Commit | Line | Data |
---|---|---|
14cf11af | 1 | /* |
14cf11af PM |
2 | * PowerPC version |
3 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) | |
4 | * | |
5 | * Derived from "arch/i386/mm/fault.c" | |
6 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
7 | * | |
8 | * Modified by Cort Dougan and Paul Mackerras. | |
9 | * | |
10 | * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com) | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version | |
15 | * 2 of the License, or (at your option) any later version. | |
16 | */ | |
17 | ||
14cf11af PM |
18 | #include <linux/signal.h> |
19 | #include <linux/sched.h> | |
68db0cf1 | 20 | #include <linux/sched/task_stack.h> |
14cf11af PM |
21 | #include <linux/kernel.h> |
22 | #include <linux/errno.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/types.h> | |
0e36b0d1 | 25 | #include <linux/pagemap.h> |
14cf11af PM |
26 | #include <linux/ptrace.h> |
27 | #include <linux/mman.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/interrupt.h> | |
30 | #include <linux/highmem.h> | |
8a39b05f | 31 | #include <linux/extable.h> |
14cf11af | 32 | #include <linux/kprobes.h> |
1eeb66a1 | 33 | #include <linux/kdebug.h> |
cdd6c482 | 34 | #include <linux/perf_event.h> |
76462232 | 35 | #include <linux/ratelimit.h> |
ba12eede | 36 | #include <linux/context_tracking.h> |
9d57472f | 37 | #include <linux/hugetlb.h> |
70ffdb93 | 38 | #include <linux/uaccess.h> |
14cf11af | 39 | |
40900194 | 40 | #include <asm/firmware.h> |
14cf11af PM |
41 | #include <asm/page.h> |
42 | #include <asm/pgtable.h> | |
43 | #include <asm/mmu.h> | |
44 | #include <asm/mmu_context.h> | |
14cf11af | 45 | #include <asm/siginfo.h> |
ae3a197e | 46 | #include <asm/debug.h> |
4f9e87c0 | 47 | |
bb4be50e | 48 | static inline bool notify_page_fault(struct pt_regs *regs) |
4f9e87c0 | 49 | { |
bb4be50e | 50 | bool ret = false; |
9f90b997 | 51 | |
bb4be50e | 52 | #ifdef CONFIG_KPROBES |
9f90b997 CH |
53 | /* kprobe_running() needs smp_processor_id() */ |
54 | if (!user_mode(regs)) { | |
55 | preempt_disable(); | |
56 | if (kprobe_running() && kprobe_fault_handler(regs, 11)) | |
bb4be50e | 57 | ret = true; |
9f90b997 CH |
58 | preempt_enable(); |
59 | } | |
bb4be50e BH |
60 | #endif /* CONFIG_KPROBES */ |
61 | ||
62 | if (unlikely(debugger_fault_handler(regs))) | |
63 | ret = true; | |
4f9e87c0 | 64 | |
9f90b997 | 65 | return ret; |
4f9e87c0 | 66 | } |
4f9e87c0 | 67 | |
14cf11af | 68 | /* |
0e36b0d1 | 69 | * Check whether the instruction inst is a store using |
14cf11af PM |
70 | * an update addressing form which will update r1. |
71 | */ | |
0e36b0d1 | 72 | static bool store_updates_sp(unsigned int inst) |
14cf11af | 73 | { |
14cf11af PM |
74 | /* check for 1 in the rA field */ |
75 | if (((inst >> 16) & 0x1f) != 1) | |
8f5ca0b3 | 76 | return false; |
14cf11af PM |
77 | /* check major opcode */ |
78 | switch (inst >> 26) { | |
8a0b1120 CL |
79 | case OP_STWU: |
80 | case OP_STBU: | |
81 | case OP_STHU: | |
82 | case OP_STFSU: | |
83 | case OP_STFDU: | |
8f5ca0b3 | 84 | return true; |
8a0b1120 | 85 | case OP_STD: /* std or stdu */ |
14cf11af | 86 | return (inst & 3) == 1; |
8a0b1120 | 87 | case OP_31: |
14cf11af PM |
88 | /* check minor opcode */ |
89 | switch ((inst >> 1) & 0x3ff) { | |
8a0b1120 CL |
90 | case OP_31_XOP_STDUX: |
91 | case OP_31_XOP_STWUX: | |
92 | case OP_31_XOP_STBUX: | |
93 | case OP_31_XOP_STHUX: | |
94 | case OP_31_XOP_STFSUX: | |
