Linux Audio

Check our new training course

Loading...
v3.1
 
  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/kernel_stat.h>
 14#include <linux/perf_event.h>
 15#include <linux/signal.h>
 16#include <linux/sched.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/module.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/system.h>
 36#include <asm/pgtable.h>
 37#include <asm/irq.h>
 38#include <asm/mmu_context.h>
 39#include <asm/compat.h>
 
 40#include "../kernel/entry.h"
 41
 42#ifndef CONFIG_64BIT
 43#define __FAIL_ADDR_MASK 0x7ffff000
 44#define __SUBCODE_MASK 0x0200
 45#define __PF_RES_FIELD 0ULL
 46#else /* CONFIG_64BIT */
 47#define __FAIL_ADDR_MASK -4096L
 48#define __SUBCODE_MASK 0x0600
 49#define __PF_RES_FIELD 0x8000000000000000ULL
 50#endif /* CONFIG_64BIT */
 51
 52#define VM_FAULT_BADCONTEXT	0x010000
 53#define VM_FAULT_BADMAP		0x020000
 54#define VM_FAULT_BADACCESS	0x040000
 
 
 
 
 
 
 
 
 55
 56static unsigned long store_indication;
 57
 58void fault_init(void)
 59{
 60	if (test_facility(2) && test_facility(75))
 61		store_indication = 0xc00;
 
 62}
 63
 64static inline int notify_page_fault(struct pt_regs *regs)
 65{
 66	int ret = 0;
 67
 68	/* kprobe_running() needs smp_processor_id() */
 69	if (kprobes_built_in() && !user_mode(regs)) {
 70		preempt_disable();
 71		if (kprobe_running() && kprobe_fault_handler(regs, 14))
 72			ret = 1;
 73		preempt_enable();
 74	}
 75	return ret;
 76}
 77
 78
 79/*
 80 * Unlock any spinlocks which will prevent us from getting the
 81 * message out.
 82 */
 83void bust_spinlocks(int yes)
 84{
 85	if (yes) {
 86		oops_in_progress = 1;
 87	} else {
 88		int loglevel_save = console_loglevel;
 89		console_unblank();
 90		oops_in_progress = 0;
 91		/*
 92		 * OK, the message is on the console.  Now we call printk()
 93		 * without oops_in_progress set so that printk will give klogd
 94		 * a poke.  Hold onto your hats...
 95		 */
 96		console_loglevel = 15;
 97		printk(" ");
 98		console_loglevel = loglevel_save;
 99	}
100}
101
102/*
103 * Returns the address space associated with the fault.
104 * Returns 0 for kernel space and 1 for user space.
105 */
106static inline int user_space_fault(unsigned long trans_exc_code)
107{
108	/*
109	 * The lowest two bits of the translation exception
110	 * identification indicate which paging table was used.
111	 */
112	trans_exc_code &= 3;
113	if (trans_exc_code == 2)
114		/* Access via secondary space, set_fs setting decides */
115		return current->thread.mm_segment.ar4;
116	if (user_mode == HOME_SPACE_MODE)
117		/* User space if the access has been done via home space. */
118		return trans_exc_code == 3;
119	/*
120	 * If the user space is not the home space the kernel runs in home
121	 * space. Access via secondary space has already been covered,
122	 * access via primary space or access register is from user space
123	 * and access via home space is from the kernel.
124	 */
125	return trans_exc_code != 3;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
126}
127
128static inline void report_user_fault(struct pt_regs *regs, long int_code,
129				     int signr, unsigned long address)
 
130{
131	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
132		return;
133	if (!unhandled_signal(current, signr))
134		return;
135	if (!printk_ratelimit())
136		return;
137	printk("User process fault: interruption code 0x%lX ", int_code);
138	print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
139	printk("\n");
140	printk("failing address: %lX\n", address);
 
 
141	show_regs(regs);
142}
143
144/*
145 * Send SIGSEGV to task.  This is an external routine
146 * to keep the stack usage of do_page_fault small.
147 */
148static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
149				int si_code, unsigned long trans_exc_code)
150{
151	struct siginfo si;
152	unsigned long address;
 
 
153
154	address = trans_exc_code & __FAIL_ADDR_MASK;
155	current->thread.prot_addr = address;
156	current->thread.trap_no = int_code;
157	report_user_fault(regs, int_code, SIGSEGV, address);
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
164static noinline void do_no_context(struct pt_regs *regs, long int_code,
165				   unsigned long trans_exc_code)
166{
167	const struct exception_table_entry *fixup;
168	unsigned long address;
169
170	/* Are we prepared to handle this kernel fault?  */
171	fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
172	if (fixup) {
173		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
174		return;
175	}
176
177	/*
178	 * Oops. The kernel tried to access some bad page. We'll have to
179	 * terminate things with extreme prejudice.
180	 */
181	address = trans_exc_code & __FAIL_ADDR_MASK;
182	if (!user_space_fault(trans_exc_code))
183		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
184		       " at virtual kernel address %p\n", (void *)address);
185	else
186		printk(KERN_ALERT "Unable to handle kernel paging request"
187		       " at virtual user address %p\n", (void *)address);
188
189	die("Oops", regs, int_code);
190	do_exit(SIGKILL);
191}
192
193static noinline void do_low_address(struct pt_regs *regs, long int_code,
194				    unsigned long trans_exc_code)
195{
196	/* Low-address protection hit in kernel mode means
197	   NULL pointer write access in kernel mode.  */
198	if (regs->psw.mask & PSW_MASK_PSTATE) {
199		/* Low-address protection hit in user mode 'cannot happen'. */
200		die ("Low-address protection", regs, int_code);
201		do_exit(SIGKILL);
202	}
203
204	do_no_context(regs, int_code, trans_exc_code);
205}
206
207static noinline void do_sigbus(struct pt_regs *regs, long int_code,
208			       unsigned long trans_exc_code)
209{
210	struct task_struct *tsk = current;
211	unsigned long address;
212	struct siginfo si;
213
214	/*
215	 * Send a sigbus, regardless of whether we were in kernel
216	 * or user mode.
217	 */
218	address = trans_exc_code & __FAIL_ADDR_MASK;
219	tsk->thread.prot_addr = address;
220	tsk->thread.trap_no = int_code;
221	si.si_signo = SIGBUS;
222	si.si_errno = 0;
223	si.si_code = BUS_ADRERR;
224	si.si_addr = (void __user *) address;
225	force_sig_info(SIGBUS, &si, tsk);
226}
227
228static noinline void do_fault_error(struct pt_regs *regs, long int_code,
229				    unsigned long trans_exc_code, int fault)
230{
231	int si_code;
232
233	switch (fault) {
234	case VM_FAULT_BADACCESS:
235	case VM_FAULT_BADMAP:
236		/* Bad memory access. Check if it is kernel or user space. */
237		if (regs->psw.mask & PSW_MASK_PSTATE) {
238			/* User mode accesses just cause a SIGSEGV */
239			si_code = (fault == VM_FAULT_BADMAP) ?
240				SEGV_MAPERR : SEGV_ACCERR;
241			do_sigsegv(regs, int_code, si_code, trans_exc_code);
242			return;
243		}
 
