blob: 14654007dce4984ad47275d0d717dd034d9d4702 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * S390 version
Heiko Carstensa53c8fa2012-07-20 11:15:04 +02004 * Copyright IBM Corp. 1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 * 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
Heiko Carstens052ff462011-01-05 12:47:28 +010012#include <linux/kernel_stat.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020013#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/signal.h>
15#include <linux/sched.h>
Ingo Molnarb17b0152017-02-08 18:51:35 +010016#include <linux/sched/debug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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>
Heiko Carstens77575912009-06-12 10:26:25 +020024#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp.h>
Christoph Hellwig1eeb66a2007-05-08 00:27:03 -070026#include <linux/kdebug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/init.h>
28#include <linux/console.h>
Paul Gortmakerdcc096c2016-09-19 17:54:56 -040029#include <linux/extable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/hardirq.h>
Michael Grundy4ba069b2006-09-20 15:58:39 +020031#include <linux/kprobes.h>
Martin Schwidefskybe5ec362007-04-27 16:01:44 +020032#include <linux/uaccess.h>
Gerald Schaefer53492b12008-04-30 13:38:46 +020033#include <linux/hugetlb.h>
Heiko Carstenscbb870c2010-02-26 22:37:43 +010034#include <asm/asm-offsets.h>
Martin Schwidefsky1ec27722015-08-20 17:28:44 +020035#include <asm/diag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/pgtable.h>
Martin Schwidefsky1e133ab2016-03-08 11:49:57 +010037#include <asm/gmap.h>
Heiko Carstensd7b250e2011-05-26 09:48:24 +020038#include <asm/irq.h>
Martin Schwidefsky6252d702008-02-09 18:24:37 +010039#include <asm/mmu_context.h>
David Howellsa0616cd2012-03-28 18:30:02 +010040#include <asm/facility.h>
Heiko Carstensa8061702008-04-17 07:46:26 +020041#include "../kernel/entry.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#define __FAIL_ADDR_MASK -4096L
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#define __SUBCODE_MASK 0x0600
45#define __PF_RES_FIELD 0x8000000000000000ULL
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Martin Schwidefsky50d72802009-12-07 12:51:45 +010047#define VM_FAULT_BADCONTEXT 0x010000
48#define VM_FAULT_BADMAP 0x020000
49#define VM_FAULT_BADACCESS 0x040000
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010050#define VM_FAULT_SIGNAL 0x080000
Dominik Dingel24eb3a82013-06-17 16:25:18 +020051#define VM_FAULT_PFAULT 0x100000
Martin Schwidefsky50d72802009-12-07 12:51:45 +010052
Martin Schwidefsky0aaba412017-08-22 12:08:22 +020053enum fault_type {
54 KERNEL_FAULT,
55 USER_FAULT,
56 VDSO_FAULT,
57 GMAP_FAULT,
58};
59
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010060static unsigned long store_indication __read_mostly;
Martin Schwidefsky92f842e2010-10-25 16:10:13 +020061
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010062static int __init fault_init(void)
Martin Schwidefsky92f842e2010-10-25 16:10:13 +020063{
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010064 if (test_facility(75))
Martin Schwidefsky92f842e2010-10-25 16:10:13 +020065 store_indication = 0xc00;
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010066 return 0;
Martin Schwidefsky92f842e2010-10-25 16:10:13 +020067}
Heiko Carstensa4f32bd2012-10-30 14:49:37 +010068early_initcall(fault_init);
Martin Schwidefsky92f842e2010-10-25 16:10:13 +020069
Martin Schwidefsky7ecb3442009-12-07 12:51:44 +010070static inline int notify_page_fault(struct pt_regs *regs)
Martin Schwidefsky10c10312007-04-27 16:01:43 +020071{
Christoph Hellwig33464e32007-05-04 18:47:46 +020072 int ret = 0;
73
74 /* kprobe_running() needs smp_processor_id() */
Heiko Carstens22e0a042010-02-26 22:37:45 +010075 if (kprobes_built_in() && !user_mode(regs)) {
Christoph Hellwig33464e32007-05-04 18:47:46 +020076 preempt_disable();
77 if (kprobe_running() && kprobe_fault_handler(regs, 14))
78 ret = 1;
79 preempt_enable();
80 }
Christoph Hellwig33464e32007-05-04 18:47:46 +020081 return ret;
Michael Grundy4ba069b2006-09-20 15:58:39 +020082}
Michael Grundy4ba069b2006-09-20 15:58:39 +020083
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85/*
86 * Unlock any spinlocks which will prevent us from getting the
Kirill Korotaevcefc8be2007-02-10 01:46:18 -080087 * message out.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
89void bust_spinlocks(int yes)
90{
91 if (yes) {
92 oops_in_progress = 1;
93 } else {
94 int loglevel_save = console_loglevel;
95 console_unblank();
96 oops_in_progress = 0;
97 /*
98 * OK, the message is on the console. Now we call printk()
99 * without oops_in_progress set so that printk will give klogd
100 * a poke. Hold onto your hats...
