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 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
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.
18 #include <linux/signal.h>
19 #include <linux/sched.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/string.h>
23 #include <linux/types.h>
24 #include <linux/ptrace.h>
25 #include <linux/mman.h>
27 #include <linux/interrupt.h>
28 #include <linux/highmem.h>
29 #include <linux/module.h>
30 #include <linux/kprobes.h>
31 #include <linux/kdebug.h>
32 #include <linux/perf_event.h>
33 #include <linux/ratelimit.h>
34 #include <linux/context_tracking.h>
36 #include <asm/firmware.h>
38 #include <asm/pgtable.h>
40 #include <asm/mmu_context.h>
41 #include <asm/uaccess.h>
42 #include <asm/tlbflush.h>
43 #include <asm/siginfo.h>
44 #include <asm/debug.h>
45 #include <mm/mmu_decl.h>
50 static inline int notify_page_fault(struct pt_regs *regs)
54 /* kprobe_running() needs smp_processor_id() */
55 if (!user_mode(regs)) {
57 if (kprobe_running() && kprobe_fault_handler(regs, 11))
65 static inline int notify_page_fault(struct pt_regs *regs)
72 * Check whether the instruction at regs->nip is a store using
73 * an update addressing form which will update r1.
75 static int store_updates_sp(struct pt_regs *regs)
79 if (get_user(inst, (unsigned int __user *)regs->nip))
81 /* check for 1 in the rA field */
82 if (((inst >> 16) & 0x1f) != 1)
84 /* check major opcode */
92 case 62: /* std or stdu */
93 return (inst & 3) == 1;
95 /* check minor opcode */
96 switch ((inst >> 1) & 0x3ff) {
100 case 439: /* sthux */
101 case 695: /* stfsux */
102 case 759: /* stfdux */
109 * do_page_fault error handling helpers
112 #define MM_FAULT_RETURN 0
113 #define MM_FAULT_CONTINUE -1
114 #define MM_FAULT_ERR(sig) (sig)
116 static int do_sigbus(struct pt_regs *regs, unsigned long address)
120 up_read(¤t->mm->mmap_sem);
122 if (user_mode(regs)) {
123 current->thread.trap_nr = BUS_ADRERR;
124 info.si_signo = SIGBUS;
126 info.si_code = BUS_ADRERR;
127 info.si_addr = (void __user *)address;
128 force_sig_info(SIGBUS, &info, current);
129 return MM_FAULT_RETURN;
131 return MM_FAULT_ERR(SIGBUS);
134 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
137 * Pagefault was interrupted by SIGKILL. We have no reason to
138 * continue the pagefault.
140 if (fatal_signal_pending(current)) {
142 * If we have retry set, the mmap semaphore will have
143 * alrady been released in __lock_page_or_retry(). Else
146 if (!(fault & VM_FAULT_RETRY))
147 up_read(¤t->mm->mmap_sem);
148 /* Coming from kernel, we need to deal with uaccess fixups */
150 return MM_FAULT_RETURN;
151 return MM_FAULT_ERR(SIGKILL);
154 /* No fault: be happy */
155 if (!(fault & VM_FAULT_ERROR))
156 return MM_FAULT_CONTINUE;
159 if (fault & VM_FAULT_OOM) {
160 up_read(¤t->mm->mmap_sem);
163 * We ran out of memory, or some other thing happened to us that
164 * made us unable to handle the page fault gracefully.
166 if (!user_mode(regs))
167 return MM_FAULT_ERR(SIGKILL);
168 pagefault_out_of_memory();
169 return MM_FAULT_RETURN;
172 /* Bus error. x86 handles HWPOISON here, we'll add this if/when
173 * we support the feature in HW
175 if (fault & VM_FAULT_SIGBUS)
176 return do_sigbus(regs, addr);
178 /* We don't understand the fault code, this is fatal */
180 return MM_FAULT_CONTINUE;
184 * For 600- and 800-family processors, the error_code parameter is DSISR
185 * for a data fault, SRR1 for an instruction fault. For 400-family processors
186 * the error_code parameter is ESR for a data fault, 0 for an instruction
188 * For 64-bit processors, the error_code parameter is
189 * - DSISR for a non-SLB data access fault,
190 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
193 * The return value is 0 if the fault was handled, or the signal
194 * number if this is a kernel fault that can't be handled here.
