Linux Audio

Check our new training course

Embedded Linux training

Mar 31-Apr 8, 2025
Register
Loading...
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  S390 version
  4 *    Copyright IBM Corp. 1999
  5 *    Author(s): Hartmut Penner (hp@de.ibm.com)
  6 *               Ulrich Weigand (uweigand@de.ibm.com)
  7 *
  8 *  Derived from "arch/i386/mm/fault.c"
  9 *    Copyright (C) 1995  Linus Torvalds
 10 */
 11
 12#include <linux/kernel_stat.h>
 13#include <linux/perf_event.h>
 14#include <linux/signal.h>
 15#include <linux/sched.h>
 16#include <linux/sched/debug.h>
 17#include <linux/kernel.h>
 18#include <linux/errno.h>
 19#include <linux/string.h>
 20#include <linux/types.h>
 21#include <linux/ptrace.h>
 22#include <linux/mman.h>
 23#include <linux/mm.h>
 24#include <linux/compat.h>
 25#include <linux/smp.h>
 26#include <linux/kdebug.h>
 27#include <linux/init.h>
 28#include <linux/console.h>
 29#include <linux/extable.h>
 30#include <linux/hardirq.h>
 31#include <linux/kprobes.h>
 32#include <linux/uaccess.h>
 33#include <linux/hugetlb.h>
 34#include <asm/asm-offsets.h>
 35#include <asm/diag.h>
 36#include <asm/pgtable.h>
 37#include <asm/gmap.h>
 38#include <asm/irq.h>
 39#include <asm/mmu_context.h>
 40#include <asm/facility.h>
 41#include "../kernel/entry.h"
 42
 
 
 
 
 
 43#define __FAIL_ADDR_MASK -4096L
 44#define __SUBCODE_MASK 0x0600
 45#define __PF_RES_FIELD 0x8000000000000000ULL
 
 46
 47#define VM_FAULT_BADCONTEXT	0x010000
 48#define VM_FAULT_BADMAP		0x020000
 49#define VM_FAULT_BADACCESS	0x040000
 50#define VM_FAULT_SIGNAL		0x080000
 51#define VM_FAULT_PFAULT		0x100000
 52
 53enum fault_type {
 54	KERNEL_FAULT,
 55	USER_FAULT,
 56	VDSO_FAULT,
 57	GMAP_FAULT,
 58};
 59
 60static unsigned long store_indication __read_mostly;
 61
 
 62static int __init fault_init(void)
 63{
 64	if (test_facility(75))
 65		store_indication = 0xc00;
 66	return 0;
 67}
 68early_initcall(fault_init);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 69
 70/*
 71 * Find out which address space caused the exception.
 
 72 */
 73static enum fault_type get_fault_type(struct pt_regs *regs)
 74{
 75	unsigned long trans_exc_code;
 76
 
 
 
 
 77	trans_exc_code = regs->int_parm_long & 3;
 78	if (likely(trans_exc_code == 0)) {
 79		/* primary space exception */
 80		if (IS_ENABLED(CONFIG_PGSTE) &&
 81		    test_pt_regs_flag(regs, PIF_GUEST_FAULT))
 82			return GMAP_FAULT;
 83		if (current->thread.mm_segment == USER_DS)
 84			return USER_FAULT;
 85		return KERNEL_FAULT;
 86	}
 87	if (trans_exc_code == 2) {
 88		/* secondary space exception */
 89		if (current->thread.mm_segment & 1) {
 90			if (current->thread.mm_segment == USER_DS_SACF)
 91				return USER_FAULT;
 92			return KERNEL_FAULT;
 93		}
 94		return VDSO_FAULT;
 95	}
 96	if (trans_exc_code == 1) {
 97		/* access register mode, not used in the kernel */
 98		return USER_FAULT;
 99	}
100	/* home space exception -> access via kernel ASCE */
101	return KERNEL_FAULT;
102}
103
104static int bad_address(void *p)
105{
106	unsigned long dummy;
107
108	return probe_kernel_address((unsigned long *)p, dummy);
109}
110
 
111static void dump_pagetable(unsigned long asce, unsigned long address)
112{
113	unsigned long *table = __va(asce & _ASCE_ORIGIN);
114
115	pr_alert("AS:%016lx ", asce);
116	switch (asce & _ASCE_TYPE_MASK) {
117	case _ASCE_TYPE_REGION1:
118		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
119		if (bad_address(table))
120			goto bad;
121		pr_cont("R1:%016lx ", *table);
122		if (*table & _REGION_ENTRY_INVALID)
123			goto out;
124		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
125		/* fallthrough */
126	case _ASCE_TYPE_REGION2:
127		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
128		if (bad_address(table))
129			goto bad;
130		pr_cont("R2:%016lx ", *table);
131		if (*table & _REGION_ENTRY_INVALID)
132			goto out;
133		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
134		/* fallthrough */
135	case _ASCE_TYPE_REGION3:
136		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
137		if (bad_address(table))
138			goto bad;
139		pr_cont("R3:%016lx ", *table);
140		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
141			goto out;
142		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
143		/* fallthrough */
144	case _ASCE_TYPE_SEGMENT:
145		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
146		if (bad_address(table))
147			goto bad;
148		pr_cont("S:%016lx ", *table);
149		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
150			goto out;
151		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
152	}
153	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
154	if (bad_address(table))
155		goto bad;
156	pr_cont("P:%016lx ", *table);
157out:
158	pr_cont("\n");
159	return;
160bad:
161	pr_cont("BAD\n");
162}
163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
164static void dump_fault_info(struct pt_regs *regs)
165{
166	unsigned long asce;
167
168	pr_alert("Failing address: %016lx TEID: %016lx\n",
169		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
170	pr_alert("Fault in ");
171	switch (regs->int_parm_long & 3) {
172	case 3:
173		pr_cont("home space ");
174		break;
175	case 2:
176		pr_cont("secondary space ");
177		break;
178	case 1:
179		pr_cont("access register ");
180		break;
181	case 0:
182		pr_cont("primary space ");
183		break;
184	}
185	pr_cont("mode while using ");
186	switch (get_fault_type(regs)) {
187	case USER_FAULT:
188		asce = S390_lowcore.user_asce;
189		pr_cont("user ");
190		break;
191	case VDSO_FAULT:
192		asce = S390_lowcore.vdso_asce;
193		pr_cont("vdso ");
194		break;
195	case GMAP_FAULT:
196		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
197		pr_cont("gmap ");
198		break;
199	case KERNEL_FAULT:
200		asce = S390_lowcore.kernel_asce;
201		pr_cont("kernel ");
202		break;
203	default:
204		unreachable();
 
