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
53 static unsigned long store_indication;
57 if (test_facility(2) && test_facility(75))
58 store_indication = 0xc00;
61 static inline int notify_page_fault(struct pt_regs *regs)
65 /* kprobe_running() needs smp_processor_id() */
66 if (kprobes_built_in() && !user_mode(regs)) {
68 if (kprobe_running() && kprobe_fault_handler(regs, 14))
77 * Unlock any spinlocks which will prevent us from getting the
80 void bust_spinlocks(int yes)
85 int loglevel_save = console_loglevel;
89 * OK, the message is on the console. Now we call printk()
90 * without oops_in_progress set so that printk will give klogd
91 * a poke. Hold onto your hats...
93 console_loglevel = 15;
95 console_loglevel = loglevel_save;
100 * Returns the address space associated with the fault.
101 * Returns 0 for kernel space and 1 for user space.
103 static inline int user_space_fault(unsigned long trans_exc_code)
106 * The lowest two bits of the translation exception
107 * identification indicate which paging table was used.
110 if (trans_exc_code == 2)
111 /* Access via secondary space, set_fs setting decides */
112 return current->thread.mm_segment.ar4;
113 if (user_mode == HOME_SPACE_MODE)
114 /* User space if the access has been done via home space. */
115 return trans_exc_code == 3;
117 * If the user space is not the home space the kernel runs in home
118 * space. Access via secondary space has already been covered,
119 * access via primary space or access register is from user space
120 * and access via home space is from the kernel.
122 return trans_exc_code != 3;
125 static inline void report_user_fault(struct pt_regs *regs, long signr)
127 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
129 if (!unhandled_signal(current, signr))
131 if (!printk_ratelimit())
133 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
135 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
136 printk(KERN_CONT "\n");
137 printk(KERN_ALERT "failing address: %lX\n",
138 regs->int_parm_long & __FAIL_ADDR_MASK);
143 * Send SIGSEGV to task. This is an external routine
144 * to keep the stack usage of do_page_fault small.
146 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
150 report_user_fault(regs, SIGSEGV);
151 si.si_signo = SIGSEGV;
152 si.si_code = si_code;
153 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
154 force_sig_info(SIGSEGV, &si, current);
157 static noinline void do_no_context(struct pt_regs *regs)
159 const struct exception_table_entry *fixup;
160 unsigned long address;
162 /* Are we prepared to handle this kernel fault? */
163 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
165 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
170 * Oops. The kernel tried to access some bad page. We'll have to
171 * terminate things with extreme prejudice.
173 address = regs->int_parm_long & __FAIL_ADDR_MASK;
174 if (!user_space_fault(regs->int_parm_long))
175 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
176 " at virtual kernel address %p\n", (void *)address);
178 printk(KERN_ALERT "Unable to handle kernel paging request"
179 " at virtual user address %p\n", (void *)address);
185 static noinline void do_low_address(struct pt_regs *regs)
187 /* Low-address protection hit in kernel mode means
188 NULL pointer write access in kernel mode. */
189 if (regs->psw.mask & PSW_MASK_PSTATE) {
190 /* Low-address protection hit in user mode 'cannot happen'. */
191 die (regs, "Low-address protection");
198 static noinline void do_sigbus(struct pt_regs *regs)
200 struct task_struct *tsk = current;
204 * Send a sigbus, regardless of whether we were in kernel
207 si.si_signo = SIGBUS;
209 si.si_code = BUS_ADRERR;
210 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
211 force_sig_info(SIGBUS, &si, tsk);
214 static noinline void do_fault_error(struct pt_regs *regs, int fault)
219 case VM_FAULT_BADACCESS:
220 case VM_FAULT_BADMAP:
221 /* Bad memory access. Check if it is kernel or user space. */
222 if (regs->psw.mask & PSW_MASK_PSTATE) {
223 /* User mode accesses just cause a SIGSEGV */
224 si_code = (fault == VM_FAULT_BADMAP) ?
