5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
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 <asm/compat.h>
40 #include "../kernel/entry.h"
43 #define __FAIL_ADDR_MASK 0x7ffff000
44 #define __SUBCODE_MASK 0x0200
45 #define __PF_RES_FIELD 0ULL
46 #else /* CONFIG_64BIT */
47 #define __FAIL_ADDR_MASK -4096L
48 #define __SUBCODE_MASK 0x0600
49 #define __PF_RES_FIELD 0x8000000000000000ULL
50 #endif /* CONFIG_64BIT */
52 #define VM_FAULT_BADCONTEXT 0x010000
53 #define VM_FAULT_BADMAP 0x020000
54 #define VM_FAULT_BADACCESS 0x040000
56 static unsigned long store_indication;
60 if (test_facility(2) && test_facility(75))
61 store_indication = 0xc00;
64 static inline int notify_page_fault(struct pt_regs *regs)
68 /* kprobe_running() needs smp_processor_id() */
69 if (kprobes_built_in() && !user_mode(regs)) {
71 if (kprobe_running() && kprobe_fault_handler(regs, 14))
80 * Unlock any spinlocks which will prevent us from getting the
83 void bust_spinlocks(int yes)
88 int loglevel_save = console_loglevel;
92 * OK, the message is on the console. Now we call printk()
93 * without oops_in_progress set so that printk will give klogd
94 * a poke. Hold onto your hats...
96 console_loglevel = 15;
98 console_loglevel = loglevel_save;
103 * Returns the address space associated with the fault.
104 * Returns 0 for kernel space and 1 for user space.
106 static inline int user_space_fault(unsigned long trans_exc_code)
109 * The lowest two bits of the translation exception
110 * identification indicate which paging table was used.
113 if (trans_exc_code == 2)
114 /* Access via secondary space, set_fs setting decides */
115 return current->thread.mm_segment.ar4;
116 if (user_mode == HOME_SPACE_MODE)
117 /* User space if the access has been done via home space. */
118 return trans_exc_code == 3;
120 * If the user space is not the home space the kernel runs in home
121 * space. Access via secondary space has already been covered,
122 * access via primary space or access register is from user space
123 * and access via home space is from the kernel.
125 return trans_exc_code != 3;
128 static inline void report_user_fault(struct pt_regs *regs, long signr)
130 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
132 if (!unhandled_signal(current, signr))
134 if (!printk_ratelimit())
136 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
138 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
139 printk(KERN_CONT "\n");
140 printk(KERN_ALERT "failing address: %lX\n",
141 regs->int_parm_long & __FAIL_ADDR_MASK);
146 * Send SIGSEGV to task. This is an external routine
147 * to keep the stack usage of do_page_fault small.
149 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
153 report_user_fault(regs, SIGSEGV);
154 si.si_signo = SIGSEGV;
155 si.si_code = si_code;
156 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
157 force_sig_info(SIGSEGV, &si, current);
160 static noinline void do_no_context(struct pt_regs *regs)
162 const struct exception_table_entry *fixup;
163 unsigned long address;
165 /* Are we prepared to handle this kernel fault? */
166 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
168 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
173 * Oops. The kernel tried to access some bad page. We'll have to
174 * terminate things with extreme prejudice.
176 address = regs->int_parm_long & __FAIL_ADDR_MASK;
177 if (!user_space_fault(regs->int_parm_long))
178 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
179 " at virtual kernel address %p\n", (void *)address);
181 printk(KERN_ALERT "Unable to handle kernel paging request"
182 " at virtual user address %p\n", (void *)address);
188 static noinline void do_low_address(struct pt_regs *regs)
190 /* Low-address protection hit in kernel mode means
191 NULL pointer write access in kernel mode. */
192 if (regs->psw.mask & PSW_MASK_PSTATE) {
193 /* Low-address protection hit in user mode 'cannot happen'. */
194 die (regs, "Low-address protection");
201 static noinline void do_sigbus(struct pt_regs *regs)
203 struct task_struct *tsk = current;
207 * Send a sigbus, regardless of whether we were in kernel
210 si.si_signo = SIGBUS;
212 si.si_code = BUS_ADRERR;
213 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
214 force_sig_info(SIGBUS, &si, tsk);
217 static noinline void do_fault_error(struct pt_regs *regs, int fault)
222 case VM_FAULT_BADACCESS:
223 case VM_FAULT_BADMAP:
224 /* Bad memory access. Check if it is kernel or user space. */
225 if (regs->psw.mask & PSW_MASK_PSTATE) {
226 /* User mode accesses just cause a SIGSEGV */
227 si_code = (fault == VM_FAULT_BADMAP) ?
