Merge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh
[linux-2.6.git] / arch / s390 / kernel / setup.c
1 /*
2  *  arch/s390/kernel/setup.c
3  *
4  *  S390 version
5  *    Copyright (C) IBM Corp. 1999,2012
6  *    Author(s): Hartmut Penner (hp@de.ibm.com),
7  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
8  *
9  *  Derived from "arch/i386/kernel/setup.c"
10  *    Copyright (C) 1995, Linus Torvalds
11  */
12
13 /*
14  * This file handles the architecture-dependent parts of initialization
15  */
16
17 #define KMSG_COMPONENT "setup"
18 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
19
20 #include <linux/errno.h>
21 #include <linux/module.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/memblock.h>
25 #include <linux/mm.h>
26 #include <linux/stddef.h>
27 #include <linux/unistd.h>
28 #include <linux/ptrace.h>
29 #include <linux/user.h>
30 #include <linux/tty.h>
31 #include <linux/ioport.h>
32 #include <linux/delay.h>
33 #include <linux/init.h>
34 #include <linux/initrd.h>
35 #include <linux/bootmem.h>
36 #include <linux/root_dev.h>
37 #include <linux/console.h>
38 #include <linux/kernel_stat.h>
39 #include <linux/device.h>
40 #include <linux/notifier.h>
41 #include <linux/pfn.h>
42 #include <linux/ctype.h>
43 #include <linux/reboot.h>
44 #include <linux/topology.h>
45 #include <linux/ftrace.h>
46 #include <linux/kexec.h>
47 #include <linux/crash_dump.h>
48 #include <linux/memory.h>
49 #include <linux/compat.h>
50
51 #include <asm/ipl.h>
52 #include <asm/uaccess.h>
53 #include <asm/smp.h>
54 #include <asm/mmu_context.h>
55 #include <asm/cpcmd.h>
56 #include <asm/lowcore.h>
57 #include <asm/irq.h>
58 #include <asm/page.h>
59 #include <asm/ptrace.h>
60 #include <asm/sections.h>
61 #include <asm/ebcdic.h>
62 #include <asm/kvm_virtio.h>
63 #include <asm/diag.h>
64 #include <asm/os_info.h>
65 #include "entry.h"
66
67 long psw_kernel_bits    = PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_ASC_PRIMARY |
68                           PSW_MASK_EA | PSW_MASK_BA;
69 long psw_user_bits      = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT |
70                           PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_MCHECK |
71                           PSW_MASK_PSTATE | PSW_ASC_HOME;
72
73 /*
74  * User copy operations.
75  */
76 struct uaccess_ops uaccess;
77 EXPORT_SYMBOL(uaccess);
78
79 /*
80  * Machine setup..
81  */
82 unsigned int console_mode = 0;
83 EXPORT_SYMBOL(console_mode);
84
85 unsigned int console_devno = -1;
86 EXPORT_SYMBOL(console_devno);
87
88 unsigned int console_irq = -1;
89 EXPORT_SYMBOL(console_irq);
90
91 unsigned long elf_hwcap = 0;
92 char elf_platform[ELF_PLATFORM_SIZE];
93
94 struct mem_chunk __initdata memory_chunk[MEMORY_CHUNKS];
95
96 int __initdata memory_end_set;
97 unsigned long __initdata memory_end;
98
99 unsigned long VMALLOC_START;
100 EXPORT_SYMBOL(VMALLOC_START);
101
102 unsigned long VMALLOC_END;
103 EXPORT_SYMBOL(VMALLOC_END);
104
105 struct page *vmemmap;
106 EXPORT_SYMBOL(vmemmap);
107
108 /* An array with a pointer to the lowcore of every CPU. */
109 struct _lowcore *lowcore_ptr[NR_CPUS];
110 EXPORT_SYMBOL(lowcore_ptr);
111
112 /*
113  * This is set up by the setup-routine at boot-time
114  * for S390 need to find out, what we have to setup
115  * using address 0x10400 ...
116  */
117
118 #include <asm/setup.h>
119
120 /*
121  * condev= and conmode= setup parameter.