 
 
 
 
 
 
 
205	}
206	pr_cont("ASCE.\n");
207	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
208}
209
210int show_unhandled_signals = 1;
211
212void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
213{
214	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
215		return;
216	if (!unhandled_signal(current, signr))
217		return;
218	if (!printk_ratelimit())
219		return;
220	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
221	       regs->int_code & 0xffff, regs->int_code >> 17);
222	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
223	printk(KERN_CONT "\n");
224	if (is_mm_fault)
225		dump_fault_info(regs);
 
226	show_regs(regs);
227}
228
229/*
230 * Send SIGSEGV to task.  This is an external routine
231 * to keep the stack usage of do_page_fault small.
232 */
233static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
234{
235	report_user_fault(regs, SIGSEGV, 1);
236	force_sig_fault(SIGSEGV, si_code,
237			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
238}
239
240const struct exception_table_entry *s390_search_extables(unsigned long addr)
241{
242	const struct exception_table_entry *fixup;
243
244	fixup = search_extable(__start_dma_ex_table,
245			       __stop_dma_ex_table - __start_dma_ex_table,
246			       addr);
247	if (!fixup)
248		fixup = search_exception_tables(addr);
249	return fixup;
250}
251
252static noinline void do_no_context(struct pt_regs *regs)
253{
254	const struct exception_table_entry *fixup;
 
255
256	/* Are we prepared to handle this kernel fault?  */
257	fixup = s390_search_extables(regs->psw.addr);
258	if (fixup) {
259		regs->psw.addr = extable_fixup(fixup);
260		return;
261	}
262
263	/*
264	 * Oops. The kernel tried to access some bad page. We'll have to
265	 * terminate things with extreme prejudice.
266	 */
267	if (get_fault_type(regs) == KERNEL_FAULT)
 
268		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
269		       " in virtual kernel address space\n");
270	else
271		printk(KERN_ALERT "Unable to handle kernel paging request"
272		       " in virtual user address space\n");
 
 
273	dump_fault_info(regs);
274	die(regs, "Oops");
275	do_exit(SIGKILL);
276}
277
278static noinline void do_low_address(struct pt_regs *regs)
279{
280	/* Low-address protection hit in kernel mode means
281	   NULL pointer write access in kernel mode.  */
282	if (regs->psw.mask & PSW_MASK_PSTATE) {
283		/* Low-address protection hit in user mode 'cannot happen'. */
284		die (regs, "Low-address protection");
285		do_exit(SIGKILL);
286	}
287
288	do_no_context(regs);
289}
290
291static noinline void do_sigbus(struct pt_regs *regs)
292{
 
 
 
293	/*
294	 * Send a sigbus, regardless of whether we were in kernel
295	 * or user mode.
296	 */
297	force_sig_fault(SIGBUS, BUS_ADRERR,
298			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
299}
300
301static noinline int signal_return(struct pt_regs *regs)
302{
303	u16 instruction;
304	int rc;
305
306	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
307	if (rc)
308		return rc;
309	if (instruction == 0x0a77) {
310		set_pt_regs_flag(regs, PIF_SYSCALL);
311		regs->int_code = 0x00040077;
312		return 0;
313	} else if (instruction == 0x0aad) {
314		set_pt_regs_flag(regs, PIF_SYSCALL);
315		regs->int_code = 0x000400ad;
316		return 0;
317	}
318	return -EACCES;
319}
320
321static noinline void do_fault_error(struct pt_regs *regs, int access,
322					vm_fault_t fault)
323{
324	int si_code;
325
326	switch (fault) {
327	case VM_FAULT_BADACCESS:
328		if (access == VM_EXEC && signal_return(regs) == 0)
329			break;
330		/* fallthrough */
331	case VM_FAULT_BADMAP:
332		/* Bad memory access. Check if it is kernel or user space. */
333		if (user_mode(regs)) {
334			/* User mode accesses just cause a SIGSEGV */
335			si_code = (fault == VM_FAULT_BADMAP) ?
336				SEGV_MAPERR : SEGV_ACCERR;
337			do_sigsegv(regs, si_code);
338			break;
339		}
340		/* fallthrough */
341	case VM_FAULT_BADCONTEXT:
342		/* fallthrough */
343	case VM_FAULT_PFAULT:
344		do_no_context(regs);
345		break;
346	case VM_FAULT_SIGNAL:
347		if (!user_mode(regs))
348			do_no_context(regs);
349		break;
350	default: /* fault & VM_FAULT_ERROR */
351		if (fault & VM_FAULT_OOM) {
352			if (!user_mode(regs))
353				do_no_context(regs);
354			else
355				pagefault_out_of_memory();
356		} else if (fault & VM_FAULT_SIGSEGV) {
357			/* Kernel mode? Handle exceptions or die */
358			if (!user_mode(regs))
359				do_no_context(regs);
360			else
361				do_sigsegv(regs, SEGV_MAPERR);
362		} else if (fault & VM_FAULT_SIGBUS) {
363			/* Kernel mode? Handle exceptions or die */
364			if (!user_mode(regs))
365				do_no_context(regs);
366			else
367				do_sigbus(regs);
368		} else
369			BUG();
370		break;
371	}
372}
373
374/*
375 * This routine handles page faults.  It determines the address,
376 * and the problem, and then passes it off to one of the appropriate
377 * routines.
378 *
379 * interruption code (int_code):
380 *   04       Protection           ->  Write-Protection  (suprression)
381 *   10       Segment translation  ->  Not present       (nullification)
382 *   11       Page translation     ->  Not present       (nullification)
383 *   3b       Region third trans.  ->  Not present       (nullification)
384 */
385static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
386{
 
387	struct gmap *gmap;
 
