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/system.h>
34 #include <asm/pgtable.h>
35 #include <asm/s390_ext.h>
36 #include <asm/mmu_context.h>
37 #include "../kernel/entry.h"
40 #define __FAIL_ADDR_MASK 0x7ffff000
41 #define __FIXUP_MASK 0x7fffffff
42 #define __SUBCODE_MASK 0x0200
43 #define __PF_RES_FIELD 0ULL
44 #else /* CONFIG_64BIT */
45 #define __FAIL_ADDR_MASK -4096L
46 #define __FIXUP_MASK ~0L
47 #define __SUBCODE_MASK 0x0600
48 #define __PF_RES_FIELD 0x8000000000000000ULL
49 #endif /* CONFIG_64BIT */
52 extern int sysctl_userprocess_debug;
55 static inline int notify_page_fault(struct pt_regs *regs)
60 /* kprobe_running() needs smp_processor_id() */
61 if (!user_mode(regs)) {
63 if (kprobe_running() && kprobe_fault_handler(regs, 14))
73 * Unlock any spinlocks which will prevent us from getting the
76 void bust_spinlocks(int yes)
81 int loglevel_save = console_loglevel;
85 * OK, the message is on the console. Now we call printk()
86 * without oops_in_progress set so that printk will give klogd
87 * a poke. Hold onto your hats...
89 console_loglevel = 15;
91 console_loglevel = loglevel_save;
96 * Returns the address space associated with the fault.
97 * Returns 0 for kernel space and 1 for user space.
99 static inline int user_space_fault(unsigned long trans_exc_code)
102 * The lowest two bits of the translation exception
103 * identification indicate which paging table was used.
106 if (trans_exc_code == 2)
107 /* Access via secondary space, set_fs setting decides */
108 return current->thread.mm_segment.ar4;
109 if (user_mode == HOME_SPACE_MODE)
110 /* User space if the access has been done via home space. */
111 return trans_exc_code == 3;
113 * If the user space is not the home space the kernel runs in home
114 * space. Access via secondary space has already been covered,
115 * access via primary space or access register is from user space
116 * and access via home space is from the kernel.
118 return trans_exc_code != 3;
122 * Send SIGSEGV to task. This is an external routine
123 * to keep the stack usage of do_page_fault small.
125 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
126 int si_code, unsigned long address)
130 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
131 #if defined(CONFIG_SYSCTL)
132 if (sysctl_userprocess_debug)
135 printk("User process fault: interruption code 0x%lX\n",
137 printk("failing address: %lX\n", address);
141 si.si_signo = SIGSEGV;
142 si.si_code = si_code;
143 si.si_addr = (void __user *) address;
144 force_sig_info(SIGSEGV, &si, current);
147 static void do_no_context(struct pt_regs *regs, unsigned long error_code,
148 unsigned long trans_exc_code)
150 const struct exception_table_entry *fixup;
151 unsigned long address;
153 /* Are we prepared to handle this kernel fault? */
154 fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
156 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
161 * Oops. The kernel tried to access some bad page. We'll have to
162 * terminate things with extreme prejudice.
