aboutsummaryrefslogtreecommitdiffstats
path: root/arch/s390/mm
diff options
context:
space:
mode:
Diffstat (limited to 'arch/s390/mm')
-rw-r--r--arch/s390/mm/Makefile7
-rw-r--r--arch/s390/mm/cmm.c443
-rw-r--r--arch/s390/mm/extmem.c588
-rw-r--r--arch/s390/mm/fault.c586
-rw-r--r--arch/s390/mm/init.c310
-rw-r--r--arch/s390/mm/ioremap.c138
-rw-r--r--arch/s390/mm/mmap.c86
7 files changed, 2158 insertions, 0 deletions
diff --git a/arch/s390/mm/Makefile b/arch/s390/mm/Makefile
new file mode 100644
index 000000000000..aa9a42b6e62d
--- /dev/null
+++ b/arch/s390/mm/Makefile
@@ -0,0 +1,7 @@
1#
2# Makefile for the linux s390-specific parts of the memory manager.
3#
4
5obj-y := init.o fault.o ioremap.o extmem.o mmap.o
6obj-$(CONFIG_CMM) += cmm.o
7
diff --git a/arch/s390/mm/cmm.c b/arch/s390/mm/cmm.c
new file mode 100644
index 000000000000..d30cdb4248a9
--- /dev/null
+++ b/arch/s390/mm/cmm.c
@@ -0,0 +1,443 @@
1/*
2 * arch/s390/mm/cmm.c
3 *
4 * S390 version
5 * Copyright (C) 2003 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
7 *
8 * Collaborative memory management interface.
9 */
10
11#include <linux/config.h>
12#include <linux/errno.h>
13#include <linux/fs.h>
14#include <linux/init.h>
15#include <linux/module.h>
16#include <linux/sched.h>
17#include <linux/sysctl.h>
18#include <linux/ctype.h>
19
20#include <asm/pgalloc.h>
21#include <asm/uaccess.h>
22
23#include "../../../drivers/s390/net/smsgiucv.h"
24
25#define CMM_NR_PAGES ((PAGE_SIZE / sizeof(unsigned long)) - 2)
26
27struct cmm_page_array {
28 struct cmm_page_array *next;
29 unsigned long index;
30 unsigned long pages[CMM_NR_PAGES];
31};
32
33static long cmm_pages = 0;
34static long cmm_timed_pages = 0;
35static volatile long cmm_pages_target = 0;
36static volatile long cmm_timed_pages_target = 0;
37static long cmm_timeout_pages = 0;
38static long cmm_timeout_seconds = 0;
39
40static struct cmm_page_array *cmm_page_list = 0;
41static struct cmm_page_array *cmm_timed_page_list = 0;
42
43static unsigned long cmm_thread_active = 0;
44static struct work_struct cmm_thread_starter;
45static wait_queue_head_t cmm_thread_wait;
46static struct timer_list cmm_timer;
47
48static void cmm_timer_fn(unsigned long);
49static void cmm_set_timer(void);
50
51static long
52cmm_strtoul(const char *cp, char **endp)
53{
54 unsigned int base = 10;
55
56 if (*cp == '0') {
57 base = 8;
58 cp++;
59 if ((*cp == 'x' || *cp == 'X') && isxdigit(cp[1])) {
60 base = 16;
61 cp++;
62 }
63 }
64 return simple_strtoul(cp, endp, base);
65}
66
67static long
68cmm_alloc_pages(long pages, long *counter, struct cmm_page_array **list)
69{
70 struct cmm_page_array *pa;
71 unsigned long page;
72
73 pa = *list;
74 while (pages) {
75 page = __get_free_page(GFP_NOIO);
76 if (!page)
77 break;
78 if (!pa || pa->index >= CMM_NR_PAGES) {
79 /* Need a new page for the page list. */
80 pa = (struct cmm_page_array *)
81 __get_free_page(GFP_NOIO);
82 if (!pa) {
83 free_page(page);
84 break;
85 }
86 pa->next = *list;
87 pa->index = 0;
88 *list = pa;
89 }
90 diag10(page);
91 pa->pages[pa->index++] = page;
92 (*counter)++;
93 pages--;
94 }
95 return pages;
96}
97
98static void
99cmm_free_pages(long pages, long *counter, struct cmm_page_array **list)
100{
101 struct cmm_page_array *pa;
102 unsigned long page;
103
104 pa = *list;
105 while (pages) {
106 if (!pa || pa->index <= 0)
107 break;
108 page = pa->pages[--pa->index];
109 if (pa->index == 0) {
110 pa = pa->next;
111 free_page((unsigned long) *list);
112 *list = pa;
113 }
114 free_page(page);
115 (*counter)--;
116 pages--;
117 }
118}
119
120static int
121cmm_thread(void *dummy)
122{
123 int rc;
124
125 daemonize("cmmthread");
126 while (1) {
127 rc = wait_event_interruptible(cmm_thread_wait,
128 (cmm_pages != cmm_pages_target ||
129 cmm_timed_pages != cmm_timed_pages_target));
130 if (rc == -ERESTARTSYS) {
131 /* Got kill signal. End thread. */
132 clear_bit(0, &cmm_thread_active);
133 cmm_pages_target = cmm_pages;
134 cmm_timed_pages_target = cmm_timed_pages;
135 break;
136 }
137 if (cmm_pages_target > cmm_pages) {
138 if (cmm_alloc_pages(1, &cmm_pages, &cmm_page_list))
139 cmm_pages_target = cmm_pages;
140 } else if (cmm_pages_target < cmm_pages) {
141 cmm_free_pages(1, &cmm_pages, &cmm_page_list);
142 }
143 if (cmm_timed_pages_target > cmm_timed_pages) {
144 if (cmm_alloc_pages(1, &cmm_timed_pages,
145 &cmm_timed_page_list))
146 cmm_timed_pages_target = cmm_timed_pages;
147 } else if (cmm_timed_pages_target < cmm_timed_pages) {
148 cmm_free_pages(1, &cmm_timed_pages,
149 &cmm_timed_page_list);
150 }
151 if (cmm_timed_pages > 0 && !timer_pending(&cmm_timer))
152 cmm_set_timer();
153 }
154 return 0;
155}
156
157static void
158cmm_start_thread(void)
159{
160 kernel_thread(cmm_thread, 0, 0);
161}
162
163static void
164cmm_kick_thread(void)
165{
166 if (!test_and_set_bit(0, &cmm_thread_active))
167 schedule_work(&cmm_thread_starter);
168 wake_up(&cmm_thread_wait);
169}
170
171static void
172cmm_set_timer(void)
173{
174 if (cmm_timed_pages_target <= 0 || cmm_timeout_seconds <= 0) {
175 if (timer_pending(&cmm_timer))
176 del_timer(&cmm_timer);
177 return;
178 }
179 if (timer_pending(&cmm_timer)) {
180 if (mod_timer(&cmm_timer, jiffies + cmm_timeout_seconds*HZ))
181 return;
182 }
183 cmm_timer.function = cmm_timer_fn;
184 cmm_timer.data = 0;
185 cmm_timer.expires = jiffies + cmm_timeout_seconds*HZ;
186 add_timer(&cmm_timer);
187}
188
189static void
190cmm_timer_fn(unsigned long ignored)
191{
192 long pages;
193
194 pages = cmm_timed_pages_target - cmm_timeout_pages;
195 if (pages < 0)
196 cmm_timed_pages_target = 0;
197 else
198 cmm_timed_pages_target = pages;
199 cmm_kick_thread();
200 cmm_set_timer();
201}
202
203void
204cmm_set_pages(long pages)
205{
206 cmm_pages_target = pages;
207 cmm_kick_thread();
208}
209
210long
211cmm_get_pages(void)
212{
213 return cmm_pages;
214}
215
216void
217cmm_add_timed_pages(long pages)
218{
219 cmm_timed_pages_target += pages;
220 cmm_kick_thread();
221}
222
223long
224cmm_get_timed_pages(void)
225{
226 return cmm_timed_pages;
227}
228
229void
230cmm_set_timeout(long pages, long seconds)
231{
232 cmm_timeout_pages = pages;
233 cmm_timeout_seconds = seconds;
234 cmm_set_timer();
235}
236
237static inline int
238cmm_skip_blanks(char *cp, char **endp)
239{
240 char *str;
241
242 for (str = cp; *str == ' ' || *str == '\t'; str++);
243 *endp = str;
244 return str != cp;
245}
246
247#ifdef CONFIG_CMM_PROC
248/* These will someday get removed. */
249#define VM_CMM_PAGES 1111
250#define VM_CMM_TIMED_PAGES 1112
251#define VM_CMM_TIMEOUT 1113
252
253static struct ctl_table cmm_table[];
254
255static int
256cmm_pages_handler(ctl_table *ctl, int write, struct file *filp,
257 void *buffer, size_t *lenp, loff_t *ppos)
258{
259 char buf[16], *p;
260 long pages;
261 int len;
262
