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/config.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/smp.h>
24 #include <linux/smp_lock.h>
25 #include <linux/init.h>
26 #include <linux/console.h>
27 #include <linux/module.h>
29 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/pgtable.h>
32 #include <asm/hardirq.h>
34 #ifndef CONFIG_ARCH_S390X
35 #define __FAIL_ADDR_MASK 0x7ffff000
36 #define __FIXUP_MASK 0x7fffffff
37 #define __SUBCODE_MASK 0x0200
38 #define __PF_RES_FIELD 0ULL
39 #else /* CONFIG_ARCH_S390X */
40 #define __FAIL_ADDR_MASK -4096L
41 #define __FIXUP_MASK ~0L
42 #define __SUBCODE_MASK 0x0600
43 #define __PF_RES_FIELD 0x8000000000000000ULL
44 #endif /* CONFIG_ARCH_S390X */
47 extern int sysctl_userprocess_debug;
50 extern void die(const char *,struct pt_regs *,long);
52 extern spinlock_t timerlist_lock;
55 * Unlock any spinlocks which will prevent us from getting the
56 * message out (timerlist_lock is acquired through the
57 * console unblank code)
59 void bust_spinlocks(int yes)
64 int loglevel_save = console_loglevel;
68 * OK, the message is on the console. Now we call printk()
69 * without oops_in_progress set so that printk will give klogd
70 * a poke. Hold onto your hats...
72 console_loglevel = 15;
74 console_loglevel = loglevel_save;
79 * Check which address space is addressed by the access
80 * register in S390_lowcore.exc_access_id.
81 * Returns 1 for user space and 0 for kernel space.
83 static int __check_access_register(struct pt_regs *regs, int error_code)
85 int areg = S390_lowcore.exc_access_id;
88 /* Access via access register 0 -> kernel address */
90 if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
92 * access register contains 0 -> kernel address,
93 * access register contains 1 -> user space address
95 return current->thread.acrs[areg];
97 /* Something unhealthy was done with the access registers... */
98 die("page fault via unknown access register", regs, error_code);
104 * Check which address space the address belongs to.
105 * Returns 1 for user space and 0 for kernel space.
107 static inline int check_user_space(struct pt_regs *regs, int error_code)
110 * The lowest two bits of S390_lowcore.trans_exc_code indicate
111 * which paging table was used:
112 * 0: Primary Segment Table Descriptor
113 * 1: STD determined via access register
114 * 2: Secondary Segment Table Descriptor
115 * 3: Home Segment Table Descriptor
117 int descriptor = S390_lowcore.trans_exc_code & 3;
118 if (descriptor == 1) {
119 save_access_regs(current->thread.acrs);
120 return __check_access_register(regs, error_code);
122 return descriptor >> 1;
126 * Send SIGSEGV to task. This is an external routine
127 * to keep the stack usage of do_page_fault small.
129 static void force_sigsegv(struct pt_regs *regs, unsigned long error_code,
130 int si_code, unsigned long address)
134 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
135 #if defined(CONFIG_SYSCTL)
136 if (sysctl_userprocess_debug)
139 printk("User process fault: interruption code 0x%lX\n",
141 printk("failing address: %lX\n", address);
145 si.si_signo = SIGSEGV;
146 si.si_code = si_code;
147 si.si_addr = (void *) address;
148 force_sig_info(SIGSEGV, &si, current);
152 * This routine handles page faults. It determines the address,
153 * and the problem, and then passes it off to one of the appropriate
157 * 04 Protection -> Write-Protection (suprression)
158 * 10 Segment translation -> Not present (nullification)
159 * 11 Page translation -> Not present (nullification)
160 * 3b Region third trans. -> Not present (nullification)
162 extern inline void do_exception(struct pt_regs *regs, unsigned long error_code)
164 struct task_struct *tsk;
165 struct mm_struct *mm;
166 struct vm_area_struct * vma;
167 unsigned long address;
169 const struct exception_table_entry *fixup;
170 int si_code = SEGV_MAPERR;
176 * Check for low-address protection. This needs to be treated
177 * as a special case because the translation exception code
178 * field is not guaranteed to contain valid data in this case.
