernel/entry.h"
 46
 47/*
 48 * Find out which address space caused the exception.
 49 */
 50static bool is_kernel_fault(struct pt_regs *regs)
 51{
 52	union teid teid = { .val = regs->int_parm_long };
 53
 54	if (user_mode(regs))
 55		return false;
 56	if (teid.as == PSW_BITS_AS_SECONDARY)
 57		return false;
 58	return true;
 59}
 60
 61static unsigned long get_fault_address(struct pt_regs *regs)
 62{
 63	union teid teid = { .val = regs->int_parm_long };
 64
 65	return teid.addr * PAGE_SIZE;
 66}
 67
 68static __always_inline bool fault_is_write(struct pt_regs *regs)
 69{
 70	union teid teid = { .val = regs->int_parm_long };
 71
 72	if (test_facility(75))
 73		return teid.fsi == TEID_FSI_STORE;
 74	return false;
 75}
 76
 77static void dump_pagetable(unsigned long asce, unsigned long address)
 78{
 79	unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
 80
 81	pr_alert("AS:%016lx ", asce);
 82	switch (asce & _ASCE_TYPE_MASK) {
 83	case _ASCE_TYPE_REGION1:
 84		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
 85		if (get_kernel_nofault(entry, table))
 86			goto bad;
 87		pr_cont("R1:%016lx ", entry);
 88		if (entry & _REGION_ENTRY_INVALID)
 89			goto out;
 90		table = __va(entry & _REGION_ENTRY_ORIGIN);
 91		fallthrough;
 92	case _ASCE_TYPE_REGION2:
 93		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
 94		if (get_kernel_nofault(entry, table))
 95			goto bad;
 96		pr_cont("R2:%016lx ", entry);
 97		if (entry & _REGION_ENTRY_INVALID)
 98			goto out;
 99		table = __va(entry & _REGION_ENTRY_ORIGIN);
100		fallthrough;
101	case _ASCE_TYPE_REGION3:
102		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
103		if (get_kernel_nofault(entry, table))
104			goto bad;
105		pr_cont("R3:%016lx ", entry);
106		if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
107			goto out;
108		table = __va(entry & _REGION_ENTRY_ORIGIN);
109		fallthrough;
110	case _ASCE_TYPE_SEGMENT:
111		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
112		if (get_kernel_nofault(entry, table))
113			goto bad;
114		pr_cont("S:%016lx ", entry);
115		if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
116			goto out;
117		table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
118	}
119	table += (address & _PAGE_INDEX) >> PAGE_SHIFT;
120	if (get_kernel_nofault(entry, table))
121		goto bad;
122	pr_cont("P:%016lx ", entry);
123out:
124	pr_cont("\n");
125	return;
126bad:
127	pr_cont("BAD\n");
128}
129
130static void dump_fault_info(struct pt_regs *regs)
131{
132	union teid teid = { .val = regs->int_parm_long };
133	unsigned long asce;
134
135	pr_alert("Failing address: %016lx TEID: %016lx",
136		 get_fault_address(regs), teid.val);
137	if (test_facility(131))
138		pr_cont(" ESOP-2");
139	else if (machine_has_esop())
140		pr_cont(" ESOP-1");
141	else
142		pr_cont(" SOP");
143	if (test_facility(75))
144		pr_cont(" FSI");
145	pr_cont("\n");
146	pr_alert("Fault in ");
147	switch (teid.as) {
148	case PSW_BITS_AS_HOME:
149		pr_cont("home space ");
150		break;
151	case PSW_BITS_AS_SECONDARY:
152		pr_cont("secondary space ");
153		break;
154	case PSW_BITS_AS_ACCREG:
155		pr_cont("access register ");
156		break;
157	case PSW_BITS_AS_PRIMARY:
158		pr_cont("primary space ");
159		break;
160	}
161	pr_cont("mode while using ");
162	if (is_kernel_fault(regs)) {
163		asce = get_lowcore()->kernel_asce.val;
164		pr_cont("kernel ");
165	} else {
166		asce = get_lowcore()->user_asce.val;
167		pr_cont("user ");
168	}
169	pr_cont("ASCE.\n");
170	dump_pagetable(asce, get_fault_address(regs));
171}
172
173int show_unhandled_signals = 1;
174
175static const struct ctl_table s390_fault_sysctl_table[] = {
176	{
177		.procname	= "userprocess_debug",
178		.data		= &show_unhandled_signals,
179		.maxlen		= sizeof(int),
180		.mode		= 0644,
181		.proc_handler	= proc_dointvec,
182	},
183};
184
185static int __init init_s390_fault_sysctls(void)
186{
187	register_sysctl_init("kernel", s390_fault_sysctl_table);
188	return 0;
189}
190arch_initcall(init_s390_fault_sysctls);
191
192void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
193{
194	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
195
196	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
197		return;
198	if (!unhandled_signal(current, signr))
199		return;
200	if (!__ratelimit(&rs))
201		return;
202	pr_alert("User process fault: interruption code %04x ilc:%d ",
203		 regs->int_code & 0xffff, regs->int_code >> 17);
204	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
205	pr_cont("\n");
206	if (is_mm_fault)
207		dump_fault_info(regs);
208	show_regs(regs);
209}
210
211static void do_sigsegv(struct pt_regs *regs, int si_code)
212{
213	report_user_fault(regs, SIGSEGV, 1);
214	force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
215}
216