263 if (!*lenp || (*ppos && !write)) {
264 *lenp = 0;
265 return 0;
266 }
267
268 if (write) {
269 len = *lenp;
270 if (copy_from_user(buf, buffer,
271 len > sizeof(buf) ? sizeof(buf) : len))
272 return -EFAULT;
273 buf[sizeof(buf) - 1] = '\0';
274 cmm_skip_blanks(buf, &p);
275 pages = cmm_strtoul(p, &p);
276 if (ctl == &cmm_table[0])
277 cmm_set_pages(pages);
278 else
279 cmm_add_timed_pages(pages);
280 } else {
281 if (ctl == &cmm_table[0])
282 pages = cmm_get_pages();
283 else
284 pages = cmm_get_timed_pages();
285 len = sprintf(buf, "%ld\n", pages);
286 if (len > *lenp)
287 len = *lenp;
288 if (copy_to_user(buffer, buf, len))
289 return -EFAULT;
290 }
291 *lenp = len;
292 *ppos += len;
293 return 0;
294}
295
296static int
297cmm_timeout_handler(ctl_table *ctl, int write, struct file *filp,
298 void *buffer, size_t *lenp, loff_t *ppos)
299{
300 char buf[64], *p;
301 long pages, seconds;
302 int len;
303
304 if (!*lenp || (*ppos && !write)) {
305 *lenp = 0;
306 return 0;
307 }
308
309 if (write) {
310 len = *lenp;
311 if (copy_from_user(buf, buffer,
312 len > sizeof(buf) ? sizeof(buf) : len))
313 return -EFAULT;
314 buf[sizeof(buf) - 1] = '\0';
315 cmm_skip_blanks(buf, &p);
316 pages = cmm_strtoul(p, &p);
317 cmm_skip_blanks(p, &p);
318 seconds = cmm_strtoul(p, &p);
319 cmm_set_timeout(pages, seconds);
320 } else {
321 len = sprintf(buf, "%ld %ld\n",
322 cmm_timeout_pages, cmm_timeout_seconds);
323 if (len > *lenp)
324 len = *lenp;
325 if (copy_to_user(buffer, buf, len))
326 return -EFAULT;
327 }
328 *lenp = len;
329 *ppos += len;
330 return 0;
331}
332
333static struct ctl_table cmm_table[] = {
334 {
335 .ctl_name = VM_CMM_PAGES,
336 .procname = "cmm_pages",
337 .mode = 0600,
338 .proc_handler = &cmm_pages_handler,
339 },
340 {
341 .ctl_name = VM_CMM_TIMED_PAGES,
342 .procname = "cmm_timed_pages",
343 .mode = 0600,
344 .proc_handler = &cmm_pages_handler,
345 },
346 {
347 .ctl_name = VM_CMM_TIMEOUT,
348 .procname = "cmm_timeout",
349 .mode = 0600,
350 .proc_handler = &cmm_timeout_handler,
351 },
352 { .ctl_name = 0 }
353};
354
355static struct ctl_table cmm_dir_table[] = {
356 {
357 .ctl_name = CTL_VM,
358 .procname = "vm",
359 .maxlen = 0,
360 .mode = 0555,
361 .child = cmm_table,
362 },
363 { .ctl_name = 0 }
364};
365#endif
366
367#ifdef CONFIG_CMM_IUCV
368#define SMSG_PREFIX "CMM"
369static void
370cmm_smsg_target(char *msg)
371{
372 long pages, seconds;
373
374 if (!cmm_skip_blanks(msg + strlen(SMSG_PREFIX), &msg))
375 return;
376 if (strncmp(msg, "SHRINK", 6) == 0) {
377 if (!cmm_skip_blanks(msg + 6, &msg))
378 return;
379 pages = cmm_strtoul(msg, &msg);
380 cmm_skip_blanks(msg, &msg);
381 if (*msg == '\0')
382 cmm_set_pages(pages);
383 } else if (strncmp(msg, "RELEASE", 7) == 0) {
384 if (!cmm_skip_blanks(msg + 7, &msg))
385 return;
386 pages = cmm_strtoul(msg, &msg);
387 cmm_skip_blanks(msg, &msg);
388 if (*msg == '\0')
389 cmm_add_timed_pages(pages);
390 } else if (strncmp(msg, "REUSE", 5) == 0) {
391 if (!cmm_skip_blanks(msg + 5, &msg))
392 return;
393 pages = cmm_strtoul(msg, &msg);
394 if (!cmm_skip_blanks(msg, &msg))
395 return;
396 seconds = cmm_strtoul(msg, &msg);
397 cmm_skip_blanks(msg, &msg);
398 if (*msg == '\0')
399 cmm_set_timeout(pages, seconds);
400 }
401}
402#endif
403
404struct ctl_table_header *cmm_sysctl_header;
405
406static int
407cmm_init (void)
408{
409#ifdef CONFIG_CMM_PROC
410 cmm_sysctl_header = register_sysctl_table(cmm_dir_table, 1);
411#endif
412#ifdef CONFIG_CMM_IUCV
413 smsg_register_callback(SMSG_PREFIX, cmm_smsg_target);
414#endif
415 INIT_WORK(&cmm_thread_starter, (void *) cmm_start_thread, 0);
416 init_waitqueue_head(&cmm_thread_wait);
417 init_timer(&cmm_timer);
418 return 0;
419}
420
421static void
422cmm_exit(void)
423{
424 cmm_free_pages(cmm_pages, &cmm_pages, &cmm_page_list);
425 cmm_free_pages(cmm_timed_pages, &cmm_timed_pages, &cmm_timed_page_list);
426#ifdef CONFIG_CMM_PROC
427 unregister_sysctl_table(cmm_sysctl_header);
428#endif
429#ifdef CONFIG_CMM_IUCV
430 smsg_unregister_callback(SMSG_PREFIX, cmm_smsg_target);
431#endif
432}
433
434module_init(cmm_init);
435module_exit(cmm_exit);
436
437EXPORT_SYMBOL(cmm_set_pages);
438EXPORT_SYMBOL(cmm_get_pages);
439EXPORT_SYMBOL(cmm_add_timed_pages);
440EXPORT_SYMBOL(cmm_get_timed_pages);
441EXPORT_SYMBOL(cmm_set_timeout);
442
443MODULE_LICENSE("GPL");
diff --git a/arch/s390/mm/extmem.c b/arch/s390/mm/extmem.c
new file mode 100644
index 000000000000..648deed17e25
--- /dev/null
+++ b/arch/s390/mm/extmem.c
@@ -0,0 +1,588 @@
1/*
2 * File...........: arch/s390/mm/extmem.c
3 * Author(s)......: Carsten Otte <cotte@de.ibm.com>
4 * Rob M van der Heij <rvdheij@nl.ibm.com>
5 * Steven Shultz <shultzss@us.ibm.com>
6 * Bugreports.to..: <Linux390@de.ibm.com>
7 * (C) IBM Corporation 2002-2004
8 */
9
10#include <linux/kernel.h>
11#include <linux/string.h>
12#include <linux/spinlock.h>
13#include <linux/list.h>
14#include <linux/slab.h>
15#include <linux/module.h>
16#include <linux/bootmem.h>
17#include <asm/page.h>
18#include <asm/ebcdic.h>
19#include <asm/errno.h>
20#include <asm/extmem.h>
21#include <asm/cpcmd.h>
22#include <linux/ctype.h>
23
24#define DCSS_DEBUG /* Debug messages on/off */
25
26#define DCSS_NAME "extmem"
27#ifdef DCSS_DEBUG
28#define PRINT_DEBUG(x...) printk(KERN_DEBUG DCSS_NAME " debug:" x)
29#else
30#define PRINT_DEBUG(x...) do {} while (0)
31#endif
32#define PRINT_INFO(x...) printk(KERN_INFO DCSS_NAME " info:" x)
33#define PRINT_WARN(x...) printk(KERN_WARNING DCSS_NAME " warning:" x)
34#define PRINT_ERR(x...) printk(KERN_ERR DCSS_NAME " error:" x)
35
36
37#define DCSS_LOADSHR 0x00
38#define DCSS_LOADNSR 0x04
39#define DCSS_PURGESEG 0x08
40#define DCSS_FINDSEG 0x0c
41#define DCSS_LOADNOLY 0x10
42#define DCSS_SEGEXT 0x18
43#define DCSS_FINDSEGA 0x0c
44
45struct qrange {
46 unsigned int start; // 3byte start address, 1 byte type
47 unsigned int end; // 3byte end address, 1 byte reserved
48};
49
50struct qout64 {
51 int segstart;
52 int segend;
53 int segcnt;
54 int segrcnt;
55 struct qrange range[6];
56};
57
58struct qin64 {
59 char qopcode;
60 char rsrv1[3];
61 char qrcode;
62 char rsrv2[3];
63 char qname[8];
64 unsigned int qoutptr;
65 short int qoutlen;
66};
67
68struct dcss_segment {
69 struct list_head list;
70 char dcss_name[8];
71 unsigned long start_addr;
72 unsigned long end;
73 atomic_t ref_count;
74 int do_nonshared;
75 unsigned int vm_segtype;
76 struct qrange range[6];
77 int segcnt;
78};
79
80static DEFINE_SPINLOCK(dcss_lock);
81static struct list_head dcss_list = LIST_HEAD_INIT(dcss_list);
82static char *segtype_string[] = { "SW", "EW", "SR", "ER", "SN", "EN", "SC",
83 "EW/EN-MIXED" };
84
85extern struct {
86 unsigned long addr, size, type;
87} memory_chunk[MEMORY_CHUNKS];
88
89/*
90 * Create the 8 bytes, ebcdic VM segment name from
91 * an ascii name.