101 */
102 console_loglevel = 15;
103 printk(" ");
104 console_loglevel = loglevel_save;
105 }
106}
107
108/*
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200109 * Find out which address space caused the exception.
110 * Access register mode is impossible, ignore space == 3.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200112static inline enum fault_type get_fault_type(struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
Heiko Carstens457f2182014-03-21 10:42:25 +0100114 unsigned long trans_exc_code;
115
Heiko Carstens457f2182014-03-21 10:42:25 +0100116 trans_exc_code = regs->int_parm_long & 3;
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200117 if (likely(trans_exc_code == 0)) {
118 /* primary space exception */
119 if (IS_ENABLED(CONFIG_PGSTE) &&
120 test_pt_regs_flag(regs, PIF_GUEST_FAULT))
121 return GMAP_FAULT;
122 if (current->thread.mm_segment == USER_DS)
123 return USER_FAULT;
124 return KERNEL_FAULT;
125 }
126 if (trans_exc_code == 2) {
127 /* secondary space exception */
128 if (current->thread.mm_segment & 1) {
129 if (current->thread.mm_segment == USER_DS_SACF)
130 return USER_FAULT;
131 return KERNEL_FAULT;
132 }
133 return VDSO_FAULT;
134 }
135 /* home space exception -> access via kernel ASCE */
136 return KERNEL_FAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Heiko Carstens3b7df342014-04-07 10:20:40 +0200139static int bad_address(void *p)
140{
141 unsigned long dummy;
142
143 return probe_kernel_address((unsigned long *)p, dummy);
144}
145
Heiko Carstens3b7df342014-04-07 10:20:40 +0200146static void dump_pagetable(unsigned long asce, unsigned long address)
147{
Heiko Carstensfe7b2742017-05-22 13:16:00 +0200148 unsigned long *table = __va(asce & _ASCE_ORIGIN);
Heiko Carstens3b7df342014-04-07 10:20:40 +0200149
150 pr_alert("AS:%016lx ", asce);
151 switch (asce & _ASCE_TYPE_MASK) {
152 case _ASCE_TYPE_REGION1:
Heiko Carstensf1c11742017-07-05 07:37:27 +0200153 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200154 if (bad_address(table))
155 goto bad;
156 pr_cont("R1:%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_REGION2:
Heiko Carstensf1c11742017-07-05 07:37:27 +0200162 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200163 if (bad_address(table))
164 goto bad;
165 pr_cont("R2:%016lx ", *table);
166 if (*table & _REGION_ENTRY_INVALID)
167 goto out;
168 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
169 /* fallthrough */
170 case _ASCE_TYPE_REGION3:
Heiko Carstensf1c11742017-07-05 07:37:27 +0200171 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200172 if (bad_address(table))
173 goto bad;
174 pr_cont("R3:%016lx ", *table);
175 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
176 goto out;
177 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
178 /* fallthrough */
179 case _ASCE_TYPE_SEGMENT:
Heiko Carstensf1c11742017-07-05 07:37:27 +0200180 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200181 if (bad_address(table))
182 goto bad;
Joe Perches91c08372015-01-05 04:29:18 -0800183 pr_cont("S:%016lx ", *table);
Heiko Carstens3b7df342014-04-07 10:20:40 +0200184 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
185 goto out;
186 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
187 }
Heiko Carstensf1c11742017-07-05 07:37:27 +0200188 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200189 if (bad_address(table))
190 goto bad;
191 pr_cont("P:%016lx ", *table);
192out:
193 pr_cont("\n");
194 return;
195bad:
196 pr_cont("BAD\n");
197}
198
Heiko Carstens3b7df342014-04-07 10:20:40 +0200199static void dump_fault_info(struct pt_regs *regs)
200{
201 unsigned long asce;
202
Heiko Carstens5d7ecce2016-02-24 14:27:46 +0100203 pr_alert("Failing address: %016lx TEID: %016lx\n",
204 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
Heiko Carstens3b7df342014-04-07 10:20:40 +0200205 pr_alert("Fault in ");
206 switch (regs->int_parm_long & 3) {
207 case 3:
208 pr_cont("home space ");
209 break;
210 case 2:
211 pr_cont("secondary space ");
212 break;
213 case 1:
214 pr_cont("access register ");
215 break;
216 case 0:
217 pr_cont("primary space ");
218 break;
219 }
220 pr_cont("mode while using ");
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200221 switch (get_fault_type(regs)) {
222 case USER_FAULT:
Heiko Carstens3b7df342014-04-07 10:20:40 +0200223 asce = S390_lowcore.user_asce;
224 pr_cont("user ");
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200225 break;
226 case VDSO_FAULT:
227 asce = S390_lowcore.vdso_asce;
228 pr_cont("vdso ");
229 break;
230 case GMAP_FAULT:
231 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
232 pr_cont("gmap ");
