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
54 static unsigned long store_indication __read_mostly;
57 static int __init fault_init(void)
59 if (test_facility(75))
60 store_indication = 0xc00;
63 early_initcall(fault_init);
66 static inline int notify_page_fault(struct pt_regs *regs)
70 /* kprobe_running() needs smp_processor_id() */
71 if (kprobes_built_in() && !user_mode(regs)) {
73 if (kprobe_running() && kprobe_fault_handler(regs, 14))
82 * Unlock any spinlocks which will prevent us from getting the
85 void bust_spinlocks(int yes)
90 int loglevel_save = console_loglevel;
94 * OK, the message is on the console. Now we call printk()
95 * without oops_in_progress set so that printk will give klogd
96 * a poke. Hold onto your hats...
98 console_loglevel = 15;
100 console_loglevel = loglevel_save;
105 * Returns the address space associated with the fault.
106 * Returns 0 for kernel space and 1 for user space.
108 static inline int user_space_fault(unsigned long trans_exc_code)
111 * The lowest two bits of the translation exception
112 * identification indicate which paging table was used.
115 if (trans_exc_code == 2)
116 /* Access via secondary space, set_fs setting decides */
117 return current->thread.mm_segment.ar4;
118 if (s390_user_mode == HOME_SPACE_MODE)
119 /* User space if the access has been done via home space. */
120 return trans_exc_code == 3;
122 * If the user space is not the home space the kernel runs in home
123 * space. Access via secondary space has already been covered,
124 * access via primary space or access register is from user space
125 * and access via home space is from the kernel.
127 return trans_exc_code != 3;
130 static inline void report_user_fault(struct pt_regs *regs, long signr)
132 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
134 if (!unhandled_signal(current, signr))
136 if (!printk_ratelimit())
138 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
140 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
141 printk(KERN_CONT "\n");
142 printk(KERN_ALERT "failing address: %lX\n",
143 regs->int_parm_long & __FAIL_ADDR_MASK);
148 * Send SIGSEGV to task. This is an external routine
149 * to keep the stack usage of do_page_fault small.
151 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
155 report_user_fault(regs, SIGSEGV);
156 si.si_signo = SIGSEGV;
157 si.si_code = si_code;
158 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
159 force_sig_info(SIGSEGV, &si, current);
162 static noinline void do_no_context(struct pt_regs *regs)
164 const struct exception_table_entry *fixup;
165 unsigned long address;
167 /* Are we prepared to handle this kernel fault? */
168 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
170 regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
175 * Oops. The kernel tried to access some bad page. We'll have to
176 * terminate things with extreme prejudice.
178 address = regs->int_parm_long & __FAIL_ADDR_MASK;
179 if (!user_space_fault(regs->int_parm_long))
180 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
181 " at virtual kernel address %p\n", (void *)address);
183 printk(KERN_ALERT "Unable to handle kernel paging request"
184 " at virtual user address %p\n", (void *)address);
190 static noinline void do_low_address(struct pt_regs *regs)
192 /* Low-address protection hit in kernel mode means
193 NULL pointer write access in kernel mode. */
194 if (regs->psw.mask & PSW_MASK_PSTATE) {
195 /* Low-address protection hit in user mode 'cannot happen'. */
196 die (regs, "Low-address protection");
203 static noinline void do_sigbus(struct pt_regs *regs)
205 struct task_struct *tsk = current;
209 * Send a sigbus, regardless of whether we were in kernel
212 si.si_signo = SIGBUS;
214 si.si_code = BUS_ADRERR;
215 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
216 force_sig_info(SIGBUS, &si, tsk);
219 static noinline void do_fault_error(struct pt_regs *regs, int fault)
224 case VM_FAULT_BADACCESS:
225 case VM_FAULT_BADMAP:
226 /* Bad memory access. Check if it is kernel or user space. */
227 if (user_mode(regs)) {
228 /* User mode accesses just cause a SIGSEGV */
229 si_code = (fault == VM_FAULT_BADMAP) ?
230 SEGV_MAPERR : SEGV_ACCERR;
231 do_sigsegv(regs, si_code);
234 case VM_FAULT_BADCONTEXT:
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)
272 struct task_struct *tsk;
273 struct mm_struct *mm;
274 struct vm_area_struct *vma;
275 unsigned long trans_exc_code;
276 unsigned long address;
282 * The instruction that caused the program check has
283 * been nullified. Don't signal single step via SIGTRAP.