95 | case OP_31_XOP_STFDUX: | |
8f5ca0b3 | 96 | return true; |
14cf11af PM |
97 | } |
98 | } | |
8f5ca0b3 | 99 | return false; |
14cf11af | 100 | } |
9be72573 BH |
101 | /* |
102 | * do_page_fault error handling helpers | |
103 | */ | |
104 | ||
c3350602 | 105 | static int |
cd60ab7a | 106 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code) |
c3350602 BH |
107 | { |
108 | /* | |
109 | * If we are in kernel mode, bail out with a SEGV, this will | |
110 | * be caught by the assembly which will restore the non-volatile | |
111 | * registers before calling bad_page_fault() | |
112 | */ | |
113 | if (!user_mode(regs)) | |
114 | return SIGSEGV; | |
115 | ||
cd60ab7a | 116 | _exception(SIGSEGV, regs, si_code, address); |
c3350602 BH |
117 | |
118 | return 0; | |
119 | } | |
120 | ||
121 | static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address) | |
122 | { | |
cd60ab7a | 123 | return __bad_area_nosemaphore(regs, address, SEGV_MAPERR); |
c3350602 BH |
124 | } |
125 | ||
9f2ee693 | 126 | static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code) |
c3350602 BH |
127 | { |
128 | struct mm_struct *mm = current->mm; | |
129 | ||
130 | /* | |
131 | * Something tried to access memory that isn't in our memory map.. | |
132 | * Fix it, but check if it's kernel or user first.. | |
133 | */ | |
134 | up_read(&mm->mmap_sem); | |
135 | ||
cd60ab7a | 136 | return __bad_area_nosemaphore(regs, address, si_code); |
c3350602 BH |
137 | } |
138 | ||
139 | static noinline int bad_area(struct pt_regs *regs, unsigned long address) | |
140 | { | |
9f2ee693 | 141 | return __bad_area(regs, address, SEGV_MAPERR); |
99cd1302 RP |
142 | } |
143 | ||
144 | static int bad_key_fault_exception(struct pt_regs *regs, unsigned long address, | |
145 | int pkey) | |
146 | { | |
8eb2ba25 EB |
147 | /* |
148 | * If we are in kernel mode, bail out with a SEGV, this will | |
149 | * be caught by the assembly which will restore the non-volatile | |
150 | * registers before calling bad_page_fault() | |
151 | */ | |
152 | if (!user_mode(regs)) | |
153 | return SIGSEGV; | |
154 | ||
5d8fb8a5 | 155 | _exception_pkey(regs, address, pkey); |
8eb2ba25 EB |
156 | |
157 | return 0; | |
c3350602 BH |
158 | } |
159 | ||
ecb101ae JS |
160 | static noinline int bad_access(struct pt_regs *regs, unsigned long address) |
161 | { | |
9f2ee693 | 162 | return __bad_area(regs, address, SEGV_ACCERR); |
ecb101ae JS |
163 | } |
164 | ||
3913fdd7 | 165 | static int do_sigbus(struct pt_regs *regs, unsigned long address, |
50a7ca3c | 166 | vm_fault_t fault) |
9be72573 | 167 | { |
63af5262 | 168 | if (!user_mode(regs)) |
b5c8f0fd | 169 | return SIGBUS; |
63af5262 AB |
170 | |
171 | current->thread.trap_nr = BUS_ADRERR; | |
3913fdd7 AB |
172 | #ifdef CONFIG_MEMORY_FAILURE |
173 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { | |
f654fc07 EB |
174 | unsigned int lsb = 0; /* shutup gcc */ |
175 | ||
3913fdd7 AB |
176 | pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", |
177 | current->comm, current->pid, address); | |
f654fc07 EB |
178 | |
179 | if (fault & VM_FAULT_HWPOISON_LARGE) | |
180 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
181 | if (fault & VM_FAULT_HWPOISON) | |
182 | lsb = PAGE_SHIFT; | |
183 | ||
184 | force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb, | |
185 | current); | |
186 | return 0; | |
3913fdd7 | 187 | } |
9d57472f | 188 | |
3913fdd7 | 189 | #endif |