233 break;
234 case KERNEL_FAULT:
235 asce = S390_lowcore.kernel_asce;
236 pr_cont("kernel ");
237 break;
Heiko Carstens3b7df342014-04-07 10:20:40 +0200238 }
239 pr_cont("ASCE.\n");
240 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
241}
242
Heiko Carstens5d7ecce2016-02-24 14:27:46 +0100243int show_unhandled_signals = 1;
244
245void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
Heiko Carstensab3c68e2010-05-17 10:00:21 +0200246{
247 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
248 return;
249 if (!unhandled_signal(current, signr))
250 return;
251 if (!printk_ratelimit())
252 return;
Hendrik Bruecknerdb1177ee2015-01-29 14:38:38 +0100253 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
Heiko Carstens413d4042014-11-19 13:31:08 +0100254 regs->int_code & 0xffff, regs->int_code >> 17);
Heiko Carstens9cb1cce2016-01-18 13:12:19 +0100255 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100256 printk(KERN_CONT "\n");
Heiko Carstens5d7ecce2016-02-24 14:27:46 +0100257 if (is_mm_fault)
258 dump_fault_info(regs);
Heiko Carstensab3c68e2010-05-17 10:00:21 +0200259 show_regs(regs);
260}
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262/*
263 * Send SIGSEGV to task. This is an external routine
264 * to keep the stack usage of do_page_fault small.
265 */
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100266static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
268 struct siginfo si;
269
Heiko Carstens5d7ecce2016-02-24 14:27:46 +0100270 report_user_fault(regs, SIGSEGV, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 si.si_signo = SIGSEGV;
Michal Hockocf0d44d2016-05-23 15:35:51 +0200272 si.si_errno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 si.si_code = si_code;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100274 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 force_sig_info(SIGSEGV, &si, current);
276}
277
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100278static noinline void do_no_context(struct pt_regs *regs)
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200279{
280 const struct exception_table_entry *fixup;
281
282 /* Are we prepared to handle this kernel fault? */
Heiko Carstens9cb1cce2016-01-18 13:12:19 +0100283 fixup = search_exception_tables(regs->psw.addr);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200284 if (fixup) {
Heiko Carstensfecc868a2016-01-18 12:49:44 +0100285 regs->psw.addr = extable_fixup(fixup);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200286 return;
287 }
288
289 /*
290 * Oops. The kernel tried to access some bad page. We'll have to
291 * terminate things with extreme prejudice.
292 */
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200293 if (get_fault_type(regs) == KERNEL_FAULT)
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200294 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
Heiko Carstens3b7df342014-04-07 10:20:40 +0200295 " in virtual kernel address space\n");
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200296 else
297 printk(KERN_ALERT "Unable to handle kernel paging request"
Heiko Carstens3b7df342014-04-07 10:20:40 +0200298 " in virtual user address space\n");
Heiko Carstens3b7df342014-04-07 10:20:40 +0200299 dump_fault_info(regs);
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100300 die(regs, "Oops");
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200301 do_exit(SIGKILL);
302}
303
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100304static noinline void do_low_address(struct pt_regs *regs)
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200305{
306 /* Low-address protection hit in kernel mode means
307 NULL pointer write access in kernel mode. */
308 if (regs->psw.mask & PSW_MASK_PSTATE) {
309 /* Low-address protection hit in user mode 'cannot happen'. */
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100310 die (regs, "Low-address protection");
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200311 do_exit(SIGKILL);
312 }
313
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100314 do_no_context(regs);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200315}
316
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100317static noinline void do_sigbus(struct pt_regs *regs)
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200318{
319 struct task_struct *tsk = current;
Martin Schwidefsky36bf9682010-10-25 16:10:35 +0200320 struct siginfo si;
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200321
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200322 /*
323 * Send a sigbus, regardless of whether we were in kernel
324 * or user mode.