225 SEGV_MAPERR : SEGV_ACCERR;
226 do_sigsegv(regs, si_code);
229 case VM_FAULT_BADCONTEXT:
232 default: /* fault & VM_FAULT_ERROR */
233 if (fault & VM_FAULT_OOM) {
234 if (!(regs->psw.mask & PSW_MASK_PSTATE))
237 pagefault_out_of_memory();
238 } else if (fault & VM_FAULT_SIGBUS) {
239 /* Kernel mode? Handle exceptions or die */
240 if (!(regs->psw.mask & PSW_MASK_PSTATE))
251 * This routine handles page faults. It determines the address,
252 * and the problem, and then passes it off to one of the appropriate
255 * interruption code (int_code):
256 * 04 Protection -> Write-Protection (suprression)
257 * 10 Segment translation -> Not present (nullification)
258 * 11 Page translation -> Not present (nullification)
259 * 3b Region third trans. -> Not present (nullification)
261 static inline int do_exception(struct pt_regs *regs, int access)
263 struct task_struct *tsk;
264 struct mm_struct *mm;
265 struct vm_area_struct *vma;
266 unsigned long trans_exc_code;
267 unsigned long address;
271 if (notify_page_fault(regs))
276 trans_exc_code = regs->int_parm_long;
279 * Verify that the fault happened in user space, that
280 * we are not in an interrupt and that there is a
283 fault = VM_FAULT_BADCONTEXT;
284 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
287 address = trans_exc_code & __FAIL_ADDR_MASK;
288 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
289 flags = FAULT_FLAG_ALLOW_RETRY;
290 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
291 flags |= FAULT_FLAG_WRITE;
292 down_read(&mm->mmap_sem);
295 if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
296 address = __gmap_fault(address,
297 (struct gmap *) S390_lowcore.gmap);
298 if (address == -EFAULT) {
299 fault = VM_FAULT_BADMAP;
302 if (address == -ENOMEM) {
303 fault = VM_FAULT_OOM;
310 fault = VM_FAULT_BADMAP;
311 vma = find_vma(mm, address);
315 if (unlikely(vma->vm_start > address)) {
316 if (!(vma->vm_flags & VM_GROWSDOWN))
318 if (expand_stack(vma, address))
323 * Ok, we have a good vm_area for this memory access, so
326 fault = VM_FAULT_BADACCESS;
327 if (unlikely(!(vma->vm_flags & access)))
330 if (is_vm_hugetlb_page(vma))
331 address &= HPAGE_MASK;
333 * If for any reason at all we couldn't handle the fault,
334 * make sure we exit gracefully rather than endlessly redo
337 fault = handle_mm_fault(mm, vma, address, flags);
338 if (unlikely(fault & VM_FAULT_ERROR))
342 * Major/minor page fault accounting is only done on the
343 * initial attempt. If we go through a retry, it is extremely
344 * likely that the page will be found in page cache at that point.
346 if (flags & FAULT_FLAG_ALLOW_RETRY) {
347 if (fault & VM_FAULT_MAJOR) {
349 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
353 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
356 if (fault & VM_FAULT_RETRY) {
357 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
359 flags &= ~FAULT_FLAG_ALLOW_RETRY;
360 down_read(&mm->mmap_sem);
365 * The instruction that caused the program check will
366 * be repeated. Don't signal single step via SIGTRAP.
368 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
371 up_read(&mm->mmap_sem);
376 void __kprobes do_protection_exception(struct pt_regs *regs)
378 unsigned long trans_exc_code;
381 trans_exc_code = regs->int_parm_long;
382 /* Protection exception is suppressing, decrement psw address. */
383 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
385 * Check for low-address protection. This needs to be treated
386 * as a special case because the translation exception code
387 * field is not guaranteed to contain valid data in this case.
389 if (unlikely(!(trans_exc_code & 4))) {
390 do_low_address(regs);
393 fault = do_exception(regs, VM_WRITE);
395 do_fault_error(regs, fault);
398 void __kprobes do_dat_exception(struct pt_regs *regs)
402 access = VM_READ | VM_EXEC | VM_WRITE;
403 fault = do_exception(regs, access);
405 do_fault_error(regs, fault);
409 void __kprobes do_asce_exception(struct pt_regs *regs)
411 struct mm_struct *mm = current->mm;
412 struct vm_area_struct *vma;
413 unsigned long trans_exc_code;
415 trans_exc_code = regs->int_parm_long;
416 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
419 down_read(&mm->mmap_sem);
420 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
421 up_read(&mm->mmap_sem);
424 update_mm(mm, current);
428 /* User mode accesses just cause a SIGSEGV */
429 if (regs->psw.mask & PSW_MASK_PSTATE) {
430 do_sigsegv(regs, SEGV_MAPERR);
439 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
444 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
445 if (!irqs_disabled())
446 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
447 regs.psw.addr = (unsigned long) __builtin_return_address(0);
448 regs.psw.addr |= PSW_ADDR_AMODE;
449 regs.int_code = pgm_int_code;
450 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
451 access = write ? VM_WRITE : VM_READ;
452 fault = do_exception(®s, access);
453 if (unlikely(fault)) {
454 if (fault & VM_FAULT_OOM)
456 else if (fault & VM_FAULT_SIGBUS)
459 return fault ? -EFAULT : 0;
464 * 'pfault' pseudo page faults routines.