92 */
93static void inline
94dcss_mkname(char *name, char *dcss_name)
95{
96 int i;
97
98 for (i = 0; i < 8; i++) {
99 if (name[i] == '\0')
100 break;
101 dcss_name[i] = toupper(name[i]);
102 };
103 for (; i < 8; i++)
104 dcss_name[i] = ' ';
105 ASCEBC(dcss_name, 8);
106}
107
108
109/*
110 * search all segments in dcss_list, and return the one
111 * namend *name. If not found, return NULL.
112 */
113static struct dcss_segment *
114segment_by_name (char *name)
115{
116 char dcss_name[9];
117 struct list_head *l;
118 struct dcss_segment *tmp, *retval = NULL;
119
120 assert_spin_locked(&dcss_lock);
121 dcss_mkname (name, dcss_name);
122 list_for_each (l, &dcss_list) {
123 tmp = list_entry (l, struct dcss_segment, list);
124 if (memcmp(tmp->dcss_name, dcss_name, 8) == 0) {
125 retval = tmp;
126 break;
127 }
128 }
129 return retval;
130}
131
132
133/*
134 * Perform a function on a dcss segment.
135 */
136static inline int
137dcss_diag (__u8 func, void *parameter,
138 unsigned long *ret1, unsigned long *ret2)
139{
140 unsigned long rx, ry;
141 int rc;
142
143 rx = (unsigned long) parameter;
144 ry = (unsigned long) func;
145 __asm__ __volatile__(
146#ifdef CONFIG_ARCH_S390X
147 " sam31\n" // switch to 31 bit
148 " diag %0,%1,0x64\n"
149 " sam64\n" // switch back to 64 bit
150#else
151 " diag %0,%1,0x64\n"
152#endif
153 " ipm %2\n"
154 " srl %2,28\n"
155 : "+d" (rx), "+d" (ry), "=d" (rc) : : "cc" );
156 *ret1 = rx;
157 *ret2 = ry;
158 return rc;
159}
160
161static inline int
162dcss_diag_translate_rc (int vm_rc) {
163 if (vm_rc == 44)
164 return -ENOENT;
165 return -EIO;
166}
167
168
169/* do a diag to get info about a segment.
170 * fills start_address, end and vm_segtype fields
171 */
172static int
173query_segment_type (struct dcss_segment *seg)
174{
175 struct qin64 *qin = kmalloc (sizeof(struct qin64), GFP_DMA);
176 struct qout64 *qout = kmalloc (sizeof(struct qout64), GFP_DMA);
177
178 int diag_cc, rc, i;
179 unsigned long dummy, vmrc;
180
181 if ((qin == NULL) || (qout == NULL)) {
182 rc = -ENOMEM;
183 goto out_free;
184 }
185
186 /* initialize diag input parameters */
187 qin->qopcode = DCSS_FINDSEGA;
188 qin->qoutptr = (unsigned long) qout;
189 qin->qoutlen = sizeof(struct qout64);
190 memcpy (qin->qname, seg->dcss_name, 8);
191
192 diag_cc = dcss_diag (DCSS_SEGEXT, qin, &dummy, &vmrc);
193
194 if (diag_cc > 1) {
195 rc = dcss_diag_translate_rc (vmrc);
196 goto out_free;
197 }
198
199 if (qout->segcnt > 6) {
200 rc = -ENOTSUPP;
201 goto out_free;
202 }
203
204 if (qout->segcnt == 1) {
205 seg->vm_segtype = qout->range[0].start & 0xff;
206 } else {
207 /* multi-part segment. only one type supported here:
208 - all parts are contiguous
209 - all parts are either EW or EN type
210 - maximum 6 parts allowed */
211 unsigned long start = qout->segstart >> PAGE_SHIFT;
212 for (i=0; i<qout->segcnt; i++) {
213 if (((qout->range[i].start & 0xff) != SEG_TYPE_EW) &&
214 ((qout->range[i].start & 0xff) != SEG_TYPE_EN)) {
215 rc = -ENOTSUPP;
216 goto out_free;
217 }
218 if (start != qout->range[i].start >> PAGE_SHIFT) {
219 rc = -ENOTSUPP;
220 goto out_free;
221 }
222 start = (qout->range[i].end >> PAGE_SHIFT) + 1;
223 }
224 seg->vm_segtype = SEG_TYPE_EWEN;
225 }
226
227 /* analyze diag output and update seg */
228 seg->start_addr = qout->segstart;
229 seg->end = qout->segend;
230
231 memcpy (seg->range, qout->range, 6*sizeof(struct qrange));
232 seg->segcnt = qout->segcnt;
233
234 rc = 0;
235
236 out_free:
237 if (qin) kfree(qin);
238 if (qout) kfree(qout);
239 return rc;
240}
241
242/*
243 * check if the given segment collides with guest storage.
244 * returns 1 if this is the case, 0 if no collision was found
245 */
246static int
247segment_overlaps_storage(struct dcss_segment *seg)
248{
249 int i;
250
251 for (i=0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
252 if (memory_chunk[i].type != 0)
253 continue;
254 if ((memory_chunk[i].addr >> 20) > (seg->end >> 20))
255 continue;
256 if (((memory_chunk[i].addr + memory_chunk[i].size - 1) >> 20)
257 < (seg->start_addr >> 20))
258 continue;
259 return 1;
260 }
261 return 0;
262}
263
264/*
265 * check if segment collides with other segments that are currently loaded
266 * returns 1 if this is the case, 0 if no collision was found
267 */
268static int
269segment_overlaps_others (struct dcss_segment *seg)
270{
271 struct list_head *l;
272 struct dcss_segment *tmp;
273
274 assert_spin_locked(&dcss_lock);
275 list_for_each(l, &dcss_list) {
276 tmp = list_entry(l, struct dcss_segment, list);
277 if ((tmp->start_addr >> 20) > (seg->end >> 20))
278 continue;
279 if ((tmp->end >> 20) < (seg->start_addr >> 20))
280 continue;
281 if (seg == tmp)
282 continue;
283 return 1;
284 }
285 return 0;
286}
287
288/*
289 * check if segment exceeds the kernel mapping range (detected or set via mem=)
290 * returns 1 if this is the case, 0 if segment fits into the range
291 */
292static inline int
293segment_exceeds_range (struct dcss_segment *seg)
294{
295 int seg_last_pfn = (seg->end) >> PAGE_SHIFT;
296 if (seg_last_pfn > max_pfn)
297 return 1;
298 return 0;
299}
300
301/*
302 * get info about a segment
303 * possible return values:
304 * -ENOSYS : we are not running on VM
305 * -EIO : could not perform query diagnose
306 * -ENOENT : no such segment
307 * -ENOTSUPP: multi-part segment cannot be used with linux
308 * -ENOSPC : segment cannot be used (overlaps with storage)
309 * -ENOMEM : out of memory
310 * 0 .. 6 : type of segment as defined in include/asm-s390/extmem.h
311 */
312int
313segment_type (char* name)
314{
315 int rc;
316 struct dcss_segment seg;
317
318 if (!MACHINE_IS_VM)
319 return -ENOSYS;
320
321 dcss_mkname(name, seg.dcss_name);
322 rc = query_segment_type (&seg);
323 if (rc < 0)
324 return rc;
325 return seg.vm_segtype;
326}
327
328/*
329 * real segment loading function, called from segment_load
