Loading...
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/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>
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
49static DEFINE_STATIC_KEY_FALSE(have_store_indication);
50
51static int __init fault_init(void)
52{
53 if (test_facility(75))
54 static_branch_enable(&have_store_indication);
55 return 0;
56}
57early_initcall(fault_init);
58
59/*
60 * Find out which address space caused the exception.
61 */
62static bool is_kernel_fault(struct pt_regs *regs)
63{
64 union teid teid = { .val = regs->int_parm_long };
65
66 if (user_mode(regs))
67 return false;
68 if (teid.as == PSW_BITS_AS_SECONDARY)
69 return false;
70 return true;
71}
72
73static unsigned long get_fault_address(struct pt_regs *regs)
74{
75 union teid teid = { .val = regs->int_parm_long };
76
77 return teid.addr * PAGE_SIZE;
78}
79
80static __always_inline bool fault_is_write(struct pt_regs *regs)
81{
82 union teid teid = { .val = regs->int_parm_long };
83
84 if (static_branch_likely(&have_store_indication))
85 return teid.fsi == TEID_FSI_STORE;
86 return false;
87}
88
89static void dump_pagetable(unsigned long asce, unsigned long address)
90{
91 unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
92
93 pr_alert("AS:%016lx ", asce);
94 switch (asce & _ASCE_TYPE_MASK) {
95 case _ASCE_TYPE_REGION1:
96 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
97 if (get_kernel_nofault(entry, table))
98 goto bad;
99 pr_cont("R1:%016lx ", entry);
100 if (entry & _REGION_ENTRY_INVALID)
101 goto out;
102 table = __va(entry & _REGION_ENTRY_ORIGIN);
103 fallthrough;
104 case _ASCE_TYPE_REGION2:
105 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
106 if (get_kernel_nofault(entry, table))
107 goto bad;
108 pr_cont("R2:%016lx ", entry);
109 if (entry & _REGION_ENTRY_INVALID)
110 goto out;
111 table = __va(entry & _REGION_ENTRY_ORIGIN);
112 fallthrough;
113 case _ASCE_TYPE_REGION3:
114 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
115 if (get_kernel_nofault(entry, table))
116 goto bad;
117 pr_cont("R3:%016lx ", entry);
118 if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
119 goto out;
120 table = __va(entry & _REGION_ENTRY_ORIGIN);
121 fallthrough;
122 case _ASCE_TYPE_SEGMENT:
123 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
124 if (get_kernel_nofault(entry, table))
125 goto bad;
126 pr_cont("S:%016lx ", entry);
127 if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
128 goto out;
129 table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
130 }
131 table += (address & _PAGE_INDEX) >> PAGE_SHIFT;
132 if (get_kernel_nofault(entry, table))
133 goto bad;
134 pr_cont("P:%016lx ", entry);
135out:
136 pr_cont("\n");
137 return;
138bad:
139 pr_cont("BAD\n");
140}
141
142static void dump_fault_info(struct pt_regs *regs)
143{
144 union teid teid = { .val = regs->int_parm_long };
145 unsigned long asce;
146
147 pr_alert("Failing address: %016lx TEID: %016lx\n",
148 get_fault_address(regs), teid.val);
149 pr_alert("Fault in ");
150 switch (teid.as) {
151 case PSW_BITS_AS_HOME:
152 pr_cont("home space ");
153 break;
154 case PSW_BITS_AS_SECONDARY:
155 pr_cont("secondary space ");
156 break;
157 case PSW_BITS_AS_ACCREG:
158 pr_cont("access register ");
159 break;
160 case PSW_BITS_AS_PRIMARY:
161 pr_cont("primary space ");
162 break;
163 }
164 pr_cont("mode while using ");
165 if (is_kernel_fault(regs)) {
166 asce = get_lowcore()->kernel_asce.val;
167 pr_cont("kernel ");
168 } else {
169 asce = get_lowcore()->user_asce.val;
170 pr_cont("user ");
171 }
172 pr_cont("ASCE.\n");
173 dump_pagetable(asce, get_fault_address(regs));
174}
175
176int show_unhandled_signals = 1;
177
178void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
179{
