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/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/errno.h>
18 #include <linux/string.h>
19 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/mman.h>
23 #include <linux/compat.h>
24 #include <linux/smp.h>
25 #include <linux/kdebug.h>
26 #include <linux/init.h>
27 #include <linux/console.h>
28 #include <linux/module.h>
29 #include <linux/hardirq.h>
30 #include <linux/kprobes.h>
31 #include <linux/uaccess.h>
32 #include <linux/hugetlb.h>
33 #include <asm/asm-offsets.h>
34 #include <asm/system.h>
35 #include <asm/pgtable.h>
36 #include <asm/s390_ext.h>
37 #include <asm/mmu_context.h>
38 #include <asm/compat.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 unsigned long long facility_list[2];
61 if (stfle(facility_list, 2) < 2)
63 if ((facility_list[0] & (1ULL << 61)) &&
64 (facility_list[1] & (1ULL << 52)))
65 store_indication = 0xc00;
68 static inline int notify_page_fault(struct pt_regs *regs)
72 /* kprobe_running() needs smp_processor_id() */
73 if (kprobes_built_in() && !user_mode(regs)) {
75 if (kprobe_running() && kprobe_fault_handler(regs, 14))
84 * Unlock any spinlocks which will prevent us from getting the
87 void bust_spinlocks(int yes)
92 int loglevel_save = console_loglevel;
96 * OK, the message is on the console. Now we call printk()
97 * without oops_in_progress set so that printk will give klogd
98 * a poke. Hold onto your hats...
100 console_loglevel = 15;
102 console_loglevel = loglevel_save;
107 * Returns the address space associated with the fault.
108 * Returns 0 for kernel space and 1 for user space.
110 static inline int user_space_fault(unsigned long trans_exc_code)
113 * The lowest two bits of the translation exception
114 * identification indicate which paging table was used.
117 if (trans_exc_code == 2)
118 /* Access via secondary space, set_fs setting decides */
119 return current->thread.mm_segment.ar4;
120 if (user_mode == HOME_SPACE_MODE)
121 /* User space if the access has been done via home space. */
122 return trans_exc_code == 3;
124 * If the user space is not the home space the kernel runs in home
125 * space. Access via secondary space has already been covered,
126 * access via primary space or access register is from user space
127 * and access via home space is from the kernel.
129 return trans_exc_code != 3;
132 static inline void report_user_fault(struct pt_regs *regs, long int_code,
133 int signr, unsigned long address)
135 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
137 if (!unhandled_signal(current, signr))
139 if (!printk_ratelimit())
141 printk("User process fault: interruption code 0x%lX ", int_code);
142 print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
144 printk("failing address: %lX\n", address);
149 * Send SIGSEGV to task. This is an external routine
150 * to keep the stack usage of do_page_fault small.
152 static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
153 int si_code, unsigned long trans_exc_code)
156 unsigned long address;
158 address = trans_exc_code & __FAIL_ADDR_MASK;
159 current->thread.prot_addr = address;
160 current->thread.trap_no = int_code;
161 report_user_fault(regs, int_code, SIGSEGV, address);
162 si.si_signo = SIGSEGV;
163 si.si_code = si_code;
164 si.si_addr = (void __user *) address;
165 force_sig_info(SIGSEGV, &si, current);
168 static noinline void do_no_context(struct pt_regs *regs, long int_code,
169 unsigned long trans_exc_code)
171 const struct exception_table_entry *fixup;
172 unsigned long address;
174 /* Are we prepared to handle this kernel fault? */
175 fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
177 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
182 * Oops. The kernel tried to access some bad page. We'll have to
183 * terminate things with extreme prejudice.
