blob: 3850d53f79fdd11afc86319be79e8f96baa12de5 [file] [log] [blame]
Gennady Sharapovea2ba7d2006-01-08 01:01:31 -08001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com)
3 * Licensed under the GPL
4 */
5
6#include "linux/kernel.h"
7#include "asm/errno.h"
8#include "linux/sched.h"
9#include "linux/mm.h"
10#include "linux/spinlock.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include "linux/init.h"
12#include "linux/ptrace.h"
13#include "asm/semaphore.h"
14#include "asm/pgtable.h"
15#include "asm/pgalloc.h"
16#include "asm/tlbflush.h"
17#include "asm/a.out.h"
18#include "asm/current.h"
19#include "asm/irq.h"
Paolo 'Blaisorblade' Giarrusso546fe1c2005-09-22 21:44:16 -070020#include "sysdep/sigcontext.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "kern_util.h"
Jeff Dike4ff83ce2007-05-06 14:51:08 -070022#include "as-layout.h"
Jeff Dikeeb830752007-05-06 14:51:07 -070023#include "arch.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "kern.h"
25#include "chan_kern.h"
26#include "mconsole_kern.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "mem.h"
28#include "mem_kern.h"
Gennady Sharapovc66fdd52006-01-08 01:01:32 -080029#include "sysdep/sigcontext.h"
30#include "sysdep/ptrace.h"
31#include "os.h"
Paolo 'Blaisorblade' Giarrussobe662a12005-09-30 11:58:59 -070032#ifdef CONFIG_MODE_SKAS
33#include "skas.h"
34#endif
Gennady Sharapovea2ba7d2006-01-08 01:01:31 -080035#include "os.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -070037/* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
Jeff Dike1d3468a2006-07-10 04:45:13 -070038int handle_page_fault(unsigned long address, unsigned long ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 int is_write, int is_user, int *code_out)
40{
41 struct mm_struct *mm = current->mm;
42 struct vm_area_struct *vma;
43 pgd_t *pgd;
44 pud_t *pud;
45 pmd_t *pmd;
46 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 int err = -EFAULT;
48
49 *code_out = SEGV_MAPERR;
Paolo 'Blaisorblade' Giarrussofea03cb2005-09-22 21:44:20 -070050
51 /* If the fault was during atomic operation, don't take the fault, just
52 * fail. */
53 if (in_atomic())
54 goto out_nosemaphore;
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 down_read(&mm->mmap_sem);
57 vma = find_vma(mm, address);
Jeff Dike1d3468a2006-07-10 04:45:13 -070058 if(!vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 goto out;
Jeff Dike1d3468a2006-07-10 04:45:13 -070060 else if(vma->vm_start <= address)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 goto good_area;
Jeff Dike1d3468a2006-07-10 04:45:13 -070062 else if(!(vma->vm_flags & VM_GROWSDOWN))
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 goto out;
Jeff Dike2d58cc92005-05-06 21:30:55 -070064 else if(is_user && !ARCH_IS_STACKGROW(address))
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 goto out;
Jeff Dike1d3468a2006-07-10 04:45:13 -070066 else if(expand_stack(vma, address))
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 goto out;
68
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -070069good_area:
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 *code_out = SEGV_ACCERR;
Jeff Dike1d3468a2006-07-10 04:45:13 -070071 if(is_write && !(vma->vm_flags & VM_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 goto out;
Jeff Dike13479d52005-05-20 13:59:08 -070073
Paolo 'Blaisorblade' Giarrussod129f312005-09-10 19:44:57 +020074 /* Don't require VM_READ|VM_EXEC for write faults! */
Jeff Dike5d864562007-05-06 14:51:24 -070075 if(!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
76 goto out;
Jeff Dike13479d52005-05-20 13:59:08 -070077
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 do {
Nick Piggin83c54072007-07-19 01:47:05 -070079 int fault;
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -070080survive:
Nick Piggin83c54072007-07-19 01:47:05 -070081 fault = handle_mm_fault(mm, vma, address, is_write);
82 if (unlikely(fault & VM_FAULT_ERROR)) {
83 if (fault & VM_FAULT_OOM) {
84 err = -ENOMEM;
85 goto out_of_memory;
86 } else if (fault & VM_FAULT_SIGBUS) {
87 err = -EACCES;
88 goto out;
89 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 BUG();
91 }
Nick Piggin83c54072007-07-19 01:47:05 -070092 if (fault & VM_FAULT_MAJOR)
93 current->maj_flt++;
94 else
95 current->min_flt++;
96
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -070097 pgd = pgd_offset(mm, address);
98 pud = pud_offset(pgd, address);
99 pmd = pmd_offset(pud, address);
100 pte = pte_offset_kernel(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 } while(!pte_present(*pte));
102 err = 0;
Paolo 'Blaisorblade' Giarrussocbc24af2005-11-13 16:07:04 -0800103 /* The below warning was added in place of
104 * pte_mkyoung(); if (is_write) pte_mkdirty();
105 * If it's triggered, we'd see normally a hang here (a clean pte is
106 * marked read-only to emulate the dirty bit).