244	case VM_FAULT_BADCONTEXT:
245		do_no_context(regs, int_code, trans_exc_code);
 
 
 
 
 
246		break;
247	default: /* fault & VM_FAULT_ERROR */
248		if (fault & VM_FAULT_OOM) {
249			if (!(regs->psw.mask & PSW_MASK_PSTATE))
250				do_no_context(regs, int_code, trans_exc_code);
251			else
252				pagefault_out_of_memory();
 
 
 
 
 
 
253		} else if (fault & VM_FAULT_SIGBUS) {
254			/* Kernel mode? Handle exceptions or die */
255			if (!(regs->psw.mask & PSW_MASK_PSTATE))
256				do_no_context(regs, int_code, trans_exc_code);
257			else
258				do_sigbus(regs, int_code, trans_exc_code);
259		} else
260			BUG();
261		break;
262	}
263}
264
265/*
266 * This routine handles page faults.  It determines the address,
267 * and the problem, and then passes it off to one of the appropriate
268 * routines.
269 *
270 * interruption code (int_code):
271 *   04       Protection           ->  Write-Protection  (suprression)
272 *   10       Segment translation  ->  Not present       (nullification)
273 *   11       Page translation     ->  Not present       (nullification)
274 *   3b       Region third trans.  ->  Not present       (nullification)
275 */
276static inline int do_exception(struct pt_regs *regs, int access,
277			       unsigned long trans_exc_code)
278{
 
279	struct task_struct *tsk;
280	struct mm_struct *mm;
281	struct vm_area_struct *vma;
 
 
282	unsigned long address;
283	unsigned int flags;
284	int fault;
285
286	if (notify_page_fault(regs))
 
 
 
 
 
 
 
287		return 0;
288
289	tsk = current;
290	mm = tsk->mm;
 
291
292	/*
293	 * Verify that the fault happened in user space, that
294	 * we are not in an interrupt and that there is a 
295	 * user context.
296	 */
297	fault = VM_FAULT_BADCONTEXT;
298	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
 
 
299		goto out;
 
 
 
 
 
 
300
301	address = trans_exc_code & __FAIL_ADDR_MASK;
302	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
303	flags = FAULT_FLAG_ALLOW_RETRY;
 
 
304	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
305		flags |= FAULT_FLAG_WRITE;
306	down_read(&mm->mmap_sem);
307
308#ifdef CONFIG_PGSTE
309	if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
310		address = gmap_fault(address,
311				     (struct gmap *) S390_lowcore.gmap);
 
 
 
312		if (address == -EFAULT) {
313			fault = VM_FAULT_BADMAP;
314			goto out_up;
315		}
316		if (address == -ENOMEM) {
317			fault = VM_FAULT_OOM;
318			goto out_up;
319		}
320	}
321#endif
322
323retry:
324	fault = VM_FAULT_BADMAP;
325	vma = find_vma(mm, address);
326	if (!vma)
327		goto out_up;
328
329	if (unlikely(vma->vm_start > address)) {
330		if (!(vma->vm_flags & VM_GROWSDOWN))
331			goto out_up;
332		if (expand_stack(vma, address))
333			goto out_up;
334	}
335
336	/*
337	 * Ok, we have a good vm_area for this memory access, so
338	 * we can handle it..
339	 */
340	fault = VM_FAULT_BADACCESS;
341	if (unlikely(!(vma->vm_flags & access)))
342		goto out_up;
343
344	if (is_vm_hugetlb_page(vma))
345		address &= HPAGE_MASK;
346	/*
347	 * If for any reason at all we couldn't handle the fault,
348	 * make sure we exit gracefully rather than endlessly redo
349	 * the fault.
350	 */
351	fault = handle_mm_fault(mm, vma, address, flags);
 
 
 
 
 
 
352	if (unlikely(fault & VM_FAULT_ERROR))
353		goto out_up;
354
355	/*
356	 * Major/minor page fault accounting is only done on the
357	 * initial attempt. If we go through a retry, it is extremely
358	 * likely that the page will be found in page cache at that point.
359	 */
360	if (flags & FAULT_FLAG_ALLOW_RETRY) {
361		if (fault & VM_FAULT_MAJOR) {
362			tsk->maj_flt++;
363			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
364				      regs, address);
365		} else {
366			tsk->min_flt++;
367			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
368				      regs, address);
369		}
370		if (fault & VM_FAULT_RETRY) {
371			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
372			 * of starvation. */
373			flags &= ~FAULT_FLAG_ALLOW_RETRY;
374			down_read(&mm->mmap_sem);
 
 
 
 
 
 
 