285 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
287 if (notify_page_fault(regs))
291 trans_exc_code = regs->int_parm_long;
294 * Verify that the fault happened in user space, that
295 * we are not in an interrupt and that there is a
298 fault = VM_FAULT_BADCONTEXT;
299 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
302 address = trans_exc_code & __FAIL_ADDR_MASK;
303 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
304 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
305 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
306 flags |= FAULT_FLAG_WRITE;
307 down_read(&mm->mmap_sem);
310 if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
311 address = __gmap_fault(address,
312 (struct gmap *) S390_lowcore.gmap);
313 if (address == -EFAULT) {
314 fault = VM_FAULT_BADMAP;
317 if (address == -ENOMEM) {
318 fault = VM_FAULT_OOM;
325 fault = VM_FAULT_BADMAP;
326 vma = find_vma(mm, address);
330 if (unlikely(vma->vm_start > address)) {
331 if (!(vma->vm_flags & VM_GROWSDOWN))
333 if (expand_stack(vma, address))
338 * Ok, we have a good vm_area for this memory access, so
341 fault = VM_FAULT_BADACCESS;
342 if (unlikely(!(vma->vm_flags & access)))
345 if (is_vm_hugetlb_page(vma))
346 address &= HPAGE_MASK;
348 * If for any reason at all we couldn't handle the fault,
349 * make sure we exit gracefully rather than endlessly redo
352 fault = handle_mm_fault(mm, vma, address, flags);
353 /* No reason to continue if interrupted by SIGKILL. */
354 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
355 fault = VM_FAULT_SIGNAL;
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 flags |= FAULT_FLAG_TRIED;
381 down_read(&mm->mmap_sem);
387 up_read(&mm->mmap_sem);
392 void __kprobes do_protection_exception(struct pt_regs *regs)
394 unsigned long trans_exc_code;
397 trans_exc_code = regs->int_parm_long;
399 * Protection exceptions are suppressing, decrement psw address.
400 * The exception to this rule are aborted transactions, for these
401 * the PSW already points to the correct location.
403 if (!(regs->int_code & 0x200))
404 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
406 * Check for low-address protection. This needs to be treated
407 * as a special case because the translation exception code
408 * field is not guaranteed to contain valid data in this case.
410 if (unlikely(!(trans_exc_code & 4))) {
411 do_low_address(regs);
414 fault = do_exception(regs, VM_WRITE);
416 do_fault_error(regs, fault);
419 void __kprobes do_dat_exception(struct pt_regs *regs)
423 access = VM_READ | VM_EXEC | VM_WRITE;
424 fault = do_exception(regs, access);
426 do_fault_error(regs, fault);
430 void __kprobes do_asce_exception(struct pt_regs *regs)
432 struct mm_struct *mm = current->mm;
433 struct vm_area_struct *vma;
434 unsigned long trans_exc_code;
437 * The instruction that caused the program check has
438 * been nullified. Don't signal single step via SIGTRAP.
440 clear_tsk_thread_flag(current, TIF_PER_TRAP);
442 trans_exc_code = regs->int_parm_long;
443 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
446 down_read(&mm->mmap_sem);
447 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
448 up_read(&mm->mmap_sem);
451 update_mm(mm, current);
455 /* User mode accesses just cause a SIGSEGV */
456 if (user_mode(regs)) {
457 do_sigsegv(regs, SEGV_MAPERR);
466 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
471 /* Emulate a uaccess fault from kernel mode. */
472 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
473 if (!irqs_disabled())
474 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
475 regs.psw.addr = (unsigned long) __builtin_return_address(0);
476 regs.psw.addr |= PSW_ADDR_AMODE;
477 regs.int_code = pgm_int_code;
478 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
479 access = write ? VM_WRITE : VM_READ;
480 fault = do_exception(®s, access);
482 * Since the fault happened in kernel mode while performing a uaccess
483 * all we need to do now is emulating a fixup in case "fault" is not
485 * For the calling uaccess functions this results always in -EFAULT.
487 return fault ? -EFAULT : 0;
492 * 'pfault' pseudo page faults routines.
