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_SIGSEGV) {
253 /* Kernel mode? Handle exceptions or die */
254 if (!user_mode(regs))
255 do_no_context(regs, int_code, trans_exc_code);
257 do_sigsegv(regs, int_code, SEGV_MAPERR,
259 } else if (fault & VM_FAULT_SIGBUS) {
260 /* Kernel mode? Handle exceptions or die */
261 if (!(regs->psw.mask & PSW_MASK_PSTATE))
262 do_no_context(regs, int_code, trans_exc_code);
264 do_sigbus(regs, int_code, trans_exc_code);
272 * This routine handles page faults. It determines the address,
273 * and the problem, and then passes it off to one of the appropriate
276 * interruption code (int_code):
277 * 04 Protection -> Write-Protection (suprression)
278 * 10 Segment translation -> Not present (nullification)
279 * 11 Page translation -> Not present (nullification)
280 * 3b Region third trans. -> Not present (nullification)
282 static inline int do_exception(struct pt_regs *regs, int access,
283 unsigned long trans_exc_code)
285 struct task_struct *tsk;
286 struct mm_struct *mm;
287 struct vm_area_struct *vma;
288 unsigned long address;
292 if (notify_page_fault(regs))
299 * Verify that the fault happened in user space, that
300 * we are not in an interrupt and that there is a
303 fault = VM_FAULT_BADCONTEXT;
304 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
307 address = trans_exc_code & __FAIL_ADDR_MASK;
308 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
309 flags = FAULT_FLAG_ALLOW_RETRY;
310 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
311 flags |= FAULT_FLAG_WRITE;
312 down_read(&mm->mmap_sem);
315 if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
316 address = __gmap_fault(address,
317 (struct gmap *) S390_lowcore.gmap);
318 if (address == -EFAULT) {
319 fault = VM_FAULT_BADMAP;
322 if (address == -ENOMEM) {
323 fault = VM_FAULT_OOM;
330 fault = VM_FAULT_BADMAP;
331 vma = find_vma(mm, address);
335 if (unlikely(vma->vm_start > address)) {
336 if (!(vma->vm_flags & VM_GROWSDOWN))
338 if (expand_stack(vma, address))
343 * Ok, we have a good vm_area for this memory access, so
346 fault = VM_FAULT_BADACCESS;
347 if (unlikely(!(vma->vm_flags & access)))
350 if (is_vm_hugetlb_page(vma))
351 address &= HPAGE_MASK;
353 * If for any reason at all we couldn't handle the fault,
354 * make sure we exit gracefully rather than endlessly redo
357 fault = handle_mm_fault(mm, vma, address, flags);
358 if (unlikely(fault & VM_FAULT_ERROR))
362 * Major/minor page fault accounting is only done on the
363 * initial attempt. If we go through a retry, it is extremely
364 * likely that the page will be found in page cache at that point.
366 if (flags & FAULT_FLAG_ALLOW_RETRY) {
367 if (fault & VM_FAULT_MAJOR) {
369 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
373 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
376 if (fault & VM_FAULT_RETRY) {
377 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
379 flags &= ~FAULT_FLAG_ALLOW_RETRY;
380 down_read(&mm->mmap_sem);
385 * The instruction that caused the program check will
386 * be repeated. Don't signal single step via SIGTRAP.
388 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
391 up_read(&mm->mmap_sem);
396 void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code,
397 unsigned long trans_exc_code)
401 /* Protection exception is suppressing, decrement psw address. */
402 regs->psw.addr = __rewind_psw(regs->psw, pgm_int_code >> 16);
404 * Check for low-address protection. This needs to be treated
405 * as a special case because the translation exception code
406 * field is not guaranteed to contain valid data in this case.
408 if (unlikely(!(trans_exc_code & 4))) {
409 do_low_address(regs, pgm_int_code, trans_exc_code);
412 fault = do_exception(regs, VM_WRITE, trans_exc_code);
414 do_fault_error(regs, 4, trans_exc_code, fault);
417 void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code,
418 unsigned long trans_exc_code)
422 access = VM_READ | VM_EXEC | VM_WRITE;
423 fault = do_exception(regs, access, trans_exc_code);
425 do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault);
429 void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code,
430 unsigned long trans_exc_code)
432 struct mm_struct *mm = current->mm;
433 struct vm_area_struct *vma;
435 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
438 down_read(&mm->mmap_sem);
439 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
440 up_read(&mm->mmap_sem);
443 update_mm(mm, current);
447 /* User mode accesses just cause a SIGSEGV */
448 if (regs->psw.mask & PSW_MASK_PSTATE) {
449 do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code);
454 do_no_context(regs, pgm_int_code, trans_exc_code);
458 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
463 /* Emulate a uaccess fault from kernel mode. */
464 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
465 if (!irqs_disabled())
466 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
467 regs.psw.addr = (unsigned long) __builtin_return_address(0);
468 regs.psw.addr |= PSW_ADDR_AMODE;
470 access = write ? VM_WRITE : VM_READ;
471 fault = do_exception(®s, access, uaddr | 2);
473 * Since the fault happened in kernel mode while performing a uaccess
474 * all we need to do now is emulating a fixup in case "fault" is not
476 * For the calling uaccess functions this results always in -EFAULT.
