Commit | Line | Data |
---|---|---|
ea2ba7dc | 1 | /* |
4c9e1385 | 2 | * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) |
1da177e4 LT |
3 | * Licensed under the GPL |
4 | */ | |
5 | ||
4c9e1385 JD |
6 | #include <linux/mm.h> |
7 | #include <linux/sched.h> | |
8 | #include <linux/hardirq.h> | |
73395a00 | 9 | #include <linux/module.h> |
4c9e1385 JD |
10 | #include <asm/current.h> |
11 | #include <asm/pgtable.h> | |
12 | #include <asm/tlbflush.h> | |
eb830759 | 13 | #include "arch.h" |
4c9e1385 JD |
14 | #include "as-layout.h" |
15 | #include "kern_util.h" | |
ea2ba7dc | 16 | #include "os.h" |
d83ecf08 | 17 | #include "skas.h" |
1da177e4 | 18 | |
4c9e1385 JD |
19 | /* |
20 | * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by | |
21 | * segv(). | |
22 | */ | |
1d3468a6 | 23 | int handle_page_fault(unsigned long address, unsigned long ip, |
1da177e4 LT |
24 | int is_write, int is_user, int *code_out) |
25 | { | |
26 | struct mm_struct *mm = current->mm; | |
27 | struct vm_area_struct *vma; | |
28 | pgd_t *pgd; | |
29 | pud_t *pud; | |
30 | pmd_t *pmd; | |
31 | pte_t *pte; | |
1da177e4 LT |
32 | int err = -EFAULT; |
33 | ||
34 | *code_out = SEGV_MAPERR; | |
fea03cb4 | 35 | |
4c9e1385 JD |
36 | /* |
37 | * If the fault was during atomic operation, don't take the fault, just | |
38 | * fail. | |
39 | */ | |
fea03cb4 PBG |
40 | if (in_atomic()) |
41 | goto out_nosemaphore; | |
42 | ||
1da177e4 LT |
43 | down_read(&mm->mmap_sem); |
44 | vma = find_vma(mm, address); | |
4c9e1385 | 45 | if (!vma) |
1da177e4 | 46 | goto out; |
4c9e1385 | 47 | else if (vma->vm_start <= address) |
1da177e4 | 48 | goto good_area; |
4c9e1385 | 49 | else if (!(vma->vm_flags & VM_GROWSDOWN)) |
1da177e4 | 50 | goto out; |
4c9e1385 | 51 | else if (is_user && !ARCH_IS_STACKGROW(address)) |
1da177e4 | 52 | goto out; |
4c9e1385 | 53 | else if (expand_stack(vma, address)) |
1da177e4 LT |
54 | goto out; |
55 | ||
3b52166c | 56 | good_area: |
1da177e4 | 57 | *code_out = SEGV_ACCERR; |
4c9e1385 | 58 | if (is_write && !(vma->vm_flags & VM_WRITE)) |
1da177e4 | 59 | goto out; |
13479d52 | 60 | |
d129f312 | 61 | /* Don't require VM_READ|VM_EXEC for write faults! */ |
4c9e1385 | 62 | if (!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC))) |
5d86456d | 63 | goto out; |
13479d52 | 64 | |
1da177e4 | 65 | do { |
83c54070 | 66 | int fault; |
1c0fe6e3 | 67 | |
d06063cc | 68 | fault = handle_mm_fault(mm, vma, address, is_write ? FAULT_FLAG_WRITE : 0); |
83c54070 NP |
69 | if (unlikely(fault & VM_FAULT_ERROR)) { |
70 | if (fault & VM_FAULT_OOM) { | |
83c54070 NP |
71 | goto out_of_memory; |
72 | } else if (fault & VM_FAULT_SIGBUS) { | |
73 | err = -EACCES; | |
74 | goto out; | |
75 | } | |
1da177e4 LT |
76 | BUG(); |
77 | } | |
83c54070 NP |
78 | if (fault & VM_FAULT_MAJOR) |
79 | current->maj_flt++; | |
80 | else | |
81 | current->min_flt++; | |
82 | ||
3b52166c PBG |
83 | pgd = pgd_offset(mm, address); |
84 | pud = pud_offset(pgd, address); | |
85 | pmd = pmd_offset(pud, address); | |
86 | pte = pte_offset_kernel(pmd, address); | |
4c9e1385 | 87 | } while (!pte_present(*pte)); |
1da177e4 | 88 | err = 0; |
4c9e1385 JD |
89 | /* |
90 | * The below warning was added in place of | |
cbc24afa PBG |
91 | * pte_mkyoung(); if (is_write) pte_mkdirty(); |
92 | * If it's triggered, we'd see normally a hang here (a clean pte is | |
93 | * marked read-only to emulate the dirty bit). | |
94 | * However, the generic code can mark a PTE writable but clean on a | |
