1/*
2 *  arch/s390/mm/fault.c
3 *
4 *  S390 version
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)
8 *
9 *  Derived from "arch/i386/mm/fault.c"
10 *    Copyright (C) 1995  Linus Torvalds
11 */
12
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>
22#include <linux/mm.h>
23#include <linux/smp.h>
24#include <linux/smp_lock.h>
25#include <linux/init.h>
26#include <linux/console.h>
27
28#include <asm/system.h>
29#include <asm/uaccess.h>
30#include <asm/pgtable.h>
31#include <asm/hardirq.h>
32
33#ifdef CONFIG_SYSCTL
34extern int sysctl_userprocess_debug;
35#endif
36
37extern void die(const char *,struct pt_regs *,long);
38
39extern spinlock_t timerlist_lock;
40
41/*
42 * Unlock any spinlocks which will prevent us from getting the
43 * message out (timerlist_lock is acquired through the
44 * console unblank code)
45 */
46void bust_spinlocks(int yes)
47{
48	spin_lock_init(&timerlist_lock);
49	if (yes) {
50		oops_in_progress = 1;
51	} else {
52		int loglevel_save = console_loglevel;
53		oops_in_progress = 0;
54		console_unblank();
55		/*
56		 * OK, the message is on the console.  Now we call printk()
57		 * without oops_in_progress set so that printk will give klogd
58		 * a poke.  Hold onto your hats...
59		 */
60		console_loglevel = 15;
61		printk(" ");
62		console_loglevel = loglevel_save;
63	}
64}
65
66/*
67 * Check which address space is addressed by the access
68 * register in S390_lowcore.exc_access_id.
69 * Returns 1 for user space and 0 for kernel space.
70 */
71static int __check_access_register(struct pt_regs *regs, int error_code)
72{
73	int areg = S390_lowcore.exc_access_id;
74
75	if (areg == 0)
76		/* Access via access register 0 -> kernel address */
77		return 0;
78	if (regs && areg < NUM_ACRS && regs->acrs[areg] <= 1)
79		/*
80		 * access register contains 0 -> kernel address,
81		 * access register contains 1 -> user space address
82		 */
83		return regs->acrs[areg];
84
85	/* Something unhealthy was done with the access registers... */
86	die("page fault via unknown access register", regs, error_code);
87	do_exit(SIGKILL);
88	return 0;
89}
90
91/*
92 * Check which address space the address belongs to.
93 * Returns 1 for user space and 0 for kernel space.
94 */
95static inline int check_user_space(struct pt_regs *regs, int error_code)
96{
97	/*
98	 * The lowest two bits of S390_lowcore.trans_exc_code indicate
99	 * which paging table was used:
100	 *   0: Primary Segment Table Descriptor
101	 *   1: STD determined via access register
102	 *   2: Secondary Segment Table Descriptor
103	 *   3: Home Segment Table Descriptor
104	 */
105	int descriptor = S390_lowcore.trans_exc_code & 3;
106	if (descriptor == 1)
107		return __check_access_register(regs, error_code);
108	return descriptor >> 1;
109}
110
111/*
112 * Send SIGSEGV to task.  This is an external routine
113 * to keep the stack usage of do_page_fault small.
114 */
115static void force_sigsegv(struct pt_regs *regs, unsigned long error_code,
116			  int si_code, unsigned long address)
117{
118	struct siginfo si;
119
120#if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
121#if defined(CONFIG_SYSCTL)
122	if (sysctl_userprocess_debug)
123#endif
124	{
125		printk("User process fault: interruption code 0x%lX\n",
126		       error_code);
127		printk("failing address: %lX\n", address);
128		show_regs(regs);
129	}
130#endif
131	si.si_signo = SIGSEGV;
132	si.si_code = si_code;
133	si.si_addr = (void *) address;
134	force_sig_info(SIGSEGV, &si, current);
135}
136
137/*
138 * This routine handles page faults.  It determines the address,
139 * and the problem, and then passes it off to one of the appropriate
140 * routines.
141 *
142 * error_code:
143 *   04       Protection           ->  Write-Protection  (suprression)
144 *   10       Segment translation  ->  Not present       (nullification)
145 *   11       Page translation     ->  Not present       (nullification)
146 *   3b       Region third trans.  ->  Not present       (nullification)
147 */
148extern inline void do_exception(struct pt_regs *regs, unsigned long error_code)
149{
150        struct task_struct *tsk;
151        struct mm_struct *mm;
152        struct vm_area_struct * vma;
153        unsigned long address;
154	int user_address;
155        unsigned long fixup;
156	int si_code = SEGV_MAPERR;
157
158        tsk = current;
159        mm = tsk->mm;
160
161	/*
162         * Check for low-address protection.  This needs to be treated
163	 * as a special case because the translation exception code
164	 * field is not guaranteed to contain valid data in this case.
