3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5 * Derived from "arch/i386/mm/fault.c"
6 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Modified by Cort Dougan and Paul Mackerras.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #include <linux/config.h>
17 #include <linux/signal.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/errno.h>
21 #include <linux/string.h>
22 #include <linux/types.h>
23 #include <linux/ptrace.h>
24 #include <linux/mman.h>
26 #include <linux/interrupt.h>
27 #include <linux/highmem.h>
28 #include <linux/module.h>
31 #include <asm/pgtable.h>
33 #include <asm/mmu_context.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
36 #include <asm/tlbflush.h>
38 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
39 extern void (*debugger)(struct pt_regs *);
40 extern void (*debugger_fault_handler)(struct pt_regs *);
41 extern int (*debugger_dabr_match)(struct pt_regs *);
42 int debugger_kernel_faults = 1;
45 unsigned long htab_reloads; /* updated by hashtable.S:hash_page() */
46 unsigned long htab_evicts; /* updated by hashtable.S:hash_page() */
47 unsigned long htab_preloads; /* updated by hashtable.S:add_hash_page() */
48 unsigned long pte_misses; /* updated by do_page_fault() */
49 unsigned long pte_errors; /* updated by do_page_fault() */
50 unsigned int probingmem;
53 * Check whether the instruction at regs->nip is a store using
54 * an update addressing form which will update r1.
56 static int store_updates_sp(struct pt_regs *regs)
60 if (get_user(inst, (unsigned int __user *)regs->nip))
62 /* check for 1 in the rA field */
63 if (((inst >> 16) & 0x1f) != 1)
65 /* check major opcode */
74 /* check minor opcode */
75 switch ((inst >> 1) & 0x3ff) {
79 case 695: /* stfsux */
80 case 759: /* stfdux */
88 * For 600- and 800-family processors, the error_code parameter is DSISR
89 * for a data fault, SRR1 for an instruction fault. For 400-family processors
90 * the error_code parameter is ESR for a data fault, 0 for an instruction
93 int do_page_fault(struct pt_regs *regs, unsigned long address,
94 unsigned long error_code)
96 struct vm_area_struct * vma;
97 struct mm_struct *mm = current->mm;
99 int code = SEGV_MAPERR;
100 #if defined(CONFIG_4xx) || defined (CONFIG_BOOKE)
101 int is_write = error_code & ESR_DST;
106 * Fortunately the bit assignments in SRR1 for an instruction
107 * fault and DSISR for a data fault are mostly the same for the
108 * bits we are interested in. But there are some bits which
109 * indicate errors in DSISR but can validly be set in SRR1.
111 if (TRAP(regs) == 0x400)
112 error_code &= 0x48200000;
114 is_write = error_code & 0x02000000;
115 #endif /* CONFIG_4xx || CONFIG_BOOKE */
117 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
118 if (debugger_fault_handler && TRAP(regs) == 0x300) {
119 debugger_fault_handler(regs);
122 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
123 if (error_code & 0x00400000) {
125 if (debugger_dabr_match(regs))
128 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
129 #endif /* CONFIG_XMON || CONFIG_KGDB */
131 if (in_atomic() || mm == NULL)
134 down_read(&mm->mmap_sem);
135 vma = find_vma(mm, address);
138 if (vma->vm_start <= address)
140 if (!(vma->vm_flags & VM_GROWSDOWN))
146 * N.B. The rs6000/xcoff ABI allows programs to access up to
147 * a few hundred bytes below the stack pointer.
148 * The kernel signal delivery code writes up to about 1.5kB
149 * below the stack pointer (r1) before decrementing it.
150 * The exec code can write slightly over 640kB to the stack
151 * before setting the user r1. Thus we allow the stack to
152 * expand to 1MB without further checks.
154 if (address + 0x100000 < vma->vm_end) {
155 /* get user regs even if this fault is in kernel mode */
156 struct pt_regs *uregs = current->thread.regs;
161 * A user-mode access to an address a long way below
162 * the stack pointer is only valid if the instruction
163 * is one which would update the stack pointer to the
164 * address accessed if the instruction completed,
165 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
166 * (or the byte, halfword, float or double forms).
168 * If we don't check this then any write to the area
169 * between the last mapped region and the stack will
170 * expand the stack rather than segfaulting.
172 if (address + 2048 < uregs->gpr[1]
173 && (!user_mode(regs) || !store_updates_sp(regs)))
176 if (expand_stack(vma, address))
181 #if defined(CONFIG_6xx)
182 if (error_code & 0x95700000)
183 /* an error such as lwarx to I/O controller space,
184 address matching DABR, eciwx, etc. */
186 #endif /* CONFIG_6xx */
187 #if defined(CONFIG_8xx)
188 /* The MPC8xx seems to always set 0x80000000, which is
189 * "undefined". Of those that can be set, this is the only
190 * one which seems bad.
