]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
14cf11af | 2 | /* |
14cf11af PM |
3 | * PowerPC version |
4 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) | |
5 | * | |
6 | * Derived from "arch/i386/mm/fault.c" | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * | |
9 | * Modified by Cort Dougan and Paul Mackerras. | |
10 | * | |
11 | * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com) | |
14cf11af PM |
12 | */ |
13 | ||
14cf11af PM |
14 | #include <linux/signal.h> |
15 | #include <linux/sched.h> | |
68db0cf1 | 16 | #include <linux/sched/task_stack.h> |
14cf11af PM |
17 | #include <linux/kernel.h> |
18 | #include <linux/errno.h> | |
19 | #include <linux/string.h> | |
20 | #include <linux/types.h> | |
0e36b0d1 | 21 | #include <linux/pagemap.h> |
14cf11af PM |
22 | #include <linux/ptrace.h> |
23 | #include <linux/mman.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/interrupt.h> | |
26 | #include <linux/highmem.h> | |
8a39b05f | 27 | #include <linux/extable.h> |
14cf11af | 28 | #include <linux/kprobes.h> |
1eeb66a1 | 29 | #include <linux/kdebug.h> |
cdd6c482 | 30 | #include <linux/perf_event.h> |
76462232 | 31 | #include <linux/ratelimit.h> |
ba12eede | 32 | #include <linux/context_tracking.h> |
9d57472f | 33 | #include <linux/hugetlb.h> |
70ffdb93 | 34 | #include <linux/uaccess.h> |
14cf11af | 35 | |
40900194 | 36 | #include <asm/firmware.h> |
14cf11af | 37 | #include <asm/page.h> |
14cf11af PM |
38 | #include <asm/mmu.h> |
39 | #include <asm/mmu_context.h> | |
14cf11af | 40 | #include <asm/siginfo.h> |
ae3a197e | 41 | #include <asm/debug.h> |
5e5be3ae | 42 | #include <asm/kup.h> |
8094892d | 43 | #include <asm/inst.h> |
4f9e87c0 | 44 | |
14cf11af | 45 | /* |
0e36b0d1 | 46 | * Check whether the instruction inst is a store using |
14cf11af PM |
47 | * an update addressing form which will update r1. |
48 | */ | |
94afd069 | 49 | static bool store_updates_sp(struct ppc_inst inst) |
14cf11af | 50 | { |
14cf11af | 51 | /* check for 1 in the rA field */ |
777e26f0 | 52 | if (((ppc_inst_val(inst) >> 16) & 0x1f) != 1) |
8f5ca0b3 | 53 | return false; |
14cf11af | 54 | /* check major opcode */ |
8094892d | 55 | switch (ppc_inst_primary_opcode(inst)) { |
8a0b1120 CL |
56 | case OP_STWU: |
57 | case OP_STBU: | |
58 | case OP_STHU: | |
59 | case OP_STFSU: | |
60 | case OP_STFDU: | |
8f5ca0b3 | 61 | return true; |
8a0b1120 | 62 | case OP_STD: /* std or stdu */ |
777e26f0 | 63 | return (ppc_inst_val(inst) & 3) == 1; |
8a0b1120 | 64 | case OP_31: |
14cf11af | 65 | /* check minor opcode */ |
777e26f0 | 66 | switch ((ppc_inst_val(inst) >> 1) & 0x3ff) { |
8a0b1120 CL |
67 | case OP_31_XOP_STDUX: |
68 | case OP_31_XOP_STWUX: | |
69 | case OP_31_XOP_STBUX: | |
70 | case OP_31_XOP_STHUX: | |
71 | case OP_31_XOP_STFSUX: | |
72 | case OP_31_XOP_STFDUX: | |
8f5ca0b3 | 73 | return true; |
14cf11af PM |
74 | } |
75 | } | |
8f5ca0b3 | 76 | return false; |
14cf11af | 77 | } |
9be72573 BH |
78 | /* |
79 | * do_page_fault error handling helpers | |
80 | */ | |
81 | ||
c3350602 | 82 | static int |
cd60ab7a | 83 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code) |
c3350602 BH |
84 | { |
85 | /* | |
86 | * If we are in kernel mode, bail out with a SEGV, this will | |
87 | * be caught by the assembly which will restore the non-volatile | |
88 | * registers before calling bad_page_fault() | |
89 | */ | |
90 | if (!user_mode(regs)) | |
91 | return SIGSEGV; | |
92 | ||
cd60ab7a | 93 | _exception(SIGSEGV, regs, si_code, address); |
c3350602 BH |
94 | |
95 | return 0; | |
96 | } | |
97 | ||
98 | static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address) | |
99 | { | |
