3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/errno.h>
13 #include <linux/sched.h>
14 #include <linux/kernel.h>
16 #include <linux/smp.h>
17 #include <linux/stddef.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/user.h>
21 #include <linux/elf.h>
22 #include <linux/security.h>
23 #include <linux/memblock.h>
25 #include <asm/pgtable.h>
26 #include <asm/processor.h>
28 #include <asm/mmu_context.h>
30 #include <asm/machdep.h>
31 #include <asm/cputable.h>
32 #include <asm/sections.h>
33 #include <asm/firmware.h>
35 #include <asm/vdso_datapage.h>
36 #include <asm/setup.h>
41 #define DBG(fmt...) printk(fmt)
46 /* Max supported size for symbol names */
47 #define MAX_SYMNAME 64
49 /* The alignment of the vDSO */
50 #define VDSO_ALIGNMENT (1 << 16)
52 static unsigned int vdso32_pages
;
53 static void *vdso32_kbase
;
54 static struct page
**vdso32_pagelist
;
55 unsigned long vdso32_sigtramp
;
56 unsigned long vdso32_rt_sigtramp
;
59 extern char vdso32_start
, vdso32_end
;
63 extern char vdso64_start
, vdso64_end
;
64 static void *vdso64_kbase
= &vdso64_start
;
65 static unsigned int vdso64_pages
;
66 static struct page
**vdso64_pagelist
;
67 unsigned long vdso64_rt_sigtramp
;
68 #endif /* CONFIG_PPC64 */
70 static int vdso_ready
;
73 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
74 * Once the early boot kernel code no longer needs to muck around
75 * with it, it will become dynamically allocated
78 struct vdso_data data
;
80 } vdso_data_store __page_aligned_data
;
81 struct vdso_data
*vdso_data
= &vdso_data_store
.data
;
83 /* Format of the patch table */
86 unsigned long ftr_mask
, ftr_value
;
91 /* Table of functions to patch based on the CPU type/revision
93 * Currently, we only change sync_dicache to do nothing on processors
94 * with a coherent icache
96 static struct vdso_patch_def vdso_patches
[] = {
98 CPU_FTR_COHERENT_ICACHE
, CPU_FTR_COHERENT_ICACHE
,
99 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
103 "__kernel_gettimeofday", NULL
107 "__kernel_clock_gettime", NULL
111 "__kernel_clock_getres", NULL
115 "__kernel_get_tbfreq", NULL
119 "__kernel_time", NULL
124 * Some infos carried around for each of them during parsing at
129 Elf32_Ehdr
*hdr
; /* ptr to ELF */
130 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
131 unsigned long dynsymsize
; /* size of .dynsym section */
132 char *dynstr
; /* ptr to .dynstr section */
133 unsigned long text
; /* offset of .text section in .so */
140 unsigned long dynsymsize
;
147 * This is called from binfmt_elf, we create the special vma for the
148 * vDSO and insert it into the mm struct tree
150 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int uses_interp
)
152 struct mm_struct
*mm
= current
->mm
;
153 struct page
**vdso_pagelist
;
154 unsigned long vdso_pages
;
155 unsigned long vdso_base
;
162 if (is_32bit_task()) {
163 vdso_pagelist
= vdso32_pagelist
;
164 vdso_pages
= vdso32_pages
;
165 vdso_base
= VDSO32_MBASE
;
167 vdso_pagelist
= vdso64_pagelist
;
168 vdso_pages
= vdso64_pages
;
170 * On 64bit we don't have a preferred map address. This
171 * allows get_unmapped_area to find an area near other mmaps
172 * and most likely share a SLB entry.
177 vdso_pagelist
= vdso32_pagelist
;
178 vdso_pages
= vdso32_pages
;
179 vdso_base
= VDSO32_MBASE
;
182 current
->mm
->context
.vdso_base
= 0;
184 /* vDSO has a problem and was disabled, just don't "enable" it for the
189 /* Add a page to the vdso size for the data page */
193 * pick a base address for the vDSO in process space. We try to put it
194 * at vdso_base which is the "natural" base for it, but we might fail
195 * and end up putting it elsewhere.
196 * Add enough to the size so that the result can be aligned.
198 if (down_write_killable(&mm
->mmap_sem
))
200 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
201 (vdso_pages
<< PAGE_SHIFT
) +
202 ((VDSO_ALIGNMENT
- 1) & PAGE_MASK
),
204 if (IS_ERR_VALUE(vdso_base
)) {
209 /* Add required alignment. */
210 vdso_base
= ALIGN(vdso_base
, VDSO_ALIGNMENT
);
213 * Put vDSO base into mm struct. We need to do this before calling
214 * install_special_mapping or the perf counter mmap tracking code
215 * will fail to recognise it as a vDSO (since arch_vma_name fails).
217 current
->mm
->context
.vdso_base
= vdso_base
;
220 * our vma flags don't have VM_WRITE so by default, the process isn't
221 * allowed to write those pages.
222 * gdb can break that with ptrace interface, and thus trigger COW on
223 * those pages but it's then your responsibility to never do that on
224 * the "data" page of the vDSO or you'll stop getting kernel updates
225 * and your nice userland gettimeofday will be totally dead.