f383d8b4 | 190 | force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, current); |
b5c8f0fd | 191 | return 0; |
9be72573 BH |
192 | } |
193 | ||
50a7ca3c SJ |
194 | static int mm_fault_error(struct pt_regs *regs, unsigned long addr, |
195 | vm_fault_t fault) | |
9be72573 BH |
196 | { |
197 | /* | |
b5c8f0fd BH |
198 | * Kernel page fault interrupted by SIGKILL. We have no reason to |
199 | * continue processing. | |
9be72573 | 200 | */ |
b5c8f0fd BH |
201 | if (fatal_signal_pending(current) && !user_mode(regs)) |
202 | return SIGKILL; | |
9be72573 BH |
203 | |
204 | /* Out of memory */ | |
c2d23f91 | 205 | if (fault & VM_FAULT_OOM) { |
c2d23f91 DR |
206 | /* |
207 | * We ran out of memory, or some other thing happened to us that | |
208 | * made us unable to handle the page fault gracefully. | |
209 | */ | |
210 | if (!user_mode(regs)) | |
b5c8f0fd | 211 | return SIGSEGV; |
c2d23f91 | 212 | pagefault_out_of_memory(); |
b5c8f0fd BH |
213 | } else { |
214 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| | |
215 | VM_FAULT_HWPOISON_LARGE)) | |
216 | return do_sigbus(regs, addr, fault); | |
217 | else if (fault & VM_FAULT_SIGSEGV) | |
218 | return bad_area_nosemaphore(regs, addr); | |
219 | else | |
220 | BUG(); | |
c2d23f91 | 221 | } |
b5c8f0fd | 222 | return 0; |
9be72573 | 223 | } |
14cf11af | 224 | |
d3ca5874 BH |
225 | /* Is this a bad kernel fault ? */ |
226 | static bool bad_kernel_fault(bool is_exec, unsigned long error_code, | |
227 | unsigned long address) | |
228 | { | |
ffca395b CL |
229 | /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */ |
230 | if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT | | |
231 | DSISR_PROTFAULT))) { | |
d3ca5874 BH |
232 | printk_ratelimited(KERN_CRIT "kernel tried to execute" |
233 | " exec-protected page (%lx) -" | |
234 | "exploit attempt? (uid: %d)\n", | |
235 | address, from_kuid(&init_user_ns, | |
236 | current_uid())); | |
237 | } | |
238 | return is_exec || (address >= TASK_SIZE); | |
239 | } | |
240 | ||
8f5ca0b3 | 241 | static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address, |
0e36b0d1 CL |
242 | struct vm_area_struct *vma, unsigned int flags, |
243 | bool *must_retry) | |
8f5ca0b3 BH |
244 | { |
245 | /* | |
246 | * N.B. The POWER/Open ABI allows programs to access up to | |
247 | * 288 bytes below the stack pointer. | |
248 | * The kernel signal delivery code writes up to about 1.5kB | |
249 | * below the stack pointer (r1) before decrementing it. | |
250 | * The exec code can write slightly over 640kB to the stack | |
251 | * before setting the user r1. Thus we allow the stack to | |
252 | * expand to 1MB without further checks. | |
253 | */ | |
254 | if (address + 0x100000 < vma->vm_end) { | |
0e36b0d1 | 255 | unsigned int __user *nip = (unsigned int __user *)regs->nip; |
8f5ca0b3 BH |
256 | /* get user regs even if this fault is in kernel mode */ |
257 | struct pt_regs *uregs = current->thread.regs; | |
258 | if (uregs == NULL) | |
259 | return true; | |
260 | ||
261 | /* | |
262 | * A user-mode access to an address a long way below | |
263 | * the stack pointer is only valid if the instruction | |
264 | * is one which would update the stack pointer to the | |
265 | * address accessed if the instruction completed, | |
266 | * i.e. either stwu rs,n(r1) or stwux rs,r1,rb | |
267 | * (or the byte, halfword, float or double forms). | |
268 | * | |
