3 * Copyright IBM Corp. 1999
4 * Author(s): Hartmut Penner (hp@de.ibm.com)
5 * Ulrich Weigand (uweigand@de.ibm.com)
7 * Derived from "arch/i386/mm/fault.c"
8 * Copyright (C) 1995 Linus Torvalds
11 #include <linux/kernel_stat.h>
12 #include <linux/perf_event.h>
13 #include <linux/signal.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/mman.h>
22 #include <linux/compat.h>
23 #include <linux/smp.h>
24 #include <linux/kdebug.h>
25 #include <linux/init.h>
26 #include <linux/console.h>
27 #include <linux/module.h>
28 #include <linux/hardirq.h>
29 #include <linux/kprobes.h>
30 #include <linux/uaccess.h>
31 #include <linux/hugetlb.h>
32 #include <asm/asm-offsets.h>
33 #include <asm/pgtable.h>
35 #include <asm/mmu_context.h>
36 #include <asm/facility.h>
37 #include "../kernel/entry.h"
40 #define __FAIL_ADDR_MASK 0x7ffff000
41 #define __SUBCODE_MASK 0x0200
42 #define __PF_RES_FIELD 0ULL
43 #else /* CONFIG_64BIT */
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #endif /* CONFIG_64BIT */
49 #define VM_FAULT_BADCONTEXT 0x010000
50 #define VM_FAULT_BADMAP 0x020000
51 #define VM_FAULT_BADACCESS 0x040000
52 #define VM_FAULT_SIGNAL 0x080000
53 #define VM_FAULT_PFAULT 0x100000
55 static unsigned long store_indication __read_mostly;
58 static int __init fault_init(void)
60 if (test_facility(75))
61 store_indication = 0xc00;
64 early_initcall(fault_init);
67 static inline int notify_page_fault(struct pt_regs *regs)
71 /* kprobe_running() needs smp_processor_id() */
72 if (kprobes_built_in() && !user_mode(regs)) {
74 if (kprobe_running() && kprobe_fault_handler(regs, 14))
83 * Unlock any spinlocks which will prevent us from getting the
86 void bust_spinlocks(int yes)
91 int loglevel_save = console_loglevel;
95 * OK, the message is on the console. Now we call printk()
96 * without oops_in_progress set so that printk will give klogd
97 * a poke. Hold onto your hats...
99 console_loglevel = 15;
101 console_loglevel = loglevel_save;
106 * Returns the address space associated with the fault.
107 * Returns 0 for kernel space and 1 for user space.
109 static inline int user_space_fault(struct pt_regs *regs)
111 unsigned long trans_exc_code;
114 * The lowest two bits of the translation exception
115 * identification indicate which paging table was used.
117 trans_exc_code = regs->int_parm_long & 3;
118 if (trans_exc_code == 3) /* home space -> kernel */
122 if (trans_exc_code == 2) /* secondary space -> set_fs */
123 return current->thread.mm_segment.ar4;
124 if (current->flags & PF_VCPU)
129 static inline void report_user_fault(struct pt_regs *regs, long signr)
131 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
133 if (!unhandled_signal(current, signr))
135 if (!printk_ratelimit())
137 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
139 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
140 printk(KERN_CONT "\n");
141 printk(KERN_ALERT "failing address: %lX\n",
142 regs->int_parm_long & __FAIL_ADDR_MASK);
147 * Send SIGSEGV to task. This is an external routine
148 * to keep the stack usage of do_page_fault small.
150 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
154 report_user_fault(regs, SIGSEGV);
155 si.si_signo = SIGSEGV;
156 si.si_code = si_code;
157 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
158 force_sig_info(SIGSEGV, &si, current);
161 static noinline void do_no_context(struct pt_regs *regs)
163 const struct exception_table_entry *fixup;
164 unsigned long address;
166 /* Are we prepared to handle this kernel fault? */
167 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
169 regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
174 * Oops. The kernel tried to access some bad page. We'll have to
175 * terminate things with extreme prejudice.
177 address = regs->int_parm_long & __FAIL_ADDR_MASK;
178 if (!user_space_fault(regs))
179 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
180 " at virtual kernel address %p\n", (void *)address);
182 printk(KERN_ALERT "Unable to handle kernel paging request"
183 " at virtual user address %p\n", (void *)address);
189 static noinline void do_low_address(struct pt_regs *regs)
191 /* Low-address protection hit in kernel mode means
192 NULL pointer write access in kernel mode. */
193 if (regs->psw.mask & PSW_MASK_PSTATE) {
194 /* Low-address protection hit in user mode 'cannot happen'. */
195 die (regs, "Low-address protection");
202 static noinline void do_sigbus(struct pt_regs *regs)
204 struct task_struct *tsk = current;
208 * Send a sigbus, regardless of whether we were in kernel
211 si.si_signo = SIGBUS;
213 si.si_code = BUS_ADRERR;
214 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
215 force_sig_info(SIGBUS, &si, tsk);
218 static noinline void do_fault_error(struct pt_regs *regs, int fault)
223 case VM_FAULT_BADACCESS:
224 case VM_FAULT_BADMAP:
225 /* Bad memory access. Check if it is kernel or user space. */
226 if (user_mode(regs)) {
227 /* User mode accesses just cause a SIGSEGV */
228 si_code = (fault == VM_FAULT_BADMAP) ?