330 */
331static int
332__segment_load (char *name, int do_nonshared, unsigned long *addr, unsigned long *end)
333{
334 struct dcss_segment *seg = kmalloc(sizeof(struct dcss_segment),
335 GFP_DMA);
336 int dcss_command, rc, diag_cc;
337
338 if (seg == NULL) {
339 rc = -ENOMEM;
340 goto out;
341 }
342 dcss_mkname (name, seg->dcss_name);
343 rc = query_segment_type (seg);
344 if (rc < 0)
345 goto out_free;
346 if (segment_exceeds_range(seg)) {
347 PRINT_WARN ("segment_load: not loading segment %s - exceeds"
348 " kernel mapping range\n",name);
349 rc = -ERANGE;
350 goto out_free;
351 }
352 if (segment_overlaps_storage(seg)) {
353 PRINT_WARN ("segment_load: not loading segment %s - overlaps"
354 " storage\n",name);
355 rc = -ENOSPC;
356 goto out_free;
357 }
358 if (segment_overlaps_others(seg)) {
359 PRINT_WARN ("segment_load: not loading segment %s - overlaps"
360 " other segments\n",name);
361 rc = -EBUSY;
362 goto out_free;
363 }
364 if (do_nonshared)
365 dcss_command = DCSS_LOADNSR;
366 else
367 dcss_command = DCSS_LOADNOLY;
368
369 diag_cc = dcss_diag(dcss_command, seg->dcss_name,
370 &seg->start_addr, &seg->end);
371 if (diag_cc > 1) {
372 PRINT_WARN ("segment_load: could not load segment %s - "
373 "diag returned error (%ld)\n",name,seg->end);
374 rc = dcss_diag_translate_rc (seg->end);
375 dcss_diag(DCSS_PURGESEG, seg->dcss_name,
376 &seg->start_addr, &seg->end);
377 goto out_free;
378 }
379 seg->do_nonshared = do_nonshared;
380 atomic_set(&seg->ref_count, 1);
381 list_add(&seg->list, &dcss_list);
382 rc = seg->vm_segtype;
383 *addr = seg->start_addr;
384 *end = seg->end;
385 if (do_nonshared)
386 PRINT_INFO ("segment_load: loaded segment %s range %p .. %p "
387 "type %s in non-shared mode\n", name,
388 (void*)seg->start_addr, (void*)seg->end,
389 segtype_string[seg->vm_segtype]);
390 else
391 PRINT_INFO ("segment_load: loaded segment %s range %p .. %p "
392 "type %s in shared mode\n", name,
393 (void*)seg->start_addr, (void*)seg->end,
394 segtype_string[seg->vm_segtype]);
395 goto out;
396 out_free:
397 kfree (seg);
398 out:
399 return rc;
400}
401
402/*
403 * this function loads a DCSS segment
404 * name : name of the DCSS
405 * do_nonshared : 0 indicates that the dcss should be shared with other linux images
406 * 1 indicates that the dcss should be exclusive for this linux image
407 * addr : will be filled with start address of the segment
408 * end : will be filled with end address of the segment
409 * return values:
410 * -ENOSYS : we are not running on VM
411 * -EIO : could not perform query or load diagnose
412 * -ENOENT : no such segment
413 * -ENOTSUPP: multi-part segment cannot be used with linux
414 * -ENOSPC : segment cannot be used (overlaps with storage)
415 * -EBUSY : segment can temporarily not be used (overlaps with dcss)
416 * -ERANGE : segment cannot be used (exceeds kernel mapping range)
417 * -EPERM : segment is currently loaded with incompatible permissions
418 * -ENOMEM : out of memory
419 * 0 .. 6 : type of segment as defined in include/asm-s390/extmem.h
420 */
421int
422segment_load (char *name, int do_nonshared, unsigned long *addr,
423 unsigned long *end)
424{
425 struct dcss_segment *seg;
426 int rc;
427
428 if (!MACHINE_IS_VM)
429 return -ENOSYS;
430
431 spin_lock (&dcss_lock);
432 seg = segment_by_name (name);
433 if (seg == NULL)
434 rc = __segment_load (name, do_nonshared, addr, end);
435 else {
436 if (do_nonshared == seg->do_nonshared) {
437 atomic_inc(&seg->ref_count);
438 *addr = seg->start_addr;
439 *end = seg->end;
440 rc = seg->vm_segtype;
441 } else {
442 *addr = *end = 0;
443 rc = -EPERM;
444 }
445 }
446 spin_unlock (&dcss_lock);
447 return rc;
448}
449
450/*
451 * this function modifies the shared state of a DCSS segment. note that
452 * name : name of the DCSS
453 * do_nonshared : 0 indicates that the dcss should be shared with other linux images
454 * 1 indicates that the dcss should be exclusive for this linux image
455 * return values:
456 * -EIO : could not perform load diagnose (segment gone!)
457 * -ENOENT : no such segment (segment gone!)
458 * -EAGAIN : segment is in use by other exploiters, try later
459 * -EINVAL : no segment with the given name is currently loaded - name invalid
460 * 0 : operation succeeded
461 */
462int
463segment_modify_shared (char *name, int do_nonshared)
464{
465 struct dcss_segment *seg;
466 unsigned long dummy;
467 int dcss_command, rc, diag_cc;
468
469 spin_lock (&dcss_lock);
470 seg = segment_by_name (name);
471 if (seg == NULL) {
472 rc = -EINVAL;
473 goto out_unlock;
474 }
475 if (do_nonshared == seg->do_nonshared) {
476 PRINT_INFO ("segment_modify_shared: not reloading segment %s"
477 " - already in requested mode\n",name);
478 rc = 0;
479 goto out_unlock;
480 }
481 if (atomic_read (&seg->ref_count) != 1) {
482 PRINT_WARN ("segment_modify_shared: not reloading segment %s - "
483 "segment is in use by other driver(s)\n",name);
484 rc = -EAGAIN;
485 goto out_unlock;
486 }
487 dcss_diag(DCSS_PURGESEG, seg->dcss_name,
488 &dummy, &dummy);
489 if (do_nonshared)
490 dcss_command = DCSS_LOADNSR;
491 else
492 dcss_command = DCSS_LOADNOLY;
493 diag_cc = dcss_diag(dcss_command, seg->dcss_name,
494 &seg->start_addr, &seg->end);
495 if (diag_cc > 1) {
496 PRINT_WARN ("segment_modify_shared: could not reload segment %s"
497 " - diag returned error (%ld)\n",name,seg->end);
498 rc = dcss_diag_translate_rc (seg->end);
499 goto out_del;
500 }
501 seg->do_nonshared = do_nonshared;
502 rc = 0;
503 goto out_unlock;
504 out_del:
505 list_del(&seg->list);
506 dcss_diag(DCSS_PURGESEG, seg->dcss_name,
507 &dummy, &dummy);
508 kfree (seg);
509 out_unlock:
510 spin_unlock(&dcss_lock);
511 return rc;
512}
513
514/*
515 * Decrease the use count of a DCSS segment and remove
516 * it from the address space if nobody is using it
517 * any longer.
