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>
28 #include <linux/hardirq.h>
30 #include <asm/system.h>
31 #include <asm/uaccess.h>
32 #include <asm/pgtable.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 save_access_regs(current->thread.acrs);
91 if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
93 * access register contains 0 -> kernel address,
94 * access register contains 1 -> user space address
96 return current->thread.acrs[areg];
98 /* Something unhealthy was done with the access registers... */
99 die("page fault via unknown access register", regs, error_code);
105 * Check which address space the address belongs to.
106 * Returns 1 for user space and 0 for kernel space.
108 static inline int check_user_space(struct pt_regs *regs, int error_code)
111 * The lowest two bits of S390_lowcore.trans_exc_code indicate
112 * which paging table was used:
113 * 0: Primary Segment Table Descriptor
114 * 1: STD determined via access register
115 * 2: Secondary Segment Table Descriptor
116 * 3: Home Segment Table Descriptor
118 int descriptor = S390_lowcore.trans_exc_code & 3;
119 if (unlikely(descriptor == 1))
120 return __check_access_register(regs, error_code);
122 return current->thread.mm_segment.ar4;
123 return descriptor != 0;
127 * Send SIGSEGV to task. This is an external routine
128 * to keep the stack usage of do_page_fault small.
130 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
131 int si_code, unsigned long address)
135 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
136 #if defined(CONFIG_SYSCTL)
137 if (sysctl_userprocess_debug)
140 printk("User process fault: interruption code 0x%lX\n",
142 printk("failing address: %lX\n", address);
146 si.si_signo = SIGSEGV;
147 si.si_code = si_code;
148 si.si_addr = (void *) address;
149 force_sig_info(SIGSEGV, &si, current);
153 * This routine handles page faults. It determines the address,
154 * and the problem, and then passes it off to one of the appropriate
158 * 04 Protection -> Write-Protection (suprression)
159 * 10 Segment translation -> Not present (nullification)
160 * 11 Page translation -> Not present (nullification)
161 * 3b Region third trans. -> Not present (nullification)
164 do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
166 struct task_struct *tsk;
167 struct mm_struct *mm;
168 struct vm_area_struct * vma;
169 unsigned long address;
171 const struct exception_table_entry *fixup;
172 int si_code = SEGV_MAPERR;
178 * Check for low-address protection. This needs to be treated
179 * as a special case because the translation exception code
180 * field is not guaranteed to contain valid data in this case.
182 if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
184 /* Low-address protection hit in kernel mode means
185 NULL pointer write access in kernel mode. */
186 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
192 /* Low-address protection hit in user mode 'cannot happen'. */
193 die ("Low-address protection", regs, error_code);
198 * get the failing address
199 * more specific the segment and page table portion of
202 address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
203 user_address = check_user_space(regs, error_code);
206 * Verify that the fault happened in user space, that
207 * we are not in an interrupt and that there is a
210 if (user_address == 0 || in_atomic() || !mm)
214 * When we get here, the fault happened in the current
215 * task's user address space, so we can switch on the
216 * interrupts again and then search the VMAs
220 down_read(&mm->mmap_sem);
222 vma = find_vma(mm, address);
225 if (vma->vm_start <= address)
227 if (!(vma->vm_flags & VM_GROWSDOWN))
229 if (expand_stack(vma, address))
232 * Ok, we have a good vm_area for this memory access, so
236 si_code = SEGV_ACCERR;
237 if (!is_protection) {
238 /* page not present, check vm flags */
239 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
242 if (!(vma->vm_flags & VM_WRITE))
248 * If for any reason at all we couldn't handle the fault,
249 * make sure we exit gracefully rather than endlessly redo
252 switch (handle_mm_fault(mm, vma, address, is_protection)) {
259 case VM_FAULT_SIGBUS:
267 up_read(&mm->mmap_sem);
269 * The instruction that caused the program check will
270 * be repeated. Don't signal single step via SIGTRAP.
272 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
276 * Something tried to access memory that isn't in our memory map..
277 * Fix it, but check if it's kernel or user first..
280 up_read(&mm->mmap_sem);
282 /* User mode accesses just cause a SIGSEGV */
283 if (regs->psw.mask & PSW_MASK_PSTATE) {
284 tsk->thread.prot_addr = address;
285 tsk->thread.trap_no = error_code;
286 do_sigsegv(regs, error_code, si_code, address);
291 /* Are we prepared to handle this kernel fault? */
292 fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
294 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
299 * Oops. The kernel tried to access some bad page. We'll have to
300 * terminate things with extreme prejudice.
302 if (user_address == 0)
303 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
304 " at virtual kernel address %p\n", (void *)address);
306 printk(KERN_ALERT "Unable to handle kernel paging request"
307 " at virtual user address %p\n", (void *)address);
309 die("Oops", regs, error_code);
314 * We ran out of memory, or some other thing happened to us that made
315 * us unable to handle the page fault gracefully.
318 up_read(&mm->mmap_sem);
323 printk("VM: killing process %s\n", tsk->comm);
324 if (regs->psw.mask & PSW_MASK_PSTATE)
329 up_read(&mm->mmap_sem);
332 * Send a sigbus, regardless of whether we were in kernel
335 tsk->thread.prot_addr = address;
336 tsk->thread.trap_no = error_code;
337 force_sig(SIGBUS, tsk);
339 /* Kernel mode? Handle exceptions or die */
340 if (!(regs->psw.mask & PSW_MASK_PSTATE))
344 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
346 regs->psw.addr -= (error_code >> 16);
347 do_exception(regs, 4, 1);
350 void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
352 do_exception(regs, error_code & 0xff, 0);
355 #ifndef CONFIG_ARCH_S390X
357 typedef struct _pseudo_wait_t {
358 struct _pseudo_wait_t *next;
359 wait_queue_head_t queue;
360 unsigned long address;
364 static pseudo_wait_t *pseudo_lock_queue = NULL;
365 static spinlock_t pseudo_wait_spinlock; /* spinlock to protect lock queue */
368 * This routine handles 'pagex' pseudo page faults.
