5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Hartmut Penner (hp@de.ibm.com)
7 * Ulrich Weigand (uweigand@de.ibm.com)
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1995 Linus Torvalds
13 #include <linux/kernel_stat.h>
14 #include <linux/perf_event.h>
15 #include <linux/signal.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
24 #include <linux/compat.h>
25 #include <linux/smp.h>
26 #include <linux/kdebug.h>
27 #include <linux/init.h>
28 #include <linux/console.h>
29 #include <linux/module.h>
30 #include <linux/hardirq.h>
31 #include <linux/kprobes.h>
32 #include <linux/uaccess.h>
33 #include <linux/hugetlb.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/system.h>
36 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
39 #include "../kernel/entry.h"
42 #define __FAIL_ADDR_MASK 0x7ffff000
43 #define __SUBCODE_MASK 0x0200
44 #define __PF_RES_FIELD 0ULL
45 #else /* CONFIG_64BIT */
46 #define __FAIL_ADDR_MASK -4096L
47 #define __SUBCODE_MASK 0x0600
48 #define __PF_RES_FIELD 0x8000000000000000ULL
49 #endif /* CONFIG_64BIT */
51 #define VM_FAULT_BADCONTEXT 0x010000
52 #define VM_FAULT_BADMAP 0x020000
53 #define VM_FAULT_BADACCESS 0x040000
55 static unsigned long store_indication;
59 if (test_facility(2) && test_facility(75))
60 store_indication = 0xc00;
63 static inline int notify_page_fault(struct pt_regs *regs)
67 /* kprobe_running() needs smp_processor_id() */
68 if (kprobes_built_in() && !user_mode(regs)) {
70 if (kprobe_running() && kprobe_fault_handler(regs, 14))
79 * Unlock any spinlocks which will prevent us from getting the
82 void bust_spinlocks(int yes)
87 int loglevel_save = console_loglevel;
91 * OK, the message is on the console. Now we call printk()
92 * without oops_in_progress set so that printk will give klogd
93 * a poke. Hold onto your hats...
95 console_loglevel = 15;
97 console_loglevel = loglevel_save;
102 * Returns the address space associated with the fault.
103 * Returns 0 for kernel space and 1 for user space.
105 static inline int user_space_fault(unsigned long trans_exc_code)
108 * The lowest two bits of the translation exception
109 * identification indicate which paging table was used.
112 if (trans_exc_code == 2)
113 /* Access via secondary space, set_fs setting decides */
114 return current->thread.mm_segment.ar4;
115 if (user_mode == HOME_SPACE_MODE)
116 /* User space if the access has been done via home space. */
117 return trans_exc_code == 3;
119 * If the user space is not the home space the kernel runs in home
120 * space. Access via secondary space has already been covered,
121 * access via primary space or access register is from user space
122 * and access via home space is from the kernel.
124 return trans_exc_code != 3;
127 static inline void report_user_fault(struct pt_regs *regs, long int_code,
128 int signr, unsigned long address)
130 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
132 if (!unhandled_signal(current, signr))
134 if (!printk_ratelimit())
136 printk("User process fault: interruption code 0x%lX ", int_code);
137 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
139 printk("failing address: %lX\n", address);
144 * Send SIGSEGV to task. This is an external routine
145 * to keep the stack usage of do_page_fault small.
147 static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
148 int si_code, unsigned long trans_exc_code)
151 unsigned long address;
153 address = trans_exc_code & __FAIL_ADDR_MASK;
154 current->thread.prot_addr = address;
155 current->thread.trap_no = int_code;
156 report_user_fault(regs, int_code, SIGSEGV, address);
157 si.si_signo = SIGSEGV;
158 si.si_code = si_code;
159 si.si_addr = (void __user *) address;
160 force_sig_info(SIGSEGV, &si, current);
163 static noinline void do_no_context(struct pt_regs *regs, long int_code,
164 unsigned long trans_exc_code)
166 const struct exception_table_entry *fixup;
167 unsigned long address;
169 /* Are we prepared to handle this kernel fault? */
170 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
172 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
177 * Oops. The kernel tried to access some bad page. We'll have to
178 * terminate things with extreme prejudice.