325 */
Martin Schwidefsky36bf9682010-10-25 16:10:35 +0200326 si.si_signo = SIGBUS;
327 si.si_errno = 0;
328 si.si_code = BUS_ADRERR;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100329 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
Martin Schwidefsky36bf9682010-10-25 16:10:35 +0200330 force_sig_info(SIGBUS, &si, tsk);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200331}
332
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100333static noinline int signal_return(struct pt_regs *regs)
334{
335 u16 instruction;
336 int rc;
337
338 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
339 if (rc)
340 return rc;
341 if (instruction == 0x0a77) {
342 set_pt_regs_flag(regs, PIF_SYSCALL);
343 regs->int_code = 0x00040077;
344 return 0;
345 } else if (instruction == 0x0aad) {
346 set_pt_regs_flag(regs, PIF_SYSCALL);
347 regs->int_code = 0x000400ad;
348 return 0;
349 }
350 return -EACCES;
351}
352
353static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100354{
355 int si_code;
356
357 switch (fault) {
358 case VM_FAULT_BADACCESS:
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100359 if (access == VM_EXEC && signal_return(regs) == 0)
360 break;
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100361 case VM_FAULT_BADMAP:
362 /* Bad memory access. Check if it is kernel or user space. */
Heiko Carstens7d256172012-07-27 10:31:12 +0200363 if (user_mode(regs)) {
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100364 /* User mode accesses just cause a SIGSEGV */
365 si_code = (fault == VM_FAULT_BADMAP) ?
366 SEGV_MAPERR : SEGV_ACCERR;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100367 do_sigsegv(regs, si_code);
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100368 break;
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100369 }
370 case VM_FAULT_BADCONTEXT:
Dominik Dingel24eb3a82013-06-17 16:25:18 +0200371 case VM_FAULT_PFAULT:
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100372 do_no_context(regs);
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100373 break;
Heiko Carstensf2c76e32012-07-27 08:54:20 +0200374 case VM_FAULT_SIGNAL:
375 if (!user_mode(regs))
376 do_no_context(regs);
377 break;
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100378 default: /* fault & VM_FAULT_ERROR */
Heiko Carstens99583182011-05-26 09:48:29 +0200379 if (fault & VM_FAULT_OOM) {
Heiko Carstens7d256172012-07-27 10:31:12 +0200380 if (!user_mode(regs))
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100381 do_no_context(regs);
Heiko Carstens99583182011-05-26 09:48:29 +0200382 else
383 pagefault_out_of_memory();
Linus Torvalds33692f22015-01-29 10:51:32 -0800384 } else if (fault & VM_FAULT_SIGSEGV) {
385 /* Kernel mode? Handle exceptions or die */
386 if (!user_mode(regs))
387 do_no_context(regs);
388 else
389 do_sigsegv(regs, SEGV_MAPERR);
Heiko Carstens99583182011-05-26 09:48:29 +0200390 } else if (fault & VM_FAULT_SIGBUS) {
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100391 /* Kernel mode? Handle exceptions or die */
Heiko Carstens7d256172012-07-27 10:31:12 +0200392 if (!user_mode(regs))
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100393 do_no_context(regs);
Martin Schwidefsky36bf9682010-10-25 16:10:35 +0200394 else
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100395 do_sigbus(regs);
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100396 } else
397 BUG();
398 break;
399 }
400}
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402/*
403 * This routine handles page faults. It determines the address,
404 * and the problem, and then passes it off to one of the appropriate
405 * routines.
406 *
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100407 * interruption code (int_code):
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 * 04 Protection -> Write-Protection (suprression)
409 * 10 Segment translation -> Not present (nullification)
410 * 11 Page translation -> Not present (nullification)
411 * 3b Region third trans. -> Not present (nullification)
412 */
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100413static inline int do_exception(struct pt_regs *regs, int access)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
Dominik Dingel24eb3a82013-06-17 16:25:18 +0200415 struct gmap *gmap;
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200416 struct task_struct *tsk;
417 struct mm_struct *mm;
418 struct vm_area_struct *vma;
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200419 enum fault_type type;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100420 unsigned long trans_exc_code;
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200421 unsigned long address;
Heiko Carstens33ce6142011-05-26 09:48:30 +0200422 unsigned int flags;
423 int fault;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Martin Schwidefsky39efd4e2012-11-21 16:36:27 +0100425 tsk = current;
426 /*
427 * The instruction that caused the program check has
428 * been nullified. Don't signal single step via SIGTRAP.
429 */
Martin Schwidefskyd3a73ac2014-04-15 12:55:07 +0200430 clear_pt_regs_flag(regs, PIF_PER_TRAP);
Martin Schwidefsky39efd4e2012-11-21 16:36:27 +0100431
Martin Schwidefsky7ecb3442009-12-07 12:51:44 +0100432 if (notify_page_fault(regs))
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100433 return 0;
Michael Grundy4ba069b2006-09-20 15:58:39 +0200434
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200435 mm = tsk->mm;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100436 trans_exc_code = regs->int_parm_long;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 /*
439 * Verify that the fault happened in user space, that
440 * we are not in an interrupt and that there is a
441 * user context.