478 return fault ? -EFAULT : 0;
483 * 'pfault' pseudo page faults routines.
485 static int pfault_disable;
487 static int __init nopfault(char *str)
493 __setup("nopfault", nopfault);
495 struct pfault_refbk {
504 } __attribute__ ((packed, aligned(8)));
506 int pfault_init(void)
508 struct pfault_refbk refbk = {
513 .refgaddr = __LC_CURRENT_PID,
514 .refselmk = 1ULL << 48,
515 .refcmpmk = 1ULL << 48,
516 .reserved = __PF_RES_FIELD };
519 if (!MACHINE_IS_VM || pfault_disable)
522 " diag %1,%0,0x258\n"
527 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
531 void pfault_fini(void)
533 struct pfault_refbk refbk = {
540 if (!MACHINE_IS_VM || pfault_disable)
546 : : "a" (&refbk), "m" (refbk) : "cc");
549 static DEFINE_SPINLOCK(pfault_lock);
550 static LIST_HEAD(pfault_list);
552 static void pfault_interrupt(unsigned int ext_int_code,
553 unsigned int param32, unsigned long param64)
555 struct task_struct *tsk;
560 * Get the external interruption subcode & pfault
561 * initial/completion signal bit. VM stores this
562 * in the 'cpu address' field associated with the
563 * external interrupt.
565 subcode = ext_int_code >> 16;
566 if ((subcode & 0xff00) != __SUBCODE_MASK)
568 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
569 if (subcode & 0x0080) {
570 /* Get the token (= pid of the affected task). */
571 pid = sizeof(void *) == 4 ? param32 : param64;
573 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
575 get_task_struct(tsk);
582 spin_lock(&pfault_lock);
583 if (subcode & 0x0080) {
584 /* signal bit is set -> a page has been swapped in by VM */
585 if (tsk->thread.pfault_wait == 1) {
586 /* Initial interrupt was faster than the completion
587 * interrupt. pfault_wait is valid. Set pfault_wait
588 * back to zero and wake up the process. This can
589 * safely be done because the task is still sleeping
590 * and can't produce new pfaults. */
591 tsk->thread.pfault_wait = 0;
592 list_del(&tsk->thread.list);
593 wake_up_process(tsk);
594 put_task_struct(tsk);
596 /* Completion interrupt was faster than initial
597 * interrupt. Set pfault_wait to -1 so the initial
598 * interrupt doesn't put the task to sleep.
599 * If the task is not running, ignore the completion
600 * interrupt since it must be a leftover of a PFAULT
601 * CANCEL operation which didn't remove all pending
602 * completion interrupts. */
603 if (tsk->state == TASK_RUNNING)
604 tsk->thread.pfault_wait = -1;
606 put_task_struct(tsk);
608 /* signal bit not set -> a real page is missing. */
609 if (tsk->thread.pfault_wait == 1) {
610 /* Already on the list with a reference: put to sleep */
611 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
612 set_tsk_need_resched(tsk);
613 } else if (tsk->thread.pfault_wait == -1) {
614 /* Completion interrupt was faster than the initial
615 * interrupt (pfault_wait == -1). Set pfault_wait
616 * back to zero and exit. */
617 tsk->thread.pfault_wait = 0;
619 /* Initial interrupt arrived before completion
620 * interrupt. Let the task sleep.
621 * An extra task reference is needed since a different
622 * cpu may set the task state to TASK_RUNNING again
623 * before the scheduler is reached. */
624 get_task_struct(tsk);
625 tsk->thread.pfault_wait = 1;
626 list_add(&tsk->thread.list, &pfault_list);
627 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
628 set_tsk_need_resched(tsk);
631 spin_unlock(&pfault_lock);
634 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
635 unsigned long action, void *hcpu)
637 struct thread_struct *thread, *next;
638 struct task_struct *tsk;
642 case CPU_DEAD_FROZEN:
643 spin_lock_irq(&pfault_lock);
644 list_for_each_entry_safe(thread, next, &pfault_list, list) {
645 thread->pfault_wait = 0;
646 list_del(&thread->list);
647 tsk = container_of(thread, struct task_struct, thread);
648 wake_up_process(tsk);
649 put_task_struct(tsk);
651 spin_unlock_irq(&pfault_lock);
659 static int __init pfault_irq_init(void)
665 rc = register_external_interrupt(0x2603, pfault_interrupt);
668 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
671 service_subclass_irq_register();
672 hotcpu_notifier(pfault_cpu_notify, 0);
676 unregister_external_interrupt(0x2603, pfault_interrupt);
681 early_initcall(pfault_irq_init);
683 #endif /* CONFIG_PFAULT */