518 */
519void
520segment_unload(char *name)
521{
522 unsigned long dummy;
523 struct dcss_segment *seg;
524
525 if (!MACHINE_IS_VM)
526 return;
527
528 spin_lock(&dcss_lock);
529 seg = segment_by_name (name);
530 if (seg == NULL) {
531 PRINT_ERR ("could not find segment %s in segment_unload, "
532 "please report to linux390@de.ibm.com\n",name);
533 goto out_unlock;
534 }
535 if (atomic_dec_return(&seg->ref_count) == 0) {
536 list_del(&seg->list);
537 dcss_diag(DCSS_PURGESEG, seg->dcss_name,
538 &dummy, &dummy);
539 kfree(seg);
540 }
541out_unlock:
542 spin_unlock(&dcss_lock);
543}
544
545/*
546 * save segment content permanently
547 */
548void
549segment_save(char *name)
550{
551 struct dcss_segment *seg;
552 int startpfn = 0;
553 int endpfn = 0;
554 char cmd1[160];
555 char cmd2[80];
556 int i;
557
558 if (!MACHINE_IS_VM)
559 return;
560
561 spin_lock(&dcss_lock);
562 seg = segment_by_name (name);
563
564 if (seg == NULL) {
565 PRINT_ERR ("could not find segment %s in segment_save, please report to linux390@de.ibm.com\n",name);
566 return;
567 }
568
569 startpfn = seg->start_addr >> PAGE_SHIFT;
570 endpfn = (seg->end) >> PAGE_SHIFT;
571 sprintf(cmd1, "DEFSEG %s", name);
572 for (i=0; i<seg->segcnt; i++) {
573 sprintf(cmd1+strlen(cmd1), " %X-%X %s",
574 seg->range[i].start >> PAGE_SHIFT,
575 seg->range[i].end >> PAGE_SHIFT,
576 segtype_string[seg->range[i].start & 0xff]);
577 }
578 sprintf(cmd2, "SAVESEG %s", name);
579 cpcmd(cmd1, NULL, 0);
580 cpcmd(cmd2, NULL, 0);
581 spin_unlock(&dcss_lock);
582}
583
584EXPORT_SYMBOL(segment_load);
585EXPORT_SYMBOL(segment_unload);
586EXPORT_SYMBOL(segment_save);
587EXPORT_SYMBOL(segment_type);
588EXPORT_SYMBOL(segment_modify_shared);
diff --git a/arch/s390/mm/fault.c b/arch/s390/mm/fault.c
new file mode 100644
index 000000000000..80306bc8c799
--- /dev/null
+++ b/arch/s390/mm/fault.c
@@ -0,0 +1,586 @@
1/*
2 * arch/s390/mm/fault.c
3 *
4 * S390 version
5 * Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Hartmut Penner (hp@de.ibm.com)
7 * Ulrich Weigand (uweigand@de.ibm.com)
8 *
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1995 Linus Torvalds
11 */
12
13#include <linux/config.h>
14#include <linux/signal.h>
15#include <linux/sched.h>
16#include <linux/kernel.h>
17#include <linux/errno.h>
18#include <linux/string.h>
19#include <linux/types.h>
20#include <linux/ptrace.h>
21#include <linux/mman.h>
22#include <linux/mm.h>
23#include <linux/smp.h>
24#include <linux/smp_lock.h>
25#include <linux/init.h>
26#include <linux/console.h>
27#include <linux/module.h>
28#include <linux/hardirq.h>
29
30#include <asm/system.h>
31#include <asm/uaccess.h>
32#include <asm/pgtable.h>
33
34#ifndef CONFIG_ARCH_S390X
35#define __FAIL_ADDR_MASK 0x7ffff000
36#define __FIXUP_MASK 0x7fffffff
37#define __SUBCODE_MASK 0x0200
38#define __PF_RES_FIELD 0ULL
39#else /* CONFIG_ARCH_S390X */
40#define __FAIL_ADDR_MASK -4096L
41#define __FIXUP_MASK ~0L
42#define __SUBCODE_MASK 0x0600
43#define __PF_RES_FIELD 0x8000000000000000ULL
44#endif /* CONFIG_ARCH_S390X */
45
46#ifdef CONFIG_SYSCTL
47extern int sysctl_userprocess_debug;
48#endif
49
50extern void die(const char *,struct pt_regs *,long);
51
52extern spinlock_t timerlist_lock;
53
54/*
55 * Unlock any spinlocks which will prevent us from getting the
56 * message out (timerlist_lock is acquired through the
57 * console unblank code)
58 */
59void bust_spinlocks(int yes)
60{
61 if (yes) {
62 oops_in_progress = 1;
63 } else {
64 int loglevel_save = console_loglevel;
65 console_unblank();
66 oops_in_progress = 0;
67 /*
68 * OK, the message is on the console. Now we call printk()
69 * without oops_in_progress set so that printk will give klogd
70 * a poke. Hold onto your hats...
71 */
72 console_loglevel = 15;
73 printk(" ");
74 console_loglevel = loglevel_save;
75 }
76}
77
78/*
79 * Check which address space is addressed by the access
80 * register in S390_lowcore.exc_access_id.
81 * Returns 1 for user space and 0 for kernel space.
82 */
83static int __check_access_register(struct pt_regs *regs, int error_code)
84{
85 int areg = S390_lowcore.exc_access_id;
86
87 if (areg == 0)
88 /* Access via access register 0 -> kernel address */
89 return 0;
90 save_access_regs(current->thread.acrs);
91 if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
92 /*
93 * access register contains 0 -> kernel address,
94 * access register contains 1 -> user space address
95 */
96 return current->thread.acrs[areg];
97
98 /* Something unhealthy was done with the access registers... */
99 die("page fault via unknown access register", regs, error_code);
100 do_exit(SIGKILL);
101 return 0;
102}
103
104/*
105 * Check which address space the address belongs to.
106 * Returns 1 for user space and 0 for kernel space.
107 */
108static inline int check_user_space(struct pt_regs *regs, int error_code)
109{
110 /*
111 * The lowest two bits of S390_lowcore.trans_exc_code indicate
112 * which paging table was used:
113 * 0: Primary Segment Table Descriptor
114 * 1: STD determined via access register
115 * 2: Secondary Segment Table Descriptor
116 * 3: Home Segment Table Descriptor
117 */
118 int descriptor = S390_lowcore.trans_exc_code & 3;
119 if (unlikely(descriptor == 1))
120 return __check_access_register(regs, error_code);
121 if (descriptor == 2)
122 return current->thread.mm_segment.ar4;
123 return descriptor != 0;
124}
125
126/*
127 * Send SIGSEGV to task. This is an external routine
128 * to keep the stack usage of do_page_fault small.
129 */
130static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
131 int si_code, unsigned long address)
132{
133 struct siginfo si;
134
135#if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
136#if defined(CONFIG_SYSCTL)
137 if (sysctl_userprocess_debug)
138#endif
139 {
140 printk("User process fault: interruption code 0x%lX\n",
141 error_code);
142 printk("failing address: %lX\n", address);
143 show_regs(regs);
144 }
145#endif
146 si.si_signo = SIGSEGV;
147 si.si_code = si_code;
148 si.si_addr = (void *) address;
149 force_sig_info(SIGSEGV, &si, current);
150}
151
152/*
153 * This routine handles page faults. It determines the address,
154 * and the problem, and then passes it off to one of the appropriate
155 * routines.
156 *
157 * error_code:
158 * 04 Protection -> Write-Protection (suprression)
159 * 10 Segment translation -> Not present (nullification)
160 * 11 Page translation -> Not present (nullification)
161 * 3b Region third trans. -> Not present (nullification)
162 */
163extern inline void
164do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
165{
166 struct task_struct *tsk;
167 struct mm_struct *mm;
168 struct vm_area_struct * vma;
169 unsigned long address;
170 int user_address;
171 const struct exception_table_entry *fixup;
172 int si_code = SEGV_MAPERR;
173
174 tsk = current;
175 mm = tsk->mm;
176
177 /*
178 * Check for low-address protection. This needs to be treated
179 * as a special case because the translation exception code
180 * field is not guaranteed to contain valid data in this case.
181 */
182 if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
183
184 /* Low-address protection hit in kernel mode means
185 NULL pointer write access in kernel mode. */
186 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
187 address = 0;
188 user_address = 0;
189 goto no_context;
190 }
191
192 /* Low-address protection hit in user mode 'cannot happen'. */
193 die ("Low-address protection", regs, error_code);
194 do_exit(SIGKILL);
195 }
196
197 /*
198 * get the failing address
199 * more specific the segment and page table portion of
200 * the address
201 */
202 address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
203 user_address = check_user_space(regs, error_code);
204
205 /*
206 * Verify that the fault happened in user space, that
207 * we are not in an interrupt and that there is a
208 * user context.
209 */
210 if (user_address == 0 || in_interrupt() || !mm)
211 goto no_context;
212
213 /*
214 * When we get here, the fault happened in the current
215 * task's user address space, so we can switch on the
216 * interrupts again and then search the VMAs
217 */
218 local_irq_enable();
219
220 down_read(&mm->mmap_sem);
221
222 vma = find_vma(mm, address);
223 if (!vma)
224 goto bad_area;
225 if (vma->vm_start <= address)
226 goto good_area;
227 if (!(vma->vm_flags & VM_GROWSDOWN))
228 goto bad_area;
229 if (expand_stack(vma, address))
230 goto bad_area;
231/*
232 * Ok, we have a good vm_area for this memory access, so
233 * we can handle it..
234 */
235good_area:
236 si_code = SEGV_ACCERR;
237 if (!is_protection) {
238 /* page not present, check vm flags */
239 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
240 goto bad_area;
241 } else {
242 if (!(vma->vm_flags & VM_WRITE))
243 goto bad_area;
244 }
245
246survive:
247 /*
248 * If for any reason at all we couldn't handle the fault,
249 * make sure we exit gracefully rather than endlessly redo
250 * the fault.
251 */
252 switch (handle_mm_fault(mm, vma, address, is_protection)) {
253 case VM_FAULT_MINOR:
254 tsk->min_flt++;
255 break;
256 case VM_FAULT_MAJOR:
257 tsk->maj_flt++;
258 break;
259 case VM_FAULT_SIGBUS:
260 goto do_sigbus;
261 case VM_FAULT_OOM:
262 goto out_of_memory;
263 default:
264 BUG();
265 }
266
267 up_read(&mm->mmap_sem);
268 /*
269 * The instruction that caused the program check will
270 * be repeated. Don't signal single step via SIGTRAP.