229 SEGV_MAPERR : SEGV_ACCERR;
230 do_sigsegv(regs, si_code);
233 case VM_FAULT_BADCONTEXT:
234 case VM_FAULT_PFAULT:
237 case VM_FAULT_SIGNAL:
238 if (!user_mode(regs))
241 default: /* fault & VM_FAULT_ERROR */
242 if (fault & VM_FAULT_OOM) {
243 if (!user_mode(regs))
246 pagefault_out_of_memory();
247 } else if (fault & VM_FAULT_SIGBUS) {
248 /* Kernel mode? Handle exceptions or die */
249 if (!user_mode(regs))
260 * This routine handles page faults. It determines the address,
261 * and the problem, and then passes it off to one of the appropriate
264 * interruption code (int_code):
265 * 04 Protection -> Write-Protection (suprression)
266 * 10 Segment translation -> Not present (nullification)
267 * 11 Page translation -> Not present (nullification)
268 * 3b Region third trans. -> Not present (nullification)
270 static inline int do_exception(struct pt_regs *regs, int access)
275 struct task_struct *tsk;
276 struct mm_struct *mm;
277 struct vm_area_struct *vma;
278 unsigned long trans_exc_code;
279 unsigned long address;
285 * The instruction that caused the program check has
286 * been nullified. Don't signal single step via SIGTRAP.
288 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
290 if (notify_page_fault(regs))
294 trans_exc_code = regs->int_parm_long;
297 * Verify that the fault happened in user space, that
298 * we are not in an interrupt and that there is a
301 fault = VM_FAULT_BADCONTEXT;
302 if (unlikely(!user_space_fault(regs) || in_atomic() || !mm))
305 address = trans_exc_code & __FAIL_ADDR_MASK;
306 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
307 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
309 flags |= FAULT_FLAG_USER;
310 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
311 flags |= FAULT_FLAG_WRITE;
312 down_read(&mm->mmap_sem);
315 gmap = (struct gmap *)
316 ((current->flags & PF_VCPU) ? S390_lowcore.gmap : 0);
318 address = __gmap_fault(address, gmap);
319 if (address == -EFAULT) {
320 fault = VM_FAULT_BADMAP;
323 if (address == -ENOMEM) {
324 fault = VM_FAULT_OOM;
327 if (gmap->pfault_enabled)
328 flags |= FAULT_FLAG_RETRY_NOWAIT;
333 fault = VM_FAULT_BADMAP;
334 vma = find_vma(mm, address);
338 if (unlikely(vma->vm_start > address)) {
339 if (!(vma->vm_flags & VM_GROWSDOWN))
341 if (expand_stack(vma, address))
346 * Ok, we have a good vm_area for this memory access, so
349 fault = VM_FAULT_BADACCESS;
350 if (unlikely(!(vma->vm_flags & access)))
353 if (is_vm_hugetlb_page(vma))
354 address &= HPAGE_MASK;
356 * If for any reason at all we couldn't handle the fault,
357 * make sure we exit gracefully rather than endlessly redo
360 fault = handle_mm_fault(mm, vma, address, flags);
361 /* No reason to continue if interrupted by SIGKILL. */
362 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
363 fault = VM_FAULT_SIGNAL;
366 if (unlikely(fault & VM_FAULT_ERROR))
370 * Major/minor page fault accounting is only done on the
371 * initial attempt. If we go through a retry, it is extremely
372 * likely that the page will be found in page cache at that point.
374 if (flags & FAULT_FLAG_ALLOW_RETRY) {
375 if (fault & VM_FAULT_MAJOR) {
377 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
381 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
384 if (fault & VM_FAULT_RETRY) {
386 if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
387 /* FAULT_FLAG_RETRY_NOWAIT has been set,
388 * mmap_sem has not been released */
389 current->thread.gmap_pfault = 1;
390 fault = VM_FAULT_PFAULT;
394 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
396 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
397 FAULT_FLAG_RETRY_NOWAIT);
398 flags |= FAULT_FLAG_TRIED;
399 down_read(&mm->mmap_sem);
405 up_read(&mm->mmap_sem);
410 void __kprobes do_protection_exception(struct pt_regs *regs)
412 unsigned long trans_exc_code;
415 trans_exc_code = regs->int_parm_long;
417 * Protection exceptions are suppressing, decrement psw address.
418 * The exception to this rule are aborted transactions, for these
419 * the PSW already points to the correct location.