95 | * concurrent read fault, triggering this harmlessly. So comment it out. | |
96 | */ | |
97 | #if 0 | |
16b03678 | 98 | WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte))); |
cbc24afa | 99 | #endif |
3b52166c PBG |
100 | flush_tlb_page(vma, address); |
101 | out: | |
1da177e4 | 102 | up_read(&mm->mmap_sem); |
fea03cb4 | 103 | out_nosemaphore: |
4c9e1385 | 104 | return err; |
1da177e4 | 105 | |
1da177e4 | 106 | out_of_memory: |
1c0fe6e3 NP |
107 | /* |
108 | * We ran out of memory, call the OOM killer, and return the userspace | |
109 | * (which will retry the fault, or kill us if we got oom-killed). | |
110 | */ | |
111 | up_read(&mm->mmap_sem); | |
112 | pagefault_out_of_memory(); | |
113 | return 0; | |
1da177e4 | 114 | } |
73395a00 | 115 | EXPORT_SYMBOL(handle_page_fault); |
1da177e4 | 116 | |
3ef6130a RW |
117 | static void show_segv_info(struct uml_pt_regs *regs) |
118 | { | |
119 | struct task_struct *tsk = current; | |
120 | struct faultinfo *fi = UPT_FAULTINFO(regs); | |
121 | ||
122 | if (!unhandled_signal(tsk, SIGSEGV)) | |
123 | return; | |
124 | ||
125 | if (!printk_ratelimit()) | |
126 | return; | |
127 | ||
128 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x", | |
129 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
130 | tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi), | |
131 | (void *)UPT_IP(regs), (void *)UPT_SP(regs), | |
132 | fi->error_code); | |
133 | ||
134 | print_vma_addr(KERN_CONT " in ", UPT_IP(regs)); | |
135 | printk(KERN_CONT "\n"); | |
136 | } | |
137 | ||
27aa6ef3 JD |
138 | static void bad_segv(struct faultinfo fi, unsigned long ip) |
139 | { | |
140 | struct siginfo si; | |
141 | ||
142 | si.si_signo = SIGSEGV; | |
143 | si.si_code = SEGV_ACCERR; | |
144 | si.si_addr = (void __user *) FAULT_ADDRESS(fi); | |
145 | current->thread.arch.faultinfo = fi; | |
146 | force_sig_info(SIGSEGV, &si, current); | |
147 | } | |
148 | ||
3e6f2ac4 JD |
149 | void fatal_sigsegv(void) |
150 | { | |
151 | force_sigsegv(SIGSEGV, current); | |
152 | do_signal(); | |
153 | /* | |
154 | * This is to tell gcc that we're not returning - do_signal | |
155 | * can, in general, return, but in this case, it's not, since | |
156 | * we just got a fatal SIGSEGV queued. | |
157 | */ | |
158 | os_dump_core(); | |
159 | } | |
160 | ||
edea1385 | 161 | void segv_handler(int sig, struct uml_pt_regs *regs) |
c66fdd5e GS |
162 | { |
163 | struct faultinfo * fi = UPT_FAULTINFO(regs); | |
164 | ||
4c9e1385 | 165 | if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) { |
3ef6130a | 166 | show_segv_info(regs); |
c66fdd5e GS |
167 | bad_segv(*fi, UPT_IP(regs)); |
168 | return; | |
169 | } | |
170 | segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs); | |
171 | } | |
172 | ||
c578455a BS |
173 | /* |
174 | * We give a *copy* of the faultinfo in the regs to segv. | |
175 | * This must be done, since nesting SEGVs could overwrite | |
176 | * the info in the regs. A pointer to the info then would | |
177 | * give us bad data! | |
178 | */ | |
5d86456d | 179 | unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user, |
77bf4400 | 180 | struct uml_pt_regs *regs) |
1da177e4 LT |
181 | { |
182 | struct siginfo si; | |
fab95c55 | 183 | jmp_buf *catcher; |
1da177e4 | 184 | int err; |
5d86456d JD |
185 | int is_write = FAULT_WRITE(fi); |
186 | unsigned long address = FAULT_ADDRESS(fi); | |
1da177e4 | 187 | |