192 if (error_code & 0x10000000)
193 /* Guarded storage error. */
195 #endif /* CONFIG_8xx */
199 if (!(vma->vm_flags & VM_WRITE))
201 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
202 /* an exec - 4xx/Book-E allows for per-page execute permission */
203 } else if (TRAP(regs) == 0x400) {
207 /* It would be nice to actually enforce the VM execute
208 permission on CPUs which can do so, but far too
209 much stuff in userspace doesn't get the permissions
210 right, so we let any page be executed for now. */
211 if (! (vma->vm_flags & VM_EXEC))
215 /* Since 4xx/Book-E supports per-page execute permission,
216 * we lazily flush dcache to icache. */
218 if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
219 struct page *page = pte_page(*ptep);
221 if (! test_bit(PG_arch_1, &page->flags)) {
222 flush_dcache_icache_page(page);
223 set_bit(PG_arch_1, &page->flags);
225 pte_update(ptep, 0, _PAGE_HWEXEC);
228 up_read(&mm->mmap_sem);
236 /* protection fault */
237 if (error_code & 0x08000000)
239 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
244 * If for any reason at all we couldn't handle the fault,
245 * make sure we exit gracefully rather than endlessly redo
249 switch (handle_mm_fault(mm, vma, address, is_write)) {
256 case VM_FAULT_SIGBUS:
264 up_read(&mm->mmap_sem);
266 * keep track of tlb+htab misses that are good addrs but
267 * just need pte's created via handle_mm_fault()
274 up_read(&mm->mmap_sem);
277 /* User mode accesses cause a SIGSEGV */
278 if (user_mode(regs)) {
279 _exception(SIGSEGV, regs, code, address);
286 * We ran out of memory, or some other thing happened to us that made
287 * us unable to handle the page fault gracefully.
290 up_read(&mm->mmap_sem);
291 if (current->pid == 1) {
293 down_read(&mm->mmap_sem);
296 printk("VM: killing process %s\n", current->comm);
302 up_read(&mm->mmap_sem);
303 info.si_signo = SIGBUS;
305 info.si_code = BUS_ADRERR;
306 info.si_addr = (void __user *)address;
307 force_sig_info (SIGBUS, &info, current);
308 if (!user_mode(regs))
314 * bad_page_fault is called when we have a bad access from the kernel.
315 * It is called from the DSI and ISI handlers in head.S and from some
316 * of the procedures in traps.c.
319 bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
321 const struct exception_table_entry *entry;
323 /* Are we prepared to handle this fault? */
324 if ((entry = search_exception_tables(regs->nip)) != NULL) {
325 regs->nip = entry->fixup;
329 /* kernel has accessed a bad area */
330 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
331 if (debugger_kernel_faults)
334 die("kernel access of bad area", regs, sig);
339 /* The pgtable.h claims some functions generically exist, but I
340 * can't find them......
342 pte_t *va_to_pte(unsigned long address)
348 if (address < TASK_SIZE)
351 dir = pgd_offset(&init_mm, address);
353 pmd = pmd_offset(dir, address & PAGE_MASK);
354 if (pmd && pmd_present(*pmd)) {
355 pte = pte_offset_kernel(pmd, address & PAGE_MASK);
356 if (pte && pte_present(*pte))
363 unsigned long va_to_phys(unsigned long address)
367 pte = va_to_pte(address);
369 return(((unsigned long)(pte_val(*pte)) & PAGE_MASK) | (address & ~(PAGE_MASK)));
374 print_8xx_pte(struct mm_struct *mm, unsigned long addr)
380 printk(" pte @ 0x%8lx: ", addr);
381 pgd = pgd_offset(mm, addr & PAGE_MASK);
383 pmd = pmd_offset(pgd, addr & PAGE_MASK);
384 if (pmd && pmd_present(*pmd)) {
385 pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
387 printk(" (0x%08lx)->(0x%08lx)->0x%08lx\n",
388 (long)pgd, (long)pte, (long)pte_val(*pte));
389 #define pp ((long)pte_val(*pte))
390 printk(" RPN: %05lx PP: %lx SPS: %lx SH: %lx "
394 (pp>>3)&1, /* small */
395 (pp>>2)&1, /* shared */
396 (pp>>1)&1, /* cache inhibit */
415 get_8xx_pte(struct mm_struct *mm, unsigned long addr)
422 pgd = pgd_offset(mm, addr & PAGE_MASK);
424 pmd = pmd_offset(pgd, addr & PAGE_MASK);
425 if (pmd && pmd_present(*pmd)) {
426 pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
428 retval = (int)pte_val(*pte);
434 #endif /* CONFIG_8xx */