Prefetch inodes in readdir
[pohmelfs.git] / arch / cris / mm / fault.c
blob9dcac8ec8fa0b8837f7faa9a08b31ccd5a256716
1 /*
2 * arch/cris/mm/fault.c
4 * Copyright (C) 2000-2010 Axis Communications AB
5 */
7 #include <linux/mm.h>
8 #include <linux/interrupt.h>
9 #include <linux/module.h>
10 #include <linux/wait.h>
11 #include <asm/uaccess.h>
13 extern int find_fixup_code(struct pt_regs *);
14 extern void die_if_kernel(const char *, struct pt_regs *, long);
15 extern void show_registers(struct pt_regs *regs);
17 /* debug of low-level TLB reload */
18 #undef DEBUG
20 #ifdef DEBUG
21 #define D(x) x
22 #else
23 #define D(x)
24 #endif
26 /* debug of higher-level faults */
27 #define DPG(x)
29 /* current active page directory */
31 DEFINE_PER_CPU(pgd_t *, current_pgd);
32 unsigned long cris_signal_return_page;
35 * This routine handles page faults. It determines the address,
36 * and the problem, and then passes it off to one of the appropriate
37 * routines.
39 * Notice that the address we're given is aligned to the page the fault
40 * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
41 * address.
43 * error_code:
44 * bit 0 == 0 means no page found, 1 means protection fault
45 * bit 1 == 0 means read, 1 means write
47 * If this routine detects a bad access, it returns 1, otherwise it
48 * returns 0.
51 asmlinkage void
52 do_page_fault(unsigned long address, struct pt_regs *regs,
53 int protection, int writeaccess)
55 struct task_struct *tsk;
56 struct mm_struct *mm;
57 struct vm_area_struct * vma;
58 siginfo_t info;
59 int fault;
61 D(printk(KERN_DEBUG
62 "Page fault for %lX on %X at %lX, prot %d write %d\n",
63 address, smp_processor_id(), instruction_pointer(regs),
64 protection, writeaccess));
66 tsk = current;
69 * We fault-in kernel-space virtual memory on-demand. The
70 * 'reference' page table is init_mm.pgd.
72 * NOTE! We MUST NOT take any locks for this case. We may
73 * be in an interrupt or a critical region, and should
74 * only copy the information from the master page table,
75 * nothing more.
77 * NOTE2: This is done so that, when updating the vmalloc
78 * mappings we don't have to walk all processes pgdirs and
79 * add the high mappings all at once. Instead we do it as they
80 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
81 * bit set so sometimes the TLB can use a lingering entry.
83 * This verifies that the fault happens in kernel space
84 * and that the fault was not a protection error (error_code & 1).
87 if (address >= VMALLOC_START &&
88 !protection &&
89 !user_mode(regs))
90 goto vmalloc_fault;
92 /* When stack execution is not allowed we store the signal
93 * trampolines in the reserved cris_signal_return_page.
94 * Handle this in the exact same way as vmalloc (we know
95 * that the mapping is there and is valid so no need to
96 * call handle_mm_fault).
98 if (cris_signal_return_page &&
99 address == cris_signal_return_page &&
100 !protection && user_mode(regs))
101 goto vmalloc_fault;
103 /* we can and should enable interrupts at this point */
104 local_irq_enable();
106 mm = tsk->mm;
107 info.si_code = SEGV_MAPERR;
110 * If we're in an interrupt or "atomic" operation or have no
111 * user context, we must not take the fault.
114 if (in_atomic() || !mm)
115 goto no_context;
117 down_read(&mm->mmap_sem);
118 vma = find_vma(mm, address);
119 if (!vma)
120 goto bad_area;
121 if (vma->vm_start <= address)
122 goto good_area;
123 if (!(vma->vm_flags & VM_GROWSDOWN))
124 goto bad_area;
125 if (user_mode(regs)) {
127 * accessing the stack below usp is always a bug.
128 * we get page-aligned addresses so we can only check
129 * if we're within a page from usp, but that might be
130 * enough to catch brutal errors at least.
132 if (address + PAGE_SIZE < rdusp())
133 goto bad_area;
135 if (expand_stack(vma, address))
136 goto bad_area;
139 * Ok, we have a good vm_area for this memory access, so
140 * we can handle it..
143 good_area:
144 info.si_code = SEGV_ACCERR;
146 /* first do some preliminary protection checks */
148 if (writeaccess == 2){
149 if (!(vma->vm_flags & VM_EXEC))
150 goto bad_area;
151 } else if (writeaccess == 1) {
152 if (!(vma->vm_flags & VM_WRITE))
153 goto bad_area;
154 } else {
155 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
156 goto bad_area;
160 * If for any reason at all we couldn't handle the fault,
161 * make sure we exit gracefully rather than endlessly redo
162 * the fault.
165 fault = handle_mm_fault(mm, vma, address, (writeaccess & 1) ? FAULT_FLAG_WRITE : 0);
166 if (unlikely(fault & VM_FAULT_ERROR)) {
167 if (fault & VM_FAULT_OOM)
168 goto out_of_memory;
169 else if (fault & VM_FAULT_SIGBUS)
170 goto do_sigbus;
171 BUG();
173 if (fault & VM_FAULT_MAJOR)
174 tsk->maj_flt++;
175 else
176 tsk->min_flt++;
178 up_read(&mm->mmap_sem);
179 return;
182 * Something tried to access memory that isn't in our memory map..
