deal with deadlock in d_walk()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / um / kernel / trap.c
CommitLineData
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>
37185b33
AV
13#include <arch.h>
14#include <as-layout.h>
15#include <kern_util.h>
16#include <os.h>
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 23int 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 32 int err = -EFAULT;
e2ec2c2b 33 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1da177e4
LT
34
35 *code_out = SEGV_MAPERR;
fea03cb4 36
4c9e1385
JD
37 /*
38 * If the fault was during atomic operation, don't take the fault, just
39 * fail.
40 */
fea03cb4
PBG
41 if (in_atomic())
42 goto out_nosemaphore;
43
e2ec2c2b
JW
44 if (is_user)
45 flags |= FAULT_FLAG_USER;
1cefe28f 46retry:
1da177e4
LT
47 down_read(&mm->mmap_sem);
48 vma = find_vma(mm, address);
4c9e1385 49 if (!vma)
1da177e4 50 goto out;
4c9e1385 51 else if (vma->vm_start <= address)
1da177e4 52 goto good_area;
4c9e1385 53 else if (!(vma->vm_flags & VM_GROWSDOWN))
1da177e4 54 goto out;
4c9e1385 55 else if (is_user && !ARCH_IS_STACKGROW(address))
1da177e4 56 goto out;
4c9e1385 57 else if (expand_stack(vma, address))
1da177e4
LT
58 goto out;
59
3b52166c 60good_area:
1da177e4 61 *code_out = SEGV_ACCERR;
e2ec2c2b
JW
62 if (is_write) {
63 if (!(vma->vm_flags & VM_WRITE))
64 goto out;
65 flags |= FAULT_FLAG_WRITE;
66 } else {
67 /* Don't require VM_READ|VM_EXEC for write faults! */
68 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
69 goto out;
70 }
13479d52 71
1da177e4 72 do {
83c54070 73 int fault;
1c0fe6e3 74
1cefe28f
KC
75 fault = handle_mm_fault(mm, vma, address, flags);
76
77 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
78 goto out_nosemaphore;
79
83c54070
NP
80 if (unlikely(fault & VM_FAULT_ERROR)) {
81 if (fault & VM_FAULT_OOM) {
83c54070
NP
82 goto out_of_memory;
83 } else if (fault & VM_FAULT_SIGBUS) {
84 err = -EACCES;
85 goto out;
86 }
1da177e4
LT
87 BUG();
88 }
1cefe28f
KC
89 if (flags & FAULT_FLAG_ALLOW_RETRY) {
90 if (fault & VM_FAULT_MAJOR)
91 current->maj_flt++;
92 else
93 current->min_flt++;
94 if (fault & VM_FAULT_RETRY) {
95 flags &= ~FAULT_FLAG_ALLOW_RETRY;
45cac65b 96 flags |= FAULT_FLAG_TRIED;
1cefe28f
KC
97
98 goto retry;
99 }
100 }
83c54070 101
3b52166c
PBG
102 pgd = pgd_offset(mm, address);
103 pud = pud_offset(pgd, address);
104 pmd = pmd_offset(pud, address);
105 pte = pte_offset_kernel(pmd, address);
4c9e1385 106 } while (!pte_present(*pte));
1da177e4 107 err = 0;
4c9e1385
JD
108 /*
109 * The below warning was added in place of
cbc24afa
PBG
110 * pte_mkyoung(); if (is_write) pte_mkdirty();
111 * If it's triggered, we'd see normally a hang here (a clean pte is
112 * marked read-only to emulate the dirty bit).
113 * However, the generic code can mark a PTE writable but clean on a
114 * concurrent read fault, triggering this harmlessly. So comment it out.
115 */
116#if 0
16b03678 117 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
cbc24afa 118#endif
3b52166c
PBG
119 flush_tlb_page(vma, address);
120out:
1da177e4 121 up_read(&mm->mmap_sem);
fea03cb4 122out_nosemaphore:
4c9e1385 123 return err;
1da177e4 124
1da177e4 125out_of_memory:
1c0fe6e3
NP
126 /*
127 * We ran out of memory, call the OOM killer, and return the userspace
128 * (which will retry the fault, or kill us if we got oom-killed).
129 */
130 up_read(&mm->mmap_sem);
086c6cc5
JW
131 if (!is_user)
132 goto out_nosemaphore;
1c0fe6e3
NP
133 pagefault_out_of_memory();
134 return 0;
1da177e4 135}
73395a00 136EXPORT_SYMBOL(handle_page_fault);
1da177e4 137
3ef6130a
RW
138static void show_segv_info(struct uml_pt_regs *regs)
139{
140 struct task_struct *tsk = current;
141 struct faultinfo *fi = UPT_FAULTINFO(regs);
142
143 if (!unhandled_signal(tsk, SIGSEGV))
144 return;
145
146 if (!printk_ratelimit())
147 return;
148
149 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x",
150 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
151 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
152 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
153 fi->error_code);
154
155 print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
156 printk(KERN_CONT "\n");
157}
158
27aa6ef3
JD
159static void bad_segv(struct faultinfo fi, unsigned long ip)
160{
161 struct siginfo si;
162
163 si.si_signo = SIGSEGV;
164 si.si_code = SEGV_ACCERR;
165 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
166 current->thread.arch.faultinfo = fi;
167 force_sig_info(SIGSEGV, &si, current);
168}
169
3e6f2ac4
JD
170void fatal_sigsegv(void)
171{
172 force_sigsegv(SIGSEGV, current);
173 do_signal();
174 /*
175 * This is to tell gcc that we're not returning - do_signal
176 * can, in general, return, but in this case, it's not, since
177 * we just got a fatal SIGSEGV queued.