375			goto retry;
376		}
377	}
378	/*
379	 * The instruction that caused the program check will
380	 * be repeated. Don't signal single step via SIGTRAP.
381	 */
382	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
 
 
 
 
 
 
 
383	fault = 0;
384out_up:
385	up_read(&mm->mmap_sem);
386out:
387	return fault;
388}
389
390void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code,
391				       unsigned long trans_exc_code)
392{
393	int fault;
 
 
394
395	/* Protection exception is suppressing, decrement psw address. */
396	regs->psw.addr -= (pgm_int_code >> 16);
 
 
 
 
 
 
397	/*
398	 * Check for low-address protection.  This needs to be treated
399	 * as a special case because the translation exception code
400	 * field is not guaranteed to contain valid data in this case.
401	 */
402	if (unlikely(!(trans_exc_code & 4))) {
403		do_low_address(regs, pgm_int_code, trans_exc_code);
404		return;
405	}
406	fault = do_exception(regs, VM_WRITE, trans_exc_code);
 
 
 
 
 
 
 
 
407	if (unlikely(fault))
408		do_fault_error(regs, 4, trans_exc_code, fault);
409}
 
410
411void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code,
412				unsigned long trans_exc_code)
413{
414	int access, fault;
 
415
416	access = VM_READ | VM_EXEC | VM_WRITE;
417	fault = do_exception(regs, access, trans_exc_code);
418	if (unlikely(fault))
419		do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault);
420}
421
422#ifdef CONFIG_64BIT
423void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code,
424				 unsigned long trans_exc_code)
425{
426	struct mm_struct *mm = current->mm;
427	struct vm_area_struct *vma;
428
429	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
430		goto no_context;
431
432	down_read(&mm->mmap_sem);
433	vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
434	up_read(&mm->mmap_sem);
435
436	if (vma) {
437		update_mm(mm, current);
438		return;
439	}
440
441	/* User mode accesses just cause a SIGSEGV */
442	if (regs->psw.mask & PSW_MASK_PSTATE) {
443		do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code);
444		return;
445	}
446
447no_context:
448	do_no_context(regs, pgm_int_code, trans_exc_code);
449}
450#endif
451
452int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
453{
454	struct pt_regs regs;
455	int access, fault;
456
457	regs.psw.mask = psw_kernel_bits;
458	if (!irqs_disabled())
459		regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
460	regs.psw.addr = (unsigned long) __builtin_return_address(0);
461	regs.psw.addr |= PSW_ADDR_AMODE;
462	uaddr &= PAGE_MASK;
463	access = write ? VM_WRITE : VM_READ;
464	fault = do_exception(&regs, access, uaddr | 2);
465	if (unlikely(fault)) {
466		if (fault & VM_FAULT_OOM)
467			return -EFAULT;
468		else if (fault & VM_FAULT_SIGBUS)
469			do_sigbus(&regs, pgm_int_code, uaddr);
470	}
471	return fault ? -EFAULT : 0;
472}
 
473
474#ifdef CONFIG_PFAULT 
475/*
476 * 'pfault' pseudo page faults routines.
477 */
478static int pfault_disable;
479
480static int __init nopfault(char *str)
481{
482	pfault_disable = 1;
483	return 1;
484}
485
486__setup("nopfault", nopfault);
487
488struct pfault_refbk {
489	u16 refdiagc;
490	u16 reffcode;
491	u16 refdwlen;
492	u16 refversn;
493	u64 refgaddr;
494	u64 refselmk;
495	u64 refcmpmk;
496	u64 reserved;
497} __attribute__ ((packed, aligned(8)));
498
 
 
 
 
 
 
 
 
 
 
 
499int pfault_init(void)
500{
501	struct pfault_refbk refbk = {
502		.refdiagc = 0x258,
503		.reffcode = 0,
504		.refdwlen = 5,
505		.refversn = 2,
506		.refgaddr = __LC_CURRENT_PID,
507		.refselmk = 1ULL << 48,
508		.refcmpmk = 1ULL << 48,
509		.reserved = __PF_RES_FIELD };
510        int rc;
511
512	if (!MACHINE_IS_VM || pfault_disable)
513		return -1;
 
514	asm volatile(
515		"	diag	%1,%0,0x258\n"
516		"0:	j	2f\n"
517		"1:	la	%0,8\n"
518		"2:\n"
519		EX_TABLE(0b,1b)
520		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
 
521        return rc;
522}
523
 
 
 
 
 
 
 
524void pfault_fini(void)
525{
526	struct pfault_refbk refbk = {
527		.refdiagc = 0x258,
528		.reffcode = 1,
529		.refdwlen = 5,
530		.refversn = 2,
531	};
532
533	if (!MACHINE_IS_VM || pfault_disable)
534		return;
 
535	asm volatile(
536		"	diag	%0,0,0x258\n"
537		"0:\n"
538		EX_TABLE(0b,0b)
539		: : "a" (&refbk), "m" (refbk) : "cc");
540}
541
542static DEFINE_SPINLOCK(pfault_lock);
543static LIST_HEAD(pfault_list);
544
545static void pfault_interrupt(unsigned int ext_int_code,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
546			     unsigned int param32, unsigned long param64)
547{
548	struct task_struct *tsk;
549	__u16 subcode;
550	pid_t pid;
551
552	/*
553	 * Get the external interruption subcode & pfault
554	 * initial/completion signal bit. VM stores this 
555	 * in the 'cpu address' field associated with the
556         * external interrupt. 
557	 */
558	subcode = ext_int_code >> 16;
559	if ((subcode & 0xff00) != __SUBCODE_MASK)
560		return;
561	kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
562	if (subcode & 0x0080) {
563		/* Get the token (= pid of the affected task). */
564		pid = sizeof(void *) == 4 ? param32 : param64;
565		rcu_read_lock();
566		tsk = find_task_by_pid_ns(pid, &init_pid_ns);
567		if (tsk)
568			get_task_struct(tsk);
569		rcu_read_unlock();
570		if (!tsk)
571			return;
572	} else {
573		tsk = current;
574	}
575	spin_lock(&pfault_lock);
576	if (subcode & 0x0080) {
577		/* signal bit is set -> a page has been swapped in by VM */
578		if (tsk->thread.pfault_wait == 1) {
579			/* Initial interrupt was faster than the completion
580			 * interrupt. pfault_wait is valid. Set pfault_wait
581			 * back to zero and wake up the process. This can
582			 * safely be done because the task is still sleeping
583			 * and can't produce new pfaults. */
584			tsk->thread.pfault_wait = 0;
585			list_del(&tsk->thread.list);
586			wake_up_process(tsk);
 