371 do_pseudo_page_fault(struct pt_regs *regs, unsigned long error_code)
373 pseudo_wait_t wait_struct;
374 pseudo_wait_t *ptr, *last, *next;
375 unsigned long address;
378 * get the failing address
379 * more specific the segment and page table portion of
382 address = S390_lowcore.trans_exc_code & 0xfffff000;
384 if (address & 0x80000000) {
385 /* high bit set -> a page has been swapped in by VM */
386 address &= 0x7fffffff;
387 spin_lock(&pseudo_wait_spinlock);
389 ptr = pseudo_lock_queue;
390 while (ptr != NULL) {
392 if (address == ptr->address) {
394 * This is one of the processes waiting
395 * for the page. Unchain from the queue.
396 * There can be more than one process
397 * waiting for the same page. VM presents
398 * an initial and a completion interrupt for
399 * every process that tries to access a
400 * page swapped out by VM.
403 pseudo_lock_queue = next;
406 /* now wake up the process */
408 wake_up(&ptr->queue);
413 spin_unlock(&pseudo_wait_spinlock);
415 /* Pseudo page faults in kernel mode is a bad idea */
416 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
418 * VM presents pseudo page faults if the interrupted
419 * state was not disabled for interrupts. So we can
420 * get pseudo page fault interrupts while running
421 * in kernel mode. We simply access the page here
422 * while we are running disabled. VM will then swap
423 * in the page synchronously.
425 if (check_user_space(regs, error_code) == 0)
426 /* dereference a virtual kernel address */
427 __asm__ __volatile__ (
429 : : "a" (address) : "0");
431 /* dereference a virtual user address */
432 __asm__ __volatile__ (
437 ".section __ex_table,\"a\"\n"
441 : : "a" (address) : "2" );
445 /* initialize and add element to pseudo_lock_queue */
446 init_waitqueue_head (&wait_struct.queue);
447 wait_struct.address = address;
448 wait_struct.resolved = 0;
449 spin_lock(&pseudo_wait_spinlock);
450 wait_struct.next = pseudo_lock_queue;
451 pseudo_lock_queue = &wait_struct;
452 spin_unlock(&pseudo_wait_spinlock);
454 * The instruction that caused the program check will
455 * be repeated. Don't signal single step via SIGTRAP.
457 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
459 wait_event(wait_struct.queue, wait_struct.resolved);
462 #endif /* CONFIG_ARCH_S390X */
466 * 'pfault' pseudo page faults routines.
468 static int pfault_disable = 0;
470 static int __init nopfault(char *str)
476 __setup("nopfault", nopfault);
487 } __attribute__ ((packed)) pfault_refbk_t;
489 int pfault_init(void)
491 pfault_refbk_t refbk =
492 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
498 __asm__ __volatile__(
499 " diag %1,%0,0x258\n"
503 ".section __ex_table,\"a\"\n"
505 #ifndef CONFIG_ARCH_S390X
507 #else /* CONFIG_ARCH_S390X */
509 #endif /* CONFIG_ARCH_S390X */
511 : "=d" (rc) : "a" (&refbk) : "cc" );
516 void pfault_fini(void)
518 pfault_refbk_t refbk =
519 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
523 __ctl_clear_bit(0,9);
524 __asm__ __volatile__(
527 ".section __ex_table,\"a\"\n"
529 #ifndef CONFIG_ARCH_S390X
531 #else /* CONFIG_ARCH_S390X */
533 #endif /* CONFIG_ARCH_S390X */
535 : : "a" (&refbk) : "cc" );
539 pfault_interrupt(struct pt_regs *regs, __u16 error_code)
541 struct task_struct *tsk;
545 * Get the external interruption subcode & pfault
546 * initial/completion signal bit. VM stores this
547 * in the 'cpu address' field associated with the
548 * external interrupt.
550 subcode = S390_lowcore.cpu_addr;
551 if ((subcode & 0xff00) != __SUBCODE_MASK)
555 * Get the token (= address of the task structure of the affected task).
557 tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
559 if (subcode & 0x0080) {
560 /* signal bit is set -> a page has been swapped in by VM */
561 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
562 /* Initial interrupt was faster than the completion
563 * interrupt. pfault_wait is valid. Set pfault_wait
564 * back to zero and wake up the process. This can
565 * safely be done because the task is still sleeping
566 * and can't produce new pfaults. */
567 tsk->thread.pfault_wait = 0;
568 wake_up_process(tsk);
569 put_task_struct(tsk);
572 /* signal bit not set -> a real page is missing. */
573 get_task_struct(tsk);
574 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
575 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
576 /* Completion interrupt was faster than the initial
577 * interrupt (swapped in a -1 for pfault_wait). Set
578 * pfault_wait back to zero and exit. This can be
579 * done safely because tsk is running in kernel
580 * mode and can't produce new pfaults. */
581 tsk->thread.pfault_wait = 0;
582 set_task_state(tsk, TASK_RUNNING);
583 put_task_struct(tsk);
585 set_tsk_need_resched(tsk);