269 | * If we don't check this then any write to the area | |
270 | * between the last mapped region and the stack will | |
271 | * expand the stack rather than segfaulting. | |
272 | */ | |
0e36b0d1 CL |
273 | if (address + 2048 >= uregs->gpr[1]) |
274 | return false; | |
275 | ||
276 | if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) && | |
96d4f267 | 277 | access_ok(nip, sizeof(*nip))) { |
0e36b0d1 CL |
278 | unsigned int inst; |
279 | int res; | |
280 | ||
281 | pagefault_disable(); | |
282 | res = __get_user_inatomic(inst, nip); | |
283 | pagefault_enable(); | |
284 | if (!res) | |
285 | return !store_updates_sp(inst); | |
286 | *must_retry = true; | |
287 | } | |
288 | return true; | |
8f5ca0b3 BH |
289 | } |
290 | return false; | |
291 | } | |
292 | ||
bd0d63f8 BH |
293 | static bool access_error(bool is_write, bool is_exec, |
294 | struct vm_area_struct *vma) | |
295 | { | |
296 | /* | |
297 | * Allow execution from readable areas if the MMU does not | |
298 | * provide separate controls over reading and executing. | |
299 | * | |
300 | * Note: That code used to not be enabled for 4xx/BookE. | |
301 | * It is now as I/D cache coherency for these is done at | |
302 | * set_pte_at() time and I see no reason why the test | |
303 | * below wouldn't be valid on those processors. This -may- | |
304 | * break programs compiled with a really old ABI though. | |
305 | */ | |
306 | if (is_exec) { | |
307 | return !(vma->vm_flags & VM_EXEC) && | |
308 | (cpu_has_feature(CPU_FTR_NOEXECUTE) || | |
309 | !(vma->vm_flags & (VM_READ | VM_WRITE))); | |
310 | } | |
311 | ||
312 | if (is_write) { | |
313 | if (unlikely(!(vma->vm_flags & VM_WRITE))) | |
314 | return true; | |
315 | return false; | |
316 | } | |
317 | ||
318 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
319 | return true; | |
f2ed480f AK |
320 | /* |
321 | * We should ideally do the vma pkey access check here. But in the | |
322 | * fault path, handle_mm_fault() also does the same check. To avoid | |
323 | * these multiple checks, we skip it here and handle access error due | |
324 | * to pkeys later. | |
325 | */ | |
bd0d63f8 BH |
326 | return false; |
327 | } | |
328 | ||
3da02648 BH |
329 | #ifdef CONFIG_PPC_SMLPAR |
330 | static inline void cmo_account_page_fault(void) | |
331 | { | |
332 | if (firmware_has_feature(FW_FEATURE_CMO)) { | |
333 | u32 page_ins; | |
334 | ||
335 | preempt_disable(); | |
336 | page_ins = be32_to_cpu(get_lppaca()->page_ins); | |
337 | page_ins += 1 << PAGE_FACTOR; | |
338 | get_lppaca()->page_ins = cpu_to_be32(page_ins); | |
339 | preempt_enable(); | |
340 | } | |
341 | } | |
342 | #else | |
343 | static inline void cmo_account_page_fault(void) { } | |
344 | #endif /* CONFIG_PPC_SMLPAR */ | |
345 | ||
5b3e84fc | 346 | #ifdef CONFIG_PPC_BOOK3S |
374f3f59 AK |
347 | static void sanity_check_fault(bool is_write, bool is_user, |
348 | unsigned long error_code, unsigned long address) | |
2865d08d | 349 | { |
374f3f59 AK |
350 | /* |
351 | * Userspace trying to access kernel address, we get PROTFAULT for that. | |
352 | */ | |
353 | if (is_user && address >= TASK_SIZE) { | |
354 | pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n", | |
355 | current->comm, current->pid, address, | |
356 | from_kuid(&init_user_ns, current_uid())); | |
357 | return; | |
358 | } | |
359 | ||
2865d08d BH |
360 | /* |
361 | * For hash translation mode, we should never get a | |
