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>
35 #include <linux/hugetlb.h>
37 #include <asm/firmware.h>
39 #include <asm/pgtable.h>
41 #include <asm/mmu_context.h>
42 #include <asm/uaccess.h>
43 #include <asm/tlbflush.h>
44 #include <asm/siginfo.h>
45 #include <asm/debug.h>
46 #include <mm/mmu_decl.h>
51 static inline int notify_page_fault(struct pt_regs *regs)
55 /* kprobe_running() needs smp_processor_id() */
56 if (!user_mode(regs)) {
58 if (kprobe_running() && kprobe_fault_handler(regs, 11))
66 static inline int notify_page_fault(struct pt_regs *regs)
73 * Check whether the instruction at regs->nip is a store using
74 * an update addressing form which will update r1.
76 static int store_updates_sp(struct pt_regs *regs)
80 if (get_user(inst, (unsigned int __user *)regs->nip))
82 /* check for 1 in the rA field */
83 if (((inst >> 16) & 0x1f) != 1)
85 /* check major opcode */
93 case 62: /* std or stdu */
94 return (inst & 3) == 1;
96 /* check minor opcode */
97 switch ((inst >> 1) & 0x3ff) {
100 case 247: /* stbux */
101 case 439: /* sthux */
102 case 695: /* stfsux */
103 case 759: /* stfdux */
110 * do_page_fault error handling helpers
113 #define MM_FAULT_RETURN 0
114 #define MM_FAULT_CONTINUE -1
115 #define MM_FAULT_ERR(sig) (sig)
117 static int do_sigbus(struct pt_regs *regs, unsigned long address,
121 unsigned int lsb = 0;
123 up_read(¤t->mm->mmap_sem);
125 if (!user_mode(regs))
126 return MM_FAULT_ERR(SIGBUS);
128 current->thread.trap_nr = BUS_ADRERR;
129 info.si_signo = SIGBUS;
131 info.si_code = BUS_ADRERR;
132 info.si_addr = (void __user *)address;
133 #ifdef CONFIG_MEMORY_FAILURE
134 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
135 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
136 current->comm, current->pid, address);
137 info.si_code = BUS_MCEERR_AR;
140 if (fault & VM_FAULT_HWPOISON_LARGE)
141 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
142 if (fault & VM_FAULT_HWPOISON)
145 info.si_addr_lsb = lsb;
146 force_sig_info(SIGBUS, &info, current);
147 return MM_FAULT_RETURN;
150 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, int fault)
153 * Pagefault was interrupted by SIGKILL. We have no reason to
154 * continue the pagefault.
156 if (fatal_signal_pending(current)) {
158 * If we have retry set, the mmap semaphore will have
159 * alrady been released in __lock_page_or_retry(). Else
162 if (!(fault & VM_FAULT_RETRY))
163 up_read(¤t->mm->mmap_sem);
164 /* Coming from kernel, we need to deal with uaccess fixups */
166 return MM_FAULT_RETURN;
167 return MM_FAULT_ERR(SIGKILL);
170 /* No fault: be happy */
171 if (!(fault & VM_FAULT_ERROR))
172 return MM_FAULT_CONTINUE;
175 if (fault & VM_FAULT_OOM) {
176 up_read(¤t->mm->mmap_sem);
179 * We ran out of memory, or some other thing happened to us that
180 * made us unable to handle the page fault gracefully.
182 if (!user_mode(regs))
183 return MM_FAULT_ERR(SIGKILL);
184 pagefault_out_of_memory();
185 return MM_FAULT_RETURN;
188 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE))
189 return do_sigbus(regs, addr, fault);
191 /* We don't understand the fault code, this is fatal */
193 return MM_FAULT_CONTINUE;
197 * For 600- and 800-family processors, the error_code parameter is DSISR
198 * for a data fault, SRR1 for an instruction fault. For 400-family processors
199 * the error_code parameter is ESR for a data fault, 0 for an instruction
201 * For 64-bit processors, the error_code parameter is
202 * - DSISR for a non-SLB data access fault,
203 * - SRR1 & 0x08000000 for a non-SLB instruction access fault
206 * The return value is 0 if the fault was handled, or the signal
207 * number if this is a kernel fault that can't be handled here.
209 int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
210 unsigned long error_code)
212 enum ctx_state prev_state = exception_enter();
213 struct vm_area_struct * vma;
214 struct mm_struct *mm = current->mm;
215 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
216 int code = SEGV_MAPERR;
218 int trap = TRAP(regs);
219 int is_exec = trap == 0x400;
221 int rc = 0, store_update_sp = 0;
223 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
225 * Fortunately the bit assignments in SRR1 for an instruction
226 * fault and DSISR for a data fault are mostly the same for the
227 * bits we are interested in. But there are some bits which
228 * indicate errors in DSISR but can validly be set in SRR1.