180 address = trans_exc_code & __FAIL_ADDR_MASK;
181 if (!user_space_fault(trans_exc_code))
182 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
183 " at virtual kernel address %p\n", (void *)address);
185 printk(KERN_ALERT "Unable to handle kernel paging request"
186 " at virtual user address %p\n", (void *)address);
188 die("Oops", regs, int_code);
192 static noinline void do_low_address(struct pt_regs *regs, long int_code,
193 unsigned long trans_exc_code)
195 /* Low-address protection hit in kernel mode means
196 NULL pointer write access in kernel mode. */
197 if (regs->psw.mask & PSW_MASK_PSTATE) {
198 /* Low-address protection hit in user mode 'cannot happen'. */
199 die ("Low-address protection", regs, int_code);
203 do_no_context(regs, int_code, trans_exc_code);
206 static noinline void do_sigbus(struct pt_regs *regs, long int_code,
207 unsigned long trans_exc_code)
209 struct task_struct *tsk = current;
210 unsigned long address;
214 * Send a sigbus, regardless of whether we were in kernel
217 address = trans_exc_code & __FAIL_ADDR_MASK;
218 tsk->thread.prot_addr = address;
219 tsk->thread.trap_no = int_code;
220 si.si_signo = SIGBUS;
222 si.si_code = BUS_ADRERR;
223 si.si_addr = (void __user *) address;
224 force_sig_info(SIGBUS, &si, tsk);
227 static noinline void do_fault_error(struct pt_regs *regs, long int_code,
228 unsigned long trans_exc_code, int fault)
233 case VM_FAULT_BADACCESS:
234 case VM_FAULT_BADMAP:
235 /* Bad memory access. Check if it is kernel or user space. */
236 if (regs->psw.mask & PSW_MASK_PSTATE) {
237 /* User mode accesses just cause a SIGSEGV */
238 si_code = (fault == VM_FAULT_BADMAP) ?
239 SEGV_MAPERR : SEGV_ACCERR;
240 do_sigsegv(regs, int_code, si_code, trans_exc_code);
243 case VM_FAULT_BADCONTEXT:
244 do_no_context(regs, int_code, trans_exc_code);
246 default: /* fault & VM_FAULT_ERROR */
247 if (fault & VM_FAULT_OOM) {
248 if (!(regs->psw.mask & PSW_MASK_PSTATE))
249 do_no_context(regs, int_code, trans_exc_code);
251 pagefault_out_of_memory();
252 } else if (fault & VM_FAULT_SIGBUS) {
253 /* Kernel mode? Handle exceptions or die */
254 if (!(regs->psw.mask & PSW_MASK_PSTATE))
255 do_no_context(regs, int_code, trans_exc_code);
257 do_sigbus(regs, int_code, trans_exc_code);
265 * This routine handles page faults. It determines the address,
266 * and the problem, and then passes it off to one of the appropriate
269 * interruption code (int_code):
270 * 04 Protection -> Write-Protection (suprression)
271 * 10 Segment translation -> Not present (nullification)
272 * 11 Page translation -> Not present (nullification)
273 * 3b Region third trans. -> Not present (nullification)
275 static inline int do_exception(struct pt_regs *regs, int access,
276 unsigned long trans_exc_code)
278 struct task_struct *tsk;
279 struct mm_struct *mm;
280 struct vm_area_struct *vma;
281 unsigned long address;
285 if (notify_page_fault(regs))
292 * Verify that the fault happened in user space, that
293 * we are not in an interrupt and that there is a
296 fault = VM_FAULT_BADCONTEXT;
297 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
300 address = trans_exc_code & __FAIL_ADDR_MASK;
301 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
302 flags = FAULT_FLAG_ALLOW_RETRY;
303 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
304 flags |= FAULT_FLAG_WRITE;
305 down_read(&mm->mmap_sem);
308 if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
309 address = __gmap_fault(address,
310 (struct gmap *) S390_lowcore.gmap);
311 if (address == -EFAULT) {
312 fault = VM_FAULT_BADMAP;
315 if (address == -ENOMEM) {
316 fault = VM_FAULT_OOM;
323 fault = VM_FAULT_BADMAP;
324 vma = find_vma(mm, address);
328 if (unlikely(vma->vm_start > address)) {
329 if (!(vma->vm_flags & VM_GROWSDOWN))
331 if (expand_stack(vma, address))
336 * Ok, we have a good vm_area for this memory access, so
339 fault = VM_FAULT_BADACCESS;
340 if (unlikely(!(vma->vm_flags & access)))
343 if (is_vm_hugetlb_page(vma))
344 address &= HPAGE_MASK;
346 * If for any reason at all we couldn't handle the fault,
347 * make sure we exit gracefully rather than endlessly redo
350 fault = handle_mm_fault(mm, vma, address, flags);
351 if (unlikely(fault & VM_FAULT_ERROR))
355 * Major/minor page fault accounting is only done on the
356 * initial attempt. If we go through a retry, it is extremely
357 * likely that the page will be found in page cache at that point.