228 SEGV_MAPERR : SEGV_ACCERR;
229 do_sigsegv(regs, si_code);
232 case VM_FAULT_BADCONTEXT:
235 default: /* fault & VM_FAULT_ERROR */
236 if (fault & VM_FAULT_OOM) {
237 if (!(regs->psw.mask & PSW_MASK_PSTATE))
240 pagefault_out_of_memory();
241 } else if (fault & VM_FAULT_SIGBUS) {
242 /* Kernel mode? Handle exceptions or die */
243 if (!(regs->psw.mask & PSW_MASK_PSTATE))
254 * This routine handles page faults. It determines the address,
255 * and the problem, and then passes it off to one of the appropriate
258 * interruption code (int_code):
259 * 04 Protection -> Write-Protection (suprression)
260 * 10 Segment translation -> Not present (nullification)
261 * 11 Page translation -> Not present (nullification)
262 * 3b Region third trans. -> Not present (nullification)
264 static inline int do_exception(struct pt_regs *regs, int access)
266 struct task_struct *tsk;
267 struct mm_struct *mm;
268 struct vm_area_struct *vma;
269 unsigned long trans_exc_code;
270 unsigned long address;
274 if (notify_page_fault(regs))
279 trans_exc_code = regs->int_parm_long;
282 * Verify that the fault happened in user space, that
283 * we are not in an interrupt and that there is a
286 fault = VM_FAULT_BADCONTEXT;
287 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
290 address = trans_exc_code & __FAIL_ADDR_MASK;
291 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
292 flags = FAULT_FLAG_ALLOW_RETRY;
293 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
294 flags |= FAULT_FLAG_WRITE;
295 down_read(&mm->mmap_sem);
298 if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
299 address = __gmap_fault(address,
300 (struct gmap *) S390_lowcore.gmap);
301 if (address == -EFAULT) {
302 fault = VM_FAULT_BADMAP;
305 if (address == -ENOMEM) {
306 fault = VM_FAULT_OOM;
313 fault = VM_FAULT_BADMAP;
314 vma = find_vma(mm, address);
318 if (unlikely(vma->vm_start > address)) {
319 if (!(vma->vm_flags & VM_GROWSDOWN))
321 if (expand_stack(vma, address))
326 * Ok, we have a good vm_area for this memory access, so
329 fault = VM_FAULT_BADACCESS;
330 if (unlikely(!(vma->vm_flags & access)))
333 if (is_vm_hugetlb_page(vma))
334 address &= HPAGE_MASK;
336 * If for any reason at all we couldn't handle the fault,
337 * make sure we exit gracefully rather than endlessly redo
340 fault = handle_mm_fault(mm, vma, address, flags);
341 if (unlikely(fault & VM_FAULT_ERROR))
345 * Major/minor page fault accounting is only done on the
346 * initial attempt. If we go through a retry, it is extremely
347 * likely that the page will be found in page cache at that point.
349 if (flags & FAULT_FLAG_ALLOW_RETRY) {
350 if (fault & VM_FAULT_MAJOR) {
352 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
356 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
359 if (fault & VM_FAULT_RETRY) {
360 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
362 flags &= ~FAULT_FLAG_ALLOW_RETRY;
363 down_read(&mm->mmap_sem);
368 * The instruction that caused the program check will
369 * be repeated. Don't signal single step via SIGTRAP.
371 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
374 up_read(&mm->mmap_sem);
379 void __kprobes do_protection_exception(struct pt_regs *regs)
381 unsigned long trans_exc_code;
384 trans_exc_code = regs->int_parm_long;
385 /* Protection exception is suppressing, decrement psw address. */
386 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
388 * Check for low-address protection. This needs to be treated
389 * as a special case because the translation exception code
390 * field is not guaranteed to contain valid data in this case.
392 if (unlikely(!(trans_exc_code & 4))) {
393 do_low_address(regs);
396 fault = do_exception(regs, VM_WRITE);
398 do_fault_error(regs, fault);
401 void __kprobes do_dat_exception(struct pt_regs *regs)
405 access = VM_READ | VM_EXEC | VM_WRITE;
406 fault = do_exception(regs, access);
408 do_fault_error(regs, fault);
412 void __kprobes do_asce_exception(struct pt_regs *regs)
414 struct mm_struct *mm = current->mm;
415 struct vm_area_struct *vma;
416 unsigned long trans_exc_code;
418 trans_exc_code = regs->int_parm_long;
419 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
422 down_read(&mm->mmap_sem);
423 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
424 up_read(&mm->mmap_sem);
427 update_mm(mm, current);
431 /* User mode accesses just cause a SIGSEGV */
432 if (regs->psw.mask & PSW_MASK_PSTATE) {
433 do_sigsegv(regs, SEGV_MAPERR);
442 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
447 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
448 if (!irqs_disabled())
449 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
450 regs.psw.addr = (unsigned long) __builtin_return_address(0);
451 regs.psw.addr |= PSW_ADDR_AMODE;
452 regs.int_code = pgm_int_code;
453 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
454 access = write ? VM_WRITE : VM_READ;
455 fault = do_exception(®s, access);
456 if (unlikely(fault)) {
457 if (fault & VM_FAULT_OOM)
459 else if (fault & VM_FAULT_SIGBUS)
462 return fault ? -EFAULT : 0;
467 * 'pfault' pseudo page faults routines.