164 address = trans_exc_code & __FAIL_ADDR_MASK;
165 if (!user_space_fault(trans_exc_code))
166 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
167 " at virtual kernel address %p\n", (void *)address);
169 printk(KERN_ALERT "Unable to handle kernel paging request"
170 " at virtual user address %p\n", (void *)address);
172 die("Oops", regs, error_code);
176 static void do_low_address(struct pt_regs *regs, unsigned long error_code,
177 unsigned long trans_exc_code)
179 /* Low-address protection hit in kernel mode means
180 NULL pointer write access in kernel mode. */
181 if (regs->psw.mask & PSW_MASK_PSTATE) {
182 /* Low-address protection hit in user mode 'cannot happen'. */
183 die ("Low-address protection", regs, error_code);
187 do_no_context(regs, error_code, trans_exc_code);
190 static void do_sigbus(struct pt_regs *regs, unsigned long error_code,
191 unsigned long trans_exc_code)
193 struct task_struct *tsk = current;
194 struct mm_struct *mm = tsk->mm;
196 up_read(&mm->mmap_sem);
198 * Send a sigbus, regardless of whether we were in kernel
201 tsk->thread.prot_addr = trans_exc_code & __FAIL_ADDR_MASK;
202 tsk->thread.trap_no = error_code;
203 force_sig(SIGBUS, tsk);
205 /* Kernel mode? Handle exceptions or die */
206 if (!(regs->psw.mask & PSW_MASK_PSTATE))
207 do_no_context(regs, error_code, trans_exc_code);
210 #ifdef CONFIG_S390_EXEC_PROTECT
211 static int signal_return(struct mm_struct *mm, struct pt_regs *regs,
212 unsigned long address, unsigned long error_code)
221 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
226 up_read(&mm->mmap_sem);
227 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
229 compat = is_compat_task();
230 if (compat && instruction == 0x0a77)
232 else if (compat && instruction == 0x0aad)
233 sys32_rt_sigreturn();
236 if (instruction == 0x0a77)
238 else if (instruction == 0x0aad)
241 current->thread.prot_addr = address;
242 current->thread.trap_no = error_code;
243 do_sigsegv(regs, error_code, SEGV_MAPERR, address);
247 #endif /* CONFIG_S390_EXEC_PROTECT */
250 * This routine handles page faults. It determines the address,
251 * and the problem, and then passes it off to one of the appropriate
255 * 04 Protection -> Write-Protection (suprression)
256 * 10 Segment translation -> Not present (nullification)
257 * 11 Page translation -> Not present (nullification)
258 * 3b Region third trans. -> Not present (nullification)
261 do_exception(struct pt_regs *regs, unsigned long error_code, int write,
262 unsigned long trans_exc_code)
264 struct task_struct *tsk;
265 struct mm_struct *mm;
266 struct vm_area_struct *vma;
267 unsigned long address;
271 if (notify_page_fault(regs))
278 * Verify that the fault happened in user space, that
279 * we are not in an interrupt and that there is a
282 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
285 address = trans_exc_code & __FAIL_ADDR_MASK;
287 * When we get here, the fault happened in the current
288 * task's user address space, so we can switch on the
289 * interrupts again and then search the VMAs
292 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
293 down_read(&mm->mmap_sem);
295 si_code = SEGV_MAPERR;
296 vma = find_vma(mm, address);
300 #ifdef CONFIG_S390_EXEC_PROTECT
301 if (unlikely((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
302 (trans_exc_code & 3) == 0 && !(vma->vm_flags & VM_EXEC)))
303 if (!signal_return(mm, regs, address, error_code))
305 * signal_return() has done an up_read(&mm->mmap_sem)
311 if (vma->vm_start <= address)
313 if (!(vma->vm_flags & VM_GROWSDOWN))
315 if (expand_stack(vma, address))
318 * Ok, we have a good vm_area for this memory access, so
322 si_code = SEGV_ACCERR;
324 /* page not present, check vm flags */
325 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
328 if (!(vma->vm_flags & VM_WRITE))
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, write ? FAULT_FLAG_WRITE : 0);
340 if (unlikely(fault & VM_FAULT_ERROR)) {
341 if (fault & VM_FAULT_OOM) {
342 up_read(&mm->mmap_sem);
343 pagefault_out_of_memory();
345 } else if (fault & VM_FAULT_SIGBUS) {
346 do_sigbus(regs, error_code, address);
351 if (fault & VM_FAULT_MAJOR) {
353 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
357 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
360 up_read(&mm->mmap_sem);
362 * The instruction that caused the program check will
363 * be repeated. Don't signal single step via SIGTRAP.
365 clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
369 * Something tried to access memory that isn't in our memory map..
370 * Fix it, but check if it's kernel or user first..
373 up_read(&mm->mmap_sem);
375 /* User mode accesses just cause a SIGSEGV */
376 if (regs->psw.mask & PSW_MASK_PSTATE) {
377 tsk->thread.prot_addr = address;
378 tsk->thread.trap_no = error_code;
379 do_sigsegv(regs, error_code, si_code, address);
384 do_no_context(regs, error_code, trans_exc_code);
387 void __kprobes do_protection_exception(struct pt_regs *regs,
390 unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
392 /* Protection exception is supressing, decrement psw address. */
393 regs->psw.addr -= (error_code >> 16);
395 * Check for low-address protection. This needs to be treated
396 * as a special case because the translation exception code
397 * field is not guaranteed to contain valid data in this case.