122  */
123
124 static int __init condev_setup(char *str)
125 {
126         int vdev;
127
128         vdev = simple_strtoul(str, &str, 0);
129         if (vdev >= 0 && vdev < 65536) {
130                 console_devno = vdev;
131                 console_irq = -1;
132         }
133         return 1;
134 }
135
136 __setup("condev=", condev_setup);
137
138 static void __init set_preferred_console(void)
139 {
140         if (MACHINE_IS_KVM)
141                 add_preferred_console("hvc", 0, NULL);
142         else if (CONSOLE_IS_3215 || CONSOLE_IS_SCLP)
143                 add_preferred_console("ttyS", 0, NULL);
144         else if (CONSOLE_IS_3270)
145                 add_preferred_console("tty3270", 0, NULL);
146 }
147
148 static int __init conmode_setup(char *str)
149 {
150 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
151         if (strncmp(str, "hwc", 4) == 0 || strncmp(str, "sclp", 5) == 0)
152                 SET_CONSOLE_SCLP;
153 #endif
154 #if defined(CONFIG_TN3215_CONSOLE)
155         if (strncmp(str, "3215", 5) == 0)
156                 SET_CONSOLE_3215;
157 #endif
158 #if defined(CONFIG_TN3270_CONSOLE)
159         if (strncmp(str, "3270", 5) == 0)
160                 SET_CONSOLE_3270;
161 #endif
162         set_preferred_console();
163         return 1;
164 }
165
166 __setup("conmode=", conmode_setup);
167
168 static void __init conmode_default(void)
169 {
170         char query_buffer[1024];
171         char *ptr;
172
173         if (MACHINE_IS_VM) {
174                 cpcmd("QUERY CONSOLE", query_buffer, 1024, NULL);
175                 console_devno = simple_strtoul(query_buffer + 5, NULL, 16);
176                 ptr = strstr(query_buffer, "SUBCHANNEL =");
177                 console_irq = simple_strtoul(ptr + 13, NULL, 16);
178                 cpcmd("QUERY TERM", query_buffer, 1024, NULL);
179                 ptr = strstr(query_buffer, "CONMODE");
180                 /*
181                  * Set the conmode to 3215 so that the device recognition 
182                  * will set the cu_type of the console to 3215. If the
183                  * conmode is 3270 and we don't set it back then both
184                  * 3215 and the 3270 driver will try to access the console
185                  * device (3215 as console and 3270 as normal tty).
186                  */
187                 cpcmd("TERM CONMODE 3215", NULL, 0, NULL);
188                 if (ptr == NULL) {
189 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
190                         SET_CONSOLE_SCLP;
191 #endif
192                         return;
193                 }
194                 if (strncmp(ptr + 8, "3270", 4) == 0) {
195 #if defined(CONFIG_TN3270_CONSOLE)
196                         SET_CONSOLE_3270;
197 #elif defined(CONFIG_TN3215_CONSOLE)
198                         SET_CONSOLE_3215;
199 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
200                         SET_CONSOLE_SCLP;
201 #endif
202                 } else if (strncmp(ptr + 8, "3215", 4) == 0) {
203 #if defined(CONFIG_TN3215_CONSOLE)
204                         SET_CONSOLE_3215;
205 #elif defined(CONFIG_TN3270_CONSOLE)
206                         SET_CONSOLE_3270;
207 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
208                         SET_CONSOLE_SCLP;
209 #endif
210                 }
211         } else {
212 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
213                 SET_CONSOLE_SCLP;
214 #endif
215         }
216 }
217
218 #ifdef CONFIG_ZFCPDUMP
219 static void __init setup_zfcpdump(unsigned int console_devno)
220 {
221         static char str[41];
222
223         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
224                 return;
225         if (OLDMEM_BASE)
226                 return;
227         if (console_devno != -1)
228                 sprintf(str, " cio_ignore=all,!0.0.%04x,!0.0.%04x",
229                         ipl_info.data.fcp.dev_id.devno, console_devno);
230         else
231                 sprintf(str, " cio_ignore=all,!0.0.%04x",
232                         ipl_info.data.fcp.dev_id.devno);
233         strcat(boot_command_line, str);
234         console_loglevel = 2;
235 }
236 #else
237 static inline void setup_zfcpdump(unsigned int console_devno) {}
238 #endif /* CONFIG_ZFCPDUMP */
239
240  /*
241  * Reboot, halt and power_off stubs. They just call _machine_restart,
242  * _machine_halt or _machine_power_off. 
243  */
244
245 void machine_restart(char *command)
246 {
247         if ((!in_interrupt() && !in_atomic()) || oops_in_progress)
248                 /*
249                  * Only unblank the console if we are called in enabled
250                  * context or a bust_spinlocks cleared the way for us.
251                  */
252                 console_unblank();
253         _machine_restart(command);
254 }
255
256 void machine_halt(void)
257 {
258         if (!in_interrupt() || oops_in_progress)
259                 /*
260                  * Only unblank the console if we are called in enabled
261                  * context or a bust_spinlocks cleared the way for us.
262                  */
263                 console_unblank();
264         _machine_halt();
265 }
266
267 void machine_power_off(void)
268 {
269         if (!in_interrupt() || oops_in_progress)
270                 /*
271                  * Only unblank the console if we are called in enabled
272                  * context or a bust_spinlocks cleared the way for us.
273                  */
274                 console_unblank();
275         _machine_power_off();
276 }
277
278 /*
279  * Dummy power off function.
280  */
281 void (*pm_power_off)(void) = machine_power_off;
282
283 static int __init early_parse_mem(char *p)
284 {
285         memory_end = memparse(p, &p);
286         memory_end_set = 1;
287         return 0;
288 }
289 early_param("mem", early_parse_mem);
290
291 static int __init parse_vmalloc(char *arg)
292 {
293         if (!arg)
294                 return -EINVAL;
295         VMALLOC_END = (memparse(arg, &arg) + PAGE_SIZE - 1) & PAGE_MASK;
296         return 0;
297 }
298 early_param("vmalloc", parse_vmalloc);
299
300 unsigned int user_mode = HOME_SPACE_MODE;
301 EXPORT_SYMBOL_GPL(user_mode);
302
303 static int set_amode_primary(void)
304 {
305         psw_kernel_bits = (psw_kernel_bits & ~PSW_MASK_ASC) | PSW_ASC_HOME;
306         psw_user_bits = (psw_user_bits & ~PSW_MASK_ASC) | PSW_ASC_PRIMARY;
307 #ifdef CONFIG_COMPAT
308         psw32_user_bits =
309                 (psw32_user_bits & ~PSW32_MASK_ASC) | PSW32_ASC_PRIMARY;
310 #endif
311
312         if (MACHINE_HAS_MVCOS) {
313                 memcpy(&uaccess, &uaccess_mvcos_switch, sizeof(uaccess));
314                 return 1;
315         } else {
316                 memcpy(&uaccess, &uaccess_pt, sizeof(uaccess));
317                 return 0;
318         }
319 }
320
321 /*
322  * Switch kernel/user addressing modes?