442 */
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100443 fault = VM_FAULT_BADCONTEXT;
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200444 type = get_fault_type(regs);
445 switch (type) {
446 case KERNEL_FAULT:
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100447 goto out;
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200448 case VDSO_FAULT:
449 fault = VM_FAULT_BADMAP;
450 goto out;
451 case USER_FAULT:
452 case GMAP_FAULT:
453 if (faulthandler_disabled() || !mm)
454 goto out;
455 break;
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Martin Schwidefsky61365e12009-12-07 12:51:42 +0100458 address = trans_exc_code & __FAIL_ADDR_MASK;
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +0200459 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
Heiko Carstensf2c76e32012-07-27 08:54:20 +0200460 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
Johannes Weiner759496b2013-09-12 15:13:39 -0700461 if (user_mode(regs))
462 flags |= FAULT_FLAG_USER;
Heiko Carstens33ce6142011-05-26 09:48:30 +0200463 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
464 flags |= FAULT_FLAG_WRITE;
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200465 down_read(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200467 gmap = NULL;
468 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
469 gmap = (struct gmap *) S390_lowcore.gmap;
Martin Schwidefsky527e30b2014-04-30 16:04:25 +0200470 current->thread.gmap_addr = address;
Martin Schwidefsky4be130a2016-03-08 12:12:18 +0100471 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
David Hildenbrand4a494432016-03-08 12:31:52 +0100472 current->thread.gmap_int_code = regs->int_code & 0xffff;
Martin Schwidefsky527e30b2014-04-30 16:04:25 +0200473 address = __gmap_translate(gmap, address);
Martin Schwidefskye5992f22011-07-24 10:48:20 +0200474 if (address == -EFAULT) {
475 fault = VM_FAULT_BADMAP;
476 goto out_up;
477 }
Dominik Dingel24eb3a82013-06-17 16:25:18 +0200478 if (gmap->pfault_enabled)
479 flags |= FAULT_FLAG_RETRY_NOWAIT;
Martin Schwidefskye5992f22011-07-24 10:48:20 +0200480 }
Martin Schwidefskye5992f22011-07-24 10:48:20 +0200481
482retry:
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100483 fault = VM_FAULT_BADMAP;
Gerald Schaefer482b05d2007-03-05 23:35:54 +0100484 vma = find_vma(mm, address);
485 if (!vma)
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100486 goto out_up;
Gerald Schaeferc1821c22007-02-05 21:18:17 +0100487
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100488 if (unlikely(vma->vm_start > address)) {
489 if (!(vma->vm_flags & VM_GROWSDOWN))
490 goto out_up;
491 if (expand_stack(vma, address))
492 goto out_up;
493 }
494
495 /*
496 * Ok, we have a good vm_area for this memory access, so
497 * we can handle it..
498 */
499 fault = VM_FAULT_BADACCESS;
Martin Schwidefsky1ab947d2009-12-07 12:51:46 +0100500 if (unlikely(!(vma->vm_flags & access)))
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100501 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Gerald Schaefer53492b12008-04-30 13:38:46 +0200503 if (is_vm_hugetlb_page(vma))
504 address &= HPAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 /*
506 * If for any reason at all we couldn't handle the fault,
507 * make sure we exit gracefully rather than endlessly redo
508 * the fault.
509 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -0700510 fault = handle_mm_fault(vma, address, flags);
Heiko Carstensf2c76e32012-07-27 08:54:20 +0200511 /* No reason to continue if interrupted by SIGKILL. */
512 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
513 fault = VM_FAULT_SIGNAL;
514 goto out;
515 }
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100516 if (unlikely(fault & VM_FAULT_ERROR))
517 goto out_up;
518
Heiko Carstens33ce6142011-05-26 09:48:30 +0200519 /*
520 * Major/minor page fault accounting is only done on the
521 * initial attempt. If we go through a retry, it is extremely
522 * likely that the page will be found in page cache at that point.