388	struct task_struct *tsk;
389	struct mm_struct *mm;
390	struct vm_area_struct *vma;
391	enum fault_type type;
392	unsigned long trans_exc_code;
393	unsigned long address;
394	unsigned int flags;
395	vm_fault_t fault;
396
397	tsk = current;
398	/*
399	 * The instruction that caused the program check has
400	 * been nullified. Don't signal single step via SIGTRAP.
401	 */
402	clear_pt_regs_flag(regs, PIF_PER_TRAP);
403
404	if (kprobe_page_fault(regs, 14))
405		return 0;
406
407	mm = tsk->mm;
408	trans_exc_code = regs->int_parm_long;
409
410	/*
411	 * Verify that the fault happened in user space, that
412	 * we are not in an interrupt and that there is a 
413	 * user context.
414	 */
415	fault = VM_FAULT_BADCONTEXT;
416	type = get_fault_type(regs);
417	switch (type) {
418	case KERNEL_FAULT:
419		goto out;
420	case VDSO_FAULT:
421		fault = VM_FAULT_BADMAP;
422		goto out;
423	case USER_FAULT:
424	case GMAP_FAULT:
425		if (faulthandler_disabled() || !mm)
426			goto out;
427		break;
428	}
429
430	address = trans_exc_code & __FAIL_ADDR_MASK;
431	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
432	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
433	if (user_mode(regs))
434		flags |= FAULT_FLAG_USER;
435	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
436		flags |= FAULT_FLAG_WRITE;
437	down_read(&mm->mmap_sem);
438
439	gmap = NULL;
440	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
441		gmap = (struct gmap *) S390_lowcore.gmap;
442		current->thread.gmap_addr = address;
443		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
444		current->thread.gmap_int_code = regs->int_code & 0xffff;
445		address = __gmap_translate(gmap, address);
446		if (address == -EFAULT) {
447			fault = VM_FAULT_BADMAP;
448			goto out_up;
449		}
 
 
 
 
450		if (gmap->pfault_enabled)
451			flags |= FAULT_FLAG_RETRY_NOWAIT;
452	}
 
453
454retry:
455	fault = VM_FAULT_BADMAP;
456	vma = find_vma(mm, address);
457	if (!vma)
458		goto out_up;
459
460	if (unlikely(vma->vm_start > address)) {
461		if (!(vma->vm_flags & VM_GROWSDOWN))
462			goto out_up;
463		if (expand_stack(vma, address))
464			goto out_up;
465	}
466
467	/*
468	 * Ok, we have a good vm_area for this memory access, so
469	 * we can handle it..
470	 */
471	fault = VM_FAULT_BADACCESS;
472	if (unlikely(!(vma->vm_flags & access)))
473		goto out_up;
474
475	if (is_vm_hugetlb_page(vma))
476		address &= HPAGE_MASK;
477	/*
478	 * If for any reason at all we couldn't handle the fault,
479	 * make sure we exit gracefully rather than endlessly redo
480	 * the fault.
481	 */
482	fault = handle_mm_fault(vma, address, flags);
483	/* No reason to continue if interrupted by SIGKILL. */
484	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
485		fault = VM_FAULT_SIGNAL;
486		if (flags & FAULT_FLAG_RETRY_NOWAIT)
487			goto out_up;
488		goto out;
489	}
490	if (unlikely(fault & VM_FAULT_ERROR))
491		goto out_up;
492
493	/*
494	 * Major/minor page fault accounting is only done on the
495	 * initial attempt. If we go through a retry, it is extremely
496	 * likely that the page will be found in page cache at that point.
497	 */
498	if (flags & FAULT_FLAG_ALLOW_RETRY) {
499		if (fault & VM_FAULT_MAJOR) {
500			tsk->maj_flt++;
501			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
502				      regs, address);
503		} else {
504			tsk->min_flt++;
505			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
506				      regs, address);
507		}
508		if (fault & VM_FAULT_RETRY) {
509			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
510			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
511				/* FAULT_FLAG_RETRY_NOWAIT has been set,
512				 * mmap_sem has not been released */
513				current->thread.gmap_pfault = 1;
514				fault = VM_FAULT_PFAULT;
515				goto out_up;
516			}
 
517			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
518			 * of starvation. */
519			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
520				   FAULT_FLAG_RETRY_NOWAIT);
521			flags |= FAULT_FLAG_TRIED;
522			down_read(&mm->mmap_sem);
523			goto retry;
524		}
525	}
526	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
527		address =  __gmap_link(gmap, current->thread.gmap_addr,
528				       address);
529		if (address == -EFAULT) {
530			fault = VM_FAULT_BADMAP;
531			goto out_up;
532		}
533		if (address == -ENOMEM) {
534			fault = VM_FAULT_OOM;
535			goto out_up;
536		}
537	}
538	fault = 0;
539out_up:
540	up_read(&mm->mmap_sem);
541out:
542	return fault;
543}
544
545void do_protection_exception(struct pt_regs *regs)
546{
547	unsigned long trans_exc_code;
548	int access;
549	vm_fault_t fault;
550
551	trans_exc_code = regs->int_parm_long;
552	/*
553	 * Protection exceptions are suppressing, decrement psw address.
554	 * The exception to this rule are aborted transactions, for these
555	 * the PSW already points to the correct location.
556	 */
557	if (!(regs->int_code & 0x200))
558		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
559	/*
560	 * Check for low-address protection.  This needs to be treated
561	 * as a special case because the translation exception code
562	 * field is not guaranteed to contain valid data in this case.
563	 */
564	if (unlikely(!(trans_exc_code & 4))) {
565		do_low_address(regs);
566		return;
567	}
568	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
569		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
570					(regs->psw.addr & PAGE_MASK);
571		access = VM_EXEC;
572		fault = VM_FAULT_BADACCESS;
573	} else {
574		access = VM_WRITE;
575		fault = do_exception(regs, access);
576	}
577	if (unlikely(fault))
578		do_fault_error(regs, access, fault);
579}
580NOKPROBE_SYMBOL(do_protection_exception);
581
582void do_dat_exception(struct pt_regs *regs)
583{
584	int access;
585	vm_fault_t fault;
586
587	access = VM_READ | VM_EXEC | VM_WRITE;
588	fault = do_exception(regs, access);
589	if (unlikely(fault))
590		do_fault_error(regs, access, fault);
591}
592NOKPROBE_SYMBOL(do_dat_exception);
593
594#ifdef CONFIG_PFAULT 
595/*
596 * 'pfault' pseudo page faults routines.
597 */
598static int pfault_disable;
599
600static int __init nopfault(char *str)
601{
602	pfault_disable = 1;
603	return 1;
604}
605
606__setup("nopfault", nopfault);
607
608struct pfault_refbk {
609	u16 refdiagc;
610	u16 reffcode;
611	u16 refdwlen;
612	u16 refversn;
613	u64 refgaddr;
614	u64 refselmk;
615	u64 refcmpmk;
616	u64 reserved;
617} __attribute__ ((packed, aligned(8)));
618
619static struct pfault_refbk pfault_init_refbk = {
620	.refdiagc = 0x258,
621	.reffcode = 0,
622	.refdwlen = 5,
623	.refversn = 2,
624	.refgaddr = __LC_LPP,
625	.refselmk = 1ULL << 48,
626	.refcmpmk = 1ULL << 48,
627	.reserved = __PF_RES_FIELD
628};
629
630int pfault_init(void)
631{
 