323  */
324 static int __init early_parse_switch_amode(char *p)
325 {
326         user_mode = PRIMARY_SPACE_MODE;
327         return 0;
328 }
329 early_param("switch_amode", early_parse_switch_amode);
330
331 static int __init early_parse_user_mode(char *p)
332 {
333         if (p && strcmp(p, "primary") == 0)
334                 user_mode = PRIMARY_SPACE_MODE;
335         else if (!p || strcmp(p, "home") == 0)
336                 user_mode = HOME_SPACE_MODE;
337         else
338                 return 1;
339         return 0;
340 }
341 early_param("user_mode", early_parse_user_mode);
342
343 static void setup_addressing_mode(void)
344 {
345         if (user_mode == PRIMARY_SPACE_MODE) {
346                 if (set_amode_primary())
347                         pr_info("Address spaces switched, "
348                                 "mvcos available\n");
349                 else
350                         pr_info("Address spaces switched, "
351                                 "mvcos not available\n");
352         }
353 }
354
355 void *restart_stack __attribute__((__section__(".data")));
356
357 static void __init setup_lowcore(void)
358 {
359         struct _lowcore *lc;
360
361         /*
362          * Setup lowcore for boot cpu
363          */
364         BUILD_BUG_ON(sizeof(struct _lowcore) != LC_PAGES * 4096);
365         lc = __alloc_bootmem_low(LC_PAGES * PAGE_SIZE, LC_PAGES * PAGE_SIZE, 0);
366         lc->restart_psw.mask = psw_kernel_bits;
367         lc->restart_psw.addr =
368                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
369         lc->external_new_psw.mask = psw_kernel_bits |
370                 PSW_MASK_DAT | PSW_MASK_MCHECK;
371         lc->external_new_psw.addr =
372                 PSW_ADDR_AMODE | (unsigned long) ext_int_handler;
373         lc->svc_new_psw.mask = psw_kernel_bits |
374                 PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
375         lc->svc_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) system_call;
376         lc->program_new_psw.mask = psw_kernel_bits |
377                 PSW_MASK_DAT | PSW_MASK_MCHECK;
378         lc->program_new_psw.addr =
379                 PSW_ADDR_AMODE | (unsigned long) pgm_check_handler;
380         lc->mcck_new_psw.mask = psw_kernel_bits;
381         lc->mcck_new_psw.addr =
382                 PSW_ADDR_AMODE | (unsigned long) mcck_int_handler;
383         lc->io_new_psw.mask = psw_kernel_bits |
384                 PSW_MASK_DAT | PSW_MASK_MCHECK;
385         lc->io_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) io_int_handler;
386         lc->clock_comparator = -1ULL;
387         lc->kernel_stack = ((unsigned long) &init_thread_union) + THREAD_SIZE;
388         lc->async_stack = (unsigned long)
389                 __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0) + ASYNC_SIZE;
390         lc->panic_stack = (unsigned long)
391                 __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, 0) + PAGE_SIZE;
392         lc->current_task = (unsigned long) init_thread_union.thread_info.task;
393         lc->thread_info = (unsigned long) &init_thread_union;
394         lc->machine_flags = S390_lowcore.machine_flags;
395         lc->stfl_fac_list = S390_lowcore.stfl_fac_list;
396         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
397                MAX_FACILITY_BIT/8);
398 #ifndef CONFIG_64BIT
399         if (MACHINE_HAS_IEEE) {
400                 lc->extended_save_area_addr = (__u32)
401                         __alloc_bootmem_low(PAGE_SIZE, PAGE_SIZE, 0);
402                 /* enable extended save area */
403                 __ctl_set_bit(14, 29);
404         }
405 #else
406         lc->vdso_per_cpu_data = (unsigned long) &lc->paste[0];
407 #endif
408         lc->sync_enter_timer = S390_lowcore.sync_enter_timer;
409         lc->async_enter_timer = S390_lowcore.async_enter_timer;
410         lc->exit_timer = S390_lowcore.exit_timer;
411         lc->user_timer = S390_lowcore.user_timer;
412         lc->system_timer = S390_lowcore.system_timer;
413         lc->steal_timer = S390_lowcore.steal_timer;
414         lc->last_update_timer = S390_lowcore.last_update_timer;
415         lc->last_update_clock = S390_lowcore.last_update_clock;
416         lc->ftrace_func = S390_lowcore.ftrace_func;
417
418         restart_stack = __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0);
419         restart_stack += ASYNC_SIZE;
420
421         /*
422          * Set up PSW restart to call ipl.c:do_restart(). Copy the relevant
423          * restart data to the absolute zero lowcore. This is necesary if
424          * PSW restart is done on an offline CPU that has lowcore zero.