466 static int pfault_disable;
468 static int __init nopfault(char *str)
474 __setup("nopfault", nopfault);
476 struct pfault_refbk {
485 } __attribute__ ((packed, aligned(8)));
487 int pfault_init(void)
489 struct pfault_refbk refbk = {
494 .refgaddr = __LC_CURRENT_PID,
495 .refselmk = 1ULL << 48,
496 .refcmpmk = 1ULL << 48,
497 .reserved = __PF_RES_FIELD };
503 " diag %1,%0,0x258\n"
508 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
512 void pfault_fini(void)
514 struct pfault_refbk refbk = {
527 : : "a" (&refbk), "m" (refbk) : "cc");
530 static DEFINE_SPINLOCK(pfault_lock);
531 static LIST_HEAD(pfault_list);
533 static void pfault_interrupt(struct ext_code ext_code,
534 unsigned int param32, unsigned long param64)
536 struct task_struct *tsk;
541 * Get the external interruption subcode & pfault
542 * initial/completion signal bit. VM stores this
543 * in the 'cpu address' field associated with the
544 * external interrupt.
546 subcode = ext_code.subcode;
547 if ((subcode & 0xff00) != __SUBCODE_MASK)
549 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
550 /* Get the token (= pid of the affected task). */
551 pid = sizeof(void *) == 4 ? param32 : param64;
553 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
555 get_task_struct(tsk);
559 spin_lock(&pfault_lock);
560 if (subcode & 0x0080) {
561 /* signal bit is set -> a page has been swapped in by VM */
562 if (tsk->thread.pfault_wait == 1) {
563 /* Initial interrupt was faster than the completion
564 * interrupt. pfault_wait is valid. Set pfault_wait
565 * back to zero and wake up the process. This can
566 * safely be done because the task is still sleeping
567 * and can't produce new pfaults. */
568 tsk->thread.pfault_wait = 0;
569 list_del(&tsk->thread.list);
570 wake_up_process(tsk);
571 put_task_struct(tsk);
573 /* Completion interrupt was faster than initial
574 * interrupt. Set pfault_wait to -1 so the initial
575 * interrupt doesn't put the task to sleep.
576 * If the task is not running, ignore the completion
577 * interrupt since it must be a leftover of a PFAULT
578 * CANCEL operation which didn't remove all pending
579 * completion interrupts. */
580 if (tsk->state == TASK_RUNNING)
581 tsk->thread.pfault_wait = -1;
584 /* signal bit not set -> a real page is missing. */
585 if (WARN_ON_ONCE(tsk != current))
587 if (tsk->thread.pfault_wait == 1) {
588 /* Already on the list with a reference: put to sleep */
589 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
590 set_tsk_need_resched(tsk);
591 } else if (tsk->thread.pfault_wait == -1) {
592 /* Completion interrupt was faster than the initial
593 * interrupt (pfault_wait == -1). Set pfault_wait
594 * back to zero and exit. */
595 tsk->thread.pfault_wait = 0;
597 /* Initial interrupt arrived before completion
598 * interrupt. Let the task sleep.
599 * An extra task reference is needed since a different
600 * cpu may set the task state to TASK_RUNNING again
601 * before the scheduler is reached. */
602 get_task_struct(tsk);
603 tsk->thread.pfault_wait = 1;
604 list_add(&tsk->thread.list, &pfault_list);
605 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
606 set_tsk_need_resched(tsk);
610 spin_unlock(&pfault_lock);
611 put_task_struct(tsk);
614 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
615 unsigned long action, void *hcpu)
617 struct thread_struct *thread, *next;
618 struct task_struct *tsk;
622 case CPU_DEAD_FROZEN:
623 spin_lock_irq(&pfault_lock);
624 list_for_each_entry_safe(thread, next, &pfault_list, list) {
625 thread->pfault_wait = 0;
626 list_del(&thread->list);
627 tsk = container_of(thread, struct task_struct, thread);
628 wake_up_process(tsk);
629 put_task_struct(tsk);
631 spin_unlock_irq(&pfault_lock);
639 static int __init pfault_irq_init(void)
643 rc = register_external_interrupt(0x2603, pfault_interrupt);
646 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
649 service_subclass_irq_register();
650 hotcpu_notifier(pfault_cpu_notify, 0);
654 unregister_external_interrupt(0x2603, pfault_interrupt);
659 early_initcall(pfault_irq_init);
661 #endif /* CONFIG_PFAULT */