cd60ab7a | 100 | return __bad_area_nosemaphore(regs, address, SEGV_MAPERR); |
c3350602 BH |
101 | } |
102 | ||
9f2ee693 | 103 | static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code) |
c3350602 BH |
104 | { |
105 | struct mm_struct *mm = current->mm; | |
106 | ||
107 | /* | |
108 | * Something tried to access memory that isn't in our memory map.. | |
109 | * Fix it, but check if it's kernel or user first.. | |
110 | */ | |
d8ed45c5 | 111 | mmap_read_unlock(mm); |
c3350602 | 112 | |
cd60ab7a | 113 | return __bad_area_nosemaphore(regs, address, si_code); |
c3350602 BH |
114 | } |
115 | ||
116 | static noinline int bad_area(struct pt_regs *regs, unsigned long address) | |
117 | { | |
9f2ee693 | 118 | return __bad_area(regs, address, SEGV_MAPERR); |
99cd1302 RP |
119 | } |
120 | ||
fe4a6856 AK |
121 | #ifdef CONFIG_PPC_MEM_KEYS |
122 | static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address, | |
123 | struct vm_area_struct *vma) | |
99cd1302 | 124 | { |
fe4a6856 AK |
125 | struct mm_struct *mm = current->mm; |
126 | int pkey; | |
127 | ||
128 | /* | |
129 | * We don't try to fetch the pkey from page table because reading | |
130 | * page table without locking doesn't guarantee stable pte value. | |
131 | * Hence the pkey value that we return to userspace can be different | |
132 | * from the pkey that actually caused access error. | |
133 | * | |
134 | * It does *not* guarantee that the VMA we find here | |
135 | * was the one that we faulted on. | |
136 | * | |
137 | * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); | |
138 | * 2. T1 : set AMR to deny access to pkey=4, touches, page | |
139 | * 3. T1 : faults... | |
140 | * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); | |
c1e8d7c6 | 141 | * 5. T1 : enters fault handler, takes mmap_lock, etc... |
fe4a6856 AK |
142 | * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really |
143 | * faulted on a pte with its pkey=4. | |
144 | */ | |
145 | pkey = vma_pkey(vma); | |
146 | ||
d8ed45c5 | 147 | mmap_read_unlock(mm); |
fe4a6856 | 148 | |
8eb2ba25 EB |
149 | /* |
150 | * If we are in kernel mode, bail out with a SEGV, this will | |
151 | * be caught by the assembly which will restore the non-volatile | |
152 | * registers before calling bad_page_fault() | |
153 | */ | |
154 | if (!user_mode(regs)) | |
155 | return SIGSEGV; | |
156 | ||
5d8fb8a5 | 157 | _exception_pkey(regs, address, pkey); |
8eb2ba25 EB |
158 | |
159 | return 0; | |
c3350602 | 160 | } |
fe4a6856 | 161 | #endif |
c3350602 | 162 | |
ecb101ae JS |
163 | static noinline int bad_access(struct pt_regs *regs, unsigned long address) |
164 | { | |
9f2ee693 | 165 | return __bad_area(regs, address, SEGV_ACCERR); |
ecb101ae JS |
166 | } |
167 | ||
3913fdd7 | 168 | static int do_sigbus(struct pt_regs *regs, unsigned long address, |
50a7ca3c | 169 | vm_fault_t fault) |
9be72573 | 170 | { |
63af5262 | 171 | if (!user_mode(regs)) |
b5c8f0fd | 172 | return SIGBUS; |
63af5262 AB |
173 | |
174 | current->thread.trap_nr = BUS_ADRERR; | |
3913fdd7 AB |
175 | #ifdef CONFIG_MEMORY_FAILURE |
176 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { | |
f654fc07 EB |
177 | unsigned int lsb = 0; /* shutup gcc */ |
178 | ||
3913fdd7 AB |
179 | pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", |
180 | current->comm, current->pid, address); | |
f654fc07 EB |
181 | |
182 | if (fault & VM_FAULT_HWPOISON_LARGE) | |
183 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
184 | if (fault & VM_FAULT_HWPOISON) | |
185 | lsb = PAGE_SHIFT; | |
186 | ||
f8eac901 | 187 | force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); |
f654fc07 | 188 | return 0; |
3913fdd7 | 189 | } |