425          */
426         lc->restart_stack = (unsigned long) restart_stack;
427         lc->restart_fn = (unsigned long) do_restart;
428         lc->restart_data = 0;
429         lc->restart_source = -1UL;
430         memcpy(&S390_lowcore.restart_stack, &lc->restart_stack,
431                4*sizeof(unsigned long));
432         copy_to_absolute_zero(&S390_lowcore.restart_psw,
433                               &lc->restart_psw, sizeof(psw_t));
434
435         set_prefix((u32)(unsigned long) lc);
436         lowcore_ptr[0] = lc;
437 }
438
439 static struct resource code_resource = {
440         .name  = "Kernel code",
441         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
442 };
443
444 static struct resource data_resource = {
445         .name = "Kernel data",
446         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
447 };
448
449 static struct resource bss_resource = {
450         .name = "Kernel bss",
451         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
452 };
453
454 static struct resource __initdata *standard_resources[] = {
455         &code_resource,
456         &data_resource,
457         &bss_resource,
458 };
459
460 static void __init setup_resources(void)
461 {
462         struct resource *res, *std_res, *sub_res;
463         int i, j;
464
465         code_resource.start = (unsigned long) &_text;
466         code_resource.end = (unsigned long) &_etext - 1;
467         data_resource.start = (unsigned long) &_etext;
468         data_resource.end = (unsigned long) &_edata - 1;
469         bss_resource.start = (unsigned long) &__bss_start;
470         bss_resource.end = (unsigned long) &__bss_stop - 1;
471
472         for (i = 0; i < MEMORY_CHUNKS; i++) {
473                 if (!memory_chunk[i].size)
474                         continue;
475                 if (memory_chunk[i].type == CHUNK_OLDMEM ||
476                     memory_chunk[i].type == CHUNK_CRASHK)
477                         continue;
478                 res = alloc_bootmem_low(sizeof(*res));
479                 res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
480                 switch (memory_chunk[i].type) {
481                 case CHUNK_READ_WRITE:
482                 case CHUNK_CRASHK:
483                         res->name = "System RAM";
484                         break;
485                 case CHUNK_READ_ONLY:
486                         res->name = "System ROM";
487                         res->flags |= IORESOURCE_READONLY;
488                         break;
489                 default:
490                         res->name = "reserved";
491                 }
492                 res->start = memory_chunk[i].addr;
493                 res->end = res->start + memory_chunk[i].size - 1;
494                 request_resource(&iomem_resource, res);
495
496                 for (j = 0; j < ARRAY_SIZE(standard_resources); j++) {
497                         std_res = standard_resources[j];
498                         if (std_res->start < res->start ||
499                             std_res->start > res->end)
500                                 continue;
501                         if (std_res->end > res->end) {
502                                 sub_res = alloc_bootmem_low(sizeof(*sub_res));
503                                 *sub_res = *std_res;
504                                 sub_res->end = res->end;
505                                 std_res->start = res->end + 1;
506                                 request_resource(res, sub_res);
507                         } else {
508                                 request_resource(res, std_res);
509                         }
510                 }
511         }
512 }
513
514 unsigned long real_memory_size;
515 EXPORT_SYMBOL_GPL(real_memory_size);
516
517 static void __init setup_memory_end(void)
518 {
519         unsigned long vmax, vmalloc_size, tmp;
520         int i;
521
522
523 #ifdef CONFIG_ZFCPDUMP
524         if (ipl_info.type == IPL_TYPE_FCP_DUMP && !OLDMEM_BASE) {
525                 memory_end = ZFCPDUMP_HSA_SIZE;
526                 memory_end_set = 1;
527         }
528 #endif
529         real_memory_size = 0;
530         memory_end &= PAGE_MASK;
531
532         /*
533          * Make sure all chunks are MAX_ORDER aligned so we don't need the
534          * extra checks that HOLES_IN_ZONE would require.