587		} else {
588			/* Completion interrupt was faster than initial
589			 * interrupt. Set pfault_wait to -1 so the initial
590			 * interrupt doesn't put the task to sleep. */
591			tsk->thread.pfault_wait = -1;
 
 
 
 
 
592		}
593		put_task_struct(tsk);
594	} else {
595		/* signal bit not set -> a real page is missing. */
596		if (tsk->thread.pfault_wait == -1) {
 
 
 
 
 
597			/* Completion interrupt was faster than the initial
598			 * interrupt (pfault_wait == -1). Set pfault_wait
599			 * back to zero and exit. */
600			tsk->thread.pfault_wait = 0;
601		} else {
602			/* Initial interrupt arrived before completion
603			 * interrupt. Let the task sleep. */
 
 
 
 
604			tsk->thread.pfault_wait = 1;
605			list_add(&tsk->thread.list, &pfault_list);
606			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 
 
 
 
607			set_tsk_need_resched(tsk);
 
608		}
609	}
 
610	spin_unlock(&pfault_lock);
 
611}
612
613static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
614				       unsigned long action, void *hcpu)
615{
616	struct thread_struct *thread, *next;
617	struct task_struct *tsk;
618
619	switch (action) {
620	case CPU_DEAD:
621	case CPU_DEAD_FROZEN:
622		spin_lock_irq(&pfault_lock);
623		list_for_each_entry_safe(thread, next, &pfault_list, list) {
624			thread->pfault_wait = 0;
625			list_del(&thread->list);
626			tsk = container_of(thread, struct task_struct, thread);
627			wake_up_process(tsk);
628		}
629		spin_unlock_irq(&pfault_lock);
630		break;
631	default:
632		break;
633	}
634	return NOTIFY_OK;
 
635}
636
637static int __init pfault_irq_init(void)
638{
639	int rc;
640
641	if (!MACHINE_IS_VM)
642		return 0;
643	rc = register_external_interrupt(0x2603, pfault_interrupt);
644	if (rc)
645		goto out_extint;
646	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
647	if (rc)
648		goto out_pfault;
649	service_subclass_irq_register();
650	hotcpu_notifier(pfault_cpu_notify, 0);
 
651	return 0;
652
653out_pfault:
654	unregister_external_interrupt(0x2603, pfault_interrupt);
655out_extint:
656	pfault_disable = 1;
657	return rc;
658}
659early_initcall(pfault_irq_init);
660
661#endif /* CONFIG_PFAULT */
v5.14.15
  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/gmap.h>
 37#include <asm/irq.h>
 38#include <asm/mmu_context.h>
 39#include <asm/facility.h>
 40#include <asm/uv.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	((__force vm_fault_t) 0x010000)
 48#define VM_FAULT_BADMAP		((__force vm_fault_t) 0x020000)
 49#define VM_FAULT_BADACCESS	((__force vm_fault_t) 0x040000)
 50#define VM_FAULT_SIGNAL		((__force vm_fault_t) 0x080000)
 51#define VM_FAULT_PFAULT		((__force vm_fault_t) 0x100000)
 52
 53enum fault_type {
 54	KERNEL_FAULT,
 55	USER_FAULT,
 56	GMAP_FAULT,
 57};
 58
 59static unsigned long store_indication __read_mostly;
 60
 61static int __init fault_init(void)
 62{
 63	if (test_facility(75))
 64		store_indication = 0xc00;
 65	return 0;
 66}
 67early_initcall(fault_init);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 68
 69/*
 70 * Find out which address space caused the exception.
 
