xref: /openbmc/linux/arch/s390/mm/fault.c (revision fd589a8f)
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/perf_counter.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/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"
38 
39 #ifndef CONFIG_64BIT
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 */
50 
51 #ifdef CONFIG_SYSCTL
52 extern int sysctl_userprocess_debug;
53 #endif
54 
55 #ifdef CONFIG_KPROBES
56 static inline int notify_page_fault(struct pt_regs *regs, long err)
57 {
58 	int ret = 0;
59 
60 	/* kprobe_running() needs smp_processor_id() */
61 	if (!user_mode(regs)) {
62 		preempt_disable();
63 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
64 			ret = 1;
65 		preempt_enable();
66 	}
67 
68 	return ret;
69 }
70 #else
71 static inline int notify_page_fault(struct pt_regs *regs, long err)
72 {
73 	return 0;
74 }
75 #endif
76 
77 
78 /*
79  * Unlock any spinlocks which will prevent us from getting the
80  * message out.
81  */
82 void bust_spinlocks(int yes)
83 {
84 	if (yes) {
85 		oops_in_progress = 1;
86 	} else {
87 		int loglevel_save = console_loglevel;
88 		console_unblank();
89 		oops_in_progress = 0;
90 		/*
91 		 * OK, the message is on the console.  Now we call printk()
92 		 * without oops_in_progress set so that printk will give klogd
93 		 * a poke.  Hold onto your hats...
94 		 */
95 		console_loglevel = 15;
96 		printk(" ");
97 		console_loglevel = loglevel_save;
98 	}
99 }
100 
101 /*
102  * Returns the address space associated with the fault.
103  * Returns 0 for kernel space, 1 for user space and
104  * 2 for code execution in user space with noexec=on.
105  */
106 static inline int check_space(struct task_struct *tsk)
107 {
108 	/*
109 	 * The lowest two bits of S390_lowcore.trans_exc_code
110 	 * indicate which paging table was used.
111 	 */
112 	int desc = S390_lowcore.trans_exc_code & 3;
113 
114 	if (desc == 3)	/* Home Segment Table Descriptor */
115 		return switch_amode == 0;
116 	if (desc == 2)	/* Secondary Segment Table Descriptor */
117 		return tsk->thread.mm_segment.ar4;
118 #ifdef CONFIG_S390_SWITCH_AMODE
119 	if (unlikely(desc == 1)) { /* STD determined via access register */
120 		/* %a0 always indicates primary space. */
121 		if (S390_lowcore.exc_access_id != 0) {
122 			save_access_regs(tsk->thread.acrs);
123 			/*
124 			 * An alet of 0 indicates primary space.
125 			 * An alet of 1 indicates secondary space.
126 			 * Any other alet values generate an
127 			 * alen-translation exception.
128 			 */
129 			if (tsk->thread.acrs[S390_lowcore.exc_access_id])
130 				return tsk->thread.mm_segment.ar4;
131 		}
132 	}
133 #endif
134 	/* Primary Segment Table Descriptor */
135 	return switch_amode << s390_noexec;
136 }
137 
138 /*
139  * Send SIGSEGV to task.  This is an external routine
140  * to keep the stack usage of do_page_fault small.
141  */
142 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
143 		       int si_code, unsigned long address)
144 {
145 	struct siginfo si;
146 
147 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
148 #if defined(CONFIG_SYSCTL)
149 	if (sysctl_userprocess_debug)
150 #endif
151 	{
152 		printk("User process fault: interruption code 0x%lX\n",
153 		       error_code);
154 		printk("failing address: %lX\n", address);
155 		show_regs(regs);
156 	}
157 #endif
158 	si.si_signo = SIGSEGV;
159 	si.si_code = si_code;
160 	si.si_addr = (void __user *) address;
161 	force_sig_info(SIGSEGV, &si, current);
162 }
163 
164 static void do_no_context(struct pt_regs *regs, unsigned long error_code,
165 			  unsigned long address)
166 {
167 	const struct exception_table_entry *fixup;
168 
169 	/* Are we prepared to handle this kernel fault?  */
170 	fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
171 	if (fixup) {
172 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
173 		return;
174 	}
175 
176 	/*
177 	 * Oops. The kernel tried to access some bad page. We'll have to
178 	 * terminate things with extreme prejudice.