 
 
 
 
 
 
 
 
632        int rc;
633
634	if (pfault_disable)
635		return -1;
636	diag_stat_inc(DIAG_STAT_X258);
637	asm volatile(
638		"	diag	%1,%0,0x258\n"
639		"0:	j	2f\n"
640		"1:	la	%0,8\n"
641		"2:\n"
642		EX_TABLE(0b,1b)
643		: "=d" (rc)
644		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
645        return rc;
646}
647
648static struct pfault_refbk pfault_fini_refbk = {
649	.refdiagc = 0x258,
650	.reffcode = 1,
651	.refdwlen = 5,
652	.refversn = 2,
653};
654
655void pfault_fini(void)
656{
 
 
 
 
 
 
657
658	if (pfault_disable)
659		return;
660	diag_stat_inc(DIAG_STAT_X258);
661	asm volatile(
662		"	diag	%0,0,0x258\n"
663		"0:	nopr	%%r7\n"
664		EX_TABLE(0b,0b)
665		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
666}
667
668static DEFINE_SPINLOCK(pfault_lock);
669static LIST_HEAD(pfault_list);
670
671#define PF_COMPLETE	0x0080
672
673/*
674 * The mechanism of our pfault code: if Linux is running as guest, runs a user
675 * space process and the user space process accesses a page that the host has
676 * paged out we get a pfault interrupt.
677 *
678 * This allows us, within the guest, to schedule a different process. Without
679 * this mechanism the host would have to suspend the whole virtual cpu until
680 * the page has been paged in.
681 *
682 * So when we get such an interrupt then we set the state of the current task
683 * to uninterruptible and also set the need_resched flag. Both happens within
684 * interrupt context(!). If we later on want to return to user space we
685 * recognize the need_resched flag and then call schedule().  It's not very
686 * obvious how this works...
687 *
688 * Of course we have a lot of additional fun with the completion interrupt (->
689 * host signals that a page of a process has been paged in and the process can
690 * continue to run). This interrupt can arrive on any cpu and, since we have
691 * virtual cpus, actually appear before the interrupt that signals that a page
692 * is missing.
693 */
694static void pfault_interrupt(struct ext_code ext_code,
695			     unsigned int param32, unsigned long param64)
696{
697	struct task_struct *tsk;
698	__u16 subcode;
699	pid_t pid;
700
701	/*
702	 * Get the external interruption subcode & pfault initial/completion
703	 * signal bit. VM stores this in the 'cpu address' field associated
704	 * with the external interrupt.
 
705	 */
706	subcode = ext_code.subcode;
707	if ((subcode & 0xff00) != __SUBCODE_MASK)
708		return;
709	inc_irq_stat(IRQEXT_PFL);
710	/* Get the token (= pid of the affected task). */
711	pid = param64 & LPP_PID_MASK;
712	rcu_read_lock();
713	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
714	if (tsk)
715		get_task_struct(tsk);
716	rcu_read_unlock();
717	if (!tsk)
718		return;
719	spin_lock(&pfault_lock);
720	if (subcode & PF_COMPLETE) {
721		/* signal bit is set -> a page has been swapped in by VM */
722		if (tsk->thread.pfault_wait == 1) {
723			/* Initial interrupt was faster than the completion
724			 * interrupt. pfault_wait is valid. Set pfault_wait
725			 * back to zero and wake up the process. This can
726			 * safely be done because the task is still sleeping
727			 * and can't produce new pfaults. */
728			tsk->thread.pfault_wait = 0;
729			list_del(&tsk->thread.list);
730			wake_up_process(tsk);
731			put_task_struct(tsk);
732		} else {
733			/* Completion interrupt was faster than initial
734			 * interrupt. Set pfault_wait to -1 so the initial
735			 * interrupt doesn't put the task to sleep.
736			 * If the task is not running, ignore the completion
737			 * interrupt since it must be a leftover of a PFAULT
738			 * CANCEL operation which didn't remove all pending
739			 * completion interrupts. */
740			if (tsk->state == TASK_RUNNING)
741				tsk->thread.pfault_wait = -1;
742		}
743	} else {
744		/* signal bit not set -> a real page is missing. */
745		if (WARN_ON_ONCE(tsk != current))
746			goto out;
747		if (tsk->thread.pfault_wait == 1) {
748			/* Already on the list with a reference: put to sleep */
749			goto block;
 
750		} else if (tsk->thread.pfault_wait == -1) {
751			/* Completion interrupt was faster than the initial
752			 * interrupt (pfault_wait == -1). Set pfault_wait
753			 * back to zero and exit. */
754			tsk->thread.pfault_wait = 0;
755		} else {
756			/* Initial interrupt arrived before completion
757			 * interrupt. Let the task sleep.
758			 * An extra task reference is needed since a different
759			 * cpu may set the task state to TASK_RUNNING again
760			 * before the scheduler is reached. */
761			get_task_struct(tsk);
762			tsk->thread.pfault_wait = 1;
763			list_add(&tsk->thread.list, &pfault_list);
764block:
765			/* Since this must be a userspace fault, there
766			 * is no kernel task state to trample. Rely on the
767			 * return to userspace schedule() to block. */
768			__set_current_state(TASK_UNINTERRUPTIBLE);
769			set_tsk_need_resched(tsk);
770			set_preempt_need_resched();
771		}
772	}
773out:
774	spin_unlock(&pfault_lock);
775	put_task_struct(tsk);
776}
777
778static int pfault_cpu_dead(unsigned int cpu)
 
779{
780	struct thread_struct *thread, *next;
781	struct task_struct *tsk;
782
783	spin_lock_irq(&pfault_lock);
784	list_for_each_entry_safe(thread, next, &pfault_list, list) {
785		thread->pfault_wait = 0;
786		list_del(&thread->list);
787		tsk = container_of(thread, struct task_struct, thread);
788		wake_up_process(tsk);
789		put_task_struct(tsk);
 