185 address = trans_exc_code & __FAIL_ADDR_MASK;
186 if (!user_space_fault(trans_exc_code))
187 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
188 " at virtual kernel address %p\n", (void *)address);
190 printk(KERN_ALERT "Unable to handle kernel paging request"
191 " at virtual user address %p\n", (void *)address);
193 die("Oops", regs, int_code);
197 static noinline void do_low_address(struct pt_regs *regs, long int_code,
198 unsigned long trans_exc_code)
200 /* Low-address protection hit in kernel mode means
201 NULL pointer write access in kernel mode. */
202 if (regs->psw.mask & PSW_MASK_PSTATE) {
203 /* Low-address protection hit in user mode 'cannot happen'. */
204 die ("Low-address protection", regs, int_code);
208 do_no_context(regs, int_code, trans_exc_code);
211 static noinline void do_sigbus(struct pt_regs *regs, long int_code,
212 unsigned long trans_exc_code)
214 struct task_struct *tsk = current;
215 unsigned long address;
219 * Send a sigbus, regardless of whether we were in kernel
222 address = trans_exc_code & __FAIL_ADDR_MASK;
223 tsk->thread.prot_addr = address;
224 tsk->thread.trap_no = int_code;
225 si.si_signo = SIGBUS;
227 si.si_code = BUS_ADRERR;
228 si.si_addr = (void __user *) address;
229 force_sig_info(SIGBUS, &si, tsk);
232 #ifdef CONFIG_S390_EXEC_PROTECT
233 static noinline int signal_return(struct pt_regs *regs, long int_code,
234 unsigned long trans_exc_code)
239 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
241 if (!rc && instruction == 0x0a77) {
242 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
243 if (is_compat_task())
247 } else if (!rc && instruction == 0x0aad) {
248 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
249 if (is_compat_task())
250 sys32_rt_sigreturn();
254 do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code);
257 #endif /* CONFIG_S390_EXEC_PROTECT */
259 static noinline void do_fault_error(struct pt_regs *regs, long int_code,
260 unsigned long trans_exc_code, int fault)
265 case VM_FAULT_BADACCESS:
266 #ifdef CONFIG_S390_EXEC_PROTECT
267 if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
268 (trans_exc_code & 3) == 0) {
269 signal_return(regs, int_code, trans_exc_code);
272 #endif /* CONFIG_S390_EXEC_PROTECT */
273 case VM_FAULT_BADMAP:
274 /* Bad memory access. Check if it is kernel or user space. */
275 if (regs->psw.mask & PSW_MASK_PSTATE) {
276 /* User mode accesses just cause a SIGSEGV */
277 si_code = (fault == VM_FAULT_BADMAP) ?
278 SEGV_MAPERR : SEGV_ACCERR;
279 do_sigsegv(regs, int_code, si_code, trans_exc_code);
282 case VM_FAULT_BADCONTEXT:
283 do_no_context(regs, int_code, trans_exc_code);
285 default: /* fault & VM_FAULT_ERROR */
286 if (fault & VM_FAULT_OOM)
287 pagefault_out_of_memory();
288 else if (fault & VM_FAULT_SIGBUS) {
289 /* Kernel mode? Handle exceptions or die */
290 if (!(regs->psw.mask & PSW_MASK_PSTATE))
291 do_no_context(regs, int_code, trans_exc_code);
293 do_sigbus(regs, int_code, trans_exc_code);
301 * This routine handles page faults. It determines the address,
302 * and the problem, and then passes it off to one of the appropriate
305 * interruption code (int_code):
306 * 04 Protection -> Write-Protection (suprression)
307 * 10 Segment translation -> Not present (nullification)
308 * 11 Page translation -> Not present (nullification)
309 * 3b Region third trans. -> Not present (nullification)
311 static inline int do_exception(struct pt_regs *regs, int access,
312 unsigned long trans_exc_code)
314 struct task_struct *tsk;
315 struct mm_struct *mm;
316 struct vm_area_struct *vma;
317 unsigned long address;
320 if (notify_page_fault(regs))
327 * Verify that the fault happened in user space, that
328 * we are not in an interrupt and that there is a
331 fault = VM_FAULT_BADCONTEXT;
332 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
335 address = trans_exc_code & __FAIL_ADDR_MASK;
337 * When we get here, the fault happened in the current
338 * task's user address space, so we can switch on the
339 * interrupts again and then search the VMAs
342 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
343 down_read(&mm->mmap_sem);
345 fault = VM_FAULT_BADMAP;
346 vma = find_vma(mm, address);
350 if (unlikely(vma->vm_start > address)) {
351 if (!(vma->vm_flags & VM_GROWSDOWN))
353 if (expand_stack(vma, address))
358 * Ok, we have a good vm_area for this memory access, so
361 fault = VM_FAULT_BADACCESS;
362 if (unlikely(!(vma->vm_flags & access)))
365 if (is_vm_hugetlb_page(vma))
366 address &= HPAGE_MASK;
368 * If for any reason at all we couldn't handle the fault,
369 * make sure we exit gracefully rather than endlessly redo
372 write = (access == VM_WRITE ||
373 (trans_exc_code & store_indication) == 0x400) ?
374 FAULT_FLAG_WRITE : 0;
375 fault = handle_mm_fault(mm, vma, address, write);
376 if (unlikely(fault & VM_FAULT_ERROR))
379 if (fault & VM_FAULT_MAJOR) {
381 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
385 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
389 * The instruction that caused the program check will
390 * be repeated. Don't signal single step via SIGTRAP.
392 clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
395 up_read(&mm->mmap_sem);
400 void __kprobes do_protection_exception(struct pt_regs *regs, long int_code)
402 unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
405 /* Protection exception is supressing, decrement psw address. */
406 regs->psw.addr -= (int_code >> 16);
408 * Check for low-address protection. This needs to be treated
409 * as a special case because the translation exception code
410 * field is not guaranteed to contain valid data in this case.
