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 signr)
129 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
131 if (!unhandled_signal(current, signr))
133 if (!printk_ratelimit())
135 printk(KERN_ALERT "User process fault: interruption code 0x%X ",
137 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
138 printk(KERN_CONT "\n");
139 printk(KERN_ALERT "failing address: %lX\n",
140 regs->int_parm_long & __FAIL_ADDR_MASK);
145 * Send SIGSEGV to task. This is an external routine
146 * to keep the stack usage of do_page_fault small.
148 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
152 report_user_fault(regs, SIGSEGV);
153 si.si_signo = SIGSEGV;
154 si.si_code = si_code;
155 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
156 force_sig_info(SIGSEGV, &si, current);
159 static noinline void do_no_context(struct pt_regs *regs)
161 const struct exception_table_entry *fixup;
162 unsigned long address;
164 /* Are we prepared to handle this kernel fault? */
165 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
167 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
172 * Oops. The kernel tried to access some bad page. We'll have to
173 * terminate things with extreme prejudice.
175 address = regs->int_parm_long & __FAIL_ADDR_MASK;
176 if (!user_space_fault(regs->int_parm_long))
177 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
178 " at virtual kernel address %p\n", (void *)address);
180 printk(KERN_ALERT "Unable to handle kernel paging request"
181 " at virtual user address %p\n", (void *)address);
187 static noinline void do_low_address(struct pt_regs *regs)
189 /* Low-address protection hit in kernel mode means
190 NULL pointer write access in kernel mode. */
191 if (regs->psw.mask & PSW_MASK_PSTATE) {
192 /* Low-address protection hit in user mode 'cannot happen'. */
193 die (regs, "Low-address protection");
200 static noinline void do_sigbus(struct pt_regs *regs)
202 struct task_struct *tsk = current;
206 * Send a sigbus, regardless of whether we were in kernel
209 si.si_signo = SIGBUS;
211 si.si_code = BUS_ADRERR;
212 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
213 force_sig_info(SIGBUS, &si, tsk);
216 static noinline void do_fault_error(struct pt_regs *regs, int fault)
221 case VM_FAULT_BADACCESS:
222 case VM_FAULT_BADMAP:
223 /* Bad memory access. Check if it is kernel or user space. */
224 if (regs->psw.mask & PSW_MASK_PSTATE) {
225 /* User mode accesses just cause a SIGSEGV */
226 si_code = (fault == VM_FAULT_BADMAP) ?
227 SEGV_MAPERR : SEGV_ACCERR;
228 do_sigsegv(regs, si_code);
231 case VM_FAULT_BADCONTEXT:
234 default: /* fault & VM_FAULT_ERROR */
235 if (fault & VM_FAULT_OOM) {
236 if (!(regs->psw.mask & PSW_MASK_PSTATE))
239 pagefault_out_of_memory();
240 } else if (fault & VM_FAULT_SIGBUS) {
241 /* Kernel mode? Handle exceptions or die */
242 if (!(regs->psw.mask & PSW_MASK_PSTATE))
253 * This routine handles page faults. It determines the address,
254 * and the problem, and then passes it off to one of the appropriate
257 * interruption code (int_code):
258 * 04 Protection -> Write-Protection (suprression)
259 * 10 Segment translation -> Not present (nullification)
260 * 11 Page translation -> Not present (nullification)
261 * 3b Region third trans. -> Not present (nullification)
263 static inline int do_exception(struct pt_regs *regs, int access)
265 struct task_struct *tsk;
266 struct mm_struct *mm;
267 struct vm_area_struct *vma;
268 unsigned long trans_exc_code;
269 unsigned long address;
273 if (notify_page_fault(regs))
278 trans_exc_code = regs->int_parm_long;
281 * Verify that the fault happened in user space, that
282 * we are not in an interrupt and that there is a
285 fault = VM_FAULT_BADCONTEXT;
286 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
289 address = trans_exc_code & __FAIL_ADDR_MASK;