535          */
536         for (i = 0; i < MEMORY_CHUNKS; i++) {
537                 unsigned long start, end;
538                 struct mem_chunk *chunk;
539                 unsigned long align;
540
541                 chunk = &memory_chunk[i];
542                 align = 1UL << (MAX_ORDER + PAGE_SHIFT - 1);
543                 start = (chunk->addr + align - 1) & ~(align - 1);
544                 end = (chunk->addr + chunk->size) & ~(align - 1);
545                 if (start >= end)
546                         memset(chunk, 0, sizeof(*chunk));
547                 else {
548                         chunk->addr = start;
549                         chunk->size = end - start;
550                 }
551                 real_memory_size = max(real_memory_size,
552                                        chunk->addr + chunk->size);
553         }
554
555         /* Choose kernel address space layout: 2, 3, or 4 levels. */
556 #ifdef CONFIG_64BIT
557         vmalloc_size = VMALLOC_END ?: 128UL << 30;
558         tmp = (memory_end ?: real_memory_size) / PAGE_SIZE;
559         tmp = tmp * (sizeof(struct page) + PAGE_SIZE) + vmalloc_size;
560         if (tmp <= (1UL << 42))
561                 vmax = 1UL << 42;       /* 3-level kernel page table */
562         else
563                 vmax = 1UL << 53;       /* 4-level kernel page table */
564 #else
565         vmalloc_size = VMALLOC_END ?: 96UL << 20;
566         vmax = 1UL << 31;               /* 2-level kernel page table */
567 #endif
568         /* vmalloc area is at the end of the kernel address space. */
569         VMALLOC_END = vmax;
570         VMALLOC_START = vmax - vmalloc_size;
571
572         /* Split remaining virtual space between 1:1 mapping & vmemmap array */
573         tmp = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
574         tmp = VMALLOC_START - tmp * sizeof(struct page);
575         tmp &= ~((vmax >> 11) - 1);     /* align to page table level */
576         tmp = min(tmp, 1UL << MAX_PHYSMEM_BITS);
577         vmemmap = (struct page *) tmp;
578
579         /* Take care that memory_end is set and <= vmemmap */
580         memory_end = min(memory_end ?: real_memory_size, tmp);
581
582         /* Fixup memory chunk array to fit into 0..memory_end */
583         for (i = 0; i < MEMORY_CHUNKS; i++) {
584                 struct mem_chunk *chunk = &memory_chunk[i];
585
586                 if (chunk->addr >= memory_end) {
587                         memset(chunk, 0, sizeof(*chunk));
588                         continue;
589                 }
590                 if (chunk->addr + chunk->size > memory_end)
591                         chunk->size = memory_end - chunk->addr;
592         }
593 }
594
595 static void __init setup_vmcoreinfo(void)
596 {
597 #ifdef CONFIG_KEXEC
598         unsigned long ptr = paddr_vmcoreinfo_note();
599
600         copy_to_absolute_zero(&S390_lowcore.vmcore_info, &ptr, sizeof(ptr));
601 #endif
602 }
603
604 #ifdef CONFIG_CRASH_DUMP
605
606 /*
607  * Find suitable location for crashkernel memory
608  */
609 static unsigned long __init find_crash_base(unsigned long crash_size,
610                                             char **msg)
611 {
612         unsigned long crash_base;
613         struct mem_chunk *chunk;
614         int i;
615
616         if (memory_chunk[0].size < crash_size) {
617                 *msg = "first memory chunk must be at least crashkernel size";
618                 return 0;
619         }
620         if (OLDMEM_BASE && crash_size == OLDMEM_SIZE)
621                 return OLDMEM_BASE;
622
623         for (i = MEMORY_CHUNKS - 1; i >= 0; i--) {
624                 chunk = &memory_chunk[i];
625                 if (chunk->size == 0)
626                         continue;
627                 if (chunk->type != CHUNK_READ_WRITE)
628                         continue;
629                 if (chunk->size < crash_size)
630                         continue;
631                 crash_base = (chunk->addr + chunk->size) - crash_size;
632                 if (crash_base < crash_size)
633                         continue;
634                 if (crash_base < ZFCPDUMP_HSA_SIZE_MAX)
635                         continue;
636                 if (crash_base < (unsigned long) INITRD_START + INITRD_SIZE)
637                         continue;
638                 return crash_base;
639         }
640         *msg = "no suitable area found";
641         return 0;
642 }
643
644 /*
645  * Check if crash_base and crash_size is valid
646  */
647 static int __init verify_crash_base(unsigned long crash_base,
648                                     unsigned long crash_size,
649                                     char **msg)
650 {
651         struct mem_chunk *chunk;
652         int i;
653
654         /*
655          * Because we do the swap to zero, we must have at least 'crash_size'
656          * bytes free space before crash_base
657          */
658         if (crash_size > crash_base) {
659                 *msg = "crashkernel offset must be greater than size";
660                 return -EINVAL;
661         }
662
663         /* First memory chunk must be at least crash_size */
664         if (memory_chunk[0].size < crash_size) {
665                 *msg = "first memory chunk must be at least crashkernel size";
666                 return -EINVAL;
667         }
668         /* Check if we fit into the respective memory chunk */
669         for (i = 0; i < MEMORY_CHUNKS; i++) {
670                 chunk = &memory_chunk[i];
671                 if (chunk->size == 0)
672                         continue;
673                 if (crash_base < chunk->addr)
674                         continue;
675                 if (crash_base >= chunk->addr + chunk->size)
676                         continue;
677                 /* we have found the memory chunk */
678                 if (crash_base + crash_size > chunk->addr + chunk->size) {
679                         *msg = "selected memory chunk is too small for "
680                                 "crashkernel memory";
681                         return -EINVAL;
682                 }
683                 return 0;
684         }
685         *msg = "invalid memory range specified";
686         return -EINVAL;
687 }
688
689 /*
690  * Reserve kdump memory by creating a memory hole in the mem_chunk array
691  */
692 static void __init reserve_kdump_bootmem(unsigned long addr, unsigned long size,
693                                          int type)
694 {
695         create_mem_hole(memory_chunk, addr, size, type);
696 }
697
698 /*
699  * When kdump is enabled, we have to ensure that no memory from
700  * the area [0 - crashkernel memory size] and
701  * [crashk_res.start - crashk_res.end] is set offline.