494 static int pfault_disable;
496 static int __init nopfault(char *str)
502 __setup("nopfault", nopfault);
504 struct pfault_refbk {
513 } __attribute__ ((packed, aligned(8)));
515 int pfault_init(void)
517 struct pfault_refbk refbk = {
522 .refgaddr = __LC_CURRENT_PID,
523 .refselmk = 1ULL << 48,
524 .refcmpmk = 1ULL << 48,
525 .reserved = __PF_RES_FIELD };
531 " diag %1,%0,0x258\n"
536 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
540 void pfault_fini(void)
542 struct pfault_refbk refbk = {
555 : : "a" (&refbk), "m" (refbk) : "cc");
558 static DEFINE_SPINLOCK(pfault_lock);
559 static LIST_HEAD(pfault_list);
561 static void pfault_interrupt(struct ext_code ext_code,
562 unsigned int param32, unsigned long param64)
564 struct task_struct *tsk;
569 * Get the external interruption subcode & pfault
570 * initial/completion signal bit. VM stores this
571 * in the 'cpu address' field associated with the
572 * external interrupt.
574 subcode = ext_code.subcode;
575 if ((subcode & 0xff00) != __SUBCODE_MASK)
577 inc_irq_stat(IRQEXT_PFL);
578 /* Get the token (= pid of the affected task). */
579 pid = sizeof(void *) == 4 ? param32 : param64;
581 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
583 get_task_struct(tsk);
587 spin_lock(&pfault_lock);
588 if (subcode & 0x0080) {
589 /* signal bit is set -> a page has been swapped in by VM */
590 if (tsk->thread.pfault_wait == 1) {
591 /* Initial interrupt was faster than the completion
592 * interrupt. pfault_wait is valid. Set pfault_wait
593 * back to zero and wake up the process. This can
594 * safely be done because the task is still sleeping
595 * and can't produce new pfaults. */
596 tsk->thread.pfault_wait = 0;
597 list_del(&tsk->thread.list);
598 wake_up_process(tsk);
599 put_task_struct(tsk);
601 /* Completion interrupt was faster than initial
602 * interrupt. Set pfault_wait to -1 so the initial
603 * interrupt doesn't put the task to sleep.
604 * If the task is not running, ignore the completion
605 * interrupt since it must be a leftover of a PFAULT
606 * CANCEL operation which didn't remove all pending
607 * completion interrupts. */
608 if (tsk->state == TASK_RUNNING)
609 tsk->thread.pfault_wait = -1;
612 /* signal bit not set -> a real page is missing. */
613 if (WARN_ON_ONCE(tsk != current))
615 if (tsk->thread.pfault_wait == 1) {
616 /* Already on the list with a reference: put to sleep */
617 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
618 set_tsk_need_resched(tsk);
619 } else if (tsk->thread.pfault_wait == -1) {
620 /* Completion interrupt was faster than the initial
621 * interrupt (pfault_wait == -1). Set pfault_wait
622 * back to zero and exit. */
623 tsk->thread.pfault_wait = 0;
625 /* Initial interrupt arrived before completion
626 * interrupt. Let the task sleep.
627 * An extra task reference is needed since a different
628 * cpu may set the task state to TASK_RUNNING again
629 * before the scheduler is reached. */
630 get_task_struct(tsk);
631 tsk->thread.pfault_wait = 1;
632 list_add(&tsk->thread.list, &pfault_list);
633 __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
634 set_tsk_need_resched(tsk);
638 spin_unlock(&pfault_lock);
639 put_task_struct(tsk);
642 static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
645 struct thread_struct *thread, *next;
646 struct task_struct *tsk;
648 switch (action & ~CPU_TASKS_FROZEN) {
650 spin_lock_irq(&pfault_lock);
651 list_for_each_entry_safe(thread, next, &pfault_list, list) {
652 thread->pfault_wait = 0;
653 list_del(&thread->list);
654 tsk = container_of(thread, struct task_struct, thread);
655 wake_up_process(tsk);
656 put_task_struct(tsk);
658 spin_unlock_irq(&pfault_lock);
666 static int __init pfault_irq_init(void)
670 rc = register_external_interrupt(0x2603, pfault_interrupt);
673 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
676 service_subclass_irq_register();
677 hotcpu_notifier(pfault_cpu_notify, 0);
681 unregister_external_interrupt(0x2603, pfault_interrupt);
686 early_initcall(pfault_irq_init);
688 #endif /* CONFIG_PFAULT */