 
 
 
 
 
 
790	}
791	spin_unlock_irq(&pfault_lock);
792	return 0;
793}
794
795static int __init pfault_irq_init(void)
796{
797	int rc;
798
799	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
800	if (rc)
801		goto out_extint;
802	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
803	if (rc)
804		goto out_pfault;
805	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
806	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
807				  NULL, pfault_cpu_dead);
808	return 0;
809
810out_pfault:
811	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
812out_extint:
813	pfault_disable = 1;
814	return rc;
815}
816early_initcall(pfault_irq_init);
817
818#endif /* CONFIG_PFAULT */
v3.15
 
  1/*
  2 *  S390 version
  3 *    Copyright IBM Corp. 1999
  4 *    Author(s): Hartmut Penner (hp@de.ibm.com)
  5 *               Ulrich Weigand (uweigand@de.ibm.com)
  6 *
  7 *  Derived from "arch/i386/mm/fault.c"
  8 *    Copyright (C) 1995  Linus Torvalds
  9 */
 10
 11#include <linux/kernel_stat.h>
 12#include <linux/perf_event.h>
 13#include <linux/signal.h>
 14#include <linux/sched.h>
 
 15#include <linux/kernel.h>
 16#include <linux/errno.h>
 17#include <linux/string.h>
 18#include <linux/types.h>
 19#include <linux/ptrace.h>
 20#include <linux/mman.h>
 21#include <linux/mm.h>
 22#include <linux/compat.h>
 23#include <linux/smp.h>
 24#include <linux/kdebug.h>
 25#include <linux/init.h>
 26#include <linux/console.h>
 27#include <linux/module.h>
 28#include <linux/hardirq.h>
 29#include <linux/kprobes.h>
 30#include <linux/uaccess.h>
 31#include <linux/hugetlb.h>
 32#include <asm/asm-offsets.h>
 
 33#include <asm/pgtable.h>
 
 34#include <asm/irq.h>
 35#include <asm/mmu_context.h>
 36#include <asm/facility.h>
 37#include "../kernel/entry.h"
 38
 39#ifndef CONFIG_64BIT
 40#define __FAIL_ADDR_MASK 0x7ffff000
 41#define __SUBCODE_MASK 0x0200
 42#define __PF_RES_FIELD 0ULL
 43#else /* CONFIG_64BIT */
 44#define __FAIL_ADDR_MASK -4096L
 45#define __SUBCODE_MASK 0x0600
 46#define __PF_RES_FIELD 0x8000000000000000ULL
 47#endif /* CONFIG_64BIT */
 48
 49#define VM_FAULT_BADCONTEXT	0x010000
 50#define VM_FAULT_BADMAP		0x020000
 51#define VM_FAULT_BADACCESS	0x040000
 52#define VM_FAULT_SIGNAL		0x080000
 53#define VM_FAULT_PFAULT		0x100000
 54
 
 
 
 
 
 
 
 55static unsigned long store_indication __read_mostly;
 56
 57#ifdef CONFIG_64BIT
 58static int __init fault_init(void)
 59{
 60	if (test_facility(75))
 61		store_indication = 0xc00;
 62	return 0;
 63}
 64early_initcall(fault_init);
 65#endif
 66
 67static inline int notify_page_fault(struct pt_regs *regs)
 68{
 69	int ret = 0;
 70
 71	/* kprobe_running() needs smp_processor_id() */
 72	if (kprobes_built_in() && !user_mode(regs)) {
 73		preempt_disable();
 74		if (kprobe_running() && kprobe_fault_handler(regs, 14))
 75			ret = 1;
 76		preempt_enable();
 77	}
 78	return ret;
 79}
 80
 81
 82/*
 83 * Unlock any spinlocks which will prevent us from getting the
 84 * message out.
 85 */
 86void bust_spinlocks(int yes)
 87{
 88	if (yes) {
 89		oops_in_progress = 1;
 90	} else {
 91		int loglevel_save = console_loglevel;
 92		console_unblank();
 93		oops_in_progress = 0;
 94		/*
 95		 * OK, the message is on the console.  Now we call printk()
 96		 * without oops_in_progress set so that printk will give klogd
 97		 * a poke.  Hold onto your hats...
 98		 */
 99		console_loglevel = 15;
100		printk(" ");
101		console_loglevel = loglevel_save;
102	}
103}
104
105/*
106 * Returns the address space associated with the fault.
107 * Returns 0 for kernel space and 1 for user space.
108 */
109static inline int user_space_fault(struct pt_regs *regs)
110{
111	unsigned long trans_exc_code;
112
113	/*
114	 * The lowest two bits of the translation exception
115	 * identification indicate which paging table was used.
116	 */
117	trans_exc_code = regs->int_parm_long & 3;
118	if (trans_exc_code == 3) /* home space -> kernel */
119		return 0;
120	if (user_mode(regs))
121		return 1;
122	if (trans_exc_code == 2) /* secondary space -> set_fs */
123		return current->thread.mm_segment.ar4;
124	if (current->flags & PF_VCPU)
125		return 1;
126	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
127}
128
129static int bad_address(void *p)
130{
131	unsigned long dummy;
132
133	return probe_kernel_address((unsigned long *)p, dummy);
134}
135
136#ifdef CONFIG_64BIT
137static void dump_pagetable(unsigned long asce, unsigned long address)
138{
139	unsigned long *table = __va(asce & PAGE_MASK);
140
141	pr_alert("AS:%016lx ", asce);
142	switch (asce & _ASCE_TYPE_MASK) {
143	case _ASCE_TYPE_REGION1:
144		table = table + ((address >> 53) & 0x7ff);
145		if (bad_address(table))
146			goto bad;
147		pr_cont("R1:%016lx ", *table);
148		if (*table & _REGION_ENTRY_INVALID)
149			goto out;
150		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
151		/* fallthrough */
152	case _ASCE_TYPE_REGION2:
153		table = table + ((address >> 42) & 0x7ff);
154		if (bad_address(table))
155			goto bad;
156		pr_cont("R2:%016lx ", *table);
157		if (*table & _REGION_ENTRY_INVALID)
158			goto out;
159		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
160		/* fallthrough */
161	case _ASCE_TYPE_REGION3:
162		table = table + ((address >> 31) & 0x7ff);
163		if (bad_address(table))
164			goto bad;
165		pr_cont("R3:%016lx ", *table);
166		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
167			goto out;
168		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
169		/* fallthrough */
170	case _ASCE_TYPE_SEGMENT:
171		table = table + ((address >> 20) & 0x7ff);
172		if (bad_address(table))
173			goto bad;
174		pr_cont(KERN_CONT "S:%016lx ", *table);
175		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
176			goto out;
177		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
178	}
179	table = table + ((address >> 12) & 0xff);
180	if (bad_address(table))
181		goto bad;
182	pr_cont("P:%016lx ", *table);
183out:
184	pr_cont("\n");
185	return;
186bad:
187	pr_cont("BAD\n");
188}
189
190#else /* CONFIG_64BIT */
191
192static void dump_pagetable(unsigned long asce, unsigned long address)
193{
194	unsigned long *table = __va(asce & PAGE_MASK);
195
196	pr_alert("AS:%08lx ", asce);
197	table = table + ((address >> 20) & 0x7ff);
198	if (bad_address(table))
199		goto bad;
200	pr_cont("S:%08lx ", *table);
201	if (*table & _SEGMENT_ENTRY_INVALID)
202		goto out;
203	table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
204	table = table + ((address >> 12) & 0xff);
205	if (bad_address(table))
206		goto bad;
207	pr_cont("P:%08lx ", *table);
208out:
209	pr_cont("\n");
210	return;
211bad:
212	pr_cont("BAD\n");
213}
214
215#endif /* CONFIG_64BIT */
216
217static void dump_fault_info(struct pt_regs *regs)
218{
219	unsigned long asce;
220
 