271 */
272 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
273 return;
274
275/*
276 * Something tried to access memory that isn't in our memory map..
277 * Fix it, but check if it's kernel or user first..
278 */
279bad_area:
280 up_read(&mm->mmap_sem);
281
282 /* User mode accesses just cause a SIGSEGV */
283 if (regs->psw.mask & PSW_MASK_PSTATE) {
284 tsk->thread.prot_addr = address;
285 tsk->thread.trap_no = error_code;
286 do_sigsegv(regs, error_code, si_code, address);
287 return;
288 }
289
290no_context:
291 /* Are we prepared to handle this kernel fault? */
292 fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
293 if (fixup) {
294 regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
295 return;
296 }
297
298/*
299 * Oops. The kernel tried to access some bad page. We'll have to
300 * terminate things with extreme prejudice.
301 */
302 if (user_address == 0)
303 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
304 " at virtual kernel address %p\n", (void *)address);
305 else
306 printk(KERN_ALERT "Unable to handle kernel paging request"
307 " at virtual user address %p\n", (void *)address);
308
309 die("Oops", regs, error_code);
310 do_exit(SIGKILL);
311
312
313/*
314 * We ran out of memory, or some other thing happened to us that made
315 * us unable to handle the page fault gracefully.
316*/
317out_of_memory:
318 up_read(&mm->mmap_sem);
319 if (tsk->pid == 1) {
320 yield();
321 goto survive;
322 }
323 printk("VM: killing process %s\n", tsk->comm);
324 if (regs->psw.mask & PSW_MASK_PSTATE)
325 do_exit(SIGKILL);
326 goto no_context;
327
328do_sigbus:
329 up_read(&mm->mmap_sem);
330
331 /*
332 * Send a sigbus, regardless of whether we were in kernel
333 * or user mode.
334 */
335 tsk->thread.prot_addr = address;
336 tsk->thread.trap_no = error_code;
337 force_sig(SIGBUS, tsk);
338
339 /* Kernel mode? Handle exceptions or die */
340 if (!(regs->psw.mask & PSW_MASK_PSTATE))
341 goto no_context;
342}
343
344void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
345{
346 regs->psw.addr -= (error_code >> 16);
347 do_exception(regs, 4, 1);
348}
349
350void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
351{
352 do_exception(regs, error_code & 0xff, 0);
353}
354
355#ifndef CONFIG_ARCH_S390X
356
357typedef struct _pseudo_wait_t {
358 struct _pseudo_wait_t *next;
359 wait_queue_head_t queue;
360 unsigned long address;
361 int resolved;
362} pseudo_wait_t;
363
364static pseudo_wait_t *pseudo_lock_queue = NULL;
365static spinlock_t pseudo_wait_spinlock; /* spinlock to protect lock queue */
366
367/*
368 * This routine handles 'pagex' pseudo page faults.
369 */
370asmlinkage void
371do_pseudo_page_fault(struct pt_regs *regs, unsigned long error_code)
372{
373 pseudo_wait_t wait_struct;
374 pseudo_wait_t *ptr, *last, *next;
375 unsigned long address;
376
377 /*
378 * get the failing address
379 * more specific the segment and page table portion of
380 * the address
381 */
382 address = S390_lowcore.trans_exc_code & 0xfffff000;
383
384 if (address & 0x80000000) {
385 /* high bit set -> a page has been swapped in by VM */
386 address &= 0x7fffffff;
387 spin_lock(&pseudo_wait_spinlock);
388 last = NULL;
389 ptr = pseudo_lock_queue;
390 while (ptr != NULL) {
391 next = ptr->next;
392 if (address == ptr->address) {
393 /*
394 * This is one of the processes waiting
395 * for the page. Unchain from the queue.
396 * There can be more than one process
397 * waiting for the same page. VM presents
398 * an initial and a completion interrupt for
399 * every process that tries to access a
400 * page swapped out by VM.
401 */
402 if (last == NULL)
403 pseudo_lock_queue = next;
404 else
405 last->next = next;
406 /* now wake up the process */
407 ptr->resolved = 1;
408 wake_up(&ptr->queue);
409 } else
410 last = ptr;
411 ptr = next;
412 }
413 spin_unlock(&pseudo_wait_spinlock);
414 } else {
415 /* Pseudo page faults in kernel mode is a bad idea */
416 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
417 /*
418 * VM presents pseudo page faults if the interrupted
419 * state was not disabled for interrupts. So we can
420 * get pseudo page fault interrupts while running
421 * in kernel mode. We simply access the page here
422 * while we are running disabled. VM will then swap
423 * in the page synchronously.
424 */
425 if (check_user_space(regs, error_code) == 0)
426 /* dereference a virtual kernel address */
427 __asm__ __volatile__ (
428 " ic 0,0(%0)"
429 : : "a" (address) : "0");
430 else
431 /* dereference a virtual user address */
432 __asm__ __volatile__ (
433 " la 2,0(%0)\n"
434 " sacf 512\n"
435 " ic 2,0(2)\n"
436 "0:sacf 0\n"
437 ".section __ex_table,\"a\"\n"
438 " .align 4\n"
439 " .long 0b,0b\n"
440 ".previous"
441 : : "a" (address) : "2" );
442
443 return;
444 }
445 /* initialize and add element to pseudo_lock_queue */
446 init_waitqueue_head (&wait_struct.queue);
447 wait_struct.address = address;
448 wait_struct.resolved = 0;
449 spin_lock(&pseudo_wait_spinlock);
450 wait_struct.next = pseudo_lock_queue;
451 pseudo_lock_queue = &wait_struct;
452 spin_unlock(&pseudo_wait_spinlock);
453 /*
454 * The instruction that caused the program check will
455 * be repeated. Don't signal single step via SIGTRAP.
456 */
457 clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
458 /* go to sleep */
459 wait_event(wait_struct.queue, wait_struct.resolved);
460 }
461}
462#endif /* CONFIG_ARCH_S390X */
463
464#ifdef CONFIG_PFAULT
465/*
466 * 'pfault' pseudo page faults routines.
467 */
468static int pfault_disable = 0;
469
470static int __init nopfault(char *str)
471{
472 pfault_disable = 1;
473 return 1;
474}
475
476__setup("nopfault", nopfault);
477
478typedef struct {
479 __u16 refdiagc;
480 __u16 reffcode;
481 __u16 refdwlen;
482 __u16 refversn;
483 __u64 refgaddr;
484 __u64 refselmk;
485 __u64 refcmpmk;
486 __u64 reserved;
487} __attribute__ ((packed)) pfault_refbk_t;
488
489int pfault_init(void)
490{
491 pfault_refbk_t refbk =
492 { 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
493 __PF_RES_FIELD };
494 int rc;
495
496 if (pfault_disable)
497 return -1;
498 __asm__ __volatile__(
499 " diag %1,%0,0x258\n"
500 "0: j 2f\n"
501 "1: la %0,8\n"
502 "2:\n"
503 ".section __ex_table,\"a\"\n"
504 " .align 4\n"
505#ifndef CONFIG_ARCH_S390X
506 " .long 0b,1b\n"
507#else /* CONFIG_ARCH_S390X */
508 " .quad 0b,1b\n"
509#endif /* CONFIG_ARCH_S390X */
510 ".previous"
511 : "=d" (rc) : "a" (&refbk) : "cc" );
512 __ctl_set_bit(0, 9);
513 return rc;
514}
515
516void pfault_fini(void)
517{
518 pfault_refbk_t refbk =
519 { 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
520
521 if (pfault_disable)
522 return;
523 __ctl_clear_bit(0,9);
524 __asm__ __volatile__(
525 " diag %0,0,0x258\n"
526 "0:\n"
527 ".section __ex_table,\"a\"\n"
528 " .align 4\n"
529#ifndef CONFIG_ARCH_S390X
530 " .long 0b,0b\n"
531#else /* CONFIG_ARCH_S390X */
532 " .quad 0b,0b\n"
533#endif /* CONFIG_ARCH_S390X */
534 ".previous"
535 : : "a" (&refbk) : "cc" );
536}
537
538asmlinkage void
539pfault_interrupt(struct pt_regs *regs, __u16 error_code)
540{
541 struct task_struct *tsk;
542 __u16 subcode;
543
544 /*
545 * Get the external interruption subcode & pfault
546 * initial/completion signal bit. VM stores this
547 * in the 'cpu address' field associated with the
548 * external interrupt.
549 */
550 subcode = S390_lowcore.cpu_addr;
551 if ((subcode & 0xff00) != __SUBCODE_MASK)
552 return;
553
554 /*
555 * Get the token (= address of the task structure of the affected task).