523 */
524 if (flags & FAULT_FLAG_ALLOW_RETRY) {
525 if (fault & VM_FAULT_MAJOR) {
526 tsk->maj_flt++;
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +0200527 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
Heiko Carstens33ce6142011-05-26 09:48:30 +0200528 regs, address);
529 } else {
530 tsk->min_flt++;
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +0200531 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
Heiko Carstens33ce6142011-05-26 09:48:30 +0200532 regs, address);
533 }
534 if (fault & VM_FAULT_RETRY) {
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200535 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
536 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
Dominik Dingel24eb3a82013-06-17 16:25:18 +0200537 /* FAULT_FLAG_RETRY_NOWAIT has been set,
538 * mmap_sem has not been released */
539 current->thread.gmap_pfault = 1;
540 fault = VM_FAULT_PFAULT;
541 goto out_up;
542 }
Heiko Carstens33ce6142011-05-26 09:48:30 +0200543 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
544 * of starvation. */
Dominik Dingel24eb3a82013-06-17 16:25:18 +0200545 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
546 FAULT_FLAG_RETRY_NOWAIT);
Shaohua Li45cac652012-10-08 16:32:19 -0700547 flags |= FAULT_FLAG_TRIED;
Martin Schwidefskye5992f22011-07-24 10:48:20 +0200548 down_read(&mm->mmap_sem);
Heiko Carstens33ce6142011-05-26 09:48:30 +0200549 goto retry;
550 }
Heiko Carstensbde69af2009-09-11 10:29:06 +0200551 }
Martin Schwidefsky0aaba412017-08-22 12:08:22 +0200552 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
Martin Schwidefsky527e30b2014-04-30 16:04:25 +0200553 address = __gmap_link(gmap, current->thread.gmap_addr,
554 address);
555 if (address == -EFAULT) {
556 fault = VM_FAULT_BADMAP;
557 goto out_up;
558 }
559 if (address == -ENOMEM) {
560 fault = VM_FAULT_OOM;
561 goto out_up;
562 }
563 }
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100564 fault = 0;
565out_up:
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200566 up_read(&mm->mmap_sem);
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100567out:
568 return fault;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569}
570
Heiko Carstens7a5388d2014-10-22 12:42:38 +0200571void do_protection_exception(struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100573 unsigned long trans_exc_code;
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100574 int access, fault;
Martin Schwidefsky61365e12009-12-07 12:51:42 +0100575
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100576 trans_exc_code = regs->int_parm_long;
Martin Schwidefskyf752ac42013-04-16 13:25:06 +0200577 /*
578 * Protection exceptions are suppressing, decrement psw address.
579 * The exception to this rule are aborted transactions, for these
580 * the PSW already points to the correct location.
581 */
582 if (!(regs->int_code & 0x200))
583 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200584 /*
585 * Check for low-address protection. This needs to be treated
586 * as a special case because the translation exception code
587 * field is not guaranteed to contain valid data in this case.
588 */
Martin Schwidefsky61365e12009-12-07 12:51:42 +0100589 if (unlikely(!(trans_exc_code & 4))) {
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100590 do_low_address(regs);
Martin Schwidefsky10c10312007-04-27 16:01:43 +0200591 return;
592 }
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100593 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
594 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
595 (regs->psw.addr & PAGE_MASK);
596 access = VM_EXEC;
597 fault = VM_FAULT_BADACCESS;
598 } else {
599 access = VM_WRITE;
600 fault = do_exception(regs, access);
601 }
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100602 if (unlikely(fault))
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100603 do_fault_error(regs, access, fault);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
Heiko Carstens7a5388d2014-10-22 12:42:38 +0200605NOKPROBE_SYMBOL(do_protection_exception);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Heiko Carstens7a5388d2014-10-22 12:42:38 +0200607void do_dat_exception(struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Martin Schwidefsky1ab947d2009-12-07 12:51:46 +0100609 int access, fault;
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100610
Martin Schwidefsky1ab947d2009-12-07 12:51:46 +0100611 access = VM_READ | VM_EXEC | VM_WRITE;
Martin Schwidefskyaa33c8c2011-12-27 11:27:18 +0100612 fault = do_exception(regs, access);
Martin Schwidefsky50d72802009-12-07 12:51:45 +0100613 if (unlikely(fault))
Martin Schwidefsky57d7f932016-03-22 10:54:24 +0100614 do_fault_error(regs, access, fault);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
Heiko Carstens7a5388d2014-10-22 12:42:38 +0200616NOKPROBE_SYMBOL(do_dat_exception);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618#ifdef CONFIG_PFAULT
619/*
620 * 'pfault' pseudo page faults routines.