107 * However, the generic code can mark a PTE writable but clean on a
108 * concurrent read fault, triggering this harmlessly. So comment it out.
109 */
110#if 0
Paolo 'Blaisorblade' Giarrusso16b03672005-09-10 19:44:58 +0200111 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
Paolo 'Blaisorblade' Giarrussocbc24af2005-11-13 16:07:04 -0800112#endif
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -0700113 flush_tlb_page(vma, address);
114out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 up_read(&mm->mmap_sem);
Paolo 'Blaisorblade' Giarrussofea03cb2005-09-22 21:44:20 -0700116out_nosemaphore:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return(err);
118
119/*
120 * We ran out of memory, or some other thing happened to us that made
121 * us unable to handle the page fault gracefully.
122 */
123out_of_memory:
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700124 if (is_init(current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 up_read(&mm->mmap_sem);
126 yield();
127 down_read(&mm->mmap_sem);
128 goto survive;
129 }
130 goto out;
131}
132
Jeff Dike27aa6ef2007-02-10 01:44:14 -0800133static void bad_segv(struct faultinfo fi, unsigned long ip)
134{
135 struct siginfo si;
136
137 si.si_signo = SIGSEGV;
138 si.si_code = SEGV_ACCERR;
139 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
140 current->thread.arch.faultinfo = fi;
141 force_sig_info(SIGSEGV, &si, current);
142}
143
144static void segv_handler(int sig, union uml_pt_regs *regs)
Gennady Sharapovc66fdd52006-01-08 01:01:32 -0800145{
146 struct faultinfo * fi = UPT_FAULTINFO(regs);
147
148 if(UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)){
149 bad_segv(*fi, UPT_IP(regs));
150 return;
151 }
152 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
153}
154
Bodo Stroesserc5784552005-05-05 16:15:31 -0700155/*
156 * We give a *copy* of the faultinfo in the regs to segv.
157 * This must be done, since nesting SEGVs could overwrite
158 * the info in the regs. A pointer to the info then would
159 * give us bad data!
160 */
Jeff Dike5d864562007-05-06 14:51:24 -0700161unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
162 union uml_pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
164 struct siginfo si;
165 void *catcher;
166 int err;
Jeff Dike5d864562007-05-06 14:51:24 -0700167 int is_write = FAULT_WRITE(fi);
168 unsigned long address = FAULT_ADDRESS(fi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Jeff Dike5d864562007-05-06 14:51:24 -0700170 if(!is_user && (address >= start_vm) && (address < end_vm)){
171 flush_tlb_kernel_vm();
172 return 0;
173 }
Jeff Dike377fad32007-05-06 14:51:25 -0700174 else if(current->mm == NULL) {
175 show_regs(container_of(regs, struct pt_regs, regs));
176 panic("Segfault with no mm");
177 }
Paolo 'Blaisorblade' Giarrusso546fe1c2005-09-22 21:44:16 -0700178
Paolo 'Blaisorblade' Giarrussobe662a12005-09-30 11:58:59 -0700179 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
Paolo 'Blaisorblade' Giarrusso546fe1c2005-09-22 21:44:16 -0700180 err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
181 else {
182 err = -EFAULT;
183 /* A thread accessed NULL, we get a fault, but CR2 is invalid.