180 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
181
182 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
183 return;
184 if (!unhandled_signal(current, signr))
185 return;
186 if (!__ratelimit(&rs))
187 return;
188 pr_alert("User process fault: interruption code %04x ilc:%d ",
189 regs->int_code & 0xffff, regs->int_code >> 17);
190 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
191 pr_cont("\n");
192 if (is_mm_fault)
193 dump_fault_info(regs);
194 show_regs(regs);
195}
196
197static void do_sigsegv(struct pt_regs *regs, int si_code)
198{
199 report_user_fault(regs, SIGSEGV, 1);
200 force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
201}
202
203static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
204{
205 unsigned long address;
206 bool is_write;
207
208 if (user_mode(regs)) {
209 if (WARN_ON_ONCE(!si_code))
210 si_code = SEGV_MAPERR;
211 return do_sigsegv(regs, si_code);
212 }
213 if (fixup_exception(regs))
214 return;
215 if (is_kernel_fault(regs)) {
216 address = get_fault_address(regs);
217 is_write = fault_is_write(regs);
218 if (kfence_handle_page_fault(address, is_write, regs))
219 return;
220 pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
221 } else {
222 pr_alert("Unable to handle kernel paging request in virtual user address space\n");
223 }
224 dump_fault_info(regs);
225 die(regs, "Oops");
226}
227
228static void handle_fault_error(struct pt_regs *regs, int si_code)
229{
230 struct mm_struct *mm = current->mm;
231
232 mmap_read_unlock(mm);
233 handle_fault_error_nolock(regs, si_code);
234}
235
236static void do_sigbus(struct pt_regs *regs)
237{
238 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
239}
240
241/*
242 * This routine handles page faults. It determines the address,
243 * and the problem, and then passes it off to one of the appropriate
244 * routines.
245 *
246 * interruption code (int_code):
247 * 04 Protection -> Write-Protection (suppression)
248 * 10 Segment translation -> Not present (nullification)
249 * 11 Page translation -> Not present (nullification)
250 * 3b Region third trans. -> Not present (nullification)
251 */
252static void do_exception(struct pt_regs *regs, int access)
253{
254 struct vm_area_struct *vma;
255 unsigned long address;
256 struct mm_struct *mm;
257 unsigned int flags;
258 vm_fault_t fault;
259 bool is_write;
260
261 /*
262 * The instruction that caused the program check has
263 * been nullified. Don't signal single step via SIGTRAP.
264 */
265 clear_thread_flag(TIF_PER_TRAP);
266 if (kprobe_page_fault(regs, 14))
267 return;
268 mm = current->mm;
269 address = get_fault_address(regs);
270 is_write = fault_is_write(regs);
271 if (is_kernel_fault(regs) || faulthandler_disabled() || !mm)
272 return handle_fault_error_nolock(regs, 0);
273 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
274 flags = FAULT_FLAG_DEFAULT;
275 if (user_mode(regs))
276 flags |= FAULT_FLAG_USER;
277 if (is_write)
278 access = VM_WRITE;
279 if (access == VM_WRITE)
280 flags |= FAULT_FLAG_WRITE;
281 if (!(flags & FAULT_FLAG_USER))
282 goto lock_mmap;
283 vma = lock_vma_under_rcu(mm, address);
284 if (!vma)
285 goto lock_mmap;
286 if (!(vma->vm_flags & access)) {
287 vma_end_read(vma);
288 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
289 return handle_fault_error_nolock(regs, SEGV_ACCERR);
290 }
291 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
292 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
293 vma_end_read(vma);
294 if (!(fault & VM_FAULT_RETRY)) {
295 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
296 goto done;
297 }
298 count_vm_vma_lock_event(VMA_LOCK_RETRY);
299 if (fault & VM_FAULT_MAJOR)
300 flags |= FAULT_FLAG_TRIED;
301 /* Quick path to respond to signals */