362 | * PROTFAULT. Any update to pte to reduce access will result in us | |
363 | * removing the hash page table entry, thus resulting in a DSISR_NOHPTE | |
364 | * fault instead of DSISR_PROTFAULT. | |
365 | * | |
366 | * A pte update to relax the access will not result in a hash page table | |
367 | * entry invalidate and hence can result in DSISR_PROTFAULT. | |
368 | * ptep_set_access_flags() doesn't do a hpte flush. This is why we have | |
369 | * the special !is_write in the below conditional. | |
370 | * | |
371 | * For platforms that doesn't supports coherent icache and do support | |
372 | * per page noexec bit, we do setup things such that we do the | |
373 | * sync between D/I cache via fault. But that is handled via low level | |
374 | * hash fault code (hash_page_do_lazy_icache()) and we should not reach | |
375 | * here in such case. | |
376 | * | |
377 | * For wrong access that can result in PROTFAULT, the above vma->vm_flags | |
378 | * check should handle those and hence we should fall to the bad_area | |
379 | * handling correctly. | |
380 | * | |
381 | * For embedded with per page exec support that doesn't support coherent | |
382 | * icache we do get PROTFAULT and we handle that D/I cache sync in | |
383 | * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON | |
384 | * is conditional for server MMU. | |
385 | * | |
386 | * For radix, we can get prot fault for autonuma case, because radix | |
387 | * page table will have them marked noaccess for user. | |
388 | */ | |
374f3f59 AK |
389 | if (radix_enabled() || is_write) |
390 | return; | |
391 | ||
392 | WARN_ON_ONCE(error_code & DSISR_PROTFAULT); | |
2865d08d BH |
393 | } |
394 | #else | |
374f3f59 AK |
395 | static void sanity_check_fault(bool is_write, bool is_user, |
396 | unsigned long error_code, unsigned long address) { } | |
5b3e84fc | 397 | #endif /* CONFIG_PPC_BOOK3S */ |
2865d08d | 398 | |
41b464e5 BH |
399 | /* |
400 | * Define the correct "is_write" bit in error_code based | |
401 | * on the processor family | |
402 | */ | |
403 | #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) | |
404 | #define page_fault_is_write(__err) ((__err) & ESR_DST) | |
f3d96e69 | 405 | #define page_fault_is_bad(__err) (0) |
41b464e5 BH |
406 | #else |
407 | #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE) | |
968159c0 | 408 | #if defined(CONFIG_PPC_8xx) |
4915349b | 409 | #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G) |
f3d96e69 BH |
410 | #elif defined(CONFIG_PPC64) |
411 | #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S) | |
412 | #else | |
413 | #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S) | |
414 | #endif | |
41b464e5 BH |
415 | #endif |
416 | ||
14cf11af PM |
417 | /* |
418 | * For 600- and 800-family processors, the error_code parameter is DSISR | |
419 | * for a data fault, SRR1 for an instruction fault. For 400-family processors | |
420 | * the error_code parameter is ESR for a data fault, 0 for an instruction | |
421 | * fault. | |
422 | * For 64-bit processors, the error_code parameter is | |
423 | * - DSISR for a non-SLB data access fault, | |
424 | * - SRR1 & 0x08000000 for a non-SLB instruction access fault | |
425 | * - 0 any SLB fault. | |
426 | * | |
427 | * The return value is 0 if the fault was handled, or the signal | |
428 | * number if this is a kernel fault that can't be handled here. | |
429 | */ | |
7afad422 BH |
430 | static int __do_page_fault(struct pt_regs *regs, unsigned long address, |