 71 */
 72static enum fault_type get_fault_type(struct pt_regs *regs)
 73{
 74	unsigned long trans_exc_code;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 75
 76	trans_exc_code = regs->int_parm_long & 3;
 77	if (likely(trans_exc_code == 0)) {
 78		/* primary space exception */
 79		if (user_mode(regs))
 80			return USER_FAULT;
 81		if (!IS_ENABLED(CONFIG_PGSTE))
 82			return KERNEL_FAULT;
 83		if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
 84			return GMAP_FAULT;
 85		return KERNEL_FAULT;
 86	}
 87	if (trans_exc_code == 2)
 88		return USER_FAULT;
 89	if (trans_exc_code == 1) {
 90		/* access register mode, not used in the kernel */
 91		return USER_FAULT;
 92	}
 93	/* home space exception -> access via kernel ASCE */
 94	return KERNEL_FAULT;
 95}
 96
 97static int bad_address(void *p)
 98{
 99	unsigned long dummy;
100
101	return get_kernel_nofault(dummy, (unsigned long *)p);
102}
103
104static void dump_pagetable(unsigned long asce, unsigned long address)
105{
106	unsigned long *table = __va(asce & _ASCE_ORIGIN);
107
108	pr_alert("AS:%016lx ", asce);
109	switch (asce & _ASCE_TYPE_MASK) {
110	case _ASCE_TYPE_REGION1:
111		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
112		if (bad_address(table))
113			goto bad;
114		pr_cont("R1:%016lx ", *table);
115		if (*table & _REGION_ENTRY_INVALID)
116			goto out;
117		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
118		fallthrough;
119	case _ASCE_TYPE_REGION2:
120		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
121		if (bad_address(table))
122			goto bad;
123		pr_cont("R2:%016lx ", *table);
124		if (*table & _REGION_ENTRY_INVALID)
125			goto out;
126		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
127		fallthrough;
128	case _ASCE_TYPE_REGION3:
129		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
130		if (bad_address(table))
131			goto bad;
132		pr_cont("R3:%016lx ", *table);
133		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
134			goto out;
135		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
136		fallthrough;
137	case _ASCE_TYPE_SEGMENT:
138		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
139		if (bad_address(table))
140			goto bad;
141		pr_cont("S:%016lx ", *table);
142		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
143			goto out;
144		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
145	}
146	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
147	if (bad_address(table))
148		goto bad;
149	pr_cont("P:%016lx ", *table);
150out:
151	pr_cont("\n");
152	return;
153bad:
154	pr_cont("BAD\n");
155}
156
157static void dump_fault_info(struct pt_regs *regs)
158{
159	unsigned long asce;
160
161	pr_alert("Failing address: %016lx TEID: %016lx\n",
162		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
163	pr_alert("Fault in ");
164	switch (regs->int_parm_long & 3) {
165	case 3:
166		pr_cont("home space ");
167		break;
168	case 2:
169		pr_cont("secondary space ");
170		break;
171	case 1:
172		pr_cont("access register ");
173		break;
174	case 0:
175		pr_cont("primary space ");
176		break;
177	}
178	pr_cont("mode while using ");
179	switch (get_fault_type(regs)) {
180	case USER_FAULT:
181		asce = S390_lowcore.user_asce;
182		pr_cont("user ");
183		break;
184	case GMAP_FAULT:
185		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
186		pr_cont("gmap ");
187		break;
188	case KERNEL_FAULT:
189		asce = S390_lowcore.kernel_asce;
190		pr_cont("kernel ");
191		break;
192	default:
193		unreachable();
194	}
195	pr_cont("ASCE.\n");
196	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
197}
198
199int show_unhandled_signals = 1;
200
201void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
202{
203	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
204		return;
205	if (!unhandled_signal(current, signr))
206		return;
207	if (!printk_ratelimit())
208		return;
209	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
210	       regs->int_code & 0xffff, regs->int_code >> 17);
211	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
212	printk(KERN_CONT "\n");
213	if (is_mm_fault)
214		dump_fault_info(regs);
215	show_regs(regs);
216}
217
218/*
219 * Send SIGSEGV to task.  This is an external routine
220 * to keep the stack usage of do_page_fault small.
221 */
222static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
 
223{
224	report_user_fault(regs, SIGSEGV, 1);
225	force_sig_fault(SIGSEGV, si_code,
226			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
227}
228
229const struct exception_table_entry *s390_search_extables(unsigned long addr)
230{
231	const struct exception_table_entry *fixup;
232
233	fixup = search_extable(__start_dma_ex_table,
234			       __stop_dma_ex_table - __start_dma_ex_table,
235			       addr);
236	if (!fixup)
237		fixup = search_exception_tables(addr);
238	return fixup;
239}
240
241static noinline void do_no_context(struct pt_regs *regs)
 
242{
243	const struct exception_table_entry *fixup;
 
244
245	/* Are we prepared to handle this kernel fault?  */
246	fixup = s390_search_extables(regs->psw.addr);
247	if (fixup && ex_handle(fixup, regs))
 
248		return;
 
249
250	/*
251	 * Oops. The kernel tried to access some bad page. We'll have to
252	 * terminate things with extreme prejudice.
253	 */
254	if (get_fault_type(regs) == KERNEL_FAULT)
 
255		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
256		       " in virtual kernel address space\n");
257	else
258		printk(KERN_ALERT "Unable to handle kernel paging request"
259		       " in virtual user address space\n");
260	dump_fault_info(regs);
261	die(regs, "Oops");
262	do_exit(SIGKILL);
263}
264
265static noinline void do_low_address(struct pt_regs *regs)
 
266{
267	/* Low-address protection hit in kernel mode means
268	   NULL pointer write access in kernel mode.  */
269	if (regs->psw.mask & PSW_MASK_PSTATE) {
270		/* Low-address protection hit in user mode 'cannot happen'. */
271		die (regs, "Low-address protection");
272		do_exit(SIGKILL);
273	}
274
275	do_no_context(regs);
276}
277
278static noinline void do_sigbus(struct pt_regs *regs)
 
279{
 
 
 
 
280	/*
281	 * Send a sigbus, regardless of whether we were in kernel
282	 * or user mode.
283	 */
284	force_sig_fault(SIGBUS, BUS_ADRERR,
285			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
 
 
 
 
 
 
286}
287
288static noinline void do_fault_error(struct pt_regs *regs, int access,
289					vm_fault_t fault)
290{
291	int si_code;
292
293	switch (fault) {
294	case VM_FAULT_BADACCESS:
295	case VM_FAULT_BADMAP:
296		/* Bad memory access. Check if it is kernel or user space. */
297		if (user_mode(regs)) {
298			/* User mode accesses just cause a SIGSEGV */
299			si_code = (fault == VM_FAULT_BADMAP) ?
300				SEGV_MAPERR : SEGV_ACCERR;
301			do_sigsegv(regs, si_code);
302			break;
303		}
304		fallthrough;
305	case VM_FAULT_BADCONTEXT:
306	case VM_FAULT_PFAULT:
307		do_no_context(regs);
308		break;
309	case VM_FAULT_SIGNAL:
310		if (!user_mode(regs))
311			do_no_context(regs);
312		break;
313	default: /* fault & VM_FAULT_ERROR */
314		if (fault & VM_FAULT_OOM) {
315			if (!user_mode(regs))
316				do_no_context(regs);
317			else
318				pagefault_out_of_memory();
319		} else if (fault & VM_FAULT_SIGSEGV) {
320			/* Kernel mode? Handle exceptions or die */
321			if (!user_mode(regs))
322				do_no_context(regs);
323			else
324				do_sigsegv(regs, SEGV_MAPERR);
325		} else if (fault & VM_FAULT_SIGBUS) {
326			/* Kernel mode? Handle exceptions or die */
327			if (!user_mode(regs))
328				do_no_context(regs);
329			else
330				do_sigbus(regs);
331		} else
332			BUG();
333		break;
334	}
335}
336
337/*
338 * This routine handles page faults.  It determines the address,
339 * and the problem, and then passes it off to one of the appropriate
340 * routines.
341 *
342 * interruption code (int_code):
343 *   04       Protection           ->  Write-Protection  (suppression)
344 *   10       Segment translation  ->  Not present       (nullification)
345 *   11       Page translation     ->  Not present       (nullification)
346 *   3b       Region third trans.  ->  Not present       (nullification)
347 */
348static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
 