302 if (fault_signal_pending(fault, regs)) {
303 if (!user_mode(regs))
304 handle_fault_error_nolock(regs, 0);
305 return;
306 }
307lock_mmap:
308retry:
309 vma = lock_mm_and_find_vma(mm, address, regs);
310 if (!vma)
311 return handle_fault_error_nolock(regs, SEGV_MAPERR);
312 if (unlikely(!(vma->vm_flags & access)))
313 return handle_fault_error(regs, SEGV_ACCERR);
314 fault = handle_mm_fault(vma, address, flags, regs);
315 if (fault_signal_pending(fault, regs)) {
316 if (!user_mode(regs))
317 handle_fault_error_nolock(regs, 0);
318 return;
319 }
320 /* The fault is fully completed (including releasing mmap lock) */
321 if (fault & VM_FAULT_COMPLETED)
322 return;
323 if (fault & VM_FAULT_RETRY) {
324 flags |= FAULT_FLAG_TRIED;
325 goto retry;
326 }
327 mmap_read_unlock(mm);
328done:
329 if (!(fault & VM_FAULT_ERROR))
330 return;
331 if (fault & VM_FAULT_OOM) {
332 if (!user_mode(regs))
333 handle_fault_error_nolock(regs, 0);
334 else
335 pagefault_out_of_memory();
336 } else if (fault & VM_FAULT_SIGSEGV) {
337 if (!user_mode(regs))
338 handle_fault_error_nolock(regs, 0);
339 else
340 do_sigsegv(regs, SEGV_MAPERR);
341 } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
342 VM_FAULT_HWPOISON_LARGE)) {
343 if (!user_mode(regs))
344 handle_fault_error_nolock(regs, 0);
345 else
346 do_sigbus(regs);
347 } else {
348 pr_emerg("Unexpected fault flags: %08x\n", fault);
349 BUG();
350 }
351}
352
353void do_protection_exception(struct pt_regs *regs)
354{
355 union teid teid = { .val = regs->int_parm_long };
356
357 /*
358 * Protection exceptions are suppressing, decrement psw address.
359 * The exception to this rule are aborted transactions, for these
360 * the PSW already points to the correct location.
361 */
362 if (!(regs->int_code & 0x200))
363 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
364 /*
365 * Check for low-address protection. This needs to be treated
366 * as a special case because the translation exception code
367 * field is not guaranteed to contain valid data in this case.
368 */
369 if (unlikely(!teid.b61)) {
370 if (user_mode(regs)) {
371 /* Low-address protection in user mode: cannot happen */
372 die(regs, "Low-address protection");
373 }
374 /*
375 * Low-address protection in kernel mode means
376 * NULL pointer write access in kernel mode.
377 */
378 return handle_fault_error_nolock(regs, 0);
379 }
380 if (unlikely(MACHINE_HAS_NX && teid.b56)) {
381 regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
382 return handle_fault_error_nolock(regs, SEGV_ACCERR);
383 }
384 do_exception(regs, VM_WRITE);
385}
386NOKPROBE_SYMBOL(do_protection_exception);
387
388void do_dat_exception(struct pt_regs *regs)
389{
390 do_exception(regs, VM_ACCESS_FLAGS);
391}
392NOKPROBE_SYMBOL(do_dat_exception);
393
394#if IS_ENABLED(CONFIG_PGSTE)
395
396void do_secure_storage_access(struct pt_regs *regs)
397{
398 union teid teid = { .val = regs->int_parm_long };
399 unsigned long addr = get_fault_address(regs);
400 struct vm_area_struct *vma;
401 struct folio_walk fw;
402 struct mm_struct *mm;
403 struct folio *folio;
404 int rc;
405
406 /*
407 * Bit 61 indicates if the address is valid, if it is not the
408 * kernel should be stopped or SIGSEGV should be sent to the
409 * process. Bit 61 is not reliable without the misc UV feature,
410 * therefore this needs to be checked too.
411 */
412 if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
413 /*
414 * When this happens, userspace did something that it
415 * was not supposed to do, e.g. branching into secure
416 * memory. Trigger a segmentation fault.
417 */
418 if (user_mode(regs)) {
419 send_sig(SIGSEGV, current, 0);
420 return;
421 }
422 /*
423 * The kernel should never run into this case and
424 * there is no way out of this situation.
