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/extable.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>
35 #include <linux/hugetlb.h>
36 #include <linux/uaccess.h>
38 #include <asm/firmware.h>
40 #include <asm/pgtable.h>
42 #include <asm/mmu_context.h>
43 #include <asm/tlbflush.h>
44 #include <asm/siginfo.h>
45 #include <asm/debug.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 unsigned int lsb = 0;
122 up_read(¤t->mm->mmap_sem);
124 if (!user_mode(regs))
125 return MM_FAULT_ERR(SIGBUS);
127 current->thread.trap_nr = BUS_ADRERR;
128 info.si_signo = SIGBUS;
130 info.si_code = BUS_ADRERR;
131 info.si_addr = (void __user *)address;
132 #ifdef CONFIG_MEMORY_FAILURE
133 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
134 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
135 current->comm, current->pid, address);
136 info.si_code = BUS_MCEERR_AR;
139 if (fault & VM_FAULT_HWPOISON_LARGE)
140 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
141 if (fault & VM_FAULT_HWPOISON)
144 info.si_addr_lsb = lsb;
145 force_sig_info(SIGBUS, &info, current);
146 return MM_FAULT_RETURN;
149 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
152 * Pagefault was interrupted by SIGKILL. We have no reason to
153 * continue the pagefault.
155 if (fatal_signal_pending(current)) {
157 * If we have retry set, the mmap semaphore will have
158 * alrady been released in __lock_page_or_retry(). Else
161 if (!(fault & VM_FAULT_RETRY))
162 up_read(¤t->mm->mmap_sem);
163 /* Coming from kernel, we need to deal with uaccess fixups */
165 return MM_FAULT_RETURN;
166 return MM_FAULT_ERR(SIGKILL);
169 /* No fault: be happy */
170 if (!(fault & VM_FAULT_ERROR))
171 return MM_FAULT_CONTINUE;
174 if (fault & VM_FAULT_OOM) {
175 up_read(¤t->mm->mmap_sem);
178 * We ran out of memory, or some other thing happened to us that
179 * made us unable to handle the page fault gracefully.
181 if (!user_mode(regs))
182 return MM_FAULT_ERR(SIGKILL);
183 pagefault_out_of_memory();
184 return MM_FAULT_RETURN;
187 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE))
188 return do_sigbus(regs, addr, fault);
190 /* We don't understand the fault code, this is fatal */
192 return MM_FAULT_CONTINUE;
196 * For 600- and 800-family processors, the error_code parameter is DSISR
197 * for a data fault, SRR1 for an instruction fault. For 400-family processors
198 * the error_code parameter is ESR for a data fault, 0 for an instruction
200 * For 64-bit processors, the error_code parameter is
201 * - DSISR for a non-SLB data access fault,
202 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
205 * The return value is 0 if the fault was handled, or the signal
206 * number if this is a kernel fault that can't be handled here.
208 int do_page_fault(struct pt_regs *regs, unsigned long address,
209 unsigned long error_code)
211 enum ctx_state prev_state = exception_enter();
212 struct vm_area_struct * vma;
213 struct mm_struct *mm = current->mm;
214 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
215 int code = SEGV_MAPERR;
217 int trap = TRAP(regs);
218 int is_exec = trap == 0x400;
220 int rc = 0, store_update_sp = 0;
222 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
224 * Fortunately the bit assignments in SRR1 for an instruction
225 * fault and DSISR for a data fault are mostly the same for the
226 * bits we are interested in. But there are some bits which
227 * indicate errors in DSISR but can validly be set in SRR1.
