2
* MMU fault handling support.
4
* Copyright (C) 1998-2002 Hewlett-Packard Co
5
* David Mosberger-Tang <davidm@hpl.hp.com>
7
#include <linux/sched.h>
8
#include <linux/kernel.h>
10
#include <linux/interrupt.h>
11
#include <linux/kprobes.h>
12
#include <linux/kdebug.h>
13
#include <linux/prefetch.h>
15
#include <asm/pgtable.h>
16
#include <asm/processor.h>
17
#include <asm/system.h>
18
#include <asm/uaccess.h>
20
extern int die(char *, struct pt_regs *, long);
23
static inline int notify_page_fault(struct pt_regs *regs, int trap)
27
if (!user_mode(regs)) {
28
/* kprobe_running() needs smp_processor_id() */
30
if (kprobe_running() && kprobe_fault_handler(regs, trap))
38
static inline int notify_page_fault(struct pt_regs *regs, int trap)
45
* Return TRUE if ADDRESS points at a page in the kernel's mapped segment
46
* (inside region 5, on ia64) and that page is present.
49
mapped_kernel_page_is_present (unsigned long address)
56
pgd = pgd_offset_k(address);
57
if (pgd_none(*pgd) || pgd_bad(*pgd))
60
pud = pud_offset(pgd, address);
61
if (pud_none(*pud) || pud_bad(*pud))
64
pmd = pmd_offset(pud, address);
65
if (pmd_none(*pmd) || pmd_bad(*pmd))
68
ptep = pte_offset_kernel(pmd, address);
73
return pte_present(pte);
77
ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs)
79
int signal = SIGSEGV, code = SEGV_MAPERR;
80
struct vm_area_struct *vma, *prev_vma;
81
struct mm_struct *mm = current->mm;
86
/* mmap_sem is performance critical.... */
87
prefetchw(&mm->mmap_sem);
90
* If we're in an interrupt or have no user context, we must not take the fault..
92
if (in_atomic() || !mm)
95
#ifdef CONFIG_VIRTUAL_MEM_MAP
97
* If fault is in region 5 and we are in the kernel, we may already
98
* have the mmap_sem (pfn_valid macro is called during mmap). There
99
* is no vma for region 5 addr's anyway, so skip getting the semaphore
100
* and go directly to the exception handling code.
103
if ((REGION_NUMBER(address) == 5) && !user_mode(regs))
108
* This is to handle the kprobes on user space access instructions
110
if (notify_page_fault(regs, TRAP_BRKPT))
113
down_read(&mm->mmap_sem);
115
vma = find_vma_prev(mm, address, &prev_vma);
116
if (!vma && !prev_vma )
120
* find_vma_prev() returns vma such that address < vma->vm_end or NULL
122
* May find no vma, but could be that the last vm area is the
123
* register backing store that needs to expand upwards, in
124
* this case vma will be null, but prev_vma will ne non-null
126
if (( !vma && prev_vma ) || (address < vma->vm_start) )
127
goto check_expansion;
132
/* OK, we've got a good vm_area for this memory area. Check the access permissions: */
134
# define VM_READ_BIT 0
135
# define VM_WRITE_BIT 1
136
# define VM_EXEC_BIT 2
138
# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \
139
|| (1 << VM_EXEC_BIT) != VM_EXEC)
140
# error File is out of sync with <linux/mm.h>. Please update.
143
if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE))))
146
mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT)
147
| (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT));
149
if ((vma->vm_flags & mask) != mask)
153
* If for any reason at all we couldn't handle the fault, make
154
* sure we exit gracefully rather than endlessly redo the
157
fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0);
158
if (unlikely(fault & VM_FAULT_ERROR)) {
160
* We ran out of memory, or some other thing happened
161
* to us that made us unable to handle the page fault
164
if (fault & VM_FAULT_OOM) {
166
} else if (fault & VM_FAULT_SIGBUS) {
172
if (fault & VM_FAULT_MAJOR)
176
up_read(&mm->mmap_sem);
180
if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) {
183
if (!(vma->vm_flags & VM_GROWSDOWN))
185
if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
186
|| REGION_OFFSET(address) >= RGN_MAP_LIMIT)
188
if (expand_stack(vma, address))
192
if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)
193
|| REGION_OFFSET(address) >= RGN_MAP_LIMIT)
196
* Since the register backing store is accessed sequentially,
197
* we disallow growing it by more than a page at a time.
199
if (address > vma->vm_end + PAGE_SIZE - sizeof(long))
201
if (expand_upwards(vma, address))
207
up_read(&mm->mmap_sem);
208
#ifdef CONFIG_VIRTUAL_MEM_MAP
211
if ((isr & IA64_ISR_SP)
212
|| ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
215
* This fault was due to a speculative load or lfetch.fault, set the "ed"
216
* bit in the psr to ensure forward progress. (Target register will get a
217
* NaT for ld.s, lfetch will be canceled.)
219
ia64_psr(regs)->ed = 1;
222
if (user_mode(regs)) {
223
si.si_signo = signal;
226
si.si_addr = (void __user *) address;
228
si.si_flags = __ISR_VALID;
229
force_sig_info(signal, &si, current);
234
if ((isr & IA64_ISR_SP)
235
|| ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH))
238
* This fault was due to a speculative load or lfetch.fault, set the "ed"
239
* bit in the psr to ensure forward progress. (Target register will get a
240
* NaT for ld.s, lfetch will be canceled.)
242
ia64_psr(regs)->ed = 1;
247
* Since we have no vma's for region 5, we might get here even if the address is
248
* valid, due to the VHPT walker inserting a non present translation that becomes
249
* stale. If that happens, the non present fault handler already purged the stale
250
* translation, which fixed the problem. So, we check to see if the translation is
251
* valid, and return if it is.
253
if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address))
256
if (ia64_done_with_exception(regs))
260
* Oops. The kernel tried to access some bad page. We'll have to terminate things
261
* with extreme prejudice.
265
if (address < PAGE_SIZE)
266
printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address);
268
printk(KERN_ALERT "Unable to handle kernel paging request at "
269
"virtual address %016lx\n", address);
270
if (die("Oops", regs, isr))
278
up_read(&mm->mmap_sem);
279
if (!user_mode(regs))
281
pagefault_out_of_memory();