184 * This code is used in __do_copy_from_user() of TT mode. */
185 address = 0;
186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 catcher = current->thread.fault_catcher;
189 if(!err)
Jeff Dike5d864562007-05-06 14:51:24 -0700190 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 else if(catcher != NULL){
192 current->thread.fault_addr = (void *) address;
193 do_longjmp(catcher, 1);
Jeff Dike1d3468a2006-07-10 04:45:13 -0700194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 else if(current->thread.fault_addr != NULL)
196 panic("fault_addr set but no fault catcher");
Jeff Dike5d864562007-05-06 14:51:24 -0700197 else if(!is_user && arch_fixup(ip, regs))
198 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Jeff Dike377fad32007-05-06 14:51:25 -0700200 if(!is_user) {
201 show_regs(container_of(regs, struct pt_regs, regs));
Jeff Dike1d3468a2006-07-10 04:45:13 -0700202 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 address, ip);
Jeff Dike377fad32007-05-06 14:51:25 -0700204 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -0700206 if (err == -EACCES) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 si.si_signo = SIGBUS;
208 si.si_errno = 0;
209 si.si_code = BUS_ADRERR;
Al Viro4d338e12006-03-31 02:30:15 -0800210 si.si_addr = (void __user *)address;
Jeff Dike5d864562007-05-06 14:51:24 -0700211 current->thread.arch.faultinfo = fi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 force_sig_info(SIGBUS, &si, current);
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -0700213 } else if (err == -ENOMEM) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 printk("VM: killing process %s\n", current->comm);
215 do_exit(SIGKILL);
Paolo 'Blaisorblade' Giarrusso3b521662005-09-03 15:57:26 -0700216 } else {
217 BUG_ON(err != -EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 si.si_signo = SIGSEGV;
Al Viro4d338e12006-03-31 02:30:15 -0800219 si.si_addr = (void __user *) address;
Jeff Dike5d864562007-05-06 14:51:24 -0700220 current->thread.arch.faultinfo = fi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 force_sig_info(SIGSEGV, &si, current);
222 }
Jeff Dike5d864562007-05-06 14:51:24 -0700223 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226void relay_signal(int sig, union uml_pt_regs *regs)
227{
Jeff Dike6edf4282006-09-25 23:33:03 -0700228 if(arch_handle_signal(sig, regs))
229 return;
230
231 if(!UPT_IS_USER(regs)){
232 if(sig == SIGBUS)
Jeff Dike83ff7df2007-05-06 14:51:50 -0700233 printk("Bus error - the host /dev/shm or /tmp mount "
234 "likely just ran out of space\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 panic("Kernel mode signal %d", sig);
Jeff Dike6edf4282006-09-25 23:33:03 -0700236 }
237
Jeff Dike5d864562007-05-06 14:51:24 -0700238 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 force_sig(sig, current);
240}
241
Jeff Dike27aa6ef2007-02-10 01:44:14 -0800242static void bus_handler(int sig, union uml_pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
244 if(current->thread.fault_catcher != NULL)
245 do_longjmp(current->thread.fault_catcher, 1);
246 else relay_signal(sig, regs);
247}
248
Jeff Dike27aa6ef2007-02-10 01:44:14 -0800249static void winch(int sig, union uml_pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 do_IRQ(WINCH_IRQ, regs);
252}
253
Jeff Dike53dd2b52006-09-27 01:50:37 -0700254const struct kern_handlers handlinfo_kern = {
255 .relay_signal = relay_signal,
256 .winch = winch,
257 .bus_handler = bus_handler,
258 .page_fault = segv_handler,
259 .sigio_handler = sigio_handler,
260 .timer_handler = timer_handler
261};
262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263void trap_init(void)
264{
265}