196 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
197 unsigned long error_code)
199 enum ctx_state prev_state = exception_enter();
200 struct vm_area_struct * vma;
201 struct mm_struct *mm = current->mm;
202 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
203 int code = SEGV_MAPERR;
205 int trap = TRAP(regs);
206 int is_exec = trap == 0x400;
208 int rc = 0, store_update_sp = 0;
210 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
212 * Fortunately the bit assignments in SRR1 for an instruction
213 * fault and DSISR for a data fault are mostly the same for the
214 * bits we are interested in. But there are some bits which
215 * indicate errors in DSISR but can validly be set in SRR1.
218 error_code &= 0x48200000;
220 is_write = error_code & DSISR_ISSTORE;
222 is_write = error_code & ESR_DST;
223 #endif /* CONFIG_4xx || CONFIG_BOOKE */
225 #ifdef CONFIG_PPC_ICSWX
227 * we need to do this early because this "data storage
228 * interrupt" does not update the DAR/DEAR so we don't want to
231 if (error_code & ICSWX_DSI_UCT) {
232 rc = acop_handle_fault(regs, address, error_code);
236 #endif /* CONFIG_PPC_ICSWX */
238 if (notify_page_fault(regs))
241 if (unlikely(debugger_fault_handler(regs)))
244 /* On a kernel SLB miss we can only check for a valid exception entry */
245 if (!user_mode(regs) && (address >= TASK_SIZE)) {
250 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
251 defined(CONFIG_PPC_BOOK3S_64))
252 if (error_code & DSISR_DABRMATCH) {
253 /* breakpoint match */
254 do_break(regs, address, error_code);
259 /* We restore the interrupt state now */
260 if (!arch_irq_disabled_regs(regs))
263 if (in_atomic() || mm == NULL) {
264 if (!user_mode(regs)) {
268 /* in_atomic() in user mode is really bad,
269 as is current->mm == NULL. */
270 printk(KERN_EMERG "Page fault in user mode with "
271 "in_atomic() = %d mm = %p\n", in_atomic(), mm);
272 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
273 regs->nip, regs->msr);
274 die("Weird page fault", regs, SIGSEGV);
277 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
280 * We want to do this outside mmap_sem, because reading code around nip
281 * can result in fault, which will cause a deadlock when called with
285 store_update_sp = store_updates_sp(regs);
288 flags |= FAULT_FLAG_USER;
290 /* When running in the kernel we expect faults to occur only to
291 * addresses in user space. All other faults represent errors in the
292 * kernel and should generate an OOPS. Unfortunately, in the case of an
293 * erroneous fault occurring in a code path which already holds mmap_sem
294 * we will deadlock attempting to validate the fault against the
295 * address space. Luckily the kernel only validly references user
296 * space from well defined areas of code, which are listed in the
299 * As the vast majority of faults will be valid we will only perform
300 * the source reference check when there is a possibility of a deadlock.
301 * Attempt to lock the address space, if we cannot we then validate the
302 * source. If this is invalid we can skip the address space check,
303 * thus avoiding the deadlock.
305 if (!down_read_trylock(&mm->mmap_sem)) {
306 if (!user_mode(regs) && !search_exception_tables(regs->nip))
307 goto bad_area_nosemaphore;
310 down_read(&mm->mmap_sem);
313 * The above down_read_trylock() might have succeeded in
314 * which case we'll have missed the might_sleep() from
320 vma = find_vma(mm, address);
323 if (vma->vm_start <= address)
325 if (!(vma->vm_flags & VM_GROWSDOWN))
329 * N.B. The POWER/Open ABI allows programs to access up to
330 * 288 bytes below the stack pointer.
331 * The kernel signal delivery code writes up to about 1.5kB
332 * below the stack pointer (r1) before decrementing it.
333 * The exec code can write slightly over 640kB to the stack
334 * before setting the user r1. Thus we allow the stack to
335 * expand to 1MB without further checks.
337 if (address + 0x100000 < vma->vm_end) {
338 /* get user regs even if this fault is in kernel mode */
339 struct pt_regs *uregs = current->thread.regs;
344 * A user-mode access to an address a long way below
345 * the stack pointer is only valid if the instruction
346 * is one which would update the stack pointer to the
347 * address accessed if the instruction completed,
348 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
349 * (or the byte, halfword, float or double forms).
351 * If we don't check this then any write to the area
352 * between the last mapped region and the stack will
353 * expand the stack rather than segfaulting.
355 if (address + 2048 < uregs->gpr[1] && !store_update_sp)
358 if (expand_stack(vma, address))
363 #if defined(CONFIG_6xx)
364 if (error_code & 0x95700000)
365 /* an error such as lwarx to I/O controller space,
366 address matching DABR, eciwx, etc. */
368 #endif /* CONFIG_6xx */
369 #if defined(CONFIG_8xx)
370 /* 8xx sometimes need to load a invalid/non-present TLBs.