425 */
426 panic("Unexpected PGM 0x3d with TEID bit 61=0");
427 }
428 if (is_kernel_fault(regs)) {
429 folio = phys_to_folio(addr);
430 if (unlikely(!folio_try_get(folio)))
431 return;
432 rc = arch_make_folio_accessible(folio);
433 folio_put(folio);
434 if (rc)
435 BUG();
436 } else {
437 mm = current->mm;
438 mmap_read_lock(mm);
439 vma = find_vma(mm, addr);
440 if (!vma)
441 return handle_fault_error(regs, SEGV_MAPERR);
442 folio = folio_walk_start(&fw, vma, addr, 0);
443 if (!folio) {
444 mmap_read_unlock(mm);
445 return;
446 }
447 /* arch_make_folio_accessible() needs a raised refcount. */
448 folio_get(folio);
449 rc = arch_make_folio_accessible(folio);
450 folio_put(folio);
451 folio_walk_end(&fw, vma);
452 if (rc)
453 send_sig(SIGSEGV, current, 0);
454 mmap_read_unlock(mm);
455 }
456}
457NOKPROBE_SYMBOL(do_secure_storage_access);
458
459#endif /* CONFIG_PGSTE */
1/*
2 * arch/s390/mm/fault.c
3 *
4 * S390 version
5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Hartmut Penner (hp@de.ibm.com)
7 * Ulrich Weigand (uweigand@de.ibm.com)
8 *
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1995 Linus Torvalds
11 */
12
13#include <linux/kernel_stat.h>
14#include <linux/perf_event.h>
15#include <linux/signal.h>
16#include <linux/sched.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>
23#include <linux/mm.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/module.h>
30#include <linux/hardirq.h>
31#include <linux/kprobes.h>
32#include <linux/uaccess.h>
33#include <linux/hugetlb.h>
34#include <asm/asm-offsets.h>
35#include <asm/pgtable.h>
36#include <asm/irq.h>
37#include <asm/mmu_context.h>
38#include <asm/facility.h>
39#include "../kernel/entry.h"
40
41#ifndef CONFIG_64BIT
42#define __FAIL_ADDR_MASK 0x7ffff000
43#define __SUBCODE_MASK 0x0200
44#define __PF_RES_FIELD 0ULL
45#else /* CONFIG_64BIT */
46#define __FAIL_ADDR_MASK -4096L
47#define __SUBCODE_MASK 0x0600
48#define __PF_RES_FIELD 0x8000000000000000ULL
49#endif /* CONFIG_64BIT */
50
51#define VM_FAULT_BADCONTEXT 0x010000
52#define VM_FAULT_BADMAP 0x020000
53#define VM_FAULT_BADACCESS 0x040000
54
55static unsigned long store_indication;
56
57void fault_init(void)
58{
59 if (test_facility(2) && test_facility(75))
60 store_indication = 0xc00;
61}
62
63static inline int notify_page_fault(struct pt_regs *regs)
64{
65 int ret = 0;
66
67 /* kprobe_running() needs smp_processor_id() */
68 if (kprobes_built_in() && !user_mode(regs)) {
69 preempt_disable();
70 if (kprobe_running() && kprobe_fault_handler(regs, 14))
71 ret = 1;
72 preempt_enable();
73 }
74 return ret;
75}
76
77
78/*
79 * Unlock any spinlocks which will prevent us from getting the
80 * message out.
81 */
82void bust_spinlocks(int yes)
83{
84 if (yes) {
85 oops_in_progress = 1;
86 } else {
87 int loglevel_save = console_loglevel;
88 console_unblank();
89 oops_in_progress = 0;
90 /*
91 * OK, the message is on the console. Now we call printk()
92 * without oops_in_progress set so that printk will give klogd
93 * a poke. Hold onto your hats...
94 */
95 console_loglevel = 15;
96 printk(" ");
97 console_loglevel = loglevel_save;
98 }
99}
100
101/*
102 * Returns the address space associated with the fault.
103 * Returns 0 for kernel space and 1 for user space.
104 */
105static inline int user_space_fault(unsigned long trans_exc_code)
106{
107 /*
108 * The lowest two bits of the translation exception
109 * identification indicate which paging table was used.
110 */
111 trans_exc_code &= 3;
112 if (trans_exc_code == 2)
113 /* Access via secondary space, set_fs setting decides */
114 return current->thread.mm_segment.ar4;
115 if (user_mode == HOME_SPACE_MODE)
116 /* User space if the access has been done via home space. */
117 return trans_exc_code == 3;
118 /*
119 * If the user space is not the home space the kernel runs in home
120 * space. Access via secondary space has already been covered,
121 * access via primary space or access register is from user space
122 * and access via home space is from the kernel.