621 */
Heiko Carstensfb0a9d72011-01-05 12:47:39 +0100622static int pfault_disable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624static int __init nopfault(char *str)
625{
626 pfault_disable = 1;
627 return 1;
628}
629
630__setup("nopfault", nopfault);
631
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200632struct pfault_refbk {
633 u16 refdiagc;
634 u16 reffcode;
635 u16 refdwlen;
636 u16 refversn;
637 u64 refgaddr;
638 u64 refselmk;
639 u64 refcmpmk;
640 u64 reserved;
641} __attribute__ ((packed, aligned(8)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643int pfault_init(void)
644{
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200645 struct pfault_refbk refbk = {
646 .refdiagc = 0x258,
647 .reffcode = 0,
648 .refdwlen = 5,
649 .refversn = 2,
Christian Borntraegere22cf8c2015-10-06 18:06:15 +0200650 .refgaddr = __LC_LPP,
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200651 .refselmk = 1ULL << 48,
652 .refcmpmk = 1ULL << 48,
653 .reserved = __PF_RES_FIELD };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 int rc;
655
Carsten Ottef32269a2011-12-27 11:27:11 +0100656 if (pfault_disable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 return -1;
Martin Schwidefsky1ec27722015-08-20 17:28:44 +0200658 diag_stat_inc(DIAG_STAT_X258);
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200659 asm volatile(
660 " diag %1,%0,0x258\n"
661 "0: j 2f\n"
662 "1: la %0,8\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 "2:\n"
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200664 EX_TABLE(0b,1b)
665 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return rc;
667}
668
669void pfault_fini(void)
670{
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200671 struct pfault_refbk refbk = {
672 .refdiagc = 0x258,
673 .reffcode = 1,
674 .refdwlen = 5,
675 .refversn = 2,
676 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Carsten Ottef32269a2011-12-27 11:27:11 +0100678 if (pfault_disable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 return;
Martin Schwidefsky1ec27722015-08-20 17:28:44 +0200680 diag_stat_inc(DIAG_STAT_X258);
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200681 asm volatile(
682 " diag %0,0,0x258\n"
Heiko Carstens6c22c982016-06-10 09:57:05 +0200683 "0: nopr %%r7\n"
Martin Schwidefsky94c12cc2006-09-28 16:56:43 +0200684 EX_TABLE(0b,0b)
685 : : "a" (&refbk), "m" (refbk) : "cc");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686}
687
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200688static DEFINE_SPINLOCK(pfault_lock);
689static LIST_HEAD(pfault_list);
690
Peter Zijlstra0227f7c2016-03-22 21:42:53 +0100691#define PF_COMPLETE 0x0080
692
693/*
694 * The mechanism of our pfault code: if Linux is running as guest, runs a user
695 * space process and the user space process accesses a page that the host has
696 * paged out we get a pfault interrupt.
697 *
698 * This allows us, within the guest, to schedule a different process. Without
699 * this mechanism the host would have to suspend the whole virtual cpu until
700 * the page has been paged in.
701 *
702 * So when we get such an interrupt then we set the state of the current task
703 * to uninterruptible and also set the need_resched flag. Both happens within
704 * interrupt context(!). If we later on want to return to user space we
705 * recognize the need_resched flag and then call schedule(). It's not very
706 * obvious how this works...
707 *
708 * Of course we have a lot of additional fun with the completion interrupt (->
709 * host signals that a page of a process has been paged in and the process can
710 * continue to run). This interrupt can arrive on any cpu and, since we have
711 * virtual cpus, actually appear before the interrupt that signals that a page
712 * is missing.
713 */
Heiko Carstensfde15c32012-03-11 11:59:31 -0400714static void pfault_interrupt(struct ext_code ext_code,
Martin Schwidefskyf6649a72010-10-25 16:10:38 +0200715 unsigned int param32, unsigned long param64)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 struct task_struct *tsk;
718 __u16 subcode;
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200719 pid_t pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721 /*
Peter Zijlstra0227f7c2016-03-22 21:42:53 +0100722 * Get the external interruption subcode & pfault initial/completion
723 * signal bit. VM stores this in the 'cpu address' field associated
724 * with the external interrupt.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 */
Heiko Carstensfde15c32012-03-11 11:59:31 -0400726 subcode = ext_code.subcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if ((subcode & 0xff00) != __SUBCODE_MASK)
728 return;
Heiko Carstens420f42e2013-01-02 15:18:18 +0100729 inc_irq_stat(IRQEXT_PFL);
Heiko Carstens54c27792012-05-10 09:44:35 +0200730 /* Get the token (= pid of the affected task). */
Christian Borntraegere22cf8c2015-10-06 18:06:15 +0200731 pid = param64 & LPP_PFAULT_PID_MASK;
Heiko Carstens54c27792012-05-10 09:44:35 +0200732 rcu_read_lock();
733 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
734 if (tsk)
735 get_task_struct(tsk);
736 rcu_read_unlock();
737 if (!tsk)
738 return;
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200739 spin_lock(&pfault_lock);
Peter Zijlstra0227f7c2016-03-22 21:42:53 +0100740 if (subcode & PF_COMPLETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 /* signal bit is set -> a page has been swapped in by VM */
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200742 if (tsk->thread.pfault_wait == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 /* Initial interrupt was faster than the completion
744 * interrupt. pfault_wait is valid. Set pfault_wait
745 * back to zero and wake up the process. This can
746 * safely be done because the task is still sleeping
Martin Schwidefskyb6d09442005-09-03 15:58:02 -0700747 * and can't produce new pfaults. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 tsk->thread.pfault_wait = 0;
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200749 list_del(&tsk->thread.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 wake_up_process(tsk);
Heiko Carstensd5e50a52012-05-09 09:37:30 +0200751 put_task_struct(tsk);
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200752 } else {
753 /* Completion interrupt was faster than initial
754 * interrupt. Set pfault_wait to -1 so the initial
Heiko Carstensfa2fb2f2011-11-14 11:19:01 +0100755 * interrupt doesn't put the task to sleep.