9d57472f | 190 | |
3913fdd7 | 191 | #endif |
2e1661d2 | 192 | force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); |
b5c8f0fd | 193 | return 0; |
9be72573 BH |
194 | } |
195 | ||
50a7ca3c SJ |
196 | static int mm_fault_error(struct pt_regs *regs, unsigned long addr, |
197 | vm_fault_t fault) | |
9be72573 BH |
198 | { |
199 | /* | |
b5c8f0fd BH |
200 | * Kernel page fault interrupted by SIGKILL. We have no reason to |
201 | * continue processing. | |
9be72573 | 202 | */ |
b5c8f0fd BH |
203 | if (fatal_signal_pending(current) && !user_mode(regs)) |
204 | return SIGKILL; | |
9be72573 BH |
205 | |
206 | /* Out of memory */ | |
c2d23f91 | 207 | if (fault & VM_FAULT_OOM) { |
c2d23f91 DR |
208 | /* |
209 | * We ran out of memory, or some other thing happened to us that | |
210 | * made us unable to handle the page fault gracefully. | |
211 | */ | |
212 | if (!user_mode(regs)) | |
b5c8f0fd | 213 | return SIGSEGV; |
c2d23f91 | 214 | pagefault_out_of_memory(); |
b5c8f0fd BH |
215 | } else { |
216 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| | |
217 | VM_FAULT_HWPOISON_LARGE)) | |
218 | return do_sigbus(regs, addr, fault); | |
219 | else if (fault & VM_FAULT_SIGSEGV) | |
220 | return bad_area_nosemaphore(regs, addr); | |
221 | else | |
222 | BUG(); | |
c2d23f91 | 223 | } |
b5c8f0fd | 224 | return 0; |
9be72573 | 225 | } |
14cf11af | 226 | |
d3ca5874 | 227 | /* Is this a bad kernel fault ? */ |
de78a9c4 | 228 | static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code, |
5e5be3ae | 229 | unsigned long address, bool is_write) |
d3ca5874 | 230 | { |
de78a9c4 CL |
231 | int is_exec = TRAP(regs) == 0x400; |
232 | ||
ffca395b CL |
233 | /* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */ |
234 | if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT | | |
235 | DSISR_PROTFAULT))) { | |
0fb1c25a CL |
236 | pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n", |
237 | address >= TASK_SIZE ? "exec-protected" : "user", | |
238 | address, | |
239 | from_kuid(&init_user_ns, current_uid())); | |
5e5be3ae ME |
240 | |
241 | // Kernel exec fault is always bad | |
242 | return true; | |
d3ca5874 | 243 | } |
de78a9c4 CL |
244 | |
245 | if (!is_exec && address < TASK_SIZE && (error_code & DSISR_PROTFAULT) && | |
246 | !search_exception_tables(regs->nip)) { | |
247 | pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n", | |
248 | address, | |
249 | from_kuid(&init_user_ns, current_uid())); | |
250 | } | |
251 | ||
5e5be3ae ME |
252 | // Kernel fault on kernel address is bad |
253 | if (address >= TASK_SIZE) | |
254 | return true; | |
255 | ||
256 | // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad | |
257 | if (!search_exception_tables(regs->nip)) | |
258 | return true; | |
259 | ||
260 | // Read/write fault in a valid region (the exception table search passed | |
261 | // above), but blocked by KUAP is bad, it can never succeed. | |
6ec20aa2 | 262 | if (bad_kuap_fault(regs, address, is_write)) |
5e5be3ae ME |
263 | return true; |
264 | ||
265 | // What's left? Kernel fault on user in well defined regions (extable | |
266 | // matched), and allowed by KUAP in the faulting context. | |
267 | return false; | |
d3ca5874 BH |
268 | } |
269 | ||
8f5ca0b3 | 270 | static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address, |
0e36b0d1 CL |
271 | struct vm_area_struct *vma, unsigned int flags, |
272 | bool *must_retry) | |
8f5ca0b3 BH |
273 | { |
274 | /* | |
275 | * N.B. The POWER/Open ABI allows programs to access up to | |
276 | * 288 bytes below the stack pointer. | |
277 | * The kernel signal delivery code writes up to about 1.5kB | |
278 | * below the stack pointer (r1) before decrementing it. | |