178 */
179 os_dump_core();
180}
181
d3c1cfcd 182void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
c66fdd5e
GS
183{
184 struct faultinfo * fi = UPT_FAULTINFO(regs);
185
4c9e1385 186 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
3ef6130a 187 show_segv_info(regs);
c66fdd5e
GS
188 bad_segv(*fi, UPT_IP(regs));
189 return;
190 }
191 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
192}
193
c578455a
BS
194/*
195 * We give a *copy* of the faultinfo in the regs to segv.
196 * This must be done, since nesting SEGVs could overwrite
197 * the info in the regs. A pointer to the info then would
198 * give us bad data!
199 */
5d86456d 200unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
77bf4400 201 struct uml_pt_regs *regs)
1da177e4
LT
202{
203 struct siginfo si;
fab95c55 204 jmp_buf *catcher;
1da177e4 205 int err;
5d86456d
JD
206 int is_write = FAULT_WRITE(fi);
207 unsigned long address = FAULT_ADDRESS(fi);
1da177e4 208
4c9e1385 209 if (!is_user && (address >= start_vm) && (address < end_vm)) {
5d86456d
JD
210 flush_tlb_kernel_vm();
211 return 0;
212 }
4c9e1385 213 else if (current->mm == NULL) {
377fad3a 214 show_regs(container_of(regs, struct pt_regs, regs));
4c9e1385 215 panic("Segfault with no mm");
377fad3a 216 }
546fe1cb 217
be662a18 218 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
4c9e1385
JD
219 err = handle_page_fault(address, ip, is_write, is_user,
220 &si.si_code);
546fe1cb
PBG
221 else {
222 err = -EFAULT;
4c9e1385
JD
223 /*
224 * A thread accessed NULL, we get a fault, but CR2 is invalid.
225 * This code is used in __do_copy_from_user() of TT mode.
226 * XXX tt mode is gone, so maybe this isn't needed any more
227 */
546fe1cb
PBG
228 address = 0;
229 }
1da177e4
LT
230
231 catcher = current->thread.fault_catcher;
4c9e1385 232 if (!err)
5d86456d 233 return 0;
4c9e1385 234 else if (catcher != NULL) {
1da177e4 235 current->thread.fault_addr = (void *) address;
fab95c55 236 UML_LONGJMP(catcher, 1);
1d3468a6 237 }
4c9e1385 238 else if (current->thread.fault_addr != NULL)
1da177e4 239 panic("fault_addr set but no fault catcher");
4c9e1385 240 else if (!is_user && arch_fixup(ip, regs))
5d86456d 241 return 0;
1da177e4 242
4c9e1385 243 if (!is_user) {
377fad3a 244 show_regs(container_of(regs, struct pt_regs, regs));
1d3468a6 245 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
1da177e4 246 address, ip);
377fad3a 247 }
1da177e4 248
3ef6130a
RW
249 show_segv_info(regs);
250
3b52166c 251 if (err == -EACCES) {
1da177e4
LT
252 si.si_signo = SIGBUS;
253 si.si_errno = 0;
254 si.si_code = BUS_ADRERR;
4d338e1a 255 si.si_addr = (void __user *)address;
5d86456d 256 current->thread.arch.faultinfo = fi;
1da177e4 257 force_sig_info(SIGBUS, &si, current);
3b52166c
PBG
258 } else {
259 BUG_ON(err != -EFAULT);
1da177e4 260 si.si_signo = SIGSEGV;
4d338e1a 261 si.si_addr = (void __user *) address;
5d86456d 262 current->thread.arch.faultinfo = fi;
1da177e4
LT
263 force_sig_info(SIGSEGV, &si, current);
264 }
5d86456d 265 return 0;
1da177e4
LT
266}
267
d3c1cfcd 268void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
1da177e4 269{
d3c1cfcd
MP
270 struct faultinfo *fi;
271 struct siginfo clean_si;
272
4c9e1385
JD
273 if (!UPT_IS_USER(regs)) {
274 if (sig == SIGBUS)
275 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
276 "mount likely just ran out of space\n");
1da177e4 277 panic("Kernel mode signal %d", sig);
6edf428e
JD
278 }
279
9226b838
JD
280 arch_examine_signal(sig, regs);
281
d3c1cfcd
MP
282 memset(&clean_si, 0, sizeof(clean_si));
283 clean_si.si_signo = si->si_signo;
284 clean_si.si_errno = si->si_errno;
285 clean_si.si_code = si->si_code;
286 switch (sig) {
287 case SIGILL:
288 case SIGFPE:
289 case SIGSEGV:
290 case SIGBUS:
291 case SIGTRAP:
292 fi = UPT_FAULTINFO(regs);
293 clean_si.si_addr = (void __user *) FAULT_ADDRESS(*fi);
294 current->thread.arch.faultinfo = *fi;
295#ifdef __ARCH_SI_TRAPNO
296 clean_si.si_trapno = si->si_trapno;
297#endif
298 break;
299 default:
300 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d)\n",
301 sig, si->si_code);
302 }
303
304 force_sig_info(sig, &clean_si, current);
1da177e4
LT
305}
306
d3c1cfcd 307void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
1da177e4 308{
4c9e1385 309 if (current->thread.fault_catcher != NULL)
fab95c55 310 UML_LONGJMP(current->thread.fault_catcher, 1);
d3c1cfcd
MP
311 else
312 relay_signal(sig, si, regs);
1da177e4
LT
313}
314
d3c1cfcd 315void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
1da177e4
LT
316{
317 do_IRQ(WINCH_IRQ, regs);
318}
319
320void trap_init(void)
321{
322}