217static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
218{
219	unsigned long address;
220	bool is_write;
221
222	if (user_mode(regs)) {
223		if (WARN_ON_ONCE(!si_code))
224			si_code = SEGV_MAPERR;
225		return do_sigsegv(regs, si_code);
226	}
227	if (fixup_exception(regs))
228		return;
229	if (is_kernel_fault(regs)) {
230		address = get_fault_address(regs);
231		is_write = fault_is_write(regs);
232		if (kfence_handle_page_fault(address, is_write, regs))
233			return;
234		pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
235	} else {
236		pr_alert("Unable to handle kernel paging request in virtual user address space\n");
237	}
238	dump_fault_info(regs);
239	die(regs, "Oops");
240}
241
242static void handle_fault_error(struct pt_regs *regs, int si_code)
243{
244	struct mm_struct *mm = current->mm;
245
246	mmap_read_unlock(mm);
247	handle_fault_error_nolock(regs, si_code);
248}
249
250static void do_sigbus(struct pt_regs *regs)
251{
252	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
253}
254
255/*
256 * This routine handles page faults.  It determines the address,
257 * and the problem, and then passes it off to one of the appropriate
258 * routines.
259 *
260 * interruption code (int_code):
261 *   04       Protection	   ->  Write-Protection  (suppression)
262 *   10       Segment translation  ->  Not present	 (nullification)
263 *   11       Page translation	   ->  Not present	 (nullification)
264 *   3b       Region third trans.  ->  Not present	 (nullification)
265 */
266static void do_exception(struct pt_regs *regs, int access)
267{
268	struct vm_area_struct *vma;
269	unsigned long address;
270	struct mm_struct *mm;
271	unsigned int flags;
272	vm_fault_t fault;
273	bool is_write;
274
275	/*
276	 * The instruction that caused the program check has
277	 * been nullified. Don't signal single step via SIGTRAP.
278	 */
279	clear_thread_flag(TIF_PER_TRAP);
280	if (kprobe_page_fault(regs, 14))
281		return;
282	mm = current->mm;
283	address = get_fault_address(regs);
284	is_write = fault_is_write(regs);
285	if (is_kernel_fault(regs) || faulthandler_disabled() || !mm)
286		return handle_fault_error_nolock(regs, 0);
287	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
288	flags = FAULT_FLAG_DEFAULT;
289	if (user_mode(regs))
290		flags |= FAULT_FLAG_USER;
291	if (is_write)
292		access = VM_WRITE;
293	if (access == VM_WRITE)
294		flags |= FAULT_FLAG_WRITE;
295	if (!(flags & FAULT_FLAG_USER))
296		goto lock_mmap;
297	vma = lock_vma_under_rcu(mm, address);
298	if (!vma)
299		goto lock_mmap;
300	if (!(vma->vm_flags & access)) {
301		vma_end_read(vma);
302		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
303		return handle_fault_error_nolock(regs, SEGV_ACCERR);
304	}
305	fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
306	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
307		vma_end_read(vma);
308	if (!(fault & VM_FAULT_RETRY)) {
309		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
310		goto done;
311	}
312	count_vm_vma_lock_event(VMA_LOCK_RETRY);
313	if (fault & VM_FAULT_MAJOR)
314		flags |= FAULT_FLAG_TRIED;
315	/* Quick path to respond to signals */
316	if (fault_signal_pending(fault, regs)) {
317		if (!user_mode(regs))
318			handle_fault_error_nolock(regs, 0);
319		return;
320	}
321lock_mmap:
322retry:
323	vma = lock_mm_and_find_vma(mm, address, regs);
324	if (!vma)
325		return handle_fault_error_nolock(regs, SEGV_MAPERR);
326	if (unlikely(!(vma->vm_flags & access)))
327		return handle_fault_error(regs, SEGV_ACCERR);
328	fault = handle_mm_fault(vma, address, flags, regs);
329	if (fault_signal_pending(fault, regs)) {
330		if (!user_mode(regs))
331			handle_fault_error_nolock(regs, 0);
332		return;
333	}
334	/* The fault is fully completed (including releasing mmap lock) */
335	if (fault & VM_FAULT_COMPLETED)
336		return;
337	if (fault & VM_FAULT_RETRY) {
338		flags |= FAULT_FLAG_TRIED;
339		goto retry;
340	}
341	mmap_read_unlock(mm);
342done:
343	if (!(fault & VM_FAULT_ERROR))
344		return;
345	if (fault & VM_FAULT_OOM) {
346		if (!user_mode(regs))
347			handle_fault_error_nolock(regs, 0);
348		else
349			pagefault_out_of_memory();
350	} else if (fault & VM_FAULT_SIGSEGV) {
351		if (!user_mode(regs))
352			handle_fault_error_nolock(regs, 0);
353		else
354			do_sigsegv(regs, SEGV_MAPERR);
355	} else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
356			    VM_FAULT_HWPOISON_LARGE)) {
357		if (!user_mode(regs))
358			handle_fault_error_nolock(regs, 0);
359		else
360			do_sigbus(regs);
361	} else {
362		pr_emerg("Unexpected fault flags: %08x\n", fault);
363		BUG();
364	}
365}
366
367void do_protection_exception(struct pt_regs *regs)
368{
369	union teid teid = { .val = regs->int_parm_long };
370
371	/*
372	 * Protection exceptions are suppressing, decrement psw address.