179 	 */
180 	if (check_space(current) == 0)
181 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
182 		       " at virtual kernel address %p\n", (void *)address);
183 	else
184 		printk(KERN_ALERT "Unable to handle kernel paging request"
185 		       " at virtual user address %p\n", (void *)address);
186 
187 	die("Oops", regs, error_code);
188 	do_exit(SIGKILL);
189 }
190 
191 static void do_low_address(struct pt_regs *regs, unsigned long error_code)
192 {
193 	/* Low-address protection hit in kernel mode means
194 	   NULL pointer write access in kernel mode.  */
195 	if (regs->psw.mask & PSW_MASK_PSTATE) {
196 		/* Low-address protection hit in user mode 'cannot happen'. */
197 		die ("Low-address protection", regs, error_code);
198 		do_exit(SIGKILL);
199 	}
200 
201 	do_no_context(regs, error_code, 0);
202 }
203 
204 static void do_sigbus(struct pt_regs *regs, unsigned long error_code,
205 		      unsigned long address)
206 {
207 	struct task_struct *tsk = current;
208 	struct mm_struct *mm = tsk->mm;
209 
210 	up_read(&mm->mmap_sem);
211 	/*
212 	 * Send a sigbus, regardless of whether we were in kernel
213 	 * or user mode.
214 	 */
215 	tsk->thread.prot_addr = address;
216 	tsk->thread.trap_no = error_code;
217 	force_sig(SIGBUS, tsk);
218 
219 	/* Kernel mode? Handle exceptions or die */
220 	if (!(regs->psw.mask & PSW_MASK_PSTATE))
221 		do_no_context(regs, error_code, address);
222 }
223 
224 #ifdef CONFIG_S390_EXEC_PROTECT
225 static int signal_return(struct mm_struct *mm, struct pt_regs *regs,
226 			 unsigned long address, unsigned long error_code)
227 {
228 	u16 instruction;
229 	int rc;
230 #ifdef CONFIG_COMPAT
231 	int compat;
232 #endif
233 
234 	pagefault_disable();
235 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
236 	pagefault_enable();
237 	if (rc)
238 		return -EFAULT;
239 
240 	up_read(&mm->mmap_sem);
241 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
242 #ifdef CONFIG_COMPAT
243 	compat = is_compat_task();
244 	if (compat && instruction == 0x0a77)
245 		sys32_sigreturn();
246 	else if (compat && instruction == 0x0aad)
247 		sys32_rt_sigreturn();
248 	else
249 #endif
250 	if (instruction == 0x0a77)
251 		sys_sigreturn();
252 	else if (instruction == 0x0aad)
253 		sys_rt_sigreturn();
254 	else {
255 		current->thread.prot_addr = address;
256 		current->thread.trap_no = error_code;
257 		do_sigsegv(regs, error_code, SEGV_MAPERR, address);
258 	}
259 	return 0;
260 }
261 #endif /* CONFIG_S390_EXEC_PROTECT */
262 
263 /*
264  * This routine handles page faults.  It determines the address,
265  * and the problem, and then passes it off to one of the appropriate
266  * routines.
267  *
268  * error_code:
269  *   04       Protection           ->  Write-Protection  (suprression)
270  *   10       Segment translation  ->  Not present       (nullification)
271  *   11       Page translation     ->  Not present       (nullification)
272  *   3b       Region third trans.  ->  Not present       (nullification)
273  */
274 static inline void
275 do_exception(struct pt_regs *regs, unsigned long error_code, int write)
276 {
277 	struct task_struct *tsk;
278 	struct mm_struct *mm;
279 	struct vm_area_struct *vma;
280 	unsigned long address;
281 	int space;
282 	int si_code;
283 	int fault;
284 
285 	if (notify_page_fault(regs, error_code))
286 		return;
287 
288 	tsk = current;
289 	mm = tsk->mm;
290 
291 	/* get the failing address and the affected space */
292 	address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
293 	space = check_space(tsk);
294 
295 	/*
296 	 * Verify that the fault happened in user space, that
297 	 * we are not in an interrupt and that there is a
298 	 * user context.
299 	 */
300 	if (unlikely(space == 0 || in_atomic() || !mm))
301 		goto no_context;
302 
303 	/*
304 	 * When we get here, the fault happened in the current
305 	 * task's user address space, so we can switch on the
306 	 * interrupts again and then search the VMAs
307 	 */
308 	local_irq_enable();
309 	perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
310 	down_read(&mm->mmap_sem);
311 
312 	si_code = SEGV_MAPERR;
313 	vma = find_vma(mm, address);
314 	if (!vma)
315 		goto bad_area;
316 
317 #ifdef CONFIG_S390_EXEC_PROTECT
318 	if (unlikely((space == 2) && !(vma->vm_flags & VM_EXEC)))
319 		if (!signal_return(mm, regs, address, error_code))
320 			/*
321 			 * signal_return() has done an up_read(&mm->mmap_sem)
322 			 * if it returns 0.