231 error_code &= 0x48200000;
233 is_write = error_code & DSISR_ISSTORE;
235 is_write = error_code & ESR_DST;
236 #endif /* CONFIG_4xx || CONFIG_BOOKE */
238 #ifdef CONFIG_PPC_ICSWX
240 * we need to do this early because this "data storage
241 * interrupt" does not update the DAR/DEAR so we don't want to
244 if (error_code & ICSWX_DSI_UCT) {
245 rc = acop_handle_fault(regs, address, error_code);
249 #endif /* CONFIG_PPC_ICSWX */
251 if (notify_page_fault(regs))
254 if (unlikely(debugger_fault_handler(regs)))
257 /* On a kernel SLB miss we can only check for a valid exception entry */
258 if (!user_mode(regs) && (address >= TASK_SIZE)) {
263 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE) || \
264 defined(CONFIG_PPC_BOOK3S_64))
265 if (error_code & DSISR_DABRMATCH) {
266 /* breakpoint match */
267 do_break(regs, address, error_code);
272 /* We restore the interrupt state now */
273 if (!arch_irq_disabled_regs(regs))
276 if (in_atomic() || mm == NULL) {
277 if (!user_mode(regs)) {
281 /* in_atomic() in user mode is really bad,
282 as is current->mm == NULL. */
283 printk(KERN_EMERG "Page fault in user mode with "
284 "in_atomic() = %d mm = %p\n", in_atomic(), 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 /* 8xx sometimes need to load a invalid/non-present TLBs.
384 * These must be invalidated separately as linux mm don't.
386 if (error_code & 0x40000000) /* no translation? */
387 _tlbil_va(address, 0, 0, 0);
389 /* The MPC8xx seems to always set 0x80000000, which is
390 * "undefined". Of those that can be set, this is the only
391 * one which seems bad.
393 if (error_code & 0x10000000)
394 /* Guarded storage error. */
396 #endif /* CONFIG_8xx */
399 #ifdef CONFIG_PPC_STD_MMU
400 /* Protection fault on exec go straight to failure on
401 * Hash based MMUs as they either don't support per-page
402 * execute permission, or if they do, it's handled already
403 * at the hash level. This test would probably have to
404 * be removed if we change the way this works to make hash
405 * processors use the same I/D cache coherency mechanism
408 if (error_code & DSISR_PROTFAULT)
410 #endif /* CONFIG_PPC_STD_MMU */
413 * Allow execution from readable areas if the MMU does not
414 * provide separate controls over reading and executing.
416 * Note: That code used to not be enabled for 4xx/BookE.
417 * It is now as I/D cache coherency for these is done at
418 * set_pte_at() time and I see no reason why the test
419 * below wouldn't be valid on those processors. This -may-
420 * break programs compiled with a really old ABI though.
422 if (!(vma->vm_flags & VM_EXEC) &&
423 (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
424 !(vma->vm_flags & (VM_READ | VM_WRITE))))
427 } else if (is_write) {
428 if (!(vma->vm_flags & VM_WRITE))
430 flags |= FAULT_FLAG_WRITE;
433 /* protection fault */
434 if (error_code & 0x08000000)
436 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
441 * If for any reason at all we couldn't handle the fault,
442 * make sure we exit gracefully rather than endlessly redo
445 fault = handle_mm_fault(mm, vma, address, flags);
446 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
447 rc = mm_fault_error(regs, address, fault);
448 if (rc >= MM_FAULT_RETURN)
455 * Major/minor page fault accounting is only done on the
456 * initial attempt. If we go through a retry, it is extremely
457 * likely that the page will be found in page cache at that point.
459 if (flags & FAULT_FLAG_ALLOW_RETRY) {
460 if (fault & VM_FAULT_MAJOR) {
462 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
464 #ifdef CONFIG_PPC_SMLPAR
465 if (firmware_has_feature(FW_FEATURE_CMO)) {
469 page_ins = be32_to_cpu(get_lppaca()->page_ins);
470 page_ins += 1 << PAGE_FACTOR;
471 get_lppaca()->page_ins = cpu_to_be32(page_ins);
474 #endif /* CONFIG_PPC_SMLPAR */
477 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
480 if (fault & VM_FAULT_RETRY) {
481 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
483 flags &= ~FAULT_FLAG_ALLOW_RETRY;
484 flags |= FAULT_FLAG_TRIED;
489 up_read(&mm->mmap_sem);
493 up_read(&mm->mmap_sem);
495 bad_area_nosemaphore:
496 /* User mode accesses cause a SIGSEGV */
497 if (user_mode(regs)) {
498 _exception(SIGSEGV, regs, code, address);
502 if (is_exec && (error_code & DSISR_PROTFAULT))
503 printk_ratelimited(KERN_CRIT "kernel tried to execute NX-protected"
504 " page (%lx) - exploit attempt? (uid: %d)\n",
505 address, from_kuid(&init_user_ns, current_uid()));
510 exception_exit(prev_state);
516 * bad_page_fault is called when we have a bad access from the kernel.
517 * It is called from the DSI and ISI handlers in head.S and from some
518 * of the procedures in traps.c.
520 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
522 const struct exception_table_entry *entry;
524 /* Are we prepared to handle this fault? */
525 if ((entry = search_exception_tables(regs->nip)) != NULL) {
526 regs->nip = entry->fixup;
530 /* kernel has accessed a bad area */
532 switch (regs->trap) {
535 printk(KERN_ALERT "Unable to handle kernel paging request for "
536 "data at address 0x%08lx\n", regs->dar);
540 printk(KERN_ALERT "Unable to handle kernel paging request for "
541 "instruction fetch\n");
544 printk(KERN_ALERT "Unable to handle kernel paging request for "
548 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
551 if (task_stack_end_corrupted(current))
552 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
554 die("Kernel access of bad area", regs, sig);