431 | unsigned long error_code) | |
14cf11af PM |
432 | { |
433 | struct vm_area_struct * vma; | |
434 | struct mm_struct *mm = current->mm; | |
9be72573 | 435 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
c433ec04 | 436 | int is_exec = TRAP(regs) == 0x400; |
da929f6a | 437 | int is_user = user_mode(regs); |
41b464e5 | 438 | int is_write = page_fault_is_write(error_code); |
50a7ca3c | 439 | vm_fault_t fault, major = 0; |
0e36b0d1 | 440 | bool must_retry = false; |
14cf11af | 441 | |
9f90b997 | 442 | if (notify_page_fault(regs)) |
65d47fd4 | 443 | return 0; |
14cf11af | 444 | |
f3d96e69 | 445 | if (unlikely(page_fault_is_bad(error_code))) { |
65d47fd4 | 446 | if (is_user) { |
f3d96e69 | 447 | _exception(SIGBUS, regs, BUS_OBJERR, address); |
65d47fd4 BH |
448 | return 0; |
449 | } | |
450 | return SIGBUS; | |
e6c8290a | 451 | } |
e6c8290a | 452 | |
2865d08d | 453 | /* Additional sanity check(s) */ |
374f3f59 | 454 | sanity_check_fault(is_write, is_user, error_code, address); |
2865d08d | 455 | |
d7df2443 BH |
456 | /* |
457 | * The kernel should never take an execute fault nor should it | |
458 | * take a page fault to a kernel address. | |
459 | */ | |
d3ca5874 | 460 | if (unlikely(!is_user && bad_kernel_fault(is_exec, error_code, address))) |
65d47fd4 | 461 | return SIGSEGV; |
14cf11af | 462 | |
11ccdd33 BH |
463 | /* |
464 | * If we're in an interrupt, have no user context or are running | |
465 | * in a region with pagefaults disabled then we must not take the fault | |
466 | */ | |
467 | if (unlikely(faulthandler_disabled() || !mm)) { | |
468 | if (is_user) | |
469 | printk_ratelimited(KERN_ERR "Page fault in user mode" | |
470 | " with faulthandler_disabled()=%d" | |
471 | " mm=%p\n", | |
472 | faulthandler_disabled(), mm); | |
473 | return bad_area_nosemaphore(regs, address); | |
474 | } | |
475 | ||
a546498f BH |
476 | /* We restore the interrupt state now */ |
477 | if (!arch_irq_disabled_regs(regs)) | |
478 | local_irq_enable(); | |
479 | ||
a8b0ca17 | 480 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
7dd1fcc2 | 481 | |
99cd1302 RP |
482 | if (error_code & DSISR_KEYFAULT) |
483 | return bad_key_fault_exception(regs, address, | |
484 | get_mm_addr_key(mm, address)); | |
e6c2a479 | 485 | |
69e044dd AK |
486 | /* |
487 | * We want to do this outside mmap_sem, because reading code around nip | |
488 | * can result in fault, which will cause a deadlock when called with | |
489 | * mmap_sem held | |
490 | */ | |
da929f6a | 491 | if (is_user) |
759496ba | 492 | flags |= FAULT_FLAG_USER; |
d2e0d2c5 BH |
493 | if (is_write) |
494 | flags |= FAULT_FLAG_WRITE; | |
495 | if (is_exec) | |
496 | flags |= FAULT_FLAG_INSTRUCTION; | |
759496ba | 497 | |
14cf11af PM |
498 | /* When running in the kernel we expect faults to occur only to |
499 | * addresses in user space. All other faults represent errors in the | |
fc5266ea AB |
500 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
501 | * erroneous fault occurring in a code path which already holds mmap_sem | |
14cf11af PM |
502 | * we will deadlock attempting to validate the fault against the |
503 | * address space. Luckily the kernel only validly references user | |
504 | * space from well defined areas of code, which are listed in the | |
505 | * exceptions table. | |
506 | * | |
507 | * As the vast majority of faults will be valid we will only perform | |
