4 * Linux architectural port borrowing liberally from similar works of
5 * others. All original copyrights apply as per the original source
8 * Modifications for the OpenRISC architecture:
9 * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
10 * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/sched.h>
23 #include <asm/uaccess.h>
24 #include <asm/siginfo.h>
25 #include <asm/signal.h>
27 #define NUM_TLB_ENTRIES 64
28 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
30 unsigned long pte_misses; /* updated by do_page_fault() */
31 unsigned long pte_errors; /* updated by do_page_fault() */
33 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
34 * - also look into include/asm-or32/mmu_context.h
36 volatile pgd_t *current_pgd;
38 extern void die(char *, struct pt_regs *, long);
41 * This routine handles page faults. It determines the address,
42 * and the problem, and then passes it off to one of the appropriate
45 * If this routine detects a bad access, it returns 1, otherwise it
49 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
50 unsigned long vector, int write_acc)
52 struct task_struct *tsk;
54 struct vm_area_struct *vma;
61 * We fault-in kernel-space virtual memory on-demand. The
62 * 'reference' page table is init_mm.pgd.
64 * NOTE! We MUST NOT take any locks for this case. We may
65 * be in an interrupt or a critical region, and should
66 * only copy the information from the master page table,
69 * NOTE2: This is done so that, when updating the vmalloc
70 * mappings we don't have to walk all processes pgdirs and
71 * add the high mappings all at once. Instead we do it as they
72 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
73 * bit set so sometimes the TLB can use a lingering entry.
75 * This verifies that the fault happens in kernel space
76 * and that the fault was not a protection error.
79 if (address >= VMALLOC_START &&
80 (vector != 0x300 && vector != 0x400) &&
84 /* If exceptions were enabled, we can reenable them here */
85 if (user_mode(regs)) {
86 /* Exception was in userspace: reenable interrupts */
89 /* If exception was in a syscall, then IRQ's may have
90 * been enabled or disabled. If they were enabled,
93 if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
98 info.si_code = SEGV_MAPERR;
101 * If we're in an interrupt or have no user
102 * context, we must not take the fault..
105 if (in_interrupt() || !mm)
108 down_read(&mm->mmap_sem);
109 vma = find_vma(mm, address);
114 if (vma->vm_start <= address)
117 if (!(vma->vm_flags & VM_GROWSDOWN))
120 if (user_mode(regs)) {
122 * accessing the stack below usp is always a bug.
123 * we get page-aligned addresses so we can only check
124 * if we're within a page from usp, but that might be
125 * enough to catch brutal errors at least.
127 if (address + PAGE_SIZE < regs->sp)
130 if (expand_stack(vma, address))
134 * Ok, we have a good vm_area for this memory access, so
139 info.si_code = SEGV_ACCERR;
141 /* first do some preliminary protection checks */
144 if (!(vma->vm_flags & VM_WRITE))
148 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
152 /* are we trying to execute nonexecutable area */
153 if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
157 * If for any reason at all we couldn't handle the fault,
158 * make sure we exit gracefully rather than endlessly redo
162 fault = handle_mm_fault(mm, vma, address, write_acc);
163 if (unlikely(fault & VM_FAULT_ERROR)) {
164 if (fault & VM_FAULT_OOM)
166 else if (fault & VM_FAULT_SIGSEGV)
168 else if (fault & VM_FAULT_SIGBUS)
172 /*RGD modeled on Cris */
173 if (fault & VM_FAULT_MAJOR)
178 up_read(&mm->mmap_sem);
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..
187 up_read(&mm->mmap_sem);
189 bad_area_nosemaphore:
191 /* User mode accesses just cause a SIGSEGV */
193 if (user_mode(regs)) {
194 info.si_signo = SIGSEGV;
196 /* info.si_code has been set above */
197 info.si_addr = (void *)address;
198 force_sig_info(SIGSEGV, &info, tsk);
204 /* Are we prepared to handle this kernel fault?
206 * (The kernel has valid exception-points in the source
207 * when it acesses user-memory. When it fails in one
208 * of those points, we find it in a table and do a jump
209 * to some fixup code that loads an appropriate error
214 const struct exception_table_entry *entry;
216 __asm__ __volatile__("l.nop 42");
218 if ((entry = search_exception_tables(regs->pc)) != NULL) {
219 /* Adjust the instruction pointer in the stackframe */
220 regs->pc = entry->fixup;
226 * Oops. The kernel tried to access some bad page. We'll have to
227 * terminate things with extreme prejudice.
230 if ((unsigned long)(address) < PAGE_SIZE)
232 "Unable to handle kernel NULL pointer dereference");
234 printk(KERN_ALERT "Unable to handle kernel access");
235 printk(" at virtual address 0x%08lx\n", address);
237 die("Oops", regs, write_acc);
242 * We ran out of memory, or some other thing happened to us that made
243 * us unable to handle the page fault gracefully.
247 __asm__ __volatile__("l.nop 42");
248 __asm__ __volatile__("l.nop 1");
250 up_read(&mm->mmap_sem);
251 printk("VM: killing process %s\n", tsk->comm);
257 up_read(&mm->mmap_sem);
260 * Send a sigbus, regardless of whether we were in kernel
263 info.si_signo = SIGBUS;
265 info.si_code = BUS_ADRERR;
266 info.si_addr = (void *)address;
267 force_sig_info(SIGBUS, &info, tsk);
269 /* Kernel mode? Handle exceptions or die */
270 if (!user_mode(regs))
277 * Synchronize this task's top level page-table
278 * with the 'reference' page table.
280 * Use current_pgd instead of tsk->active_mm->pgd
281 * since the latter might be unavailable if this
282 * code is executed in a misfortunately run irq
283 * (like inside schedule() between switch_mm and
287 int offset = pgd_index(address);
294 phx_warn("do_page_fault(): vmalloc_fault will not work, "
295 "since current_pgd assign a proper value somewhere\n"
296 "anyhow we don't need this at the moment\n");
298 phx_mmu("vmalloc_fault");
300 pgd = (pgd_t *)current_pgd + 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
315 pud = pud_offset(pgd, address);
316 pud_k = pud_offset(pgd_k, address);
317 if (!pud_present(*pud_k))
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))