 
221	pr_alert("Fault in ");
222	switch (regs->int_parm_long & 3) {
223	case 3:
224		pr_cont("home space ");
225		break;
226	case 2:
227		pr_cont("secondary space ");
228		break;
229	case 1:
230		pr_cont("access register ");
231		break;
232	case 0:
233		pr_cont("primary space ");
234		break;
235	}
236	pr_cont("mode while using ");
237	if (!user_space_fault(regs)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
238		asce = S390_lowcore.kernel_asce;
239		pr_cont("kernel ");
240	}
241#ifdef CONFIG_PGSTE
242	else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
243		struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
244		asce = gmap->asce;
245		pr_cont("gmap ");
246	}
247#endif
248	else {
249		asce = S390_lowcore.user_asce;
250		pr_cont("user ");
251	}
252	pr_cont("ASCE.\n");
253	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
254}
255
256static inline void report_user_fault(struct pt_regs *regs, long signr)
 
 
257{
258	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
259		return;
260	if (!unhandled_signal(current, signr))
261		return;
262	if (!printk_ratelimit())
263		return;
264	printk(KERN_ALERT "User process fault: interruption code 0x%X ",
265	       regs->int_code);
266	print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
267	printk(KERN_CONT "\n");
268	printk(KERN_ALERT "failing address: %016lx TEID: %016lx\n",
269	       regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
270	dump_fault_info(regs);
271	show_regs(regs);
272}
273
274/*
275 * Send SIGSEGV to task.  This is an external routine
276 * to keep the stack usage of do_page_fault small.
277 */
278static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
279{
280	struct siginfo si;
 
 
 
 
 
 
 
281
282	report_user_fault(regs, SIGSEGV);
283	si.si_signo = SIGSEGV;
284	si.si_code = si_code;
285	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
286	force_sig_info(SIGSEGV, &si, current);
 
287}
288
289static noinline void do_no_context(struct pt_regs *regs)
290{
291	const struct exception_table_entry *fixup;
292	unsigned long address;
293
294	/* Are we prepared to handle this kernel fault?  */
295	fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
296	if (fixup) {
297		regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
298		return;
299	}
300
301	/*
302	 * Oops. The kernel tried to access some bad page. We'll have to
303	 * terminate things with extreme prejudice.
304	 */
305	address = regs->int_parm_long & __FAIL_ADDR_MASK;
306	if (!user_space_fault(regs))
307		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
308		       " in virtual kernel address space\n");
309	else
310		printk(KERN_ALERT "Unable to handle kernel paging request"
311		       " in virtual user address space\n");
312	printk(KERN_ALERT "failing address: %016lx TEID: %016lx\n",
313	       regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
314	dump_fault_info(regs);
315	die(regs, "Oops");
316	do_exit(SIGKILL);
317}
318
319static noinline void do_low_address(struct pt_regs *regs)
320{
321	/* Low-address protection hit in kernel mode means
322	   NULL pointer write access in kernel mode.  */
323	if (regs->psw.mask & PSW_MASK_PSTATE) {
324		/* Low-address protection hit in user mode 'cannot happen'. */
325		die (regs, "Low-address protection");
326		do_exit(SIGKILL);
327	}
328
329	do_no_context(regs);
330}
331
332static noinline void do_sigbus(struct pt_regs *regs)
333{
334	struct task_struct *tsk = current;
335	struct siginfo si;
336
337	/*
338	 * Send a sigbus, regardless of whether we were in kernel
339	 * or user mode.
340	 */
341	si.si_signo = SIGBUS;
342	si.si_errno = 0;
343	si.si_code = BUS_ADRERR;
344	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
345	force_sig_info(SIGBUS, &si, tsk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
346}
347
348static noinline void do_fault_error(struct pt_regs *regs, int fault)
 
349{
350	int si_code;
351
352	switch (fault) {
353	case VM_FAULT_BADACCESS:
 
 
 
354	case VM_FAULT_BADMAP:
355		/* Bad memory access. Check if it is kernel or user space. */
356		if (user_mode(regs)) {
357			/* User mode accesses just cause a SIGSEGV */
358			si_code = (fault == VM_FAULT_BADMAP) ?
359				SEGV_MAPERR : SEGV_ACCERR;
360			do_sigsegv(regs, si_code);
361			return;
362		}
 
363	case VM_FAULT_BADCONTEXT:
 
364	case VM_FAULT_PFAULT:
365		do_no_context(regs);
366		break;
367	case VM_FAULT_SIGNAL:
368		if (!user_mode(regs))
369			do_no_context(regs);
370		break;
371	default: /* fault & VM_FAULT_ERROR */
372		if (fault & VM_FAULT_OOM) {
373			if (!user_mode(regs))
374				do_no_context(regs);
375			else
376				pagefault_out_of_memory();
 
 
 
 
 