702  */
703 static int kdump_mem_notifier(struct notifier_block *nb,
704                               unsigned long action, void *data)
705 {
706         struct memory_notify *arg = data;
707
708         if (arg->start_pfn < PFN_DOWN(resource_size(&crashk_res)))
709                 return NOTIFY_BAD;
710         if (arg->start_pfn > PFN_DOWN(crashk_res.end))
711                 return NOTIFY_OK;
712         if (arg->start_pfn + arg->nr_pages - 1 < PFN_DOWN(crashk_res.start))
713                 return NOTIFY_OK;
714         return NOTIFY_BAD;
715 }
716
717 static struct notifier_block kdump_mem_nb = {
718         .notifier_call = kdump_mem_notifier,
719 };
720
721 #endif
722
723 /*
724  * Make sure that oldmem, where the dump is stored, is protected
725  */
726 static void reserve_oldmem(void)
727 {
728 #ifdef CONFIG_CRASH_DUMP
729         if (!OLDMEM_BASE)
730                 return;
731
732         reserve_kdump_bootmem(OLDMEM_BASE, OLDMEM_SIZE, CHUNK_OLDMEM);
733         reserve_kdump_bootmem(OLDMEM_SIZE, memory_end - OLDMEM_SIZE,
734                               CHUNK_OLDMEM);
735         if (OLDMEM_BASE + OLDMEM_SIZE == real_memory_size)
736                 saved_max_pfn = PFN_DOWN(OLDMEM_BASE) - 1;
737         else
738                 saved_max_pfn = PFN_DOWN(real_memory_size) - 1;
739 #endif
740 }
741
742 /*
743  * Reserve memory for kdump kernel to be loaded with kexec
744  */
745 static void __init reserve_crashkernel(void)
746 {
747 #ifdef CONFIG_CRASH_DUMP
748         unsigned long long crash_base, crash_size;
749         char *msg = NULL;
750         int rc;
751
752         rc = parse_crashkernel(boot_command_line, memory_end, &crash_size,
753                                &crash_base);
754         if (rc || crash_size == 0)
755                 return;
756         crash_base = ALIGN(crash_base, KEXEC_CRASH_MEM_ALIGN);
757         crash_size = ALIGN(crash_size, KEXEC_CRASH_MEM_ALIGN);
758         if (register_memory_notifier(&kdump_mem_nb))
759                 return;
760         if (!crash_base)
761                 crash_base = find_crash_base(crash_size, &msg);
762         if (!crash_base) {
763                 pr_info("crashkernel reservation failed: %s\n", msg);
764                 unregister_memory_notifier(&kdump_mem_nb);
765                 return;
766         }
767         if (verify_crash_base(crash_base, crash_size, &msg)) {
768                 pr_info("crashkernel reservation failed: %s\n", msg);
769                 unregister_memory_notifier(&kdump_mem_nb);
770                 return;
771         }
772         if (!OLDMEM_BASE && MACHINE_IS_VM)
773                 diag10_range(PFN_DOWN(crash_base), PFN_DOWN(crash_size));
774         crashk_res.start = crash_base;
775         crashk_res.end = crash_base + crash_size - 1;
776         insert_resource(&iomem_resource, &crashk_res);
777         reserve_kdump_bootmem(crash_base, crash_size, CHUNK_CRASHK);
778         pr_info("Reserving %lluMB of memory at %lluMB "
779                 "for crashkernel (System RAM: %luMB)\n",
780                 crash_size >> 20, crash_base >> 20, memory_end >> 20);
781         os_info_crashkernel_add(crash_base, crash_size);
782 #endif
783 }
784
785 static void __init setup_memory(void)
786 {
787         unsigned long bootmap_size;
788         unsigned long start_pfn, end_pfn;
789         int i;
790
791         /*
792          * partially used pages are not usable - thus
793          * we are rounding upwards:
794          */
795         start_pfn = PFN_UP(__pa(&_end));
796         end_pfn = max_pfn = PFN_DOWN(memory_end);
797
798 #ifdef CONFIG_BLK_DEV_INITRD
799         /*
800          * Move the initrd in case the bitmap of the bootmem allocater
801          * would overwrite it.
802          */
803
804         if (INITRD_START && INITRD_SIZE) {
805                 unsigned long bmap_size;
806                 unsigned long start;
807
808                 bmap_size = bootmem_bootmap_pages(end_pfn - start_pfn + 1);
809                 bmap_size = PFN_PHYS(bmap_size);
810
811                 if (PFN_PHYS(start_pfn) + bmap_size > INITRD_START) {
812                         start = PFN_PHYS(start_pfn) + bmap_size + PAGE_SIZE;
813
814 #ifdef CONFIG_CRASH_DUMP
815                         if (OLDMEM_BASE) {
816                                 /* Move initrd behind kdump oldmem */
817                                 if (start + INITRD_SIZE > OLDMEM_BASE &&
818                                     start < OLDMEM_BASE + OLDMEM_SIZE)
819                                         start = OLDMEM_BASE + OLDMEM_SIZE;
820                         }
821 #endif
822                         if (start + INITRD_SIZE > memory_end) {
823                                 pr_err("initrd extends beyond end of "
824                                        "memory (0x%08lx > 0x%08lx) "
825                                        "disabling initrd\n",
826                                        start + INITRD_SIZE, memory_end);
827                                 INITRD_START = INITRD_SIZE = 0;
828                         } else {
829                                 pr_info("Moving initrd (0x%08lx -> "
830                                         "0x%08lx, size: %ld)\n",
831                                         INITRD_START, start, INITRD_SIZE);
832                                 memmove((void *) start, (void *) INITRD_START,
833                                         INITRD_SIZE);
834                                 INITRD_START = start;
835                         }
836                 }
837         }
838 #endif
839
840         /*
841          * Initialize the boot-time allocator
842          */
843         bootmap_size = init_bootmem(start_pfn, end_pfn);
844
845         /*
846          * Register RAM areas with the bootmem allocator.