469 static int pfault_disable;
471 static int __init nopfault(char *str)
477 __setup("nopfault", nopfault);
479 struct pfault_refbk {
488 } __attribute__ ((packed, aligned(8)));
490 int pfault_init(void)
492 struct pfault_refbk refbk = {
497 .refgaddr = __LC_CURRENT_PID,
498 .refselmk = 1ULL << 48,
499 .refcmpmk = 1ULL << 48,
500 .reserved = __PF_RES_FIELD };
506 " diag %1,%0,0x258\n"
511 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
515 void pfault_fini(void)
517 struct pfault_refbk refbk = {
530 : : "a" (&refbk), "m" (refbk) : "cc");
533 static DEFINE_SPINLOCK(pfault_lock);
534 static LIST_HEAD(pfault_list);
536 static void pfault_interrupt(unsigned int ext_int_code,
537 unsigned int param32, unsigned long param64)
539 struct task_struct *tsk;
544 * Get the external interruption subcode & pfault
545 * initial/completion signal bit. VM stores this
546 * in the 'cpu address' field associated with the
547 * external interrupt.
549 subcode = ext_int_code >> 16;
550 if ((subcode & 0xff00) != __SUBCODE_MASK)
552 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
553 if (subcode & 0x0080) {
554 /* Get the token (= pid of the affected task). */
555 pid = sizeof(void *) == 4 ? param32 : param64;
557 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
559 get_task_struct(tsk);
566 spin_lock(&pfault_lock);
567 if (subcode & 0x0080) {
568 /* signal bit is set -> a page has been swapped in by VM */
569 if (tsk->thread.pfault_wait == 1) {
570 /* Initial interrupt was faster than the completion
571 * interrupt. pfault_wait is valid. Set pfault_wait
572 * back to zero and wake up the process. This can
573 * safely be done because the task is still sleeping
574 * and can't produce new pfaults. */
575 tsk->thread.pfault_wait = 0;
576 list_del(&tsk->thread.list);
577 wake_up_process(tsk);
579 /* Completion interrupt was faster than initial
580 * interrupt. Set pfault_wait to -1 so the initial
581 * interrupt doesn't put the task to sleep.
582 * If the task is not running, ignore the completion
583 * interrupt since it must be a leftover of a PFAULT
584 * CANCEL operation which didn't remove all pending
585 * completion interrupts. */
586 if (tsk->state == TASK_RUNNING)
587 tsk->thread.pfault_wait = -1;
589 put_task_struct(tsk);
591 /* signal bit not set -> a real page is missing. */
592 if (tsk->thread.pfault_wait == -1) {
593 /* Completion interrupt was faster than the initial
594 * interrupt (pfault_wait == -1). Set pfault_wait
595 * back to zero and exit. */
596 tsk->thread.pfault_wait = 0;
598 /* Initial interrupt arrived before completion
599 * interrupt. Let the task sleep. */
600 tsk->thread.pfault_wait = 1;
601 list_add(&tsk->thread.list, &pfault_list);
602 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
603 set_tsk_need_resched(tsk);
606 spin_unlock(&pfault_lock);
609 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
610 unsigned long action, void *hcpu)
612 struct thread_struct *thread, *next;
613 struct task_struct *tsk;
617 case CPU_DEAD_FROZEN:
618 spin_lock_irq(&pfault_lock);
619 list_for_each_entry_safe(thread, next, &pfault_list, list) {
620 thread->pfault_wait = 0;
621 list_del(&thread->list);
622 tsk = container_of(thread, struct task_struct, thread);
623 wake_up_process(tsk);
625 spin_unlock_irq(&pfault_lock);
633 static int __init pfault_irq_init(void)
637 rc = register_external_interrupt(0x2603, pfault_interrupt);
640 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
643 service_subclass_irq_register();
644 hotcpu_notifier(pfault_cpu_notify, 0);
648 unregister_external_interrupt(0x2603, pfault_interrupt);
653 early_initcall(pfault_irq_init);
655 #endif /* CONFIG_PFAULT */