4c9e1385 | 188 | if (!is_user && (address >= start_vm) && (address < end_vm)) { |
5d86456d JD |
189 | flush_tlb_kernel_vm(); |
190 | return 0; | |
191 | } | |
4c9e1385 | 192 | else if (current->mm == NULL) { |
377fad3a | 193 | show_regs(container_of(regs, struct pt_regs, regs)); |
4c9e1385 | 194 | panic("Segfault with no mm"); |
377fad3a | 195 | } |
546fe1cb | 196 | |
be662a18 | 197 | if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi)) |
4c9e1385 JD |
198 | err = handle_page_fault(address, ip, is_write, is_user, |
199 | &si.si_code); | |
546fe1cb PBG |
200 | else { |
201 | err = -EFAULT; | |
4c9e1385 JD |
202 | /* |
203 | * A thread accessed NULL, we get a fault, but CR2 is invalid. | |
204 | * This code is used in __do_copy_from_user() of TT mode. | |
205 | * XXX tt mode is gone, so maybe this isn't needed any more | |
206 | */ | |
546fe1cb PBG |
207 | address = 0; |
208 | } | |
1da177e4 LT |
209 | |
210 | catcher = current->thread.fault_catcher; | |
4c9e1385 | 211 | if (!err) |
5d86456d | 212 | return 0; |
4c9e1385 | 213 | else if (catcher != NULL) { |
1da177e4 | 214 | current->thread.fault_addr = (void *) address; |
fab95c55 | 215 | UML_LONGJMP(catcher, 1); |
1d3468a6 | 216 | } |
4c9e1385 | 217 | else if (current->thread.fault_addr != NULL) |
1da177e4 | 218 | panic("fault_addr set but no fault catcher"); |
4c9e1385 | 219 | else if (!is_user && arch_fixup(ip, regs)) |
5d86456d | 220 | return 0; |
1da177e4 | 221 | |
4c9e1385 | 222 | if (!is_user) { |
377fad3a | 223 | show_regs(container_of(regs, struct pt_regs, regs)); |
1d3468a6 | 224 | panic("Kernel mode fault at addr 0x%lx, ip 0x%lx", |
1da177e4 | 225 | address, ip); |
377fad3a | 226 | } |
1da177e4 | 227 | |
3ef6130a RW |
228 | show_segv_info(regs); |
229 | ||
3b52166c | 230 | if (err == -EACCES) { |
1da177e4 LT |
231 | si.si_signo = SIGBUS; |
232 | si.si_errno = 0; | |
233 | si.si_code = BUS_ADRERR; | |
4d338e1a | 234 | si.si_addr = (void __user *)address; |
5d86456d | 235 | current->thread.arch.faultinfo = fi; |
1da177e4 | 236 | force_sig_info(SIGBUS, &si, current); |
3b52166c PBG |
237 | } else { |
238 | BUG_ON(err != -EFAULT); | |
1da177e4 | 239 | si.si_signo = SIGSEGV; |
4d338e1a | 240 | si.si_addr = (void __user *) address; |
5d86456d | 241 | current->thread.arch.faultinfo = fi; |
1da177e4 LT |
242 | force_sig_info(SIGSEGV, &si, current); |
243 | } | |
5d86456d | 244 | return 0; |
1da177e4 LT |
245 | } |
246 | ||
77bf4400 | 247 | void relay_signal(int sig, struct uml_pt_regs *regs) |
1da177e4 | 248 | { |
4c9e1385 JD |
249 | if (!UPT_IS_USER(regs)) { |
250 | if (sig == SIGBUS) | |
251 | printk(KERN_ERR "Bus error - the host /dev/shm or /tmp " | |
252 | "mount likely just ran out of space\n"); | |
1da177e4 | 253 | panic("Kernel mode signal %d", sig); |
6edf428e JD |
254 | } |
255 | ||
9226b838 JD |
256 | arch_examine_signal(sig, regs); |
257 | ||
5d86456d | 258 | current->thread.arch.faultinfo = *UPT_FAULTINFO(regs); |
1da177e4 LT |
259 | force_sig(sig, current); |
260 | } | |
261 | ||
edea1385 | 262 | void bus_handler(int sig, struct uml_pt_regs *regs) |
1da177e4 | 263 | { |
4c9e1385 | 264 | if (current->thread.fault_catcher != NULL) |
fab95c55 | 265 | UML_LONGJMP(current->thread.fault_catcher, 1); |
1da177e4 LT |
266 | else relay_signal(sig, regs); |
267 | } | |
268 | ||
edea1385 | 269 | void winch(int sig, struct uml_pt_regs *regs) |
1da177e4 LT |
270 | { |
271 | do_IRQ(WINCH_IRQ, regs); | |
272 | } | |
273 | ||
274 | void trap_init(void) | |
275 | { | |
276 | } |