 
377		} else if (fault & VM_FAULT_SIGBUS) {
378			/* Kernel mode? Handle exceptions or die */
379			if (!user_mode(regs))
380				do_no_context(regs);
381			else
382				do_sigbus(regs);
383		} else
384			BUG();
385		break;
386	}
387}
388
389/*
390 * This routine handles page faults.  It determines the address,
391 * and the problem, and then passes it off to one of the appropriate
392 * routines.
393 *
394 * interruption code (int_code):
395 *   04       Protection           ->  Write-Protection  (suprression)
396 *   10       Segment translation  ->  Not present       (nullification)
397 *   11       Page translation     ->  Not present       (nullification)
398 *   3b       Region third trans.  ->  Not present       (nullification)
399 */
400static inline int do_exception(struct pt_regs *regs, int access)
401{
402#ifdef CONFIG_PGSTE
403	struct gmap *gmap;
404#endif
405	struct task_struct *tsk;
406	struct mm_struct *mm;
407	struct vm_area_struct *vma;
 
408	unsigned long trans_exc_code;
409	unsigned long address;
410	unsigned int flags;
411	int fault;
412
413	tsk = current;
414	/*
415	 * The instruction that caused the program check has
416	 * been nullified. Don't signal single step via SIGTRAP.
417	 */
418	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
419
420	if (notify_page_fault(regs))
421		return 0;
422
423	mm = tsk->mm;
424	trans_exc_code = regs->int_parm_long;
425
426	/*
427	 * Verify that the fault happened in user space, that
428	 * we are not in an interrupt and that there is a 
429	 * user context.
430	 */
431	fault = VM_FAULT_BADCONTEXT;
432	if (unlikely(!user_space_fault(regs) || in_atomic() || !mm))
 
 
 
 
 
433		goto out;
 
 
 
 
 
 
434
435	address = trans_exc_code & __FAIL_ADDR_MASK;
436	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
437	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
438	if (user_mode(regs))
439		flags |= FAULT_FLAG_USER;
440	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
441		flags |= FAULT_FLAG_WRITE;
442	down_read(&mm->mmap_sem);
443
444#ifdef CONFIG_PGSTE
445	gmap = (struct gmap *)
446		((current->flags & PF_VCPU) ? S390_lowcore.gmap : 0);
447	if (gmap) {
448		address = __gmap_fault(address, gmap);
 
 
449		if (address == -EFAULT) {
450			fault = VM_FAULT_BADMAP;
451			goto out_up;
452		}
453		if (address == -ENOMEM) {
454			fault = VM_FAULT_OOM;
455			goto out_up;
456		}
457		if (gmap->pfault_enabled)
458			flags |= FAULT_FLAG_RETRY_NOWAIT;
459	}
460#endif
461
462retry:
463	fault = VM_FAULT_BADMAP;
464	vma = find_vma(mm, address);
465	if (!vma)
466		goto out_up;
467
468	if (unlikely(vma->vm_start > address)) {
469		if (!(vma->vm_flags & VM_GROWSDOWN))
470			goto out_up;
471		if (expand_stack(vma, address))
472			goto out_up;
473	}
474
475	/*
476	 * Ok, we have a good vm_area for this memory access, so
477	 * we can handle it..
478	 */
479	fault = VM_FAULT_BADACCESS;
480	if (unlikely(!(vma->vm_flags & access)))
481		goto out_up;
482
483	if (is_vm_hugetlb_page(vma))
484		address &= HPAGE_MASK;
485	/*
486	 * If for any reason at all we couldn't handle the fault,
487	 * make sure we exit gracefully rather than endlessly redo
488	 * the fault.
489	 */
490	fault = handle_mm_fault(mm, vma, address, flags);
491	/* No reason to continue if interrupted by SIGKILL. */
492	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
493		fault = VM_FAULT_SIGNAL;
 
 
494		goto out;
495	}
496	if (unlikely(fault & VM_FAULT_ERROR))
497		goto out_up;
498
499	/*
500	 * Major/minor page fault accounting is only done on the
501	 * initial attempt. If we go through a retry, it is extremely
502	 * likely that the page will be found in page cache at that point.
503	 */
504	if (flags & FAULT_FLAG_ALLOW_RETRY) {
505		if (fault & VM_FAULT_MAJOR) {
506			tsk->maj_flt++;
507			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
508				      regs, address);
509		} else {
510			tsk->min_flt++;
511			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
512				      regs, address);
513		}
514		if (fault & VM_FAULT_RETRY) {
515#ifdef CONFIG_PGSTE
516			if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
517				/* FAULT_FLAG_RETRY_NOWAIT has been set,
518				 * mmap_sem has not been released */
519				current->thread.gmap_pfault = 1;
520				fault = VM_FAULT_PFAULT;
521				goto out_up;
522			}
523#endif
524			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
525			 * of starvation. */
526			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
527				   FAULT_FLAG_RETRY_NOWAIT);
528			flags |= FAULT_FLAG_TRIED;
529			down_read(&mm->mmap_sem);
530			goto retry;
531		}
532	}
 
 
 
 
 
 
 
 
 
 
 
 
533	fault = 0;
534out_up:
535	up_read(&mm->mmap_sem);
536out:
537	return fault;
538}
539
540void __kprobes do_protection_exception(struct pt_regs *regs)
541{
542	unsigned long trans_exc_code;
543	int fault;
 
544
545	trans_exc_code = regs->int_parm_long;
546	/*
547	 * Protection exceptions are suppressing, decrement psw address.
548	 * The exception to this rule are aborted transactions, for these
549	 * the PSW already points to the correct location.
550	 */
551	if (!(regs->int_code & 0x200))
552		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
553	/*
554	 * Check for low-address protection.  This needs to be treated
555	 * as a special case because the translation exception code
556	 * field is not guaranteed to contain valid data in this case.
557	 */
558	if (unlikely(!(trans_exc_code & 4))) {
559		do_low_address(regs);
560		return;
561	}
562	fault = do_exception(regs, VM_WRITE);
 
 
 
 
 
 
 
 
563	if (unlikely(fault))
564		do_fault_error(regs, fault);
565}
 
566
567void __kprobes do_dat_exception(struct pt_regs *regs)
568{
569	int access, fault;
 
570
571	access = VM_READ | VM_EXEC | VM_WRITE;
572	fault = do_exception(regs, access);
573	if (unlikely(fault))
574		do_fault_error(regs, fault);
575}
 