123 */
124 return trans_exc_code != 3;
125}
126
127static inline void report_user_fault(struct pt_regs *regs, long signr)
128{
129 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
130 return;
131 if (!unhandled_signal(current, signr))
132 return;
133 if (!printk_ratelimit())
134 return;
135 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
136 regs->int_code);
137 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
138 printk(KERN_CONT "\n");
139 printk(KERN_ALERT "failing address: %lX\n",
140 regs->int_parm_long & __FAIL_ADDR_MASK);
141 show_regs(regs);
142}
143
144/*
145 * Send SIGSEGV to task. This is an external routine
146 * to keep the stack usage of do_page_fault small.
147 */
148static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
149{
150 struct siginfo si;
151
152 report_user_fault(regs, SIGSEGV);
153 si.si_signo = SIGSEGV;
154 si.si_code = si_code;
155 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
156 force_sig_info(SIGSEGV, &si, current);
157}
158
159static noinline void do_no_context(struct pt_regs *regs)
160{
161 const struct exception_table_entry *fixup;
162 unsigned long address;
163
164 /* Are we prepared to handle this kernel fault? */
165 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
166 if (fixup) {
167 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
168 return;
169 }
170
171 /*
172 * Oops. The kernel tried to access some bad page. We'll have to
173 * terminate things with extreme prejudice.
174 */
175 address = regs->int_parm_long & __FAIL_ADDR_MASK;
176 if (!user_space_fault(regs->int_parm_long))
177 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
178 " at virtual kernel address %p\n", (void *)address);
179 else
180 printk(KERN_ALERT "Unable to handle kernel paging request"
181 " at virtual user address %p\n", (void *)address);
182
183 die(regs, "Oops");
184 do_exit(SIGKILL);
185}
186
187static noinline void do_low_address(struct pt_regs *regs)
188{
189 /* Low-address protection hit in kernel mode means
190 NULL pointer write access in kernel mode. */
191 if (regs->psw.mask & PSW_MASK_PSTATE) {
192 /* Low-address protection hit in user mode 'cannot happen'. */
193 die (regs, "Low-address protection");
194 do_exit(SIGKILL);
195 }
196
197 do_no_context(regs);
198}
199
200static noinline void do_sigbus(struct pt_regs *regs)
201{
202 struct task_struct *tsk = current;
203 struct siginfo si;
204
205 /*
206 * Send a sigbus, regardless of whether we were in kernel
207 * or user mode.
208 */
209 si.si_signo = SIGBUS;
210 si.si_errno = 0;
211 si.si_code = BUS_ADRERR;
212 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
213 force_sig_info(SIGBUS, &si, tsk);
214}
215
216static noinline void do_fault_error(struct pt_regs *regs, int fault)
217{
218 int si_code;
219
220 switch (fault) {
221 case VM_FAULT_BADACCESS:
222 case VM_FAULT_BADMAP:
223 /* Bad memory access. Check if it is kernel or user space. */
224 if (regs->psw.mask & PSW_MASK_PSTATE) {
225 /* User mode accesses just cause a SIGSEGV */
226 si_code = (fault == VM_FAULT_BADMAP) ?
227 SEGV_MAPERR : SEGV_ACCERR;
228 do_sigsegv(regs, si_code);
229 return;
230 }
231 case VM_FAULT_BADCONTEXT:
232 do_no_context(regs);
233 break;
234 default: /* fault & VM_FAULT_ERROR */
235 if (fault & VM_FAULT_OOM) {
236 if (!(regs->psw.mask & PSW_MASK_PSTATE))
237 do_no_context(regs);
238 else
239 pagefault_out_of_memory();
240 } else if (fault & VM_FAULT_SIGBUS) {
241 /* Kernel mode? Handle exceptions or die */
242 if (!(regs->psw.mask & PSW_MASK_PSTATE))
243 do_no_context(regs);
244 else
245 do_sigbus(regs);
246 } else
247 BUG();
248 break;
249 }
250}
251
252/*
253 * This routine handles page faults. It determines the address,
254 * and the problem, and then passes it off to one of the appropriate
255 * routines.