290 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
291 flags = FAULT_FLAG_ALLOW_RETRY;
292 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
293 flags |= FAULT_FLAG_WRITE;
294 down_read(&mm->mmap_sem);
297 if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
298 address = __gmap_fault(address,
299 (struct gmap *) S390_lowcore.gmap);
300 if (address == -EFAULT) {
301 fault = VM_FAULT_BADMAP;
304 if (address == -ENOMEM) {
305 fault = VM_FAULT_OOM;
312 fault = VM_FAULT_BADMAP;
313 vma = find_vma(mm, address);
317 if (unlikely(vma->vm_start > address)) {
318 if (!(vma->vm_flags & VM_GROWSDOWN))
320 if (expand_stack(vma, address))
325 * Ok, we have a good vm_area for this memory access, so
328 fault = VM_FAULT_BADACCESS;
329 if (unlikely(!(vma->vm_flags & access)))
332 if (is_vm_hugetlb_page(vma))
333 address &= HPAGE_MASK;
335 * If for any reason at all we couldn't handle the fault,
336 * make sure we exit gracefully rather than endlessly redo
339 fault = handle_mm_fault(mm, vma, address, flags);
340 if (unlikely(fault & VM_FAULT_ERROR))
344 * Major/minor page fault accounting is only done on the
345 * initial attempt. If we go through a retry, it is extremely
346 * likely that the page will be found in page cache at that point.
348 if (flags & FAULT_FLAG_ALLOW_RETRY) {
349 if (fault & VM_FAULT_MAJOR) {
351 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
355 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
358 if (fault & VM_FAULT_RETRY) {
359 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
361 flags &= ~FAULT_FLAG_ALLOW_RETRY;
362 down_read(&mm->mmap_sem);
367 * The instruction that caused the program check will
368 * be repeated. Don't signal single step via SIGTRAP.
370 clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
373 up_read(&mm->mmap_sem);
378 void __kprobes do_protection_exception(struct pt_regs *regs)
380 unsigned long trans_exc_code;
383 trans_exc_code = regs->int_parm_long;
384 /* Protection exception is suppressing, decrement psw address. */
385 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
387 * Check for low-address protection. This needs to be treated
388 * as a special case because the translation exception code
389 * field is not guaranteed to contain valid data in this case.
391 if (unlikely(!(trans_exc_code & 4))) {
392 do_low_address(regs);
395 fault = do_exception(regs, VM_WRITE);
397 do_fault_error(regs, fault);
400 void __kprobes do_dat_exception(struct pt_regs *regs)
404 access = VM_READ | VM_EXEC | VM_WRITE;
405 fault = do_exception(regs, access);
407 do_fault_error(regs, fault);
411 void __kprobes do_asce_exception(struct pt_regs *regs)
413 struct mm_struct *mm = current->mm;
414 struct vm_area_struct *vma;
415 unsigned long trans_exc_code;
417 trans_exc_code = regs->int_parm_long;
418 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
421 down_read(&mm->mmap_sem);
422 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
423 up_read(&mm->mmap_sem);
426 update_mm(mm, current);
430 /* User mode accesses just cause a SIGSEGV */
431 if (regs->psw.mask & PSW_MASK_PSTATE) {
432 do_sigsegv(regs, SEGV_MAPERR);
441 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
446 regs.psw.mask = psw_kernel_bits | PSW_MASK_DAT | PSW_MASK_MCHECK;
447 if (!irqs_disabled())
448 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
449 regs.psw.addr = (unsigned long) __builtin_return_address(0);
450 regs.psw.addr |= PSW_ADDR_AMODE;
451 regs.int_code = pgm_int_code;
452 regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
453 access = write ? VM_WRITE : VM_READ;
454 fault = do_exception(®s, access);
455 if (unlikely(fault)) {
456 if (fault & VM_FAULT_OOM)
458 else if (fault & VM_FAULT_SIGBUS)
461 return fault ? -EFAULT : 0;
466 * 'pfault' pseudo page faults routines.