279 | * The exec code can write slightly over 640kB to the stack | |
280 | * before setting the user r1. Thus we allow the stack to | |
281 | * expand to 1MB without further checks. | |
282 | */ | |
283 | if (address + 0x100000 < vma->vm_end) { | |
7ba68b21 | 284 | struct ppc_inst __user *nip = (struct ppc_inst __user *)regs->nip; |
8f5ca0b3 BH |
285 | /* get user regs even if this fault is in kernel mode */ |
286 | struct pt_regs *uregs = current->thread.regs; | |
287 | if (uregs == NULL) | |
288 | return true; | |
289 | ||
290 | /* | |
291 | * A user-mode access to an address a long way below | |
292 | * the stack pointer is only valid if the instruction | |
293 | * is one which would update the stack pointer to the | |
294 | * address accessed if the instruction completed, | |
295 | * i.e. either stwu rs,n(r1) or stwux rs,r1,rb | |
296 | * (or the byte, halfword, float or double forms). | |
297 | * | |
298 | * If we don't check this then any write to the area | |
299 | * between the last mapped region and the stack will | |
300 | * expand the stack rather than segfaulting. | |
301 | */ | |
0e36b0d1 CL |
302 | if (address + 2048 >= uregs->gpr[1]) |
303 | return false; | |
304 | ||
305 | if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) && | |
96d4f267 | 306 | access_ok(nip, sizeof(*nip))) { |
94afd069 | 307 | struct ppc_inst inst; |
0e36b0d1 | 308 | |
7ba68b21 | 309 | if (!probe_user_read_inst(&inst, nip)) |
0e36b0d1 CL |
310 | return !store_updates_sp(inst); |
311 | *must_retry = true; | |
312 | } | |
313 | return true; | |
8f5ca0b3 BH |
314 | } |
315 | return false; | |
316 | } | |
317 | ||
fe4a6856 AK |
318 | #ifdef CONFIG_PPC_MEM_KEYS |
319 | static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey, | |
320 | struct vm_area_struct *vma) | |
321 | { | |
fe4a6856 AK |
322 | /* |
323 | * Make sure to check the VMA so that we do not perform | |
324 | * faults just to hit a pkey fault as soon as we fill in a | |
325 | * page. Only called for current mm, hence foreign == 0 | |
326 | */ | |
327 | if (!arch_vma_access_permitted(vma, is_write, is_exec, 0)) | |
328 | return true; | |
329 | ||
330 | return false; | |
331 | } | |
332 | #endif | |
333 | ||
334 | static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma) | |
bd0d63f8 BH |
335 | { |
336 | /* | |
337 | * Allow execution from readable areas if the MMU does not | |
338 | * provide separate controls over reading and executing. | |
339 | * | |
340 | * Note: That code used to not be enabled for 4xx/BookE. | |
341 | * It is now as I/D cache coherency for these is done at | |
342 | * set_pte_at() time and I see no reason why the test | |
343 | * below wouldn't be valid on those processors. This -may- | |
344 | * break programs compiled with a really old ABI though. | |
345 | */ | |
346 | if (is_exec) { | |
347 | return !(vma->vm_flags & VM_EXEC) && | |
348 | (cpu_has_feature(CPU_FTR_NOEXECUTE) || | |
349 | !(vma->vm_flags & (VM_READ | VM_WRITE))); | |
350 | } | |
351 | ||
352 | if (is_write) { | |
353 | if (unlikely(!(vma->vm_flags & VM_WRITE))) | |
354 | return true; | |
355 | return false; | |
356 | } | |
357 | ||
3122e80e | 358 | if (unlikely(!vma_is_accessible(vma))) |
bd0d63f8 | 359 | return true; |
f2ed480f AK |
360 | /* |
361 | * We should ideally do the vma pkey access check here. But in the | |
362 | * fault path, handle_mm_fault() also does the same check. To avoid | |
363 | * these multiple checks, we skip it here and handle access error due | |
364 | * to pkeys later. | |
365 | */ | |
bd0d63f8 BH |
366 | return false; |
367 | } | |
368 | ||
3da02648 BH |
369 | #ifdef CONFIG_PPC_SMLPAR |