349{
350	struct gmap *gmap;
351	struct task_struct *tsk;
352	struct mm_struct *mm;
353	struct vm_area_struct *vma;
354	enum fault_type type;
355	unsigned long trans_exc_code;
356	unsigned long address;
357	unsigned int flags;
358	vm_fault_t fault;
359
360	tsk = current;
361	/*
362	 * The instruction that caused the program check has
363	 * been nullified. Don't signal single step via SIGTRAP.
364	 */
365	clear_thread_flag(TIF_PER_TRAP);
366
367	if (kprobe_page_fault(regs, 14))
368		return 0;
369
 
370	mm = tsk->mm;
371	trans_exc_code = regs->int_parm_long;
372
373	/*
374	 * Verify that the fault happened in user space, that
375	 * we are not in an interrupt and that there is a 
376	 * user context.
377	 */
378	fault = VM_FAULT_BADCONTEXT;
379	type = get_fault_type(regs);
380	switch (type) {
381	case KERNEL_FAULT:
382		goto out;
383	case USER_FAULT:
384	case GMAP_FAULT:
385		if (faulthandler_disabled() || !mm)
386			goto out;
387		break;
388	}
389
390	address = trans_exc_code & __FAIL_ADDR_MASK;
391	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
392	flags = FAULT_FLAG_DEFAULT;
393	if (user_mode(regs))
394		flags |= FAULT_FLAG_USER;
395	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
396		flags |= FAULT_FLAG_WRITE;
397	mmap_read_lock(mm);
398
399	gmap = NULL;
400	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
401		gmap = (struct gmap *) S390_lowcore.gmap;
402		current->thread.gmap_addr = address;
403		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
404		current->thread.gmap_int_code = regs->int_code & 0xffff;
405		address = __gmap_translate(gmap, address);
406		if (address == -EFAULT) {
407			fault = VM_FAULT_BADMAP;
408			goto out_up;
409		}
410		if (gmap->pfault_enabled)
411			flags |= FAULT_FLAG_RETRY_NOWAIT;
 
 
412	}
 
413
414retry:
415	fault = VM_FAULT_BADMAP;
416	vma = find_vma(mm, address);
417	if (!vma)
418		goto out_up;
419
420	if (unlikely(vma->vm_start > address)) {
421		if (!(vma->vm_flags & VM_GROWSDOWN))
422			goto out_up;
423		if (expand_stack(vma, address))
424			goto out_up;
425	}
426
427	/*
428	 * Ok, we have a good vm_area for this memory access, so
429	 * we can handle it..
430	 */
431	fault = VM_FAULT_BADACCESS;
432	if (unlikely(!(vma->vm_flags & access)))
433		goto out_up;
434
435	if (is_vm_hugetlb_page(vma))
436		address &= HPAGE_MASK;
437	/*
438	 * If for any reason at all we couldn't handle the fault,
439	 * make sure we exit gracefully rather than endlessly redo
440	 * the fault.
441	 */
442	fault = handle_mm_fault(vma, address, flags, regs);
443	if (fault_signal_pending(fault, regs)) {
444		fault = VM_FAULT_SIGNAL;
445		if (flags & FAULT_FLAG_RETRY_NOWAIT)
446			goto out_up;
447		goto out;
448	}
449	if (unlikely(fault & VM_FAULT_ERROR))
450		goto out_up;
451
 
 
 
 
 
452	if (flags & FAULT_FLAG_ALLOW_RETRY) {
 
 
 
 
 
 
 
 
 
453		if (fault & VM_FAULT_RETRY) {
454			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
455			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
456				/* FAULT_FLAG_RETRY_NOWAIT has been set,
457				 * mmap_lock has not been released */
458				current->thread.gmap_pfault = 1;
459				fault = VM_FAULT_PFAULT;
460				goto out_up;
461			}
462			flags &= ~FAULT_FLAG_RETRY_NOWAIT;
463			flags |= FAULT_FLAG_TRIED;
464			mmap_read_lock(mm);
465			goto retry;
466		}
467	}
468	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
469		address =  __gmap_link(gmap, current->thread.gmap_addr,
470				       address);
471		if (address == -EFAULT) {
472			fault = VM_FAULT_BADMAP;
473			goto out_up;
474		}
475		if (address == -ENOMEM) {
476			fault = VM_FAULT_OOM;
477			goto out_up;
478		}
479	}
480	fault = 0;
481out_up:
482	mmap_read_unlock(mm);
483out:
484	return fault;
485}
486
487void do_protection_exception(struct pt_regs *regs)
 