fc5266ea | 508 | * the source reference check when there is a possibility of a deadlock. |
14cf11af PM |
509 | * Attempt to lock the address space, if we cannot we then validate the |
510 | * source. If this is invalid we can skip the address space check, | |
511 | * thus avoiding the deadlock. | |
512 | */ | |
b15021d9 | 513 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
da929f6a | 514 | if (!is_user && !search_exception_tables(regs->nip)) |
c3350602 | 515 | return bad_area_nosemaphore(regs, address); |
14cf11af | 516 | |
9be72573 | 517 | retry: |
14cf11af | 518 | down_read(&mm->mmap_sem); |
a546498f BH |
519 | } else { |
520 | /* | |
521 | * The above down_read_trylock() might have succeeded in | |
522 | * which case we'll have missed the might_sleep() from | |
523 | * down_read(): | |
524 | */ | |
525 | might_sleep(); | |
14cf11af PM |
526 | } |
527 | ||
528 | vma = find_vma(mm, address); | |
b15021d9 | 529 | if (unlikely(!vma)) |
c3350602 | 530 | return bad_area(regs, address); |
b15021d9 | 531 | if (likely(vma->vm_start <= address)) |
14cf11af | 532 | goto good_area; |
b15021d9 | 533 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) |
c3350602 | 534 | return bad_area(regs, address); |
14cf11af | 535 | |
8f5ca0b3 | 536 | /* The stack is being expanded, check if it's valid */ |
0e36b0d1 CL |
537 | if (unlikely(bad_stack_expansion(regs, address, vma, flags, |
538 | &must_retry))) { | |
539 | if (!must_retry) | |
540 | return bad_area(regs, address); | |
541 | ||
542 | up_read(&mm->mmap_sem); | |
543 | if (fault_in_pages_readable((const char __user *)regs->nip, | |
544 | sizeof(unsigned int))) | |
545 | return bad_area_nosemaphore(regs, address); | |
546 | goto retry; | |
547 | } | |
14cf11af | 548 | |
8f5ca0b3 | 549 | /* Try to expand it */ |
b15021d9 | 550 | if (unlikely(expand_stack(vma, address))) |
c3350602 | 551 | return bad_area(regs, address); |
14cf11af PM |
552 | |
553 | good_area: | |
bd0d63f8 | 554 | if (unlikely(access_error(is_write, is_exec, vma))) |
ecb101ae | 555 | return bad_access(regs, address); |
14cf11af PM |
556 | |
557 | /* | |
558 | * If for any reason at all we couldn't handle the fault, | |
559 | * make sure we exit gracefully rather than endlessly redo | |
560 | * the fault. | |
561 | */ | |
dcddffd4 | 562 | fault = handle_mm_fault(vma, address, flags); |
e6c2a479 RP |
563 | |
564 | #ifdef CONFIG_PPC_MEM_KEYS | |
565 | /* | |
f2ed480f AK |
566 | * we skipped checking for access error due to key earlier. |
567 | * Check that using handle_mm_fault error return. | |
e6c2a479 RP |
568 | */ |
569 | if (unlikely(fault & VM_FAULT_SIGSEGV) && | |
f2ed480f AK |
570 | !arch_vma_access_permitted(vma, is_write, is_exec, 0)) { |
571 | ||
e6c2a479 RP |
572 | int pkey = vma_pkey(vma); |
573 | ||
f2ed480f AK |
574 | up_read(&mm->mmap_sem); |
575 | return bad_key_fault_exception(regs, address, pkey); | |
e6c2a479 RP |
576 | } |
577 | #endif /* CONFIG_PPC_MEM_KEYS */ | |
578 | ||
f43bb27e | 579 | major |= fault & VM_FAULT_MAJOR; |
14c02e41 LD |
580 | |
581 | /* | |
582 | * Handle the retry right now, the mmap_sem has been released in that | |
583 | * case. | |
584 | */ | |
585 | if (unlikely(fault & VM_FAULT_RETRY)) { | |
586 | /* We retry only once */ | |
587 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
588 | /* | |
589 | * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
590 | * of starvation. | |
591 | */ | |