370 | static inline void cmo_account_page_fault(void) | |
371 | { | |
372 | if (firmware_has_feature(FW_FEATURE_CMO)) { | |
373 | u32 page_ins; | |
374 | ||
375 | preempt_disable(); | |
376 | page_ins = be32_to_cpu(get_lppaca()->page_ins); | |
377 | page_ins += 1 << PAGE_FACTOR; | |
378 | get_lppaca()->page_ins = cpu_to_be32(page_ins); | |
379 | preempt_enable(); | |
380 | } | |
381 | } | |
382 | #else | |
383 | static inline void cmo_account_page_fault(void) { } | |
384 | #endif /* CONFIG_PPC_SMLPAR */ | |
385 | ||
5b3e84fc | 386 | #ifdef CONFIG_PPC_BOOK3S |
374f3f59 AK |
387 | static void sanity_check_fault(bool is_write, bool is_user, |
388 | unsigned long error_code, unsigned long address) | |
2865d08d | 389 | { |
374f3f59 AK |
390 | /* |
391 | * Userspace trying to access kernel address, we get PROTFAULT for that. | |
392 | */ | |
393 | if (is_user && address >= TASK_SIZE) { | |
0f9aee0c CL |
394 | if ((long)address == -1) |
395 | return; | |
396 | ||
374f3f59 AK |
397 | pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n", |
398 | current->comm, current->pid, address, | |
399 | from_kuid(&init_user_ns, current_uid())); | |
400 | return; | |
401 | } | |
402 | ||
2865d08d BH |
403 | /* |
404 | * For hash translation mode, we should never get a | |
405 | * PROTFAULT. Any update to pte to reduce access will result in us | |
406 | * removing the hash page table entry, thus resulting in a DSISR_NOHPTE | |
407 | * fault instead of DSISR_PROTFAULT. | |
408 | * | |
409 | * A pte update to relax the access will not result in a hash page table | |
410 | * entry invalidate and hence can result in DSISR_PROTFAULT. | |
411 | * ptep_set_access_flags() doesn't do a hpte flush. This is why we have | |
412 | * the special !is_write in the below conditional. | |
413 | * | |
414 | * For platforms that doesn't supports coherent icache and do support | |
415 | * per page noexec bit, we do setup things such that we do the | |
416 | * sync between D/I cache via fault. But that is handled via low level | |
417 | * hash fault code (hash_page_do_lazy_icache()) and we should not reach | |
418 | * here in such case. | |
419 | * | |
420 | * For wrong access that can result in PROTFAULT, the above vma->vm_flags | |
421 | * check should handle those and hence we should fall to the bad_area | |
422 | * handling correctly. | |
423 | * | |
424 | * For embedded with per page exec support that doesn't support coherent | |
425 | * icache we do get PROTFAULT and we handle that D/I cache sync in | |
426 | * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON | |
427 | * is conditional for server MMU. | |
428 | * | |
429 | * For radix, we can get prot fault for autonuma case, because radix | |
430 | * page table will have them marked noaccess for user. | |
431 | */ | |
374f3f59 AK |
432 | if (radix_enabled() || is_write) |
433 | return; | |
434 | ||
435 | WARN_ON_ONCE(error_code & DSISR_PROTFAULT); | |
2865d08d BH |
436 | } |
437 | #else | |
374f3f59 AK |
438 | static void sanity_check_fault(bool is_write, bool is_user, |
439 | unsigned long error_code, unsigned long address) { } | |
5b3e84fc | 440 | #endif /* CONFIG_PPC_BOOK3S */ |
2865d08d | 441 | |
41b464e5 BH |
442 | /* |
443 | * Define the correct "is_write" bit in error_code based | |
444 | * on the processor family | |
445 | */ | |
446 | #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) | |
447 | #define page_fault_is_write(__err) ((__err) & ESR_DST) | |
f3d96e69 | 448 | #define page_fault_is_bad(__err) (0) |
41b464e5 BH |
449 | #else |
450 | #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE) | |
968159c0 | 451 | #if defined(CONFIG_PPC_8xx) |