488{
489	unsigned long trans_exc_code;
490	int access;
491	vm_fault_t fault;
492
493	trans_exc_code = regs->int_parm_long;
494	/*
495	 * Protection exceptions are suppressing, decrement psw address.
496	 * The exception to this rule are aborted transactions, for these
497	 * the PSW already points to the correct location.
498	 */
499	if (!(regs->int_code & 0x200))
500		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
501	/*
502	 * Check for low-address protection.  This needs to be treated
503	 * as a special case because the translation exception code
504	 * field is not guaranteed to contain valid data in this case.
505	 */
506	if (unlikely(!(trans_exc_code & 4))) {
507		do_low_address(regs);
508		return;
509	}
510	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
511		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
512					(regs->psw.addr & PAGE_MASK);
513		access = VM_EXEC;
514		fault = VM_FAULT_BADACCESS;
515	} else {
516		access = VM_WRITE;
517		fault = do_exception(regs, access);
518	}
519	if (unlikely(fault))
520		do_fault_error(regs, access, fault);
521}
522NOKPROBE_SYMBOL(do_protection_exception);
523
524void do_dat_exception(struct pt_regs *regs)
 
525{
526	int access;
527	vm_fault_t fault;
528
529	access = VM_ACCESS_FLAGS;
530	fault = do_exception(regs, access);
531	if (unlikely(fault))
532		do_fault_error(regs, access, fault);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
533}
534NOKPROBE_SYMBOL(do_dat_exception);
535
536#ifdef CONFIG_PFAULT 
537/*
538 * 'pfault' pseudo page faults routines.
539 */
540static int pfault_disable;
541
542static int __init nopfault(char *str)
543{
544	pfault_disable = 1;
545	return 1;
546}
547
548__setup("nopfault", nopfault);
549
550struct pfault_refbk {
551	u16 refdiagc;
552	u16 reffcode;
553	u16 refdwlen;
554	u16 refversn;
555	u64 refgaddr;
556	u64 refselmk;
557	u64 refcmpmk;
558	u64 reserved;
559} __attribute__ ((packed, aligned(8)));
560
561static struct pfault_refbk pfault_init_refbk = {
562	.refdiagc = 0x258,
563	.reffcode = 0,
564	.refdwlen = 5,
565	.refversn = 2,
566	.refgaddr = __LC_LPP,
567	.refselmk = 1ULL << 48,
568	.refcmpmk = 1ULL << 48,
569	.reserved = __PF_RES_FIELD
570};
571
572int pfault_init(void)
573{
 
 
 
 
 
 
 
 
 
574        int rc;
575
576	if (pfault_disable)
577		return -1;
578	diag_stat_inc(DIAG_STAT_X258);
579	asm volatile(
580		"	diag	%1,%0,0x258\n"
581		"0:	j	2f\n"
582		"1:	la	%0,8\n"
583		"2:\n"
584		EX_TABLE(0b,1b)
585		: "=d" (rc)
586		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
587        return rc;
588}
589
590static struct pfault_refbk pfault_fini_refbk = {
591	.refdiagc = 0x258,
592	.reffcode = 1,
593	.refdwlen = 5,
594	.refversn = 2,
595};
596
597void pfault_fini(void)
598{
 
 
 
 
 
 
599
600	if (pfault_disable)
601		return;
602	diag_stat_inc(DIAG_STAT_X258);
603	asm volatile(
604		"	diag	%0,0,0x258\n"
605		"0:	nopr	%%r7\n"
606		EX_TABLE(0b,0b)
607		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
608}
609
610static DEFINE_SPINLOCK(pfault_lock);
611static LIST_HEAD(pfault_list);
612
613#define PF_COMPLETE	0x0080
614
615/*
616 * The mechanism of our pfault code: if Linux is running as guest, runs a user
617 * space process and the user space process accesses a page that the host has
618 * paged out we get a pfault interrupt.
619 *
620 * This allows us, within the guest, to schedule a different process. Without
621 * this mechanism the host would have to suspend the whole virtual cpu until
622 * the page has been paged in.
623 *
624 * So when we get such an interrupt then we set the state of the current task
625 * to uninterruptible and also set the need_resched flag. Both happens within
626 * interrupt context(!). If we later on want to return to user space we
627 * recognize the need_resched flag and then call schedule().  It's not very
628 * obvious how this works...
629 *
630 * Of course we have a lot of additional fun with the completion interrupt (->
631 * host signals that a page of a process has been paged in and the process can
632 * continue to run). This interrupt can arrive on any cpu and, since we have
633 * virtual cpus, actually appear before the interrupt that signals that a page
634 * is missing.
635 */
636static void pfault_interrupt(struct ext_code ext_code,
637			     unsigned int param32, unsigned long param64)
638{
639	struct task_struct *tsk;
640	__u16 subcode;
641	pid_t pid;
642
643	/*
644	 * Get the external interruption subcode & pfault initial/completion
645	 * signal bit. VM stores this in the 'cpu address' field associated
646	 * with the external interrupt.
 
647	 */
648	subcode = ext_code.subcode;
649	if ((subcode & 0xff00) != __SUBCODE_MASK)
650		return;
651	inc_irq_stat(IRQEXT_PFL);
652	/* Get the token (= pid of the affected task). */
653	pid = param64 & LPP_PID_MASK;
654	rcu_read_lock();
655	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
656	if (tsk)
657		get_task_struct(tsk);
658	rcu_read_unlock();
659	if (!tsk)
660		return;
 
 
 
 
661	spin_lock(&pfault_lock);
662	if (subcode & PF_COMPLETE) {
663		/* signal bit is set -> a page has been swapped in by VM */
664		if (tsk->thread.pfault_wait == 1) {
665			/* Initial interrupt was faster than the completion
666			 * interrupt. pfault_wait is valid. Set pfault_wait
667			 * back to zero and wake up the process. This can
668			 * safely be done because the task is still sleeping
669			 * and can't produce new pfaults. */
670			tsk->thread.pfault_wait = 0;
671			list_del(&tsk->thread.list);
672			wake_up_process(tsk);
673			put_task_struct(tsk);
674		} else {
675			/* Completion interrupt was faster than initial
676			 * interrupt. Set pfault_wait to -1 so the initial
677			 * interrupt doesn't put the task to sleep.
678			 * If the task is not running, ignore the completion
679			 * interrupt since it must be a leftover of a PFAULT
680			 * CANCEL operation which didn't remove all pending
681			 * completion interrupts. */
682			if (task_is_running(tsk))
683				tsk->thread.pfault_wait = -1;
684		}
 