256 *
257 * interruption code (int_code):
258 * 04 Protection -> Write-Protection (suprression)
259 * 10 Segment translation -> Not present (nullification)
260 * 11 Page translation -> Not present (nullification)
261 * 3b Region third trans. -> Not present (nullification)
262 */
263static inline int do_exception(struct pt_regs *regs, int access)
264{
265 struct task_struct *tsk;
266 struct mm_struct *mm;
267 struct vm_area_struct *vma;
268 unsigned long trans_exc_code;
269 unsigned long address;
270 unsigned int flags;
271 int fault;
272
273 if (notify_page_fault(regs))
274 return 0;
275
276 tsk = current;
277 mm = tsk->mm;
278 trans_exc_code = regs->int_parm_long;
279
280 /*
281 * Verify that the fault happened in user space, that
282 * we are not in an interrupt and that there is a
283 * user context.
284 */
285 fault = VM_FAULT_BADCONTEXT;
286 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
287 goto out;
288
289 address = trans_exc_code & __FAIL_ADDR_MASK;
290 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
291 flags = FAULT_FLAG_ALLOW_RETRY;
292 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
293 flags |= FAULT_FLAG_WRITE;
294 down_read(&mm->mmap_sem);
295
296#ifdef CONFIG_PGSTE
297 if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
298 address = __gmap_fault(address,
299 (struct gmap *) S390_lowcore.gmap);
300 if (address == -EFAULT) {
301 fault = VM_FAULT_BADMAP;
302 goto out_up;
303 }
304 if (address == -ENOMEM) {
305 fault = VM_FAULT_OOM;
306 goto out_up;
307 }
308 }
309#endif
310
311retry:
312 fault = VM_FAULT_BADMAP;
313 vma = find_vma(mm, address);
314 if (!vma)
315 goto out_up;
316
317 if (unlikely(vma->vm_start > address)) {
318 if (!(vma->vm_flags & VM_GROWSDOWN))
319 goto out_up;
320 if (expand_stack(vma, address))
321 goto out_up;
322 }
323
324 /*
325 * Ok, we have a good vm_area for this memory access, so
326 * we can handle it..
327 */
328 fault = VM_FAULT_BADACCESS;
329 if (unlikely(!(vma->vm_flags & access)))
330 goto out_up;
331
332 if (is_vm_hugetlb_page(vma))
333 address &= HPAGE_MASK;
334 /*
335 * If for any reason at all we couldn't handle the fault,
336 * make sure we exit gracefully rather than endlessly redo
337 * the fault.
338 */
339 fault = handle_mm_fault(mm, vma, address, flags);
340 if (unlikely(fault & VM_FAULT_ERROR))
341 goto out_up;
342
343 /*
344 * Major/minor page fault accounting is only done on the
345 * initial attempt. If we go through a retry, it is extremely
346 * likely that the page will be found in page cache at that point.
347 */
348 if (flags & FAULT_FLAG_ALLOW_RETRY) {
349 if (fault & VM_FAULT_MAJOR) {
350 tsk->maj_flt++;
351 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
352 regs, address);
353 } else {
354 tsk->min_flt++;
355 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
356 regs, address);
357 }
358 if (fault & VM_FAULT_RETRY) {
359 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
360 * of starvation. */
361 flags &= ~FAULT_FLAG_ALLOW_RETRY;
362 down_read(&mm->mmap_sem);
363 goto retry;
364 }
365 }
366 /*
367 * The instruction that caused the program check will
368 * be repeated. Don't signal single step via SIGTRAP.
369 */
370 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
371 fault = 0;
372out_up:
373 up_read(&mm->mmap_sem);
374out:
375 return fault;
376}
377
378void __kprobes do_protection_exception(struct pt_regs *regs)
379{
380 unsigned long trans_exc_code;
381 int fault;
382
383 trans_exc_code = regs->int_parm_long;
384 /* Protection exception is suppressing, decrement psw address. */
385 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
386 /*
387 * Check for low-address protection. This needs to be treated
388 * as a special case because the translation exception code
389 * field is not guaranteed to contain valid data in this case.