4915349b | 452 | #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G) |
f3d96e69 BH |
453 | #elif defined(CONFIG_PPC64) |
454 | #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S) | |
455 | #else | |
456 | #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S) | |
457 | #endif | |
41b464e5 BH |
458 | #endif |
459 | ||
14cf11af PM |
460 | /* |
461 | * For 600- and 800-family processors, the error_code parameter is DSISR | |
462 | * for a data fault, SRR1 for an instruction fault. For 400-family processors | |
463 | * the error_code parameter is ESR for a data fault, 0 for an instruction | |
464 | * fault. | |
465 | * For 64-bit processors, the error_code parameter is | |
466 | * - DSISR for a non-SLB data access fault, | |
467 | * - SRR1 & 0x08000000 for a non-SLB instruction access fault | |
468 | * - 0 any SLB fault. | |
469 | * | |
470 | * The return value is 0 if the fault was handled, or the signal | |
471 | * number if this is a kernel fault that can't be handled here. | |
472 | */ | |
7afad422 BH |
473 | static int __do_page_fault(struct pt_regs *regs, unsigned long address, |
474 | unsigned long error_code) | |
14cf11af PM |
475 | { |
476 | struct vm_area_struct * vma; | |
477 | struct mm_struct *mm = current->mm; | |
dde16072 | 478 | unsigned int flags = FAULT_FLAG_DEFAULT; |
c433ec04 | 479 | int is_exec = TRAP(regs) == 0x400; |
da929f6a | 480 | int is_user = user_mode(regs); |
41b464e5 | 481 | int is_write = page_fault_is_write(error_code); |
50a7ca3c | 482 | vm_fault_t fault, major = 0; |
0e36b0d1 | 483 | bool must_retry = false; |
b98cca44 | 484 | bool kprobe_fault = kprobe_page_fault(regs, 11); |
14cf11af | 485 | |
b98cca44 | 486 | if (unlikely(debugger_fault_handler(regs) || kprobe_fault)) |
65d47fd4 | 487 | return 0; |
14cf11af | 488 | |
f3d96e69 | 489 | if (unlikely(page_fault_is_bad(error_code))) { |
65d47fd4 | 490 | if (is_user) { |
f3d96e69 | 491 | _exception(SIGBUS, regs, BUS_OBJERR, address); |
65d47fd4 BH |
492 | return 0; |
493 | } | |
494 | return SIGBUS; | |
e6c8290a | 495 | } |
e6c8290a | 496 | |
2865d08d | 497 | /* Additional sanity check(s) */ |
374f3f59 | 498 | sanity_check_fault(is_write, is_user, error_code, address); |
2865d08d | 499 | |
d7df2443 BH |
500 | /* |
501 | * The kernel should never take an execute fault nor should it | |
de78a9c4 CL |
502 | * take a page fault to a kernel address or a page fault to a user |
503 | * address outside of dedicated places | |
d7df2443 | 504 | */ |
5e5be3ae | 505 | if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write))) |
65d47fd4 | 506 | return SIGSEGV; |
14cf11af | 507 | |
11ccdd33 BH |
508 | /* |
509 | * If we're in an interrupt, have no user context or are running | |
510 | * in a region with pagefaults disabled then we must not take the fault | |
511 | */ | |
512 | if (unlikely(faulthandler_disabled() || !mm)) { | |
513 | if (is_user) | |
514 | printk_ratelimited(KERN_ERR "Page fault in user mode" | |
515 | " with faulthandler_disabled()=%d" | |
516 | " mm=%p\n", | |
517 | faulthandler_disabled(), mm); | |
518 | return bad_area_nosemaphore(regs, address); | |
519 | } | |
520 | ||
a546498f BH |
521 | /* We restore the interrupt state now */ |
522 | if (!arch_irq_disabled_regs(regs)) | |
523 | local_irq_enable(); | |
524 | ||
a8b0ca17 | 525 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
7dd1fcc2 | 526 | |
69e044dd | 527 | /* |
c1e8d7c6 | 528 | * We want to do this outside mmap_lock, because reading code around nip |
69e044dd | 529 | * can result in fault, which will cause a deadlock when called with |
c1e8d7c6 | 530 | * mmap_lock held |
69e044dd | 531 | */ |
da929f6a | 532 | if (is_user) |