359 if (flags & FAULT_FLAG_ALLOW_RETRY) {
360 if (fault & VM_FAULT_MAJOR) {
362 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
366 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
369 if (fault & VM_FAULT_RETRY) {
370 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
372 flags &= ~FAULT_FLAG_ALLOW_RETRY;
373 down_read(&mm->mmap_sem);
378 * The instruction that caused the program check will
379 * be repeated. Don't signal single step via SIGTRAP.
381 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
384 up_read(&mm->mmap_sem);
389 void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code,
390 unsigned long trans_exc_code)
394 /* Protection exception is suppressing, decrement psw address. */
395 regs->psw.addr = __rewind_psw(regs->psw, pgm_int_code >> 16);
397 * Check for low-address protection. This needs to be treated
398 * as a special case because the translation exception code
399 * field is not guaranteed to contain valid data in this case.
401 if (unlikely(!(trans_exc_code & 4))) {
402 do_low_address(regs, pgm_int_code, trans_exc_code);
405 fault = do_exception(regs, VM_WRITE, trans_exc_code);
407 do_fault_error(regs, 4, trans_exc_code, fault);
410 void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code,
411 unsigned long trans_exc_code)
415 access = VM_READ | VM_EXEC | VM_WRITE;
416 fault = do_exception(regs, access, trans_exc_code);
418 do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault);
422 void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code,
423 unsigned long trans_exc_code)
425 struct mm_struct *mm = current->mm;
426 struct vm_area_struct *vma;
428 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
431 down_read(&mm->mmap_sem);
432 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
433 up_read(&mm->mmap_sem);
436 update_mm(mm, current);
440 /* User mode accesses just cause a SIGSEGV */
441 if (regs->psw.mask & PSW_MASK_PSTATE) {
442 do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code);
447 do_no_context(regs, pgm_int_code, trans_exc_code);
451 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
456 /* Emulate a uaccess fault from kernel mode. */
457 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
458 if (!irqs_disabled())
459 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
460 regs.psw.addr = (unsigned long) __builtin_return_address(0);
461 regs.psw.addr |= PSW_ADDR_AMODE;
463 access = write ? VM_WRITE : VM_READ;
464 fault = do_exception(®s, access, uaddr | 2);
466 * Since the fault happened in kernel mode while performing a uaccess
467 * all we need to do now is emulating a fixup in case "fault" is not
469 * For the calling uaccess functions this results always in -EFAULT.
471 return fault ? -EFAULT : 0;
476 * 'pfault' pseudo page faults routines.