576
577#ifdef CONFIG_PFAULT 
578/*
579 * 'pfault' pseudo page faults routines.
580 */
581static int pfault_disable;
582
583static int __init nopfault(char *str)
584{
585	pfault_disable = 1;
586	return 1;
587}
588
589__setup("nopfault", nopfault);
590
591struct pfault_refbk {
592	u16 refdiagc;
593	u16 reffcode;
594	u16 refdwlen;
595	u16 refversn;
596	u64 refgaddr;
597	u64 refselmk;
598	u64 refcmpmk;
599	u64 reserved;
600} __attribute__ ((packed, aligned(8)));
601
 
 
 
 
 
 
 
 
 
 
 
602int pfault_init(void)
603{
604	struct pfault_refbk refbk = {
605		.refdiagc = 0x258,
606		.reffcode = 0,
607		.refdwlen = 5,
608		.refversn = 2,
609		.refgaddr = __LC_CURRENT_PID,
610		.refselmk = 1ULL << 48,
611		.refcmpmk = 1ULL << 48,
612		.reserved = __PF_RES_FIELD };
613        int rc;
614
615	if (pfault_disable)
616		return -1;
 
617	asm volatile(
618		"	diag	%1,%0,0x258\n"
619		"0:	j	2f\n"
620		"1:	la	%0,8\n"
621		"2:\n"
622		EX_TABLE(0b,1b)
623		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 
624        return rc;
625}
626
 
 
 
 
 
 
 
627void pfault_fini(void)
628{
629	struct pfault_refbk refbk = {
630		.refdiagc = 0x258,
631		.reffcode = 1,
632		.refdwlen = 5,
633		.refversn = 2,
634	};
635
636	if (pfault_disable)
637		return;
 
638	asm volatile(
639		"	diag	%0,0,0x258\n"
640		"0:\n"
641		EX_TABLE(0b,0b)
642		: : "a" (&refbk), "m" (refbk) : "cc");
643}
644
645static DEFINE_SPINLOCK(pfault_lock);
646static LIST_HEAD(pfault_list);
647
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
648static void pfault_interrupt(struct ext_code ext_code,
649			     unsigned int param32, unsigned long param64)
650{
651	struct task_struct *tsk;
652	__u16 subcode;
653	pid_t pid;
654
655	/*
656	 * Get the external interruption subcode & pfault
657	 * initial/completion signal bit. VM stores this 
658	 * in the 'cpu address' field associated with the
659         * external interrupt. 
660	 */
661	subcode = ext_code.subcode;
662	if ((subcode & 0xff00) != __SUBCODE_MASK)
663		return;
664	inc_irq_stat(IRQEXT_PFL);
665	/* Get the token (= pid of the affected task). */
666	pid = sizeof(void *) == 4 ? param32 : param64;
667	rcu_read_lock();
668	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
669	if (tsk)
670		get_task_struct(tsk);
671	rcu_read_unlock();
672	if (!tsk)
673		return;
674	spin_lock(&pfault_lock);
675	if (subcode & 0x0080) {
676		/* signal bit is set -> a page has been swapped in by VM */
677		if (tsk->thread.pfault_wait == 1) {
678			/* Initial interrupt was faster than the completion
679			 * interrupt. pfault_wait is valid. Set pfault_wait
680			 * back to zero and wake up the process. This can
681			 * safely be done because the task is still sleeping
682			 * and can't produce new pfaults. */
683			tsk->thread.pfault_wait = 0;
684			list_del(&tsk->thread.list);
685			wake_up_process(tsk);
686			put_task_struct(tsk);
687		} else {
688			/* Completion interrupt was faster than initial
689			 * interrupt. Set pfault_wait to -1 so the initial
690			 * interrupt doesn't put the task to sleep.
691			 * If the task is not running, ignore the completion
692			 * interrupt since it must be a leftover of a PFAULT
693			 * CANCEL operation which didn't remove all pending
694			 * completion interrupts. */
695			if (tsk->state == TASK_RUNNING)
696				tsk->thread.pfault_wait = -1;
697		}
698	} else {
699		/* signal bit not set -> a real page is missing. */
700		if (WARN_ON_ONCE(tsk != current))
701			goto out;
702		if (tsk->thread.pfault_wait == 1) {
703			/* Already on the list with a reference: put to sleep */
704			__set_task_state(tsk, TASK_UNINTERRUPTIBLE);
705			set_tsk_need_resched(tsk);
706		} else if (tsk->thread.pfault_wait == -1) {
707			/* Completion interrupt was faster than the initial
708			 * interrupt (pfault_wait == -1). Set pfault_wait
709			 * back to zero and exit. */
710			tsk->thread.pfault_wait = 0;
711		} else {
712			/* Initial interrupt arrived before completion
713			 * interrupt. Let the task sleep.
714			 * An extra task reference is needed since a different
715			 * cpu may set the task state to TASK_RUNNING again
716			 * before the scheduler is reached. */
717			get_task_struct(tsk);
718			tsk->thread.pfault_wait = 1;
719			list_add(&tsk->thread.list, &pfault_list);
720			__set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 
 
 
 
721			set_tsk_need_resched(tsk);
 
722		}
723	}
724out:
725	spin_unlock(&pfault_lock);
726	put_task_struct(tsk);
727}
728
729static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
730			     void *hcpu)
731{
732	struct thread_struct *thread, *next;
733	struct task_struct *tsk;
734
735	switch (action & ~CPU_TASKS_FROZEN) {
736	case CPU_DEAD:
737		spin_lock_irq(&pfault_lock);
738		list_for_each_entry_safe(thread, next, &pfault_list, list) {
739			thread->pfault_wait = 0;
740			list_del(&thread->list);
741			tsk = container_of(thread, struct task_struct, thread);
742			wake_up_process(tsk);
743			put_task_struct(tsk);
744		}
745		spin_unlock_irq(&pfault_lock);
746		break;
747	default:
748		break;
749	}
750	return NOTIFY_OK;
 
751}
752
753static int __init pfault_irq_init(void)
754{
755	int rc;
756
757	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
758	if (rc)
759		goto out_extint;
760	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
761	if (rc)
762		goto out_pfault;
763	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
764	hotcpu_notifier(pfault_cpu_notify, 0);
 
765	return 0;
766
767out_pfault:
768	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
769out_extint:
770	pfault_disable = 1;
771	return rc;
772}
773early_initcall(pfault_irq_init);
774
775#endif /* CONFIG_PFAULT */