759496ba | 533 | flags |= FAULT_FLAG_USER; |
d2e0d2c5 BH |
534 | if (is_write) |
535 | flags |= FAULT_FLAG_WRITE; | |
536 | if (is_exec) | |
537 | flags |= FAULT_FLAG_INSTRUCTION; | |
759496ba | 538 | |
14cf11af PM |
539 | /* When running in the kernel we expect faults to occur only to |
540 | * addresses in user space. All other faults represent errors in the | |
fc5266ea | 541 | * kernel and should generate an OOPS. Unfortunately, in the case of an |
c1e8d7c6 | 542 | * erroneous fault occurring in a code path which already holds mmap_lock |
14cf11af PM |
543 | * we will deadlock attempting to validate the fault against the |
544 | * address space. Luckily the kernel only validly references user | |
545 | * space from well defined areas of code, which are listed in the | |
546 | * exceptions table. | |
547 | * | |
548 | * As the vast majority of faults will be valid we will only perform | |
fc5266ea | 549 | * the source reference check when there is a possibility of a deadlock. |
14cf11af PM |
550 | * Attempt to lock the address space, if we cannot we then validate the |
551 | * source. If this is invalid we can skip the address space check, | |
552 | * thus avoiding the deadlock. | |
553 | */ | |
d8ed45c5 | 554 | if (unlikely(!mmap_read_trylock(mm))) { |
da929f6a | 555 | if (!is_user && !search_exception_tables(regs->nip)) |
c3350602 | 556 | return bad_area_nosemaphore(regs, address); |
14cf11af | 557 | |
9be72573 | 558 | retry: |
d8ed45c5 | 559 | mmap_read_lock(mm); |
a546498f BH |
560 | } else { |
561 | /* | |
562 | * The above down_read_trylock() might have succeeded in | |
563 | * which case we'll have missed the might_sleep() from | |
564 | * down_read(): | |
565 | */ | |
566 | might_sleep(); | |
14cf11af PM |
567 | } |
568 | ||
569 | vma = find_vma(mm, address); | |
b15021d9 | 570 | if (unlikely(!vma)) |
c3350602 | 571 | return bad_area(regs, address); |
b15021d9 | 572 | if (likely(vma->vm_start <= address)) |
14cf11af | 573 | goto good_area; |
b15021d9 | 574 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) |
c3350602 | 575 | return bad_area(regs, address); |
14cf11af | 576 | |
8f5ca0b3 | 577 | /* The stack is being expanded, check if it's valid */ |
0e36b0d1 CL |
578 | if (unlikely(bad_stack_expansion(regs, address, vma, flags, |
579 | &must_retry))) { | |
580 | if (!must_retry) | |
581 | return bad_area(regs, address); | |
582 | ||
d8ed45c5 | 583 | mmap_read_unlock(mm); |
0e36b0d1 CL |
584 | if (fault_in_pages_readable((const char __user *)regs->nip, |
585 | sizeof(unsigned int))) | |
586 | return bad_area_nosemaphore(regs, address); | |
587 | goto retry; | |
588 | } | |
14cf11af | 589 | |
8f5ca0b3 | 590 | /* Try to expand it */ |
b15021d9 | 591 | if (unlikely(expand_stack(vma, address))) |
c3350602 | 592 | return bad_area(regs, address); |
14cf11af PM |
593 | |
594 | good_area: | |
fe4a6856 AK |
595 | |
596 | #ifdef CONFIG_PPC_MEM_KEYS | |
597 | if (unlikely(access_pkey_error(is_write, is_exec, | |
598 | (error_code & DSISR_KEYFAULT), vma))) | |
599 | return bad_access_pkey(regs, address, vma); | |
600 | #endif /* CONFIG_PPC_MEM_KEYS */ | |
601 | ||
bd0d63f8 | 602 | if (unlikely(access_error(is_write, is_exec, vma))) |
ecb101ae | 603 | return bad_access(regs, address); |
14cf11af PM |
604 | |
605 | /* | |
606 | * If for any reason at all we couldn't handle the fault, | |
607 | * make sure we exit gracefully rather than endlessly redo | |
608 | * the fault. | |
609 | */ | |
dcddffd4 | 610 | fault = handle_mm_fault(vma, address, flags); |
e6c2a479 | 611 | |
f43bb27e | 612 | major |= fault & VM_FAULT_MAJOR; |
14c02e41 | 613 | |
c9a0dad1 PX |
614 | if (fault_signal_pending(fault, regs)) |