421 if (!(regs->int_code & 0x200))
422 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
424 * Check for low-address protection. This needs to be treated
425 * as a special case because the translation exception code
426 * field is not guaranteed to contain valid data in this case.
428 if (unlikely(!(trans_exc_code & 4))) {
429 do_low_address(regs);
432 fault = do_exception(regs, VM_WRITE);
434 do_fault_error(regs, fault);
437 void __kprobes do_dat_exception(struct pt_regs *regs)
441 access = VM_READ | VM_EXEC | VM_WRITE;
442 fault = do_exception(regs, access);
444 do_fault_error(regs, fault);
449 * 'pfault' pseudo page faults routines.
451 static int pfault_disable;
453 static int __init nopfault(char *str)
459 __setup("nopfault", nopfault);
461 struct pfault_refbk {
470 } __attribute__ ((packed, aligned(8)));
472 int pfault_init(void)
474 struct pfault_refbk refbk = {
479 .refgaddr = __LC_CURRENT_PID,
480 .refselmk = 1ULL << 48,
481 .refcmpmk = 1ULL << 48,
482 .reserved = __PF_RES_FIELD };
488 " diag %1,%0,0x258\n"
493 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
497 void pfault_fini(void)
499 struct pfault_refbk refbk = {
512 : : "a" (&refbk), "m" (refbk) : "cc");
515 static DEFINE_SPINLOCK(pfault_lock);
516 static LIST_HEAD(pfault_list);
518 static void pfault_interrupt(struct ext_code ext_code,
519 unsigned int param32, unsigned long param64)
521 struct task_struct *tsk;
526 * Get the external interruption subcode & pfault
527 * initial/completion signal bit. VM stores this
528 * in the 'cpu address' field associated with the
529 * external interrupt.
531 subcode = ext_code.subcode;
532 if ((subcode & 0xff00) != __SUBCODE_MASK)
534 inc_irq_stat(IRQEXT_PFL);
535 /* Get the token (= pid of the affected task). */
536 pid = sizeof(void *) == 4 ? param32 : param64;
538 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
540 get_task_struct(tsk);
544 spin_lock(&pfault_lock);
545 if (subcode & 0x0080) {
546 /* signal bit is set -> a page has been swapped in by VM */
547 if (tsk->thread.pfault_wait == 1) {
548 /* Initial interrupt was faster than the completion
549 * interrupt. pfault_wait is valid. Set pfault_wait
550 * back to zero and wake up the process. This can
551 * safely be done because the task is still sleeping
552 * and can't produce new pfaults. */
553 tsk->thread.pfault_wait = 0;
554 list_del(&tsk->thread.list);
555 wake_up_process(tsk);
556 put_task_struct(tsk);
558 /* Completion interrupt was faster than initial
559 * interrupt. Set pfault_wait to -1 so the initial
560 * interrupt doesn't put the task to sleep.
561 * If the task is not running, ignore the completion
562 * interrupt since it must be a leftover of a PFAULT
563 * CANCEL operation which didn't remove all pending
564 * completion interrupts. */
565 if (tsk->state == TASK_RUNNING)
566 tsk->thread.pfault_wait = -1;
569 /* signal bit not set -> a real page is missing. */
570 if (WARN_ON_ONCE(tsk != current))
572 if (tsk->thread.pfault_wait == 1) {
573 /* Already on the list with a reference: put to sleep */
574 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
575 set_tsk_need_resched(tsk);
576 } else if (tsk->thread.pfault_wait == -1) {
577 /* Completion interrupt was faster than the initial
578 * interrupt (pfault_wait == -1). Set pfault_wait
579 * back to zero and exit. */
580 tsk->thread.pfault_wait = 0;
582 /* Initial interrupt arrived before completion
583 * interrupt. Let the task sleep.
584 * An extra task reference is needed since a different
585 * cpu may set the task state to TASK_RUNNING again
586 * before the scheduler is reached. */
587 get_task_struct(tsk);
588 tsk->thread.pfault_wait = 1;
589 list_add(&tsk->thread.list, &pfault_list);
590 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
591 set_tsk_need_resched(tsk);
595 spin_unlock(&pfault_lock);
596 put_task_struct(tsk);
599 static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
602 struct thread_struct *thread, *next;
603 struct task_struct *tsk;
605 switch (action & ~CPU_TASKS_FROZEN) {
607 spin_lock_irq(&pfault_lock);
608 list_for_each_entry_safe(thread, next, &pfault_list, list) {
609 thread->pfault_wait = 0;
610 list_del(&thread->list);
611 tsk = container_of(thread, struct task_struct, thread);
612 wake_up_process(tsk);
613 put_task_struct(tsk);
615 spin_unlock_irq(&pfault_lock);
623 static int __init pfault_irq_init(void)
627 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
630 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
633 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
634 hotcpu_notifier(pfault_cpu_notify, 0);
638 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
643 early_initcall(pfault_irq_init);
645 #endif /* CONFIG_PFAULT */