390 */
391 if (unlikely(!(trans_exc_code & 4))) {
392 do_low_address(regs);
393 return;
394 }
395 fault = do_exception(regs, VM_WRITE);
396 if (unlikely(fault))
397 do_fault_error(regs, fault);
398}
399
400void __kprobes do_dat_exception(struct pt_regs *regs)
401{
402 int access, fault;
403
404 access = VM_READ | VM_EXEC | VM_WRITE;
405 fault = do_exception(regs, access);
406 if (unlikely(fault))
407 do_fault_error(regs, fault);
408}
409
410#ifdef CONFIG_64BIT
411void __kprobes do_asce_exception(struct pt_regs *regs)
412{
413 struct mm_struct *mm = current->mm;
414 struct vm_area_struct *vma;
415 unsigned long trans_exc_code;
416
417 trans_exc_code = regs->int_parm_long;
418 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
419 goto no_context;
420
421 down_read(&mm->mmap_sem);
422 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
423 up_read(&mm->mmap_sem);
424
425 if (vma) {
426 update_mm(mm, current);
427 return;
428 }
429
430 /* User mode accesses just cause a SIGSEGV */
431 if (regs->psw.mask & PSW_MASK_PSTATE) {
432 do_sigsegv(regs, SEGV_MAPERR);
433 return;
434 }
435
436no_context:
437 do_no_context(regs);
438}
439#endif
440
441int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
442{
443 struct pt_regs regs;
444 int access, fault;
445
446 /* Emulate a uaccess fault from kernel mode. */
447 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
448 if (!irqs_disabled())
449 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
450 regs.psw.addr = (unsigned long) __builtin_return_address(0);
451 regs.psw.addr |= PSW_ADDR_AMODE;
452 regs.int_code = pgm_int_code;
453 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
454 access = write ? VM_WRITE : VM_READ;
455 fault = do_exception(®s, access);
456 /*
457 * Since the fault happened in kernel mode while performing a uaccess
458 * all we need to do now is emulating a fixup in case "fault" is not
459 * zero.
460 * For the calling uaccess functions this results always in -EFAULT.
461 */
462 return fault ? -EFAULT : 0;
463}
464
465#ifdef CONFIG_PFAULT
466/*
467 * 'pfault' pseudo page faults routines.
468 */
469static int pfault_disable;
470
471static int __init nopfault(char *str)
472{
473 pfault_disable = 1;
474 return 1;
475}
476
477__setup("nopfault", nopfault);
478
479struct pfault_refbk {
480 u16 refdiagc;
481 u16 reffcode;
482 u16 refdwlen;
483 u16 refversn;
484 u64 refgaddr;
485 u64 refselmk;
486 u64 refcmpmk;
487 u64 reserved;
488} __attribute__ ((packed, aligned(8)));
489
490int pfault_init(void)
491{
492 struct pfault_refbk refbk = {
493 .refdiagc = 0x258,
494 .reffcode = 0,
495 .refdwlen = 5,
496 .refversn = 2,
497 .refgaddr = __LC_CURRENT_PID,
498 .refselmk = 1ULL << 48,
499 .refcmpmk = 1ULL << 48,
500 .reserved = __PF_RES_FIELD };
501 int rc;
502
503 if (pfault_disable)
504 return -1;
505 asm volatile(
506 " diag %1,%0,0x258\n"
507 "0: j 2f\n"
508 "1: la %0,8\n"
509 "2:\n"
510 EX_TABLE(0b,1b)
511 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
512 return rc;
513}
514
515void pfault_fini(void)
516{
517 struct pfault_refbk refbk = {
518 .refdiagc = 0x258,
519 .reffcode = 1,
520 .refdwlen = 5,
521 .refversn = 2,
522 };
523
524 if (pfault_disable)
525 return;
526 asm volatile(
527 " diag %0,0,0x258\n"
528 "0:\n"
529 EX_TABLE(0b,0b)
530 : : "a" (&refbk), "m" (refbk) : "cc");
531}
532
533static DEFINE_SPINLOCK(pfault_lock);
534static LIST_HEAD(pfault_list);
535
536static void pfault_interrupt(struct ext_code ext_code,
537 unsigned int param32, unsigned long param64)
538{
539 struct task_struct *tsk;
540 __u16 subcode;
541 pid_t pid;
542
543 /*
544 * Get the external interruption subcode & pfault
545 * initial/completion signal bit. VM stores this
546 * in the 'cpu address' field associated with the
547 * external interrupt.