323 			 */
324 			return;
325 #endif
326 
327 	if (vma->vm_start <= address)
328 		goto good_area;
329 	if (!(vma->vm_flags & VM_GROWSDOWN))
330 		goto bad_area;
331 	if (expand_stack(vma, address))
332 		goto bad_area;
333 /*
334  * Ok, we have a good vm_area for this memory access, so
335  * we can handle it..
336  */
337 good_area:
338 	si_code = SEGV_ACCERR;
339 	if (!write) {
340 		/* page not present, check vm flags */
341 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
342 			goto bad_area;
343 	} else {
344 		if (!(vma->vm_flags & VM_WRITE))
345 			goto bad_area;
346 	}
347 
348 	if (is_vm_hugetlb_page(vma))
349 		address &= HPAGE_MASK;
350 	/*
351 	 * If for any reason at all we couldn't handle the fault,
352 	 * make sure we exit gracefully rather than endlessly redo
353 	 * the fault.
354 	 */
355 	fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
356 	if (unlikely(fault & VM_FAULT_ERROR)) {
357 		if (fault & VM_FAULT_OOM) {
358 			up_read(&mm->mmap_sem);
359 			pagefault_out_of_memory();
360 			return;
361 		} else if (fault & VM_FAULT_SIGBUS) {
362 			do_sigbus(regs, error_code, address);
363 			return;
364 		}
365 		BUG();
366 	}
367 	if (fault & VM_FAULT_MAJOR) {
368 		tsk->maj_flt++;
369 		perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
370 				     regs, address);
371 	} else {
372 		tsk->min_flt++;
373 		perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
374 				     regs, address);
375 	}
376         up_read(&mm->mmap_sem);
377 	/*
378 	 * The instruction that caused the program check will
379 	 * be repeated. Don't signal single step via SIGTRAP.
380 	 */
381 	clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
382         return;
383 
384 /*
385  * Something tried to access memory that isn't in our memory map..
386  * Fix it, but check if it's kernel or user first..
387  */
388 bad_area:
389 	up_read(&mm->mmap_sem);
390 
391 	/* User mode accesses just cause a SIGSEGV */
392 	if (regs->psw.mask & PSW_MASK_PSTATE) {
393 		tsk->thread.prot_addr = address;
394 		tsk->thread.trap_no = error_code;
395 		do_sigsegv(regs, error_code, si_code, address);
396 		return;
397 	}
398 
399 no_context:
400 	do_no_context(regs, error_code, address);
401 }
402 
403 void __kprobes do_protection_exception(struct pt_regs *regs,
404 				       long error_code)
405 {
406 	/* Protection exception is supressing, decrement psw address. */
407 	regs->psw.addr -= (error_code >> 16);
408 	/*
409 	 * Check for low-address protection.  This needs to be treated
410 	 * as a special case because the translation exception code
411 	 * field is not guaranteed to contain valid data in this case.
412 	 */
413 	if (unlikely(!(S390_lowcore.trans_exc_code & 4))) {
414 		do_low_address(regs, error_code);
415 		return;
416 	}
417 	do_exception(regs, 4, 1);
418 }
419 
420 void __kprobes do_dat_exception(struct pt_regs *regs, long error_code)
421 {
422 	do_exception(regs, error_code & 0xff, 0);
423 }
424 
425 #ifdef CONFIG_64BIT
426 void __kprobes do_asce_exception(struct pt_regs *regs, unsigned long error_code)
427 {
428 	struct mm_struct *mm;
429 	struct vm_area_struct *vma;
430 	unsigned long address;
431 	int space;
432 
433 	mm = current->mm;
434 	address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
435 	space = check_space(current);
436 
437 	if (unlikely(space == 0 || in_atomic() || !mm))
438 		goto no_context;
439 
440 	local_irq_enable();
441 
442 	down_read(&mm->mmap_sem);
443 	vma = find_vma(mm, address);
444 	up_read(&mm->mmap_sem);
445 
446 	if (vma) {
447 		update_mm(mm, current);
448 		return;
449 	}
450 
451 	/* User mode accesses just cause a SIGSEGV */
452 	if (regs->psw.mask & PSW_MASK_PSTATE) {
453 		current->thread.prot_addr = address;
454 		current->thread.trap_no = error_code;
455 		do_sigsegv(regs, error_code, SEGV_MAPERR, address);
456 		return;
457 	}
458 
459 no_context:
460 	do_no_context(regs, error_code, address);
461 }
462 #endif
463 
464 #ifdef CONFIG_PFAULT
465 /*
466  * 'pfault' pseudo page faults routines.