847          */
848
849         for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
850                 unsigned long start_chunk, end_chunk, pfn;
851
852                 if (memory_chunk[i].type != CHUNK_READ_WRITE &&
853                     memory_chunk[i].type != CHUNK_CRASHK)
854                         continue;
855                 start_chunk = PFN_DOWN(memory_chunk[i].addr);
856                 end_chunk = start_chunk + PFN_DOWN(memory_chunk[i].size);
857                 end_chunk = min(end_chunk, end_pfn);
858                 if (start_chunk >= end_chunk)
859                         continue;
860                 memblock_add_node(PFN_PHYS(start_chunk),
861                                   PFN_PHYS(end_chunk - start_chunk), 0);
862                 pfn = max(start_chunk, start_pfn);
863                 for (; pfn < end_chunk; pfn++)
864                         page_set_storage_key(PFN_PHYS(pfn),
865                                              PAGE_DEFAULT_KEY, 0);
866         }
867
868         psw_set_key(PAGE_DEFAULT_KEY);
869
870         free_bootmem_with_active_regions(0, max_pfn);
871
872         /*
873          * Reserve memory used for lowcore/command line/kernel image.
874          */
875         reserve_bootmem(0, (unsigned long)_ehead, BOOTMEM_DEFAULT);
876         reserve_bootmem((unsigned long)_stext,
877                         PFN_PHYS(start_pfn) - (unsigned long)_stext,
878                         BOOTMEM_DEFAULT);
879         /*
880          * Reserve the bootmem bitmap itself as well. We do this in two
881          * steps (first step was init_bootmem()) because this catches
882          * the (very unlikely) case of us accidentally initializing the
883          * bootmem allocator with an invalid RAM area.
884          */
885         reserve_bootmem(start_pfn << PAGE_SHIFT, bootmap_size,
886                         BOOTMEM_DEFAULT);
887
888 #ifdef CONFIG_CRASH_DUMP
889         if (crashk_res.start)
890                 reserve_bootmem(crashk_res.start,
891                                 crashk_res.end - crashk_res.start + 1,
892                                 BOOTMEM_DEFAULT);
893         if (is_kdump_kernel())
894                 reserve_bootmem(elfcorehdr_addr - OLDMEM_BASE,
895                                 PAGE_ALIGN(elfcorehdr_size), BOOTMEM_DEFAULT);
896 #endif
897 #ifdef CONFIG_BLK_DEV_INITRD
898         if (INITRD_START && INITRD_SIZE) {
899                 if (INITRD_START + INITRD_SIZE <= memory_end) {
900                         reserve_bootmem(INITRD_START, INITRD_SIZE,
901                                         BOOTMEM_DEFAULT);
902                         initrd_start = INITRD_START;
903                         initrd_end = initrd_start + INITRD_SIZE;
904                 } else {
905                         pr_err("initrd extends beyond end of "
906                                "memory (0x%08lx > 0x%08lx) "
907                                "disabling initrd\n",
908                                initrd_start + INITRD_SIZE, memory_end);
909                         initrd_start = initrd_end = 0;
910                 }
911         }
912 #endif
913 }
914
915 /*
916  * Setup hardware capabilities.
917  */
918 static void __init setup_hwcaps(void)
919 {
920         static const int stfl_bits[6] = { 0, 2, 7, 17, 19, 21 };
921         struct cpuid cpu_id;
922         int i;
923
924         /*
925          * The store facility list bits numbers as found in the principles
926          * of operation are numbered with bit 1UL<<31 as number 0 to
927          * bit 1UL<<0 as number 31.
928          *   Bit 0: instructions named N3, "backported" to esa-mode
929          *   Bit 2: z/Architecture mode is active
930          *   Bit 7: the store-facility-list-extended facility is installed
931          *   Bit 17: the message-security assist is installed
932          *   Bit 19: the long-displacement facility is installed
933          *   Bit 21: the extended-immediate facility is installed
934          *   Bit 22: extended-translation facility 3 is installed
935          *   Bit 30: extended-translation facility 3 enhancement facility
936          * These get translated to:
937          *   HWCAP_S390_ESAN3 bit 0, HWCAP_S390_ZARCH bit 1,
938          *   HWCAP_S390_STFLE bit 2, HWCAP_S390_MSA bit 3,
939          *   HWCAP_S390_LDISP bit 4, HWCAP_S390_EIMM bit 5 and
940          *   HWCAP_S390_ETF3EH bit 8 (22 && 30).
941          */
942         for (i = 0; i < 6; i++)
943                 if (test_facility(stfl_bits[i]))
944                         elf_hwcap |= 1UL << i;
945
946         if (test_facility(22) && test_facility(30))
947                 elf_hwcap |= HWCAP_S390_ETF3EH;
948
949         /*
950          * Check for additional facilities with store-facility-list-extended.