183 * Fix it, but check if it's kernel or user first..
186 bad_area:
187 up_read(&mm->mmap_sem);
189 bad_area_nosemaphore:
190 DPG(show_registers(regs));
192 /* User mode accesses just cause a SIGSEGV */
194 if (user_mode(regs)) {
195 printk(KERN_NOTICE "%s (pid %d) segfaults for page "
196 "address %08lx at pc %08lx\n",
197 tsk->comm, tsk->pid,
198 address, instruction_pointer(regs));
200 /* With DPG on, we've already dumped registers above. */
201 DPG(if (0))
202 show_registers(regs);
204 #ifdef CONFIG_NO_SEGFAULT_TERMINATION
205 DECLARE_WAIT_QUEUE_HEAD(wq);
206 wait_event_interruptible(wq, 0 == 1);
207 #else
208 info.si_signo = SIGSEGV;
209 info.si_errno = 0;
210 /* info.si_code has been set above */
211 info.si_addr = (void *)address;
212 force_sig_info(SIGSEGV, &info, tsk);
213 #endif
214 return;
217 no_context:
219 /* Are we prepared to handle this kernel fault?
221 * (The kernel has valid exception-points in the source
222 * when it accesses user-memory. When it fails in one
223 * of those points, we find it in a table and do a jump
224 * to some fixup code that loads an appropriate error
225 * code)
228 if (find_fixup_code(regs))
229 return;
232 * Oops. The kernel tried to access some bad page. We'll have to
233 * terminate things with extreme prejudice.
236 if (!oops_in_progress) {
237 oops_in_progress = 1;
238 if ((unsigned long) (address) < PAGE_SIZE)
239 printk(KERN_ALERT "Unable to handle kernel NULL "
240 "pointer dereference");
241 else
242 printk(KERN_ALERT "Unable to handle kernel access"
243 " at virtual address %08lx\n", address);
245 die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
246 oops_in_progress = 0;
249 do_exit(SIGKILL);
252 * We ran out of memory, or some other thing happened to us that made
253 * us unable to handle the page fault gracefully.
256 out_of_memory:
257 up_read(&mm->mmap_sem);
258 if (!user_mode(regs))
259 goto no_context;
260 pagefault_out_of_memory();
261 return;
263 do_sigbus:
264 up_read(&mm->mmap_sem);
267 * Send a sigbus, regardless of whether we were in kernel
268 * or user mode.
270 info.si_signo = SIGBUS;
271 info.si_errno = 0;
272 info.si_code = BUS_ADRERR;
273 info.si_addr = (void *)address;
274 force_sig_info(SIGBUS, &info, tsk);
276 /* Kernel mode? Handle exceptions or die */
277 if (!user_mode(regs))
278 goto no_context;
279 return;
281 vmalloc_fault:
284 * Synchronize this task's top level page-table
285 * with the 'reference' page table.
287 * Use current_pgd instead of tsk->active_mm->pgd
288 * since the latter might be unavailable if this
289 * code is executed in a misfortunately run irq
290 * (like inside schedule() between switch_mm and
291 * switch_to...).
294 int offset = pgd_index(address);
295 pgd_t *pgd, *pgd_k;
296 pud_t *pud, *pud_k;
297 pmd_t *pmd, *pmd_k;
298 pte_t *pte_k;
300 pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
301 pgd_k = init_mm.pgd + offset;
303 /* Since we're two-level, we don't need to do both
304 * set_pgd and set_pmd (they do the same thing). If
305 * we go three-level at some point, do the right thing
306 * with pgd_present and set_pgd here.
308 * Also, since the vmalloc area is global, we don't
309 * need to copy individual PTE's, it is enough to
310 * copy the pgd pointer into the pte page of the
311 * root task. If that is there, we'll find our pte if
312 * it exists.
315 pud = pud_offset(pgd, address);
316 pud_k = pud_offset(pgd_k, address);
317 if (!pud_present(*pud_k))
318 goto no_context;
320 pmd = pmd_offset(pud, address);
321 pmd_k = pmd_offset(pud_k, address);
323 if (!pmd_present(*pmd_k))
324 goto bad_area_nosemaphore;
326 set_pmd(pmd, *pmd_k);
328 /* Make sure the actual PTE exists as well to
329 * catch kernel vmalloc-area accesses to non-mapped
330 * addresses. If we don't do this, this will just
331 * silently loop forever.
334 pte_k = pte_offset_kernel(pmd_k, address);
335 if (!pte_present(*pte_k))
336 goto no_context;
338 return;
342 /* Find fixup code. */
344 find_fixup_code(struct pt_regs *regs)
346 const struct exception_table_entry *fixup;
347 /* in case of delay slot fault (v32) */
348 unsigned long ip = (instruction_pointer(regs) & ~0x1);
350 fixup = search_exception_tables(ip);
351 if (fixup != 0) {
352 /* Adjust the instruction pointer in the stackframe. */
353 instruction_pointer(regs) = fixup->fixup;
354 arch_fixup(regs);
355 return 1;
358 return 0;