371 * These must be invalidated separately as linux mm don't.
373 if (error_code & 0x40000000) /* no translation? */
374 _tlbil_va(address, 0, 0, 0);
376 /* The MPC8xx seems to always set 0x80000000, which is
377 * "undefined". Of those that can be set, this is the only
378 * one which seems bad.
380 if (error_code & 0x10000000)
381 /* Guarded storage error. */
383 #endif /* CONFIG_8xx */
386 #ifdef CONFIG_PPC_STD_MMU
387 /* Protection fault on exec go straight to failure on
388 * Hash based MMUs as they either don't support per-page
389 * execute permission, or if they do, it's handled already
390 * at the hash level. This test would probably have to
391 * be removed if we change the way this works to make hash
392 * processors use the same I/D cache coherency mechanism
395 if (error_code & DSISR_PROTFAULT)
397 #endif /* CONFIG_PPC_STD_MMU */
400 * Allow execution from readable areas if the MMU does not
401 * provide separate controls over reading and executing.
403 * Note: That code used to not be enabled for 4xx/BookE.
404 * It is now as I/D cache coherency for these is done at
405 * set_pte_at() time and I see no reason why the test
406 * below wouldn't be valid on those processors. This -may-
407 * break programs compiled with a really old ABI though.
409 if (!(vma->vm_flags & VM_EXEC) &&
410 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
411 !(vma->vm_flags & (VM_READ | VM_WRITE))))
414 } else if (is_write) {
415 if (!(vma->vm_flags & VM_WRITE))
417 flags |= FAULT_FLAG_WRITE;
420 /* protection fault */
421 if (error_code & 0x08000000)
423 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
428 * If for any reason at all we couldn't handle the fault,
429 * make sure we exit gracefully rather than endlessly redo
432 fault = handle_mm_fault(mm, vma, address, flags);
433 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
434 rc = mm_fault_error(regs, address, fault);
435 if (rc >= MM_FAULT_RETURN)
442 * Major/minor page fault accounting is only done on the
443 * initial attempt. If we go through a retry, it is extremely
444 * likely that the page will be found in page cache at that point.
446 if (flags & FAULT_FLAG_ALLOW_RETRY) {
447 if (fault & VM_FAULT_MAJOR) {
449 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
451 #ifdef CONFIG_PPC_SMLPAR
452 if (firmware_has_feature(FW_FEATURE_CMO)) {
456 page_ins = be32_to_cpu(get_lppaca()->page_ins);
457 page_ins += 1 << PAGE_FACTOR;
458 get_lppaca()->page_ins = cpu_to_be32(page_ins);
461 #endif /* CONFIG_PPC_SMLPAR */
464 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
467 if (fault & VM_FAULT_RETRY) {
468 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
470 flags &= ~FAULT_FLAG_ALLOW_RETRY;
471 flags |= FAULT_FLAG_TRIED;
476 up_read(&mm->mmap_sem);
480 up_read(&mm->mmap_sem);
482 bad_area_nosemaphore:
483 /* User mode accesses cause a SIGSEGV */
484 if (user_mode(regs)) {
485 _exception(SIGSEGV, regs, code, address);
489 if (is_exec && (error_code & DSISR_PROTFAULT))
490 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
491 " page (%lx) - exploit attempt? (uid: %d)\n",
492 address, from_kuid(&init_user_ns, current_uid()));
497 exception_exit(prev_state);
503 * bad_page_fault is called when we have a bad access from the kernel.
504 * It is called from the DSI and ISI handlers in head.S and from some
505 * of the procedures in traps.c.
507 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
509 const struct exception_table_entry *entry;
511 /* Are we prepared to handle this fault? */
512 if ((entry = search_exception_tables(regs->nip)) != NULL) {
513 regs->nip = entry->fixup;
517 /* kernel has accessed a bad area */
519 switch (regs->trap) {
522 printk(KERN_ALERT "Unable to handle kernel paging request for "
523 "data at address 0x%08lx\n", regs->dar);
527 printk(KERN_ALERT "Unable to handle kernel paging request for "
528 "instruction fetch\n");
531 printk(KERN_ALERT "Unable to handle kernel paging request for "
535 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
538 if (task_stack_end_corrupted(current))
539 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
541 die("Kernel access of bad area", regs, sig);