180 if (error_code == 4 && !(S390_lowcore.trans_exc_code & 4)) {
182 /* Low-address protection hit in kernel mode means
183 NULL pointer write access in kernel mode. */
184 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
190 /* Low-address protection hit in user mode 'cannot happen'. */
191 die ("Low-address protection", regs, error_code);
196 * get the failing address
197 * more specific the segment and page table portion of
200 address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
201 user_address = check_user_space(regs, error_code);
204 * Verify that the fault happened in user space, that
205 * we are not in an interrupt and that there is a
208 if (user_address == 0 || in_interrupt() || !mm)
212 * When we get here, the fault happened in the current
213 * task's user address space, so we can switch on the
214 * interrupts again and then search the VMAs
218 down_read(&mm->mmap_sem);
220 vma = find_vma(mm, address);
223 if (vma->vm_start <= address)
225 if (!(vma->vm_flags & VM_GROWSDOWN))
227 if (expand_stack(vma, address))
230 * Ok, we have a good vm_area for this memory access, so
234 si_code = SEGV_ACCERR;
235 if (error_code != 4) {
236 /* page not present, check vm flags */
237 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
240 if (!(vma->vm_flags & VM_WRITE))
246 * If for any reason at all we couldn't handle the fault,
247 * make sure we exit gracefully rather than endlessly redo
250 switch (handle_mm_fault(mm, vma, address, error_code == 4)) {
257 case VM_FAULT_SIGBUS:
265 up_read(&mm->mmap_sem);
269 * Something tried to access memory that isn't in our memory map..
270 * Fix it, but check if it's kernel or user first..
273 up_read(&mm->mmap_sem);
275 /* User mode accesses just cause a SIGSEGV */
276 if (regs->psw.mask & PSW_MASK_PSTATE) {
277 tsk->thread.prot_addr = address;
278 tsk->thread.trap_no = error_code;
279 force_sigsegv(regs, error_code, si_code, address);
284 /* Are we prepared to handle this kernel fault? */
285 fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
287 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
292 * Oops. The kernel tried to access some bad page. We'll have to
293 * terminate things with extreme prejudice.
295 if (user_address == 0)
296 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
297 " at virtual kernel address %p\n", (void *)address);
299 printk(KERN_ALERT "Unable to handle kernel paging request"
300 " at virtual user address %p\n", (void *)address);
302 die("Oops", regs, error_code);
307 * We ran out of memory, or some other thing happened to us that made
308 * us unable to handle the page fault gracefully.
311 up_read(&mm->mmap_sem);
316 printk("VM: killing process %s\n", tsk->comm);
317 if (regs->psw.mask & PSW_MASK_PSTATE)
322 up_read(&mm->mmap_sem);
325 * Send a sigbus, regardless of whether we were in kernel
328 tsk->thread.prot_addr = address;
329 tsk->thread.trap_no = error_code;
330 force_sig(SIGBUS, tsk);
332 /* Kernel mode? Handle exceptions or die */
333 if (!(regs->psw.mask & PSW_MASK_PSTATE))
337 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
339 regs->psw.addr -= (error_code >> 16);
340 do_exception(regs, 4);
343 void do_segment_exception(struct pt_regs *regs, unsigned long error_code)
345 do_exception(regs, 0x10);
348 void do_page_exception(struct pt_regs *regs, unsigned long error_code)
350 do_exception(regs, 0x11);
353 #ifdef CONFIG_ARCH_S390X
356 do_region_exception(struct pt_regs *regs, unsigned long error_code)
358 do_exception(regs, 0x3b);
361 #else /* CONFIG_ARCH_S390X */
363 typedef struct _pseudo_wait_t {
364 struct _pseudo_wait_t *next;
365 wait_queue_head_t queue;
366 unsigned long address;
370 static pseudo_wait_t *pseudo_lock_queue = NULL;
371 static spinlock_t pseudo_wait_spinlock; /* spinlock to protect lock queue */
374 * This routine handles 'pagex' pseudo page faults.
377 do_pseudo_page_fault(struct pt_regs *regs, unsigned long error_code)
379 pseudo_wait_t wait_struct;
380 pseudo_wait_t *ptr, *last, *next;
381 unsigned long address;
384 * get the failing address
385 * more specific the segment and page table portion of
388 address = S390_lowcore.trans_exc_code & 0xfffff000;
390 if (address & 0x80000000) {
391 /* high bit set -> a page has been swapped in by VM */
392 address &= 0x7fffffff;
393 spin_lock(&pseudo_wait_spinlock);
395 ptr = pseudo_lock_queue;
396 while (ptr != NULL) {
398 if (address == ptr->address) {
400 * This is one of the processes waiting
401 * for the page. Unchain from the queue.