412 if (unlikely(!(trans_exc_code & 4))) {
413 do_low_address(regs, int_code, trans_exc_code);
416 fault = do_exception(regs, VM_WRITE, trans_exc_code);
418 do_fault_error(regs, 4, trans_exc_code, fault);
421 void __kprobes do_dat_exception(struct pt_regs *regs, long int_code)
423 unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
426 access = VM_READ | VM_EXEC | VM_WRITE;
427 #ifdef CONFIG_S390_EXEC_PROTECT
428 if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
429 (trans_exc_code & 3) == 0)
432 fault = do_exception(regs, access, trans_exc_code);
434 do_fault_error(regs, int_code & 255, trans_exc_code, fault);
438 void __kprobes do_asce_exception(struct pt_regs *regs, long int_code)
440 unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
441 struct mm_struct *mm = current->mm;
442 struct vm_area_struct *vma;
444 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
449 down_read(&mm->mmap_sem);
450 vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
451 up_read(&mm->mmap_sem);
454 update_mm(mm, current);
458 /* User mode accesses just cause a SIGSEGV */
459 if (regs->psw.mask & PSW_MASK_PSTATE) {
460 do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code);
465 do_no_context(regs, int_code, trans_exc_code);
469 int __handle_fault(unsigned long uaddr, unsigned long int_code, int write_user)
474 regs.psw.mask = psw_kernel_bits;
475 if (!irqs_disabled())
476 regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
477 regs.psw.addr = (unsigned long) __builtin_return_address(0);
478 regs.psw.addr |= PSW_ADDR_AMODE;
480 access = write_user ? VM_WRITE : VM_READ;
481 fault = do_exception(®s, access, uaddr | 2);
482 if (unlikely(fault)) {
483 if (fault & VM_FAULT_OOM) {
484 pagefault_out_of_memory();
486 } else if (fault & VM_FAULT_SIGBUS)
487 do_sigbus(®s, int_code, uaddr);
489 return fault ? -EFAULT : 0;
494 * 'pfault' pseudo page faults routines.
496 static ext_int_info_t ext_int_pfault;
497 static int pfault_disable = 0;
499 static int __init nopfault(char *str)
505 __setup("nopfault", nopfault);
516 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
518 int pfault_init(void)
520 pfault_refbk_t refbk =
521 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
525 if (!MACHINE_IS_VM || pfault_disable)
528 " diag %1,%0,0x258\n"
533 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
538 void pfault_fini(void)
540 pfault_refbk_t refbk =
541 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
543 if (!MACHINE_IS_VM || pfault_disable)
545 __ctl_clear_bit(0,9);
550 : : "a" (&refbk), "m" (refbk) : "cc");
553 static void pfault_interrupt(__u16 int_code)
555 struct task_struct *tsk;
559 * Get the external interruption subcode & pfault
560 * initial/completion signal bit. VM stores this
561 * in the 'cpu address' field associated with the
562 * external interrupt.
564 subcode = S390_lowcore.cpu_addr;
565 if ((subcode & 0xff00) != __SUBCODE_MASK)
569 * Get the token (= address of the task structure of the affected task).
571 tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
573 if (subcode & 0x0080) {
574 /* signal bit is set -> a page has been swapped in by VM */
575 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
576 /* Initial interrupt was faster than the completion
577 * interrupt. pfault_wait is valid. Set pfault_wait
578 * back to zero and wake up the process. This can
579 * safely be done because the task is still sleeping
580 * and can't produce new pfaults. */
581 tsk->thread.pfault_wait = 0;
582 wake_up_process(tsk);
583 put_task_struct(tsk);
586 /* signal bit not set -> a real page is missing. */
587 get_task_struct(tsk);
588 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
589 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
590 /* Completion interrupt was faster than the initial
591 * interrupt (swapped in a -1 for pfault_wait). Set
592 * pfault_wait back to zero and exit. This can be
593 * done safely because tsk is running in kernel
594 * mode and can't produce new pfaults. */
595 tsk->thread.pfault_wait = 0;
596 set_task_state(tsk, TASK_RUNNING);
597 put_task_struct(tsk);
599 set_tsk_need_resched(tsk);
603 void __init pfault_irq_init(void)
609 * Try to get pfault pseudo page faults going.
611 if (register_early_external_interrupt(0x2603, pfault_interrupt,
612 &ext_int_pfault) != 0)
613 panic("Couldn't request external interrupt 0x2603");
615 if (pfault_init() == 0)
618 /* Tough luck, no pfault. */
620 unregister_early_external_interrupt(0x2603, pfault_interrupt,