615 | return user_mode(regs) ? 0 : SIGBUS; | |
616 | ||
14c02e41 | 617 | /* |
c1e8d7c6 | 618 | * Handle the retry right now, the mmap_lock has been released in that |
14c02e41 LD |
619 | * case. |
620 | */ | |
621 | if (unlikely(fault & VM_FAULT_RETRY)) { | |
14c02e41 | 622 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
14c02e41 | 623 | flags |= FAULT_FLAG_TRIED; |
c9a0dad1 | 624 | goto retry; |
14c02e41 | 625 | } |
14cf11af | 626 | } |
9be72573 | 627 | |
d8ed45c5 | 628 | mmap_read_unlock(current->mm); |
b5c8f0fd BH |
629 | |
630 | if (unlikely(fault & VM_FAULT_ERROR)) | |
631 | return mm_fault_error(regs, address, fault); | |
632 | ||
9be72573 | 633 | /* |
14c02e41 | 634 | * Major/minor page fault accounting. |
9be72573 | 635 | */ |
f43bb27e | 636 | if (major) { |
14c02e41 | 637 | current->maj_flt++; |
04aafdc6 | 638 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); |
3da02648 | 639 | cmo_account_page_fault(); |
14c02e41 LD |
640 | } else { |
641 | current->min_flt++; | |
04aafdc6 | 642 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); |
ac17dc8e | 643 | } |
c3350602 | 644 | return 0; |
7afad422 BH |
645 | } |
646 | NOKPROBE_SYMBOL(__do_page_fault); | |
647 | ||
648 | int do_page_fault(struct pt_regs *regs, unsigned long address, | |
649 | unsigned long error_code) | |
650 | { | |
651 | enum ctx_state prev_state = exception_enter(); | |
652 | int rc = __do_page_fault(regs, address, error_code); | |
ba12eede LZ |
653 | exception_exit(prev_state); |
654 | return rc; | |
14cf11af | 655 | } |
03465f89 | 656 | NOKPROBE_SYMBOL(do_page_fault); |
14cf11af PM |
657 | |
658 | /* | |
659 | * bad_page_fault is called when we have a bad access from the kernel. | |
660 | * It is called from the DSI and ISI handlers in head.S and from some | |
661 | * of the procedures in traps.c. | |
662 | */ | |
663 | void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig) | |
664 | { | |
665 | const struct exception_table_entry *entry; | |
46ddcb39 | 666 | int is_write = page_fault_is_write(regs->dsisr); |
14cf11af PM |
667 | |
668 | /* Are we prepared to handle this fault? */ | |
669 | if ((entry = search_exception_tables(regs->nip)) != NULL) { | |
61a92f70 | 670 | regs->nip = extable_fixup(entry); |
14cf11af PM |
671 | return; |
672 | } | |
673 | ||
674 | /* kernel has accessed a bad area */ | |
723925b7 | 675 | |
2271db20 | 676 | switch (TRAP(regs)) { |
a416dd8d ME |
677 | case 0x300: |
678 | case 0x380: | |
d7b45615 | 679 | case 0xe00: |
46ddcb39 | 680 | pr_alert("BUG: %s on %s at 0x%08lx\n", |
49a502ea | 681 | regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" : |
46ddcb39 CL |
682 | "Unable to handle kernel data access", |
683 | is_write ? "write" : "read", regs->dar); | |
a416dd8d ME |
684 | break; |
685 | case 0x400: | |
686 | case 0x480: | |
49a502ea CL |
687 | pr_alert("BUG: Unable to handle kernel instruction fetch%s", |
688 | regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n"); | |
a416dd8d | 689 | break; |
eab861a7 | 690 | case 0x600: |
49a502ea CL |
691 | pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n", |
692 | regs->dar); | |
eab861a7 | 693 | break; |
a416dd8d | 694 | default: |
49a502ea CL |
695 | pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n", |
696 | regs->dar); | |
a416dd8d | 697 | break; |
723925b7 OJ |
698 | } |
699 | printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", | |
700 | regs->nip); | |
701 | ||
a70857e4 | 702 | if (task_stack_end_corrupted(current)) |
28b54990 AB |
703 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); |
704 | ||
14cf11af PM |
705 | die("Kernel access of bad area", regs, sig); |
706 | } |