230 error_code &= 0x48200000;
232 is_write = error_code & DSISR_ISSTORE;
234 is_write = error_code & ESR_DST;
235 #endif /* CONFIG_4xx || CONFIG_BOOKE */
237 #ifdef CONFIG_PPC_ICSWX
239 * we need to do this early because this "data storage
240 * interrupt" does not update the DAR/DEAR so we don't want to
243 if (error_code & ICSWX_DSI_UCT) {
244 rc = acop_handle_fault(regs, address, error_code);
248 #endif /* CONFIG_PPC_ICSWX */
250 if (notify_page_fault(regs))
253 if (unlikely(debugger_fault_handler(regs)))
256 /* On a kernel SLB miss we can only check for a valid exception entry */
257 if (!user_mode(regs) && (address >= TASK_SIZE)) {
262 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
263 defined(CONFIG_PPC_BOOK3S_64))
264 if (error_code & DSISR_DABRMATCH) {
265 /* breakpoint match */
266 do_break(regs, address, error_code);
271 /* We restore the interrupt state now */
272 if (!arch_irq_disabled_regs(regs))
275 if (faulthandler_disabled() || mm == NULL) {
276 if (!user_mode(regs)) {
280 /* faulthandler_disabled() in user mode is really bad,
281 as is current->mm == NULL. */
282 printk(KERN_EMERG "Page fault in user mode with "
283 "faulthandler_disabled() = %d mm = %p\n",
284 faulthandler_disabled(), mm);
285 printk(KERN_EMERG "NIP = %lx MSR = %lx\n",
286 regs->nip, regs->msr);
287 die("Weird page fault", regs, SIGSEGV);
290 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
293 * We want to do this outside mmap_sem, because reading code around nip
294 * can result in fault, which will cause a deadlock when called with
298 store_update_sp = store_updates_sp(regs);
301 flags |= FAULT_FLAG_USER;
303 /* When running in the kernel we expect faults to occur only to
304 * addresses in user space. All other faults represent errors in the
305 * kernel and should generate an OOPS. Unfortunately, in the case of an
306 * erroneous fault occurring in a code path which already holds mmap_sem
307 * we will deadlock attempting to validate the fault against the
308 * address space. Luckily the kernel only validly references user
309 * space from well defined areas of code, which are listed in the
312 * As the vast majority of faults will be valid we will only perform
313 * the source reference check when there is a possibility of a deadlock.
314 * Attempt to lock the address space, if we cannot we then validate the
315 * source. If this is invalid we can skip the address space check,
316 * thus avoiding the deadlock.
318 if (!down_read_trylock(&mm->mmap_sem)) {
319 if (!user_mode(regs) && !search_exception_tables(regs->nip))
320 goto bad_area_nosemaphore;
323 down_read(&mm->mmap_sem);
326 * The above down_read_trylock() might have succeeded in
327 * which case we'll have missed the might_sleep() from
333 vma = find_vma(mm, address);
336 if (vma->vm_start <= address)
338 if (!(vma->vm_flags & VM_GROWSDOWN))
342 * N.B. The POWER/Open ABI allows programs to access up to
343 * 288 bytes below the stack pointer.
344 * The kernel signal delivery code writes up to about 1.5kB
345 * below the stack pointer (r1) before decrementing it.
346 * The exec code can write slightly over 640kB to the stack
347 * before setting the user r1. Thus we allow the stack to
348 * expand to 1MB without further checks.
350 if (address + 0x100000 < vma->vm_end) {
351 /* get user regs even if this fault is in kernel mode */
352 struct pt_regs *uregs = current->thread.regs;
357 * A user-mode access to an address a long way below
358 * the stack pointer is only valid if the instruction
359 * is one which would update the stack pointer to the
360 * address accessed if the instruction completed,
361 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
362 * (or the byte, halfword, float or double forms).
364 * If we don't check this then any write to the area
365 * between the last mapped region and the stack will
366 * expand the stack rather than segfaulting.
368 if (address + 2048 < uregs->gpr[1] && !store_update_sp)
371 if (expand_stack(vma, address))
376 #if defined(CONFIG_6xx)
377 if (error_code & 0x95700000)
378 /* an error such as lwarx to I/O controller space,
379 address matching DABR, eciwx, etc. */
381 #endif /* CONFIG_6xx */
382 #if defined(CONFIG_8xx)
383 /* The MPC8xx seems to always set 0x80000000, which is
384 * "undefined". Of those that can be set, this is the only
385 * one which seems bad.
387 if (error_code & 0x10000000)
388 /* Guarded storage error. */
390 #endif /* CONFIG_8xx */
394 * An execution fault + no execute ?
396 * On CPUs that don't have CPU_FTR_COHERENT_ICACHE we
397 * deliberately create NX mappings, and use the fault to do the
398 * cache flush. This is usually handled in hash_page_do_lazy_icache()
399 * but we could end up here if that races with a concurrent PTE
400 * update. In that case we need to fall through here to the VMA
403 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
404 (regs->msr & SRR1_ISI_N_OR_G))
408 * Allow execution from readable areas if the MMU does not
409 * provide separate controls over reading and executing.
411 * Note: That code used to not be enabled for 4xx/BookE.
412 * It is now as I/D cache coherency for these is done at
413 * set_pte_at() time and I see no reason why the test
414 * below wouldn't be valid on those processors. This -may-
415 * break programs compiled with a really old ABI though.