165	 */
166	if (error_code == 4 && !(S390_lowcore.trans_exc_code & 4)) {
167
168		/* Low-address protection hit in kernel mode means
169		   NULL pointer write access in kernel mode.  */
170 		if (!(regs->psw.mask & PSW_PROBLEM_STATE)) {
171			address = 0;
172			user_address = 0;
173			goto no_context;
174		}
175
176		/* Low-address protection hit in user mode 'cannot happen'.  */
177		die ("Low-address protection", regs, error_code);
178        	do_exit(SIGKILL);
179	}
180
181        /*
182         * get the failing address
183         * more specific the segment and page table portion of
184         * the address
185         */
186        address = S390_lowcore.trans_exc_code & -4096L;
187	user_address = check_user_space(regs, error_code);
188
189	/*
190	 * Verify that the fault happened in user space, that
191	 * we are not in an interrupt and that there is a
192	 * user context.
193	 */
194        if (user_address == 0 || in_interrupt() || !mm)
195                goto no_context;
196
197	/*
198	 * When we get here, the fault happened in the current
199	 * task's user address space, so we can switch on the
200	 * interrupts again and then search the VMAs
201	 */
202	__sti();
203
204        down_read(&mm->mmap_sem);
205
206        vma = find_vma(mm, address);
207        if (!vma)
208                goto bad_area;
209        if (vma->vm_start <= address)
210                goto good_area;
211        if (!(vma->vm_flags & VM_GROWSDOWN))
212                goto bad_area;
213        if (expand_stack(vma, address))
214                goto bad_area;
215/*
216 * Ok, we have a good vm_area for this memory access, so
217 * we can handle it..
218 */
219good_area:
220	si_code = SEGV_ACCERR;
221	if (error_code != 4) {
222		/* page not present, check vm flags */
223		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
224			goto bad_area;
225	} else {
226		if (!(vma->vm_flags & VM_WRITE))
227			goto bad_area;
228	}
229
230survive:
231	/*
232	 * If for any reason at all we couldn't handle the fault,
233	 * make sure we exit gracefully rather than endlessly redo
234	 * the fault.
235	 */
236	switch (handle_mm_fault(mm, vma, address, error_code == 4)) {
237	case 1:
238		tsk->min_flt++;
239		break;
240	case 2:
241		tsk->maj_flt++;
242		break;
243	case 0:
244		goto do_sigbus;
245	default:
246		goto out_of_memory;
247	}
248
249        up_read(&mm->mmap_sem);
250        return;
251
252/*
253 * Something tried to access memory that isn't in our memory map..
254 * Fix it, but check if it's kernel or user first..
255 */
256bad_area:
257        up_read(&mm->mmap_sem);
258
259        /* User mode accesses just cause a SIGSEGV */
260        if (regs->psw.mask & PSW_PROBLEM_STATE) {
261                tsk->thread.prot_addr = address;
262                tsk->thread.trap_no = error_code;
263		force_sigsegv(regs, error_code, si_code, address);
264                return;
265	}
266
267no_context:
268        /* Are we prepared to handle this kernel fault?  */
269        if ((fixup = search_exception_table(regs->psw.addr)) != 0) {
270                regs->psw.addr = fixup;
271                return;
272        }
273
274/*
275 * Oops. The kernel tried to access some bad page. We'll have to
276 * terminate things with extreme prejudice.
277 */
278        if (user_address == 0)
279                printk(KERN_ALERT "Unable to handle kernel pointer dereference"
280        	       " at virtual kernel address %016lx\n", address);
281        else
282                printk(KERN_ALERT "Unable to handle kernel paging request"
283		       " at virtual user address %016lx\n", address);
284
285        die("Oops", regs, error_code);
286        do_exit(SIGKILL);
287
288
289/*
290 * We ran out of memory, or some other thing happened to us that made
291 * us unable to handle the page fault gracefully.
292*/
293out_of_memory:
294	if (tsk->pid == 1) {
295		yield();
296		goto survive;
297	}
298	up_read(&mm->mmap_sem);
299	printk("VM: killing process %s\n", tsk->comm);
300	if (regs->psw.mask & PSW_PROBLEM_STATE)
301		do_exit(SIGKILL);
302	goto no_context;
303
304do_sigbus:
305	up_read(&mm->mmap_sem);
306
307	/*
308	 * Send a sigbus, regardless of whether we were in kernel
309	 * or user mode.
310	 */
311        tsk->thread.prot_addr = address;
312        tsk->thread.trap_no = error_code;
313	force_sig(SIGBUS, tsk);
314
315	/* Kernel mode? Handle exceptions or die */
316	if (!(regs->psw.mask & PSW_PROBLEM_STATE))
317		goto no_context;
318}
319
320void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
321{
322	regs->psw.addr -= (error_code >> 16);
323	do_exception(regs, 4);
324}
325
326void do_segment_exception(struct pt_regs *regs, unsigned long error_code)
327{
328	do_exception(regs, 0x10);
329}
330
331void do_page_exception(struct pt_regs *regs, unsigned long error_code)
332{
333	do_exception(regs, 0x11);
334}
335
336void do_region_exception(struct pt_regs *regs, unsigned long error_code)
337{
338	do_exception(regs, 0x3b);
339}
340
341#ifdef CONFIG_PFAULT
342/*
343 * 'pfault' pseudo page faults routines.