756 * If the task is not running, ignore the completion
757 * interrupt since it must be a leftover of a PFAULT
758 * CANCEL operation which didn't remove all pending
759 * completion interrupts. */
760 if (tsk->state == TASK_RUNNING)
761 tsk->thread.pfault_wait = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763 } else {
764 /* signal bit not set -> a real page is missing. */
Heiko Carstensd49f47f2012-05-10 10:47:21 +0200765 if (WARN_ON_ONCE(tsk != current))
766 goto out;
Heiko Carstensd5e50a52012-05-09 09:37:30 +0200767 if (tsk->thread.pfault_wait == 1) {
768 /* Already on the list with a reference: put to sleep */
Peter Zijlstra0227f7c2016-03-22 21:42:53 +0100769 goto block;
Heiko Carstensd5e50a52012-05-09 09:37:30 +0200770 } else if (tsk->thread.pfault_wait == -1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 /* Completion interrupt was faster than the initial
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200772 * interrupt (pfault_wait == -1). Set pfault_wait
773 * back to zero and exit. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 tsk->thread.pfault_wait = 0;
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200775 } else {
776 /* Initial interrupt arrived before completion
Heiko Carstensd5e50a52012-05-09 09:37:30 +0200777 * interrupt. Let the task sleep.
778 * An extra task reference is needed since a different
779 * cpu may set the task state to TASK_RUNNING again
780 * before the scheduler is reached. */
781 get_task_struct(tsk);
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200782 tsk->thread.pfault_wait = 1;
783 list_add(&tsk->thread.list, &pfault_list);
Peter Zijlstra0227f7c2016-03-22 21:42:53 +0100784block:
785 /* Since this must be a userspace fault, there
786 * is no kernel task state to trample. Rely on the
787 * return to userspace schedule() to block. */
788 __set_current_state(TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 set_tsk_need_resched(tsk);
Martin Schwidefskyc3601922016-10-25 12:21:44 +0200790 set_preempt_need_resched();
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 }
Heiko Carstensd49f47f2012-05-10 10:47:21 +0200793out:
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200794 spin_unlock(&pfault_lock);
Heiko Carstens54c27792012-05-10 09:44:35 +0200795 put_task_struct(tsk);
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200796}
797
Sebastian Andrzej Siewior84c9cee2016-09-06 19:04:53 +0200798static int pfault_cpu_dead(unsigned int cpu)
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200799{
800 struct thread_struct *thread, *next;
801 struct task_struct *tsk;
802
Sebastian Andrzej Siewior84c9cee2016-09-06 19:04:53 +0200803 spin_lock_irq(&pfault_lock);
804 list_for_each_entry_safe(thread, next, &pfault_list, list) {
805 thread->pfault_wait = 0;
806 list_del(&thread->list);
807 tsk = container_of(thread, struct task_struct, thread);
808 wake_up_process(tsk);
809 put_task_struct(tsk);
Heiko Carstensf2db2e62011-05-23 10:24:34 +0200810 }
Sebastian Andrzej Siewior84c9cee2016-09-06 19:04:53 +0200811 spin_unlock_irq(&pfault_lock);
812 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Heiko Carstensfb0a9d72011-01-05 12:47:39 +0100815static int __init pfault_irq_init(void)
Heiko Carstens29b08d22006-12-04 15:40:40 +0100816{
Heiko Carstensfb0a9d72011-01-05 12:47:39 +0100817 int rc;
Heiko Carstens29b08d22006-12-04 15:40:40 +0100818
Thomas Huth1dad0932014-03-31 15:24:08 +0200819 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200820 if (rc)
821 goto out_extint;
822 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
823 if (rc)
824 goto out_pfault;
Heiko Carstens82003c32013-09-04 13:35:45 +0200825 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
Sebastian Andrzej Siewior84c9cee2016-09-06 19:04:53 +0200826 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
827 NULL, pfault_cpu_dead);
Heiko Carstensfb0a9d72011-01-05 12:47:39 +0100828 return 0;
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200829
830out_pfault:
Thomas Huth1dad0932014-03-31 15:24:08 +0200831 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200832out_extint:
833 pfault_disable = 1;
834 return rc;
Heiko Carstens29b08d22006-12-04 15:40:40 +0100835}
Heiko Carstensfb0a9d72011-01-05 12:47:39 +0100836early_initcall(pfault_irq_init);
837
Heiko Carstens7dd8fe12011-05-23 10:24:35 +0200838#endif /* CONFIG_PFAULT */