417 if (!(vma->vm_flags & VM_EXEC) &&
418 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
419 !(vma->vm_flags & (VM_READ | VM_WRITE))))
422 } else if (is_write) {
423 if (!(vma->vm_flags & VM_WRITE))
425 flags |= FAULT_FLAG_WRITE;
428 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
431 #ifdef CONFIG_PPC_STD_MMU
433 * For hash translation mode, we should never get a
434 * PROTFAULT. Any update to pte to reduce access will result in us
435 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE
436 * fault instead of DSISR_PROTFAULT.
438 * A pte update to relax the access will not result in a hash page table
439 * entry invalidate and hence can result in DSISR_PROTFAULT.
440 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have
441 * the special !is_write in the below conditional.
443 * For platforms that doesn't supports coherent icache and do support
444 * per page noexec bit, we do setup things such that we do the
445 * sync between D/I cache via fault. But that is handled via low level
446 * hash fault code (hash_page_do_lazy_icache()) and we should not reach
449 * For wrong access that can result in PROTFAULT, the above vma->vm_flags
450 * check should handle those and hence we should fall to the bad_area
451 * handling correctly.
453 * For embedded with per page exec support that doesn't support coherent
454 * icache we do get PROTFAULT and we handle that D/I cache sync in
455 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON
456 * is conditional for server MMU.
458 * For radix, we can get prot fault for autonuma case, because radix
459 * page table will have them marked noaccess for user.
461 if (!radix_enabled() && !is_write)
462 WARN_ON_ONCE(error_code & DSISR_PROTFAULT);
463 #endif /* CONFIG_PPC_STD_MMU */
466 * If for any reason at all we couldn't handle the fault,
467 * make sure we exit gracefully rather than endlessly redo
470 fault = handle_mm_fault(vma, address, flags);
471 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
472 if (fault & VM_FAULT_SIGSEGV)
474 rc = mm_fault_error(regs, address, fault);
475 if (rc >= MM_FAULT_RETURN)
482 * Major/minor page fault accounting is only done on the
483 * initial attempt. If we go through a retry, it is extremely
484 * likely that the page will be found in page cache at that point.
486 if (flags & FAULT_FLAG_ALLOW_RETRY) {
487 if (fault & VM_FAULT_MAJOR) {
489 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
491 #ifdef CONFIG_PPC_SMLPAR
492 if (firmware_has_feature(FW_FEATURE_CMO)) {
496 page_ins = be32_to_cpu(get_lppaca()->page_ins);
497 page_ins += 1 << PAGE_FACTOR;
498 get_lppaca()->page_ins = cpu_to_be32(page_ins);
501 #endif /* CONFIG_PPC_SMLPAR */
504 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
507 if (fault & VM_FAULT_RETRY) {
508 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
510 flags &= ~FAULT_FLAG_ALLOW_RETRY;
511 flags |= FAULT_FLAG_TRIED;
516 up_read(&mm->mmap_sem);
520 up_read(&mm->mmap_sem);
522 bad_area_nosemaphore:
523 /* User mode accesses cause a SIGSEGV */
524 if (user_mode(regs)) {
525 _exception(SIGSEGV, regs, code, address);
529 if (is_exec && (error_code & DSISR_PROTFAULT))
530 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
531 " page (%lx) - exploit attempt? (uid: %d)\n",
532 address, from_kuid(&init_user_ns, current_uid()));
537 exception_exit(prev_state);
540 NOKPROBE_SYMBOL(do_page_fault);
543 * bad_page_fault is called when we have a bad access from the kernel.
544 * It is called from the DSI and ISI handlers in head.S and from some
545 * of the procedures in traps.c.
547 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
549 const struct exception_table_entry *entry;
551 /* Are we prepared to handle this fault? */
552 if ((entry = search_exception_tables(regs->nip)) != NULL) {
553 regs->nip = extable_fixup(entry);
557 /* kernel has accessed a bad area */
559 switch (regs->trap) {
562 printk(KERN_ALERT "Unable to handle kernel paging request for "
563 "data at address 0x%08lx\n", regs->dar);
567 printk(KERN_ALERT "Unable to handle kernel paging request for "
568 "instruction fetch\n");
571 printk(KERN_ALERT "Unable to handle kernel paging request for "
572 "unaligned access at address 0x%08lx\n", regs->dar);
575 printk(KERN_ALERT "Unable to handle kernel paging request for "
579 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
582 if (task_stack_end_corrupted(current))
583 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
585 die("Kernel access of bad area", regs, sig);