556 */
557 tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
558
559 if (subcode & 0x0080) {
560 /* signal bit is set -> a page has been swapped in by VM */
561 if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
562 /* Initial interrupt was faster than the completion
563 * interrupt. pfault_wait is valid. Set pfault_wait
564 * back to zero and wake up the process. This can
565 * safely be done because the task is still sleeping
566 * and can't procude new pfaults. */
567 tsk->thread.pfault_wait = 0;
568 wake_up_process(tsk);
569 }
570 } else {
571 /* signal bit not set -> a real page is missing. */
572 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
573 if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
574 /* Completion interrupt was faster than the initial
575 * interrupt (swapped in a -1 for pfault_wait). Set
576 * pfault_wait back to zero and exit. This can be
577 * done safely because tsk is running in kernel
578 * mode and can't produce new pfaults. */
579 tsk->thread.pfault_wait = 0;
580 set_task_state(tsk, TASK_RUNNING);
581 } else
582 set_tsk_need_resched(tsk);
583 }
584}
585#endif
586
diff --git a/arch/s390/mm/init.c b/arch/s390/mm/init.c
new file mode 100644
index 000000000000..8e723bc7f795
--- /dev/null
+++ b/arch/s390/mm/init.c
@@ -0,0 +1,310 @@
1/*
2 * arch/s390/mm/init.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 *
8 * Derived from "arch/i386/mm/init.c"
9 * Copyright (C) 1995 Linus Torvalds
10 */
11
12#include <linux/config.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/swap.h>
23#include <linux/smp.h>
24#include <linux/init.h>
25#include <linux/pagemap.h>
26#include <linux/bootmem.h>
27
28#include <asm/processor.h>
29#include <asm/system.h>
30#include <asm/uaccess.h>
31#include <asm/pgtable.h>
32#include <asm/pgalloc.h>
33#include <asm/dma.h>
34#include <asm/lowcore.h>
35#include <asm/tlb.h>
36#include <asm/tlbflush.h>
37
38DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
39
40pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__((__aligned__(PAGE_SIZE)));
41char empty_zero_page[PAGE_SIZE] __attribute__((__aligned__(PAGE_SIZE)));
42
43void diag10(unsigned long addr)
44{
45 if (addr >= 0x7ff00000)
46 return;
47#ifdef __s390x__
48 asm volatile (
49 " sam31\n"
50 " diag %0,%0,0x10\n"
51 "0: sam64\n"
52 ".section __ex_table,\"a\"\n"
53 " .align 8\n"
54 " .quad 0b, 0b\n"
55 ".previous\n"
56 : : "a" (addr));
57#else
58 asm volatile (
59 " diag %0,%0,0x10\n"
60 "0:\n"
61 ".section __ex_table,\"a\"\n"
62 " .align 4\n"
63 " .long 0b, 0b\n"
64 ".previous\n"
65 : : "a" (addr));
66#endif
67}
68
69void show_mem(void)
70{
71 int i, total = 0, reserved = 0;
72 int shared = 0, cached = 0;
73
74 printk("Mem-info:\n");
75 show_free_areas();
76 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
77 i = max_mapnr;
78 while (i-- > 0) {
79 total++;
80 if (PageReserved(mem_map+i))
81 reserved++;
82 else if (PageSwapCache(mem_map+i))
83 cached++;
84 else if (page_count(mem_map+i))
85 shared += page_count(mem_map+i) - 1;
86 }
87 printk("%d pages of RAM\n",total);
88 printk("%d reserved pages\n",reserved);
89 printk("%d pages shared\n",shared);
90 printk("%d pages swap cached\n",cached);
91}
92
93/* References to section boundaries */
94
95extern unsigned long _text;
96extern unsigned long _etext;
97extern unsigned long _edata;
98extern unsigned long __bss_start;
99extern unsigned long _end;
100
101extern unsigned long __init_begin;
102extern unsigned long __init_end;
103
104/*
105 * paging_init() sets up the page tables
106 */
107
108#ifndef CONFIG_ARCH_S390X
109void __init paging_init(void)
110{
111 pgd_t * pg_dir;
112 pte_t * pg_table;
113 pte_t pte;
114 int i;
115 unsigned long tmp;
116 unsigned long pfn = 0;
117 unsigned long pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) | _KERNSEG_TABLE;
118 static const int ssm_mask = 0x04000000L;
119
120 /* unmap whole virtual address space */
121
122 pg_dir = swapper_pg_dir;
123
124 for (i=0;i<KERNEL_PGD_PTRS;i++)
125 pmd_clear((pmd_t*)pg_dir++);
126
127 /*
128 * map whole physical memory to virtual memory (identity mapping)
129 */
130
131 pg_dir = swapper_pg_dir;
132
133 while (pfn < max_low_pfn) {
134 /*
135 * pg_table is physical at this point
136 */
137 pg_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
138
139 pg_dir->pgd0 = (_PAGE_TABLE | __pa(pg_table));
140 pg_dir->pgd1 = (_PAGE_TABLE | (__pa(pg_table)+1024));
141 pg_dir->pgd2 = (_PAGE_TABLE | (__pa(pg_table)+2048));
142 pg_dir->pgd3 = (_PAGE_TABLE | (__pa(pg_table)+3072));
143 pg_dir++;
144
145 for (tmp = 0 ; tmp < PTRS_PER_PTE ; tmp++,pg_table++) {
146 pte = pfn_pte(pfn, PAGE_KERNEL);
147 if (pfn >= max_low_pfn)
148 pte_clear(&init_mm, 0, &pte);
149 set_pte(pg_table, pte);
150 pfn++;
151 }
152 }
153
154 S390_lowcore.kernel_asce = pgdir_k;
155
156 /* enable virtual mapping in kernel mode */
157 __asm__ __volatile__(" LCTL 1,1,%0\n"
158 " LCTL 7,7,%0\n"
159 " LCTL 13,13,%0\n"
160 " SSM %1"
161 : : "m" (pgdir_k), "m" (ssm_mask));
162
163 local_flush_tlb();
164
165 {
166 unsigned long zones_size[MAX_NR_ZONES] = { 0, 0, 0};
167
168 zones_size[ZONE_DMA] = max_low_pfn;
169 free_area_init(zones_size);
170 }
171 return;
172}
173
174#else /* CONFIG_ARCH_S390X */
175void __init paging_init(void)
176{
177 pgd_t * pg_dir;
178 pmd_t * pm_dir;
179 pte_t * pt_dir;
180 pte_t pte;
181 int i,j,k;
182 unsigned long pfn = 0;
183 unsigned long pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) |
184 _KERN_REGION_TABLE;
185 static const int ssm_mask = 0x04000000L;
186
187 unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
188 unsigned long dma_pfn, high_pfn;
189
190 dma_pfn = MAX_DMA_ADDRESS >> PAGE_SHIFT;
191 high_pfn = max_low_pfn;
192
193 if (dma_pfn > high_pfn)
194 zones_size[ZONE_DMA] = high_pfn;
195 else {
196 zones_size[ZONE_DMA] = dma_pfn;
197 zones_size[ZONE_NORMAL] = high_pfn - dma_pfn;
198 }
199
200 /* Initialize mem_map[]. */
201 free_area_init(zones_size);
202
203
204 /*
205 * map whole physical memory to virtual memory (identity mapping)
206 */
207
208 pg_dir = swapper_pg_dir;
209
210 for (i = 0 ; i < PTRS_PER_PGD ; i++,pg_dir++) {
211
212 if (pfn >= max_low_pfn) {
213 pgd_clear(pg_dir);
214 continue;
215 }
216
217 pm_dir = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE*4);
218 pgd_populate(&init_mm, pg_dir, pm_dir);
219
220 for (j = 0 ; j < PTRS_PER_PMD ; j++,pm_dir++) {
221 if (pfn >= max_low_pfn) {
222 pmd_clear(pm_dir);
223 continue;
224 }
225
226 pt_dir = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
227 pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
228
229 for (k = 0 ; k < PTRS_PER_PTE ; k++,pt_dir++) {
230 pte = pfn_pte(pfn, PAGE_KERNEL);
231 if (pfn >= max_low_pfn) {
232 pte_clear(&init_mm, 0, &pte);
233 continue;
234 }
235 set_pte(pt_dir, pte);
236 pfn++;
237 }
238 }
239 }
240
241 S390_lowcore.kernel_asce = pgdir_k;
242
243 /* enable virtual mapping in kernel mode */
244 __asm__ __volatile__("lctlg 1,1,%0\n\t"
245 "lctlg 7,7,%0\n\t"
246 "lctlg 13,13,%0\n\t"
247 "ssm %1"
248 : :"m" (pgdir_k), "m" (ssm_mask));
249
250 local_flush_tlb();
251
252 return;
253}
254#endif /* CONFIG_ARCH_S390X */
255
256void __init mem_init(void)
257{
258 unsigned long codesize, reservedpages, datasize, initsize;
259
260 max_mapnr = num_physpages = max_low_pfn;
261 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
262
263 /* clear the zero-page */
264 memset(empty_zero_page, 0, PAGE_SIZE);
265
266 /* this will put all low memory onto the freelists */
267 totalram_pages += free_all_bootmem();
268
269 reservedpages = 0;
270
271 codesize = (unsigned long) &_etext - (unsigned long) &_text;
272 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
273 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
274 printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n",
275 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
276 max_mapnr << (PAGE_SHIFT-10),
277 codesize >> 10,
278 reservedpages << (PAGE_SHIFT-10),
279 datasize >>10,
280 initsize >> 10);
281}
282
283void free_initmem(void)
284{
285 unsigned long addr;
286
287 addr = (unsigned long)(&__init_begin);
288 for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
289 ClearPageReserved(virt_to_page(addr));
290 set_page_count(virt_to_page(addr), 1);
291 free_page(addr);
292 totalram_pages++;
293 }
294 printk ("Freeing unused kernel memory: %ldk freed\n",
295 ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10);
296}
297
298#ifdef CONFIG_BLK_DEV_INITRD
299void free_initrd_mem(unsigned long start, unsigned long end)
300{
301 if (start < end)
302 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
303 for (; start < end; start += PAGE_SIZE) {
304 ClearPageReserved(virt_to_page(start));
305 set_page_count(virt_to_page(start), 1);
306 free_page(start);
307 totalram_pages++;
308 }
309}
310#endif
diff --git a/arch/s390/mm/ioremap.c b/arch/s390/mm/ioremap.c
new file mode 100644
index 000000000000..c6c39d868bc8
--- /dev/null
+++ b/arch/s390/mm/ioremap.c
@@ -0,0 +1,138 @@
1/*
2 * arch/s390/mm/ioremap.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 *
8 * Derived from "arch/i386/mm/extable.c"
9 * (C) Copyright 1995 1996 Linus Torvalds
10 *
11 * Re-map IO memory to kernel address space so that we can access it.