478 static int pfault_disable;
480 static int __init nopfault(char *str)
486 __setup("nopfault", nopfault);
488 struct pfault_refbk {
497 } __attribute__ ((packed, aligned(8)));
499 int pfault_init(void)
501 struct pfault_refbk refbk = {
506 .refgaddr = __LC_CURRENT_PID,
507 .refselmk = 1ULL << 48,
508 .refcmpmk = 1ULL << 48,
509 .reserved = __PF_RES_FIELD };
512 if (!MACHINE_IS_VM || pfault_disable)
515 " diag %1,%0,0x258\n"
520 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
524 void pfault_fini(void)
526 struct pfault_refbk refbk = {
533 if (!MACHINE_IS_VM || pfault_disable)
539 : : "a" (&refbk), "m" (refbk) : "cc");
542 static DEFINE_SPINLOCK(pfault_lock);
543 static LIST_HEAD(pfault_list);
545 static void pfault_interrupt(unsigned int ext_int_code,
546 unsigned int param32, unsigned long param64)
548 struct task_struct *tsk;
553 * Get the external interruption subcode & pfault
554 * initial/completion signal bit. VM stores this
555 * in the 'cpu address' field associated with the
556 * external interrupt.
558 subcode = ext_int_code >> 16;
559 if ((subcode & 0xff00) != __SUBCODE_MASK)
561 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
562 if (subcode & 0x0080) {
563 /* Get the token (= pid of the affected task). */
564 pid = sizeof(void *) == 4 ? param32 : param64;
566 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
568 get_task_struct(tsk);
575 spin_lock(&pfault_lock);
576 if (subcode & 0x0080) {
577 /* signal bit is set -> a page has been swapped in by VM */
578 if (tsk->thread.pfault_wait == 1) {
579 /* Initial interrupt was faster than the completion
580 * interrupt. pfault_wait is valid. Set pfault_wait
581 * back to zero and wake up the process. This can
582 * safely be done because the task is still sleeping
583 * and can't produce new pfaults. */
584 tsk->thread.pfault_wait = 0;
585 list_del(&tsk->thread.list);
586 wake_up_process(tsk);
587 put_task_struct(tsk);
589 /* Completion interrupt was faster than initial
590 * interrupt. Set pfault_wait to -1 so the initial
591 * interrupt doesn't put the task to sleep.
592 * If the task is not running, ignore the completion
593 * interrupt since it must be a leftover of a PFAULT
594 * CANCEL operation which didn't remove all pending
595 * completion interrupts. */
596 if (tsk->state == TASK_RUNNING)
597 tsk->thread.pfault_wait = -1;
599 put_task_struct(tsk);
601 /* signal bit not set -> a real page is missing. */
602 if (tsk->thread.pfault_wait == 1) {
603 /* Already on the list with a reference: put to sleep */
604 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
605 set_tsk_need_resched(tsk);
606 } else if (tsk->thread.pfault_wait == -1) {
607 /* Completion interrupt was faster than the initial
608 * interrupt (pfault_wait == -1). Set pfault_wait
609 * back to zero and exit. */
610 tsk->thread.pfault_wait = 0;
612 /* Initial interrupt arrived before completion
613 * interrupt. Let the task sleep.
614 * An extra task reference is needed since a different
615 * cpu may set the task state to TASK_RUNNING again
616 * before the scheduler is reached. */
617 get_task_struct(tsk);
618 tsk->thread.pfault_wait = 1;
619 list_add(&tsk->thread.list, &pfault_list);
620 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
621 set_tsk_need_resched(tsk);
624 spin_unlock(&pfault_lock);
627 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
628 unsigned long action, void *hcpu)
630 struct thread_struct *thread, *next;
631 struct task_struct *tsk;
635 case CPU_DEAD_FROZEN:
636 spin_lock_irq(&pfault_lock);
637 list_for_each_entry_safe(thread, next, &pfault_list, list) {
638 thread->pfault_wait = 0;
639 list_del(&thread->list);
640 tsk = container_of(thread, struct task_struct, thread);
641 wake_up_process(tsk);
642 put_task_struct(tsk);
644 spin_unlock_irq(&pfault_lock);
652 static int __init pfault_irq_init(void)
658 rc = register_external_interrupt(0x2603, pfault_interrupt);
661 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
664 service_subclass_irq_register();
665 hotcpu_notifier(pfault_cpu_notify, 0);
669 unregister_external_interrupt(0x2603, pfault_interrupt);
674 early_initcall(pfault_irq_init);
676 #endif /* CONFIG_PFAULT */