468 static int pfault_disable;
470 static int __init nopfault(char *str)
476 __setup("nopfault", nopfault);
478 struct pfault_refbk {
487 } __attribute__ ((packed, aligned(8)));
489 int pfault_init(void)
491 struct pfault_refbk refbk = {
496 .refgaddr = __LC_CURRENT_PID,
497 .refselmk = 1ULL << 48,
498 .refcmpmk = 1ULL << 48,
499 .reserved = __PF_RES_FIELD };
505 " diag %1,%0,0x258\n"
510 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
514 void pfault_fini(void)
516 struct pfault_refbk refbk = {
529 : : "a" (&refbk), "m" (refbk) : "cc");
532 static DEFINE_SPINLOCK(pfault_lock);
533 static LIST_HEAD(pfault_list);
535 static void pfault_interrupt(unsigned int ext_int_code,
536 unsigned int param32, unsigned long param64)
538 struct task_struct *tsk;
543 * Get the external interruption subcode & pfault
544 * initial/completion signal bit. VM stores this
545 * in the 'cpu address' field associated with the
546 * external interrupt.
548 subcode = ext_int_code >> 16;
549 if ((subcode & 0xff00) != __SUBCODE_MASK)
551 kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
552 if (subcode & 0x0080) {
553 /* Get the token (= pid of the affected task). */
554 pid = sizeof(void *) == 4 ? param32 : param64;
556 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
558 get_task_struct(tsk);
565 spin_lock(&pfault_lock);
566 if (subcode & 0x0080) {
567 /* signal bit is set -> a page has been swapped in by VM */
568 if (tsk->thread.pfault_wait == 1) {
569 /* Initial interrupt was faster than the completion
570 * interrupt. pfault_wait is valid. Set pfault_wait
571 * back to zero and wake up the process. This can
572 * safely be done because the task is still sleeping
573 * and can't produce new pfaults. */
574 tsk->thread.pfault_wait = 0;
575 list_del(&tsk->thread.list);
576 wake_up_process(tsk);
578 /* Completion interrupt was faster than initial
579 * interrupt. Set pfault_wait to -1 so the initial
580 * interrupt doesn't put the task to sleep.
581 * If the task is not running, ignore the completion
582 * interrupt since it must be a leftover of a PFAULT
583 * CANCEL operation which didn't remove all pending
584 * completion interrupts. */
585 if (tsk->state == TASK_RUNNING)
586 tsk->thread.pfault_wait = -1;
588 put_task_struct(tsk);
590 /* signal bit not set -> a real page is missing. */
591 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 tsk->thread.pfault_wait = 1;
600 list_add(&tsk->thread.list, &pfault_list);
601 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
602 set_tsk_need_resched(tsk);
605 spin_unlock(&pfault_lock);
608 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
609 unsigned long action, void *hcpu)
611 struct thread_struct *thread, *next;
612 struct task_struct *tsk;
616 case CPU_DEAD_FROZEN:
617 spin_lock_irq(&pfault_lock);
618 list_for_each_entry_safe(thread, next, &pfault_list, list) {
619 thread->pfault_wait = 0;
620 list_del(&thread->list);
621 tsk = container_of(thread, struct task_struct, thread);
622 wake_up_process(tsk);
624 spin_unlock_irq(&pfault_lock);
632 static int __init pfault_irq_init(void)
636 rc = register_external_interrupt(0x2603, pfault_interrupt);
639 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
642 service_subclass_irq_register();
643 hotcpu_notifier(pfault_cpu_notify, 0);
647 unregister_external_interrupt(0x2603, pfault_interrupt);
652 early_initcall(pfault_irq_init);
654 #endif /* CONFIG_PFAULT */