402 * There can be more than one process
403 * waiting for the same page. VM presents
404 * an initial and a completion interrupt for
405 * every process that tries to access a
406 * page swapped out by VM.
409 pseudo_lock_queue = next;
412 /* now wake up the process */
414 wake_up(&ptr->queue);
419 spin_unlock(&pseudo_wait_spinlock);
421 /* Pseudo page faults in kernel mode is a bad idea */
422 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
424 * VM presents pseudo page faults if the interrupted
425 * state was not disabled for interrupts. So we can
426 * get pseudo page fault interrupts while running
427 * in kernel mode. We simply access the page here
428 * while we are running disabled. VM will then swap
429 * in the page synchronously.
431 if (check_user_space(regs, error_code) == 0)
432 /* dereference a virtual kernel address */
433 __asm__ __volatile__ (
435 : : "a" (address) : "0");
437 /* dereference a virtual user address */
438 __asm__ __volatile__ (
443 ".section __ex_table,\"a\"\n"
447 : : "a" (address) : "2" );
451 /* initialize and add element to pseudo_lock_queue */
452 init_waitqueue_head (&wait_struct.queue);
453 wait_struct.address = address;
454 wait_struct.resolved = 0;
455 spin_lock(&pseudo_wait_spinlock);
456 wait_struct.next = pseudo_lock_queue;
457 pseudo_lock_queue = &wait_struct;
458 spin_unlock(&pseudo_wait_spinlock);
460 wait_event(wait_struct.queue, wait_struct.resolved);
463 #endif /* CONFIG_ARCH_S390X */
467 * 'pfault' pseudo page faults routines.
469 static int pfault_disable = 0;
471 static int __init nopfault(char *str)
477 __setup("nopfault", nopfault);
488 } __attribute__ ((packed)) pfault_refbk_t;
490 int pfault_init(void)
492 pfault_refbk_t refbk =
493 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
499 __asm__ __volatile__(
500 " diag %1,%0,0x258\n"
504 ".section __ex_table,\"a\"\n"
506 #ifndef CONFIG_ARCH_S390X
508 #else /* CONFIG_ARCH_S390X */
510 #endif /* CONFIG_ARCH_S390X */
512 : "=d" (rc) : "a" (&refbk) : "cc" );
517 void pfault_fini(void)
519 pfault_refbk_t refbk =
520 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
524 __ctl_clear_bit(0,9);
525 __asm__ __volatile__(
528 ".section __ex_table,\"a\"\n"
530 #ifndef CONFIG_ARCH_S390X
532 #else /* CONFIG_ARCH_S390X */
534 #endif /* CONFIG_ARCH_S390X */
536 : : "a" (&refbk) : "cc" );
540 pfault_interrupt(struct pt_regs *regs, __u16 error_code)
542 struct task_struct *tsk;
546 * Get the external interruption subcode & pfault
547 * initial/completion signal bit. VM stores this
548 * in the 'cpu address' field associated with the
549 * external interrupt.
551 subcode = S390_lowcore.cpu_addr;
552 if ((subcode & 0xff00) != __SUBCODE_MASK)
556 * Get the token (= address of the task structure of the affected task).
558 tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
560 if (subcode & 0x0080) {
561 /* signal bit is set -> a page has been swapped in by VM */
562 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
563 /* Initial interrupt was faster than the completion
564 * interrupt. pfault_wait is valid. Set pfault_wait
565 * back to zero and wake up the process. This can
566 * safely be done because the task is still sleeping
567 * and can't procude new pfaults. */
568 tsk->thread.pfault_wait = 0;
569 wake_up_process(tsk);
572 /* signal bit not set -> a real page is missing. */
573 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
574 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
575 /* Completion interrupt was faster than the initial
576 * interrupt (swapped in a -1 for pfault_wait). Set
577 * pfault_wait back to zero and exit. This can be
578 * done safely because tsk is running in kernel
579 * mode and can't produce new pfaults. */
580 tsk->thread.pfault_wait = 0;
581 set_task_state(tsk, TASK_RUNNING);
583 set_tsk_need_resched(tsk);