399 if (unlikely(!(trans_exc_code & 4))) {
400 do_low_address(regs, error_code, trans_exc_code);
403 do_exception(regs, 4, 1, trans_exc_code);
406 void __kprobes do_dat_exception(struct pt_regs *regs, long error_code)
408 do_exception(regs, error_code & 0xff, 0, S390_lowcore.trans_exc_code);
412 void __kprobes do_asce_exception(struct pt_regs *regs, unsigned long error_code)
414 unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
415 struct mm_struct *mm;
416 struct vm_area_struct *vma;
417 unsigned long address;
420 address = trans_exc_code & __FAIL_ADDR_MASK;
422 if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
427 down_read(&mm->mmap_sem);
428 vma = find_vma(mm, address);
429 up_read(&mm->mmap_sem);
432 update_mm(mm, current);
436 /* User mode accesses just cause a SIGSEGV */
437 if (regs->psw.mask & PSW_MASK_PSTATE) {
438 current->thread.prot_addr = address;
439 current->thread.trap_no = error_code;
440 do_sigsegv(regs, error_code, SEGV_MAPERR, address);
445 do_no_context(regs, error_code, trans_exc_code);
451 * 'pfault' pseudo page faults routines.
453 static ext_int_info_t ext_int_pfault;
454 static int pfault_disable = 0;
456 static int __init nopfault(char *str)
462 __setup("nopfault", nopfault);
473 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
475 int pfault_init(void)
477 pfault_refbk_t refbk =
478 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
482 if (!MACHINE_IS_VM || pfault_disable)
485 " diag %1,%0,0x258\n"
490 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
495 void pfault_fini(void)
497 pfault_refbk_t refbk =
498 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
500 if (!MACHINE_IS_VM || pfault_disable)
502 __ctl_clear_bit(0,9);
507 : : "a" (&refbk), "m" (refbk) : "cc");
510 static void pfault_interrupt(__u16 error_code)
512 struct task_struct *tsk;
516 * Get the external interruption subcode & pfault
517 * initial/completion signal bit. VM stores this
518 * in the 'cpu address' field associated with the
519 * external interrupt.
521 subcode = S390_lowcore.cpu_addr;
522 if ((subcode & 0xff00) != __SUBCODE_MASK)
526 * Get the token (= address of the task structure of the affected task).
528 tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
530 if (subcode & 0x0080) {
531 /* signal bit is set -> a page has been swapped in by VM */
532 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
533 /* Initial interrupt was faster than the completion
534 * interrupt. pfault_wait is valid. Set pfault_wait
535 * back to zero and wake up the process. This can
536 * safely be done because the task is still sleeping
537 * and can't produce new pfaults. */
538 tsk->thread.pfault_wait = 0;
539 wake_up_process(tsk);
540 put_task_struct(tsk);
543 /* signal bit not set -> a real page is missing. */
544 get_task_struct(tsk);
545 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
546 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
547 /* Completion interrupt was faster than the initial
548 * interrupt (swapped in a -1 for pfault_wait). Set
549 * pfault_wait back to zero and exit. This can be
550 * done safely because tsk is running in kernel
551 * mode and can't produce new pfaults. */
552 tsk->thread.pfault_wait = 0;
553 set_task_state(tsk, TASK_RUNNING);
554 put_task_struct(tsk);
556 set_tsk_need_resched(tsk);
560 void __init pfault_irq_init(void)
566 * Try to get pfault pseudo page faults going.
568 if (register_early_external_interrupt(0x2603, pfault_interrupt,
569 &ext_int_pfault) != 0)
570 panic("Couldn't request external interrupt 0x2603");
572 if (pfault_init() == 0)
575 /* Tough luck, no pfault. */
577 unregister_early_external_interrupt(0x2603, pfault_interrupt,