467  */
468 static ext_int_info_t ext_int_pfault;
469 static int pfault_disable = 0;
470 
471 static int __init nopfault(char *str)
472 {
473 	pfault_disable = 1;
474 	return 1;
475 }
476 
477 __setup("nopfault", nopfault);
478 
479 typedef struct {
480 	__u16 refdiagc;
481 	__u16 reffcode;
482 	__u16 refdwlen;
483 	__u16 refversn;
484 	__u64 refgaddr;
485 	__u64 refselmk;
486 	__u64 refcmpmk;
487 	__u64 reserved;
488 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
489 
490 int pfault_init(void)
491 {
492 	pfault_refbk_t refbk =
493 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
494 		  __PF_RES_FIELD };
495         int rc;
496 
497 	if (!MACHINE_IS_VM || pfault_disable)
498 		return -1;
499 	asm volatile(
500 		"	diag	%1,%0,0x258\n"
501 		"0:	j	2f\n"
502 		"1:	la	%0,8\n"
503 		"2:\n"
504 		EX_TABLE(0b,1b)
505 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
506         __ctl_set_bit(0, 9);
507         return rc;
508 }
509 
510 void pfault_fini(void)
511 {
512 	pfault_refbk_t refbk =
513 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
514 
515 	if (!MACHINE_IS_VM || pfault_disable)
516 		return;
517 	__ctl_clear_bit(0,9);
518 	asm volatile(
519 		"	diag	%0,0,0x258\n"
520 		"0:\n"
521 		EX_TABLE(0b,0b)
522 		: : "a" (&refbk), "m" (refbk) : "cc");
523 }
524 
525 static void pfault_interrupt(__u16 error_code)
526 {
527 	struct task_struct *tsk;
528 	__u16 subcode;
529 
530 	/*
531 	 * Get the external interruption subcode & pfault
532 	 * initial/completion signal bit. VM stores this
533 	 * in the 'cpu address' field associated with the
534          * external interrupt.
535 	 */
536 	subcode = S390_lowcore.cpu_addr;
537 	if ((subcode & 0xff00) != __SUBCODE_MASK)
538 		return;
539 
540 	/*
541 	 * Get the token (= address of the task structure of the affected task).
542 	 */
543 	tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
544 
545 	if (subcode & 0x0080) {
546 		/* signal bit is set -> a page has been swapped in by VM */
547 		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
548 			/* Initial interrupt was faster than the completion
549 			 * interrupt. pfault_wait is valid. Set pfault_wait
550 			 * back to zero and wake up the process. This can
551 			 * safely be done because the task is still sleeping
552 			 * and can't produce new pfaults. */
553 			tsk->thread.pfault_wait = 0;
554 			wake_up_process(tsk);
555 			put_task_struct(tsk);
556 		}
557 	} else {
558 		/* signal bit not set -> a real page is missing. */
559 		get_task_struct(tsk);
560 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
561 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
562 			/* Completion interrupt was faster than the initial
563 			 * interrupt (swapped in a -1 for pfault_wait). Set
564 			 * pfault_wait back to zero and exit. This can be
565 			 * done safely because tsk is running in kernel
566 			 * mode and can't produce new pfaults. */
567 			tsk->thread.pfault_wait = 0;
568 			set_task_state(tsk, TASK_RUNNING);
569 			put_task_struct(tsk);
570 		} else
571 			set_tsk_need_resched(tsk);
572 	}
573 }
574 
575 void __init pfault_irq_init(void)
576 {
577 	if (!MACHINE_IS_VM)
578 		return;
579 
580 	/*
581 	 * Try to get pfault pseudo page faults going.
582 	 */
583 	if (register_early_external_interrupt(0x2603, pfault_interrupt,
584 					      &ext_int_pfault) != 0)
585 		panic("Couldn't request external interrupt 0x2603");
586 
587 	if (pfault_init() == 0)
588 		return;
589 
590 	/* Tough luck, no pfault. */
591 	pfault_disable = 1;
592 	unregister_early_external_interrupt(0x2603, pfault_interrupt,
593 					    &ext_int_pfault);
594 }
595 #endif
596