685	} else {
686		/* signal bit not set -> a real page is missing. */
687		if (WARN_ON_ONCE(tsk != current))
688			goto out;
689		if (tsk->thread.pfault_wait == 1) {
690			/* Already on the list with a reference: put to sleep */
691			goto block;
692		} else if (tsk->thread.pfault_wait == -1) {
693			/* Completion interrupt was faster than the initial
694			 * interrupt (pfault_wait == -1). Set pfault_wait
695			 * back to zero and exit. */
696			tsk->thread.pfault_wait = 0;
697		} else {
698			/* Initial interrupt arrived before completion
699			 * interrupt. Let the task sleep.
700			 * An extra task reference is needed since a different
701			 * cpu may set the task state to TASK_RUNNING again
702			 * before the scheduler is reached. */
703			get_task_struct(tsk);
704			tsk->thread.pfault_wait = 1;
705			list_add(&tsk->thread.list, &pfault_list);
706block:
707			/* Since this must be a userspace fault, there
708			 * is no kernel task state to trample. Rely on the
709			 * return to userspace schedule() to block. */
710			__set_current_state(TASK_UNINTERRUPTIBLE);
711			set_tsk_need_resched(tsk);
712			set_preempt_need_resched();
713		}
714	}
715out:
716	spin_unlock(&pfault_lock);
717	put_task_struct(tsk);
718}
719
720static int pfault_cpu_dead(unsigned int cpu)
 
721{
722	struct thread_struct *thread, *next;
723	struct task_struct *tsk;
724
725	spin_lock_irq(&pfault_lock);
726	list_for_each_entry_safe(thread, next, &pfault_list, list) {
727		thread->pfault_wait = 0;
728		list_del(&thread->list);
729		tsk = container_of(thread, struct task_struct, thread);
730		wake_up_process(tsk);
731		put_task_struct(tsk);
 
 
 
 
 
 
 
732	}
733	spin_unlock_irq(&pfault_lock);
734	return 0;
735}
736
737static int __init pfault_irq_init(void)
738{
739	int rc;
740
741	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
 
 
742	if (rc)
743		goto out_extint;
744	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
745	if (rc)
746		goto out_pfault;
747	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
748	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
749				  NULL, pfault_cpu_dead);
750	return 0;
751
752out_pfault:
753	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
754out_extint:
755	pfault_disable = 1;
756	return rc;
757}
758early_initcall(pfault_irq_init);
759
760#endif /* CONFIG_PFAULT */
761
762#if IS_ENABLED(CONFIG_PGSTE)
763
764void do_secure_storage_access(struct pt_regs *regs)
765{
766	unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
767	struct vm_area_struct *vma;
768	struct mm_struct *mm;
769	struct page *page;
770	int rc;
771
772	/*
773	 * bit 61 tells us if the address is valid, if it's not we
774	 * have a major problem and should stop the kernel or send a
775	 * SIGSEGV to the process. Unfortunately bit 61 is not
776	 * reliable without the misc UV feature so we need to check
777	 * for that as well.
778	 */
779	if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
780	    !test_bit_inv(61, &regs->int_parm_long)) {
781		/*
782		 * When this happens, userspace did something that it
783		 * was not supposed to do, e.g. branching into secure
784		 * memory. Trigger a segmentation fault.
785		 */
786		if (user_mode(regs)) {
787			send_sig(SIGSEGV, current, 0);
788			return;
789		}
790
791		/*
792		 * The kernel should never run into this case and we
793		 * have no way out of this situation.
794		 */
795		panic("Unexpected PGM 0x3d with TEID bit 61=0");
796	}
797
798	switch (get_fault_type(regs)) {
799	case USER_FAULT:
800		mm = current->mm;
801		mmap_read_lock(mm);
802		vma = find_vma(mm, addr);
803		if (!vma) {
804			mmap_read_unlock(mm);
805			do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
806			break;
807		}
808		page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
809		if (IS_ERR_OR_NULL(page)) {
810			mmap_read_unlock(mm);
811			break;
812		}
813		if (arch_make_page_accessible(page))
814			send_sig(SIGSEGV, current, 0);
815		put_page(page);
816		mmap_read_unlock(mm);
817		break;
818	case KERNEL_FAULT:
819		page = phys_to_page(addr);
820		if (unlikely(!try_get_page(page)))
821			break;
822		rc = arch_make_page_accessible(page);
823		put_page(page);
824		if (rc)
825			BUG();
826		break;
827	case GMAP_FAULT:
828	default:
829		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
830		WARN_ON_ONCE(1);
831	}
832}
833NOKPROBE_SYMBOL(do_secure_storage_access);
834
835void do_non_secure_storage_access(struct pt_regs *regs)
836{
837	unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
838	struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
839
840	if (get_fault_type(regs) != GMAP_FAULT) {
841		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
842		WARN_ON_ONCE(1);
843		return;
844	}
845
846	if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
847		send_sig(SIGSEGV, current, 0);
848}
849NOKPROBE_SYMBOL(do_non_secure_storage_access);
850
851void do_secure_storage_violation(struct pt_regs *regs)
852{
853	/*
854	 * Either KVM messed up the secure guest mapping or the same
855	 * page is mapped into multiple secure guests.
856	 *
857	 * This exception is only triggered when a guest 2 is running
858	 * and can therefore never occur in kernel context.
859	 */
860	printk_ratelimited(KERN_WARNING
861			   "Secure storage violation in task: %s, pid %d\n",
862			   current->comm, current->pid);
863	send_sig(SIGSEGV, current, 0);
864}
865
866#endif /* CONFIG_PGSTE */