592 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
593 | flags |= FAULT_FLAG_TRIED; | |
594 | if (!fatal_signal_pending(current)) | |
595 | goto retry; | |
596 | } | |
14c02e41 | 597 | |
b5c8f0fd BH |
598 | /* |
599 | * User mode? Just return to handle the fatal exception otherwise | |
600 | * return to bad_page_fault | |
601 | */ | |
602 | return is_user ? 0 : SIGBUS; | |
14cf11af | 603 | } |
9be72573 | 604 | |
b5c8f0fd BH |
605 | up_read(¤t->mm->mmap_sem); |
606 | ||
607 | if (unlikely(fault & VM_FAULT_ERROR)) | |
608 | return mm_fault_error(regs, address, fault); | |
609 | ||
9be72573 | 610 | /* |
14c02e41 | 611 | * Major/minor page fault accounting. |
9be72573 | 612 | */ |
f43bb27e | 613 | if (major) { |
14c02e41 | 614 | current->maj_flt++; |
04aafdc6 | 615 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
3da02648 | 616 | cmo_account_page_fault(); |
14c02e41 LD |
617 | } else { |
618 | current->min_flt++; | |
04aafdc6 | 619 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
ac17dc8e | 620 | } |
c3350602 | 621 | return 0; |
7afad422 BH |
622 | } |
623 | NOKPROBE_SYMBOL(__do_page_fault); | |
624 | ||
625 | int do_page_fault(struct pt_regs *regs, unsigned long address, | |
626 | unsigned long error_code) | |
627 | { | |
628 | enum ctx_state prev_state = exception_enter(); | |
629 | int rc = __do_page_fault(regs, address, error_code); | |
ba12eede LZ |
630 | exception_exit(prev_state); |
631 | return rc; | |
14cf11af | 632 | } |
03465f89 | 633 | NOKPROBE_SYMBOL(do_page_fault); |
14cf11af PM |
634 | |
635 | /* | |
636 | * bad_page_fault is called when we have a bad access from the kernel. | |
637 | * It is called from the DSI and ISI handlers in head.S and from some | |
638 | * of the procedures in traps.c. | |
639 | */ | |
640 | void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) | |
641 | { | |
642 | const struct exception_table_entry *entry; | |
643 | ||
644 | /* Are we prepared to handle this fault? */ | |
645 | if ((entry = search_exception_tables(regs->nip)) != NULL) { | |
61a92f70 | 646 | regs->nip = extable_fixup(entry); |
14cf11af PM |
647 | return; |
648 | } | |
649 | ||
650 | /* kernel has accessed a bad area */ | |
723925b7 | 651 | |
2271db20 | 652 | switch (TRAP(regs)) { |
a416dd8d ME |
653 | case 0x300: |
654 | case 0x380: | |
d7b45615 | 655 | case 0xe00: |
49a502ea CL |
656 | pr_alert("BUG: %s at 0x%08lx\n", |
657 | regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" : | |
658 | "Unable to handle kernel data access", regs->dar); | |
a416dd8d ME |
659 | break; |
660 | case 0x400: | |
661 | case 0x480: | |
49a502ea CL |
662 | pr_alert("BUG: Unable to handle kernel instruction fetch%s", |
663 | regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n"); | |
a416dd8d | 664 | break; |
eab861a7 | 665 | case 0x600: |
49a502ea CL |
666 | pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n", |
667 | regs->dar); | |
eab861a7 | 668 | break; |
a416dd8d | 669 | default: |
49a502ea CL |
670 | pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n", |
671 | regs->dar); | |
a416dd8d | 672 | break; |
723925b7 OJ |
673 | } |
674 | printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", | |
675 | regs->nip); | |
676 | ||
a70857e4 | 677 | if (task_stack_end_corrupted(current)) |
28b54990 AB |
678 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); |
679 | ||
14cf11af PM |
680 | die("Kernel access of bad area", regs, sig); |
681 | } |