548 */
549 subcode = ext_code.subcode;
550 if ((subcode & 0xff00) != __SUBCODE_MASK)
551 return;
552 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
553 /* Get the token (= pid of the affected task). */
554 pid = sizeof(void *) == 4 ? param32 : param64;
555 rcu_read_lock();
556 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
557 if (tsk)
558 get_task_struct(tsk);
559 rcu_read_unlock();
560 if (!tsk)
561 return;
562 spin_lock(&pfault_lock);
563 if (subcode & 0x0080) {
564 /* signal bit is set -> a page has been swapped in by VM */
565 if (tsk->thread.pfault_wait == 1) {
566 /* Initial interrupt was faster than the completion
567 * interrupt. pfault_wait is valid. Set pfault_wait
568 * back to zero and wake up the process. This can
569 * safely be done because the task is still sleeping
570 * and can't produce new pfaults. */
571 tsk->thread.pfault_wait = 0;
572 list_del(&tsk->thread.list);
573 wake_up_process(tsk);
574 put_task_struct(tsk);
575 } else {
576 /* Completion interrupt was faster than initial
577 * interrupt. Set pfault_wait to -1 so the initial
578 * interrupt doesn't put the task to sleep.
579 * If the task is not running, ignore the completion
580 * interrupt since it must be a leftover of a PFAULT
581 * CANCEL operation which didn't remove all pending
582 * completion interrupts. */
583 if (tsk->state == TASK_RUNNING)
584 tsk->thread.pfault_wait = -1;
585 }
586 } else {
587 /* signal bit not set -> a real page is missing. */
588 if (WARN_ON_ONCE(tsk != current))
589 goto out;
590 if (tsk->thread.pfault_wait == 1) {
591 /* Already on the list with a reference: put to sleep */
592 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
593 set_tsk_need_resched(tsk);
594 } else if (tsk->thread.pfault_wait == -1) {
595 /* Completion interrupt was faster than the initial
596 * interrupt (pfault_wait == -1). Set pfault_wait
597 * back to zero and exit. */
598 tsk->thread.pfault_wait = 0;
599 } else {
600 /* Initial interrupt arrived before completion
601 * interrupt. Let the task sleep.
602 * An extra task reference is needed since a different
603 * cpu may set the task state to TASK_RUNNING again
604 * before the scheduler is reached. */
605 get_task_struct(tsk);
606 tsk->thread.pfault_wait = 1;
607 list_add(&tsk->thread.list, &pfault_list);
608 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
609 set_tsk_need_resched(tsk);
610 }
611 }
612out:
613 spin_unlock(&pfault_lock);
614 put_task_struct(tsk);
615}
616
617static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
618 unsigned long action, void *hcpu)
619{
620 struct thread_struct *thread, *next;
621 struct task_struct *tsk;
622
623 switch (action) {
624 case CPU_DEAD:
625 case CPU_DEAD_FROZEN:
626 spin_lock_irq(&pfault_lock);
627 list_for_each_entry_safe(thread, next, &pfault_list, list) {
628 thread->pfault_wait = 0;
629 list_del(&thread->list);
630 tsk = container_of(thread, struct task_struct, thread);
631 wake_up_process(tsk);
632 put_task_struct(tsk);
633 }
634 spin_unlock_irq(&pfault_lock);
635 break;
636 default:
637 break;
638 }
639 return NOTIFY_OK;
640}
641
642static int __init pfault_irq_init(void)
643{
644 int rc;
645
646 rc = register_external_interrupt(0x2603, pfault_interrupt);
647 if (rc)
648 goto out_extint;
649 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
650 if (rc)
651 goto out_pfault;
652 service_subclass_irq_register();
653 hotcpu_notifier(pfault_cpu_notify, 0);
654 return 0;
655
656out_pfault:
657 unregister_external_interrupt(0x2603, pfault_interrupt);
658out_extint:
659 pfault_disable = 1;
660 return rc;
661}
662early_initcall(pfault_irq_init);
663
664#endif /* CONFIG_PFAULT */