951          * stfle stores doublewords (8 byte) with bit 1ULL<<63 as bit 0
952          * and 1ULL<<0 as bit 63. Bits 0-31 contain the same information
953          * as stored by stfl, bits 32-xxx contain additional facilities.
954          * How many facility words are stored depends on the number of
955          * doublewords passed to the instruction. The additional facilities
956          * are:
957          *   Bit 42: decimal floating point facility is installed
958          *   Bit 44: perform floating point operation facility is installed
959          * translated to:
960          *   HWCAP_S390_DFP bit 6 (42 && 44).
961          */
962         if ((elf_hwcap & (1UL << 2)) && test_facility(42) && test_facility(44))
963                 elf_hwcap |= HWCAP_S390_DFP;
964
965         /*
966          * Huge page support HWCAP_S390_HPAGE is bit 7.
967          */
968         if (MACHINE_HAS_HPAGE)
969                 elf_hwcap |= HWCAP_S390_HPAGE;
970
971         /*
972          * 64-bit register support for 31-bit processes
973          * HWCAP_S390_HIGH_GPRS is bit 9.
974          */
975         elf_hwcap |= HWCAP_S390_HIGH_GPRS;
976
977         get_cpu_id(&cpu_id);
978         switch (cpu_id.machine) {
979         case 0x9672:
980 #if !defined(CONFIG_64BIT)
981         default:        /* Use "g5" as default for 31 bit kernels. */
982 #endif
983                 strcpy(elf_platform, "g5");
984                 break;
985         case 0x2064:
986         case 0x2066:
987 #if defined(CONFIG_64BIT)
988         default:        /* Use "z900" as default for 64 bit kernels. */
989 #endif
990                 strcpy(elf_platform, "z900");
991                 break;
992         case 0x2084:
993         case 0x2086:
994                 strcpy(elf_platform, "z990");
995                 break;
996         case 0x2094:
997         case 0x2096:
998                 strcpy(elf_platform, "z9-109");
999                 break;
1000         case 0x2097:
1001         case 0x2098:
1002                 strcpy(elf_platform, "z10");
1003                 break;
1004         case 0x2817:
1005         case 0x2818:
1006                 strcpy(elf_platform, "z196");
1007                 break;
1008         }
1009 }
1010
1011 /*
1012  * Setup function called from init/main.c just after the banner
1013  * was printed.
1014  */
1015
1016 void __init setup_arch(char **cmdline_p)
1017 {
1018         /*
1019          * print what head.S has found out about the machine
1020          */
1021 #ifndef CONFIG_64BIT
1022         if (MACHINE_IS_VM)
1023                 pr_info("Linux is running as a z/VM "
1024                         "guest operating system in 31-bit mode\n");
1025         else if (MACHINE_IS_LPAR)
1026                 pr_info("Linux is running natively in 31-bit mode\n");
1027         if (MACHINE_HAS_IEEE)
1028                 pr_info("The hardware system has IEEE compatible "
1029                         "floating point units\n");
1030         else
1031                 pr_info("The hardware system has no IEEE compatible "
1032                         "floating point units\n");
1033 #else /* CONFIG_64BIT */
1034         if (MACHINE_IS_VM)
1035                 pr_info("Linux is running as a z/VM "
1036                         "guest operating system in 64-bit mode\n");
1037         else if (MACHINE_IS_KVM)
1038                 pr_info("Linux is running under KVM in 64-bit mode\n");
1039         else if (MACHINE_IS_LPAR)
1040                 pr_info("Linux is running natively in 64-bit mode\n");
1041 #endif /* CONFIG_64BIT */
1042
1043         /* Have one command line that is parsed and saved in /proc/cmdline */
1044         /* boot_command_line has been already set up in early.c */
1045         *cmdline_p = boot_command_line;
1046
1047         ROOT_DEV = Root_RAM0;
1048
1049         init_mm.start_code = PAGE_OFFSET;
1050         init_mm.end_code = (unsigned long) &_etext;
1051         init_mm.end_data = (unsigned long) &_edata;
1052         init_mm.brk = (unsigned long) &_end;
1053
1054         if (MACHINE_HAS_MVCOS)
1055                 memcpy(&uaccess, &uaccess_mvcos, sizeof(uaccess));
1056         else
1057                 memcpy(&uaccess, &uaccess_std, sizeof(uaccess));
1058
1059         parse_early_param();
1060
1061         os_info_init();
1062         setup_ipl();
1063         setup_memory_end();
1064         setup_addressing_mode();
1065         reserve_oldmem();
1066         reserve_crashkernel();
1067         setup_memory();
1068         setup_resources();
1069         setup_vmcoreinfo();
1070         setup_lowcore();
1071
1072         cpu_init();
1073         s390_init_cpu_topology();
1074
1075         /*
1076          * Setup capabilities (ELF_HWCAP & ELF_PLATFORM).
1077          */
1078         setup_hwcaps();
1079
1080         /*
1081          * Create kernel page tables and switch to virtual addressing.
1082          */
1083         paging_init();
1084
1085         /* Setup default console */
1086         conmode_default();
1087         set_preferred_console();
1088
1089         /* Setup zfcpdump support */
1090         setup_zfcpdump(console_devno);
1091 }