344 */
345static int pfault_disable = 0;
346
347static int __init nopfault(char *str)
348{
349	pfault_disable = 1;
350	return 1;
351}
352
353__setup("nopfault", nopfault);
354
355typedef struct {
356	__u16 refdiagc;
357	__u16 reffcode;
358	__u16 refdwlen;
359	__u16 refversn;
360	__u64 refgaddr;
361	__u64 refselmk;
362	__u64 refcmpmk;
363	__u64 reserved;
364} __attribute__ ((packed)) pfault_refbk_t;
365
366typedef struct _pseudo_wait_t {
367       struct _pseudo_wait_t *next;
368       wait_queue_head_t queue;
369       unsigned long address;
370       int resolved;
371} pseudo_wait_t;
372
373int pfault_init(void)
374{
375	pfault_refbk_t refbk =
376	{ 0x258, 0, 5, 2, __LC_KERNEL_STACK, 1ULL << 48, 1ULL << 48,
377          0x8000000000000000ULL };
378        int rc;
379
380	if (pfault_disable)
381		return -1;
382        __asm__ __volatile__(
383                "    diag  %1,%0,0x258\n"
384		"0:  j     2f\n"
385		"1:  la    %0,8\n"
386		"2:\n"
387		".section __ex_table,\"a\"\n"
388		"   .align 4\n"
389		"   .quad  0b,1b\n"
390		".previous"
391                : "=d" (rc) : "a" (&refbk) : "cc" );
392	__ctl_set_bit(0, 9);
393        return rc;
394}
395
396void pfault_fini(void)
397{
398	pfault_refbk_t refbk =
399	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
400
401	if (pfault_disable)
402		return;
403	__ctl_clear_bit(0, 9);
404        __asm__ __volatile__(
405                "    diag  %0,0,0x258\n"
406		"0:\n"
407		".section __ex_table,\"a\"\n"
408		"   .align 4\n"
409		"   .quad  0b,0b\n"
410		".previous"
411		: : "a" (&refbk) : "cc" );
412}
413
414asmlinkage void
415pfault_interrupt(struct pt_regs *regs, __u16 error_code)
416{
417	struct task_struct *tsk;
418	wait_queue_head_t queue;
419	wait_queue_head_t *qp;
420	__u16 subcode;
421
422	/*
423	 * Get the external interruption subcode & pfault
424	 * initial/completion signal bit. VM stores this
425	 * in the 'cpu address' field associated with the
426         * external interrupt.
427	 */
428	subcode = S390_lowcore.cpu_addr;
429	if ((subcode & 0xff00) != 0x0600)
430		return;
431
432	/*
433	 * Get the token (= address of kernel stack of affected task).
434	 */
435	tsk = (struct task_struct *)
436		(*((unsigned long *) __LC_PFAULT_INTPARM) - THREAD_SIZE);
437
438	/*
439	 * We got all needed information from the lowcore and can
440	 * now safely switch on interrupts.
441	 */
442	if (regs->psw.mask & PSW_PROBLEM_STATE)
443		__sti();
444
445	if (subcode & 0x0080) {
446		/* signal bit is set -> a page has been swapped in by VM */
447		qp = (wait_queue_head_t *)
448			xchg(&tsk->thread.pfault_wait, -1);
449		if (qp != NULL) {
450			/* Initial interrupt was faster than the completion
451			 * interrupt. pfault_wait is valid. Set pfault_wait
452			 * back to zero and wake up the process. This can
453			 * safely be done because the task is still sleeping
454			 * and can't procude new pfaults. */
455			tsk->thread.pfault_wait = 0ULL;
456			wake_up(qp);
457		}
458	} else {
459		/* signal bit not set -> a real page is missing. */
460                init_waitqueue_head (&queue);
461		qp = (wait_queue_head_t *)
462			xchg(&tsk->thread.pfault_wait, (addr_t) &queue);
463		if (qp != NULL) {
464			/* Completion interrupt was faster than the initial
465			 * interrupt (swapped in a -1 for pfault_wait). Set
466			 * pfault_wait back to zero and exit. This can be
467			 * done safely because tsk is running in kernel
468			 * mode and can't produce new pfaults. */
469			tsk->thread.pfault_wait = 0ULL;
470		}
471
472                /* go to sleep */
473                wait_event(queue, tsk->thread.pfault_wait == 0ULL);
474	}
475}
476#endif
477
478