373	 * The exception to this rule are aborted transactions, for these
374	 * the PSW already points to the correct location.
375	 */
376	if (!(regs->int_code & 0x200)) {
377		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
378		set_pt_regs_flag(regs, PIF_PSW_ADDR_ADJUSTED);
379	}
380	/*
381	 * If bit 61 if the TEID is not set, the remainder of the
382	 * TEID is unpredictable. Special handling is required.
383	 */
384	if (unlikely(!teid.b61)) {
385		if (user_mode(regs)) {
386			dump_fault_info(regs);
387			die(regs, "Unexpected TEID");
388		}
389		/* Assume low-address protection in kernel mode. */
390		return handle_fault_error_nolock(regs, 0);
391	}
392	if (unlikely(cpu_has_nx() && teid.b56)) {
393		regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
394		return handle_fault_error_nolock(regs, SEGV_ACCERR);
395	}
396	do_exception(regs, VM_WRITE);
397}
398NOKPROBE_SYMBOL(do_protection_exception);
399
400void do_dat_exception(struct pt_regs *regs)
401{
402	do_exception(regs, VM_ACCESS_FLAGS);
403}
404NOKPROBE_SYMBOL(do_dat_exception);
405
406#if IS_ENABLED(CONFIG_PGSTE)
407
408void do_secure_storage_access(struct pt_regs *regs)
409{
410	union teid teid = { .val = regs->int_parm_long };
411	unsigned long addr = get_fault_address(regs);
412	struct vm_area_struct *vma;
413	struct folio_walk fw;
414	struct mm_struct *mm;
415	struct folio *folio;
416	int rc;
417
418	/*
419	 * Bit 61 indicates if the address is valid, if it is not the
420	 * kernel should be stopped or SIGSEGV should be sent to the
421	 * process. Bit 61 is not reliable without the misc UV feature,
422	 * therefore this needs to be checked too.
423	 */
424	if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
425		/*
426		 * When this happens, userspace did something that it
427		 * was not supposed to do, e.g. branching into secure
428		 * memory. Trigger a segmentation fault.
429		 */
430		if (user_mode(regs)) {
431			send_sig(SIGSEGV, current, 0);
432			return;
433		}
434		/*
435		 * The kernel should never run into this case and
436		 * there is no way out of this situation.
437		 */
438		panic("Unexpected PGM 0x3d with TEID bit 61=0");
439	}
440	if (is_kernel_fault(regs)) {
441		folio = phys_to_folio(addr);
442		if (unlikely(!folio_try_get(folio)))
443			return;
444		rc = uv_convert_from_secure(folio_to_phys(folio));
445		if (!rc)
446			clear_bit(PG_arch_1, &folio->flags.f);
447		folio_put(folio);
448		/*
449		 * There are some valid fixup types for kernel
450		 * accesses to donated secure memory. zeropad is one
451		 * of them.
452		 */
453		if (rc)
454			return handle_fault_error_nolock(regs, 0);
455	} else {
456		if (faulthandler_disabled())
457			return handle_fault_error_nolock(regs, 0);
458		mm = current->mm;
459		mmap_read_lock(mm);
460		vma = find_vma(mm, addr);
461		if (!vma)
462			return handle_fault_error(regs, SEGV_MAPERR);
463		folio = folio_walk_start(&fw, vma, addr, 0);
464		if (!folio) {
465			mmap_read_unlock(mm);
466			return;
467		}
468		/* arch_make_folio_accessible() needs a raised refcount. */
469		folio_get(folio);
470		rc = arch_make_folio_accessible(folio);
471		folio_put(folio);
472		folio_walk_end(&fw, vma);
473		if (rc)
474			send_sig(SIGSEGV, current, 0);
475		mmap_read_unlock(mm);
476	}
477}
478NOKPROBE_SYMBOL(do_secure_storage_access);
479
480#endif /* CONFIG_PGSTE */