12 * This is needed for high PCI addresses that aren't mapped in the
13 * 640k-1MB IO memory area on PC's
14 */
15
16#include <linux/vmalloc.h>
17#include <linux/mm.h>
18#include <asm/io.h>
19#include <asm/pgalloc.h>
20#include <asm/cacheflush.h>
21#include <asm/tlbflush.h>
22
23static inline void remap_area_pte(pte_t * pte, unsigned long address, unsigned long size,
24 unsigned long phys_addr, unsigned long flags)
25{
26 unsigned long end;
27 unsigned long pfn;
28
29 address &= ~PMD_MASK;
30 end = address + size;
31 if (end > PMD_SIZE)
32 end = PMD_SIZE;
33 if (address >= end)
34 BUG();
35 pfn = phys_addr >> PAGE_SHIFT;
36 do {
37 if (!pte_none(*pte)) {
38 printk("remap_area_pte: page already exists\n");
39 BUG();
40 }
41 set_pte(pte, pfn_pte(pfn, __pgprot(flags)));
42 address += PAGE_SIZE;
43 pfn++;
44 pte++;
45 } while (address && (address < end));
46}
47
48static inline int remap_area_pmd(pmd_t * pmd, unsigned long address, unsigned long size,
49 unsigned long phys_addr, unsigned long flags)
50{
51 unsigned long end;
52
53 address &= ~PGDIR_MASK;
54 end = address + size;
55 if (end > PGDIR_SIZE)
56 end = PGDIR_SIZE;
57 phys_addr -= address;
58 if (address >= end)
59 BUG();
60 do {
61 pte_t * pte = pte_alloc_kernel(&init_mm, pmd, address);
62 if (!pte)
63 return -ENOMEM;
64 remap_area_pte(pte, address, end - address, address + phys_addr, flags);
65 address = (address + PMD_SIZE) & PMD_MASK;
66 pmd++;
67 } while (address && (address < end));
68 return 0;
69}
70
71static int remap_area_pages(unsigned long address, unsigned long phys_addr,
72 unsigned long size, unsigned long flags)
73{
74 int error;
75 pgd_t * dir;
76 unsigned long end = address + size;
77
78 phys_addr -= address;
79 dir = pgd_offset(&init_mm, address);
80 flush_cache_all();
81 if (address >= end)
82 BUG();
83 spin_lock(&init_mm.page_table_lock);
84 do {
85 pmd_t *pmd;
86 pmd = pmd_alloc(&init_mm, dir, address);
87 error = -ENOMEM;
88 if (!pmd)
89 break;
90 if (remap_area_pmd(pmd, address, end - address,
91 phys_addr + address, flags))
92 break;
93 error = 0;
94 address = (address + PGDIR_SIZE) & PGDIR_MASK;
95 dir++;
96 } while (address && (address < end));
97 spin_unlock(&init_mm.page_table_lock);
98 flush_tlb_all();
99 return 0;
100}
101
102/*
103 * Generic mapping function (not visible outside):
104 */
105
106/*
107 * Remap an arbitrary physical address space into the kernel virtual
108 * address space. Needed when the kernel wants to access high addresses
109 * directly.
110 */
111void * __ioremap(unsigned long phys_addr, unsigned long size, unsigned long flags)
112{
113 void * addr;
114 struct vm_struct * area;
115
116 if (phys_addr < virt_to_phys(high_memory))
117 return phys_to_virt(phys_addr);
118 if (phys_addr & ~PAGE_MASK)
119 return NULL;
120 size = PAGE_ALIGN(size);
121 if (!size || size > phys_addr + size)
122 return NULL;
123 area = get_vm_area(size, VM_IOREMAP);
124 if (!area)
125 return NULL;
126 addr = area->addr;
127 if (remap_area_pages((unsigned long) addr, phys_addr, size, flags)) {
128 vfree(addr);
129 return NULL;
130 }
131 return addr;
132}
133
134void iounmap(void *addr)
135{
136 if (addr > high_memory)
137 vfree(addr);
138}
diff --git a/arch/s390/mm/mmap.c b/arch/s390/mm/mmap.c
new file mode 100644
index 000000000000..fb187e5a54b4
--- /dev/null
+++ b/arch/s390/mm/mmap.c
@@ -0,0 +1,86 @@
1/*
2 * linux/arch/s390/mm/mmap.c
3 *
4 * flexible mmap layout support
5 *
6 * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
7 * All Rights Reserved.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 *
23 *
24 * Started by Ingo Molnar <mingo@elte.hu>
25 */
26
27#include <linux/personality.h>
28#include <linux/mm.h>
29#include <linux/module.h>
30
31/*
32 * Top of mmap area (just below the process stack).
33 *
34 * Leave an at least ~128 MB hole.
35 */
36#define MIN_GAP (128*1024*1024)
37#define MAX_GAP (TASK_SIZE/6*5)
38
39static inline unsigned long mmap_base(void)
40{
41 unsigned long gap = current->signal->rlim[RLIMIT_STACK].rlim_cur;
42
43 if (gap < MIN_GAP)
44 gap = MIN_GAP;
45 else if (gap > MAX_GAP)
46 gap = MAX_GAP;
47
48 return TASK_SIZE - (gap & PAGE_MASK);
49}
50
51static inline int mmap_is_legacy(void)
52{
53#ifdef CONFIG_ARCH_S390X
54 /*
55 * Force standard allocation for 64 bit programs.
56 */
57 if (!test_thread_flag(TIF_31BIT))
58 return 1;
59#endif
60 return sysctl_legacy_va_layout ||
61 (current->personality & ADDR_COMPAT_LAYOUT) ||
62 current->signal->rlim[RLIMIT_STACK].rlim_cur == RLIM_INFINITY;
63}
64
65/*
66 * This function, called very early during the creation of a new
67 * process VM image, sets up which VM layout function to use:
68 */
69void arch_pick_mmap_layout(struct mm_struct *mm)
70{
71 /*
72 * Fall back to the standard layout if the personality
73 * bit is set, or if the expected stack growth is unlimited:
74 */
75 if (mmap_is_legacy()) {
76 mm->mmap_base = TASK_UNMAPPED_BASE;
77 mm->get_unmapped_area = arch_get_unmapped_area;
78 mm->unmap_area = arch_unmap_area;
79 } else {
80 mm->mmap_base = mmap_base();
81 mm->get_unmapped_area = arch_get_unmapped_area_topdown;
82 mm->unmap_area = arch_unmap_area_topdown;
83 }
84}
85EXPORT_SYMBOL_GPL(arch_pick_mmap_layout);
86
generated by cgit v1.2.2 (git 2.25.0) at 2024-09-23 18:18:23 -0400