226 * It's fine to use that for setting breakpoints in the vDSO code
229 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
231 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
,
234 current
->mm
->context
.vdso_base
= 0;
238 up_write(&mm
->mmap_sem
);
242 up_write(&mm
->mmap_sem
);
246 const char *arch_vma_name(struct vm_area_struct
*vma
)
248 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
256 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
263 /* Grab section headers and strings so we can tell who is who */
264 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
265 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
267 /* Find the section they want */
268 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
269 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
271 *size
= sechdrs
[i
].sh_size
;
272 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
279 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
283 char name
[MAX_SYMNAME
], *c
;
285 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
286 if (lib
->dynsym
[i
].st_name
== 0)
288 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
290 c
= strchr(name
, '@');
293 if (strcmp(symname
, name
) == 0)
294 return &lib
->dynsym
[i
];
299 /* Note that we assume the section is .text and the symbol is relative to
302 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
305 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
308 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
312 return sym
->st_value
- VDSO32_LBASE
;
315 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
316 struct lib64_elfinfo
*v64
,
317 const char *orig
, const char *fix
)
319 Elf32_Sym
*sym32_gen
, *sym32_fix
;
321 sym32_gen
= find_symbol32(v32
, orig
);
322 if (sym32_gen
== NULL
) {
323 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
327 sym32_gen
->st_name
= 0;
330 sym32_fix
= find_symbol32(v32
, fix
);
331 if (sym32_fix
== NULL
) {
332 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
335 sym32_gen
->st_value
= sym32_fix
->st_value
;
336 sym32_gen
->st_size
= sym32_fix
->st_size
;
337 sym32_gen
->st_info
= sym32_fix
->st_info
;
338 sym32_gen
->st_other
= sym32_fix
->st_other
;
339 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
343 #else /* !CONFIG_VDSO32 */
344 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
350 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
351 struct lib64_elfinfo
*v64
,
352 const char *orig
, const char *fix
)
356 #endif /* CONFIG_VDSO32 */
361 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
368 /* Grab section headers and strings so we can tell who is who */
369 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
370 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
372 /* Find the section they want */
373 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
374 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
376 *size
= sechdrs
[i
].sh_size
;
377 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
385 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
389 char name
[MAX_SYMNAME
], *c
;
391 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
392 if (lib
->dynsym
[i
].st_name
== 0)
394 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
396 c
= strchr(name
, '@');
399 if (strcmp(symname
, name
) == 0)
400 return &lib
->dynsym
[i
];
405 /* Note that we assume the section is .text and the symbol is relative to
408 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
411 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
414 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
418 #ifdef VDS64_HAS_DESCRIPTORS
419 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
422 return sym
->st_value
- VDSO64_LBASE
;
426 static int __init
vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
427 struct lib64_elfinfo
*v64
,
428 const char *orig
, const char *fix
)
430 Elf64_Sym
*sym64_gen
, *sym64_fix
;
432 sym64_gen
= find_symbol64(v64
, orig
);
433 if (sym64_gen
== NULL
) {
434 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
438 sym64_gen
->st_name
= 0;
441 sym64_fix
= find_symbol64(v64
, fix
);
442 if (sym64_fix
== NULL
) {
443 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
446 sym64_gen
->st_value
= sym64_fix
->st_value
;
447 sym64_gen
->st_size
= sym64_fix
->st_size
;
448 sym64_gen
->st_info
= sym64_fix
->st_info
;
449 sym64_gen
->st_other
= sym64_fix
->st_other
;
450 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
455 #endif /* CONFIG_PPC64 */
458 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
459 struct lib64_elfinfo
*v64
)
464 * Locate symbol tables & text section
468 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
469 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
470 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
471 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
474 sect
= find_section32(v32
->hdr
, ".text", NULL
);
476 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
479 v32
->text
= sect
- vdso32_kbase
;
483 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
484 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
485 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
486 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
489 sect
= find_section64(v64
->hdr
, ".text", NULL
);
491 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
494 v64
->text
= sect
- vdso64_kbase
;
495 #endif /* CONFIG_PPC64 */
500 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
501 struct lib64_elfinfo
*v64
)
504 * Find signal trampolines
508 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
510 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
511 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
514 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
515 struct lib64_elfinfo
*v64
)
523 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
525 printk(KERN_ERR
"vDSO64: Can't find symbol "
526 "__kernel_datapage_offset !\n");
529 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
530 (vdso64_pages
<< PAGE_SHIFT
) -
531 (sym64
->st_value
- VDSO64_LBASE
);
532 #endif /* CONFIG_PPC64 */
535 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
537 printk(KERN_ERR
"vDSO32: Can't find symbol "
538 "__kernel_datapage_offset !\n");
541 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
542 (vdso32_pages
<< PAGE_SHIFT
) -
543 (sym32
->st_value
- VDSO32_LBASE
);
550 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
551 struct lib64_elfinfo
*v64
)
557 start
= find_section64(v64
->hdr
, "__ftr_fixup", &size
);
559 do_feature_fixups(cur_cpu_spec
->cpu_features
,
560 start
, start
+ size
);
562 start
= find_section64(v64
->hdr
, "__mmu_ftr_fixup", &size
);
564 do_feature_fixups(cur_cpu_spec
->mmu_features
,
565 start
, start
+ size
);
567 start
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size
);
569 do_feature_fixups(powerpc_firmware_features
,
570 start
, start
+ size
);
572 start
= find_section64(v64
->hdr
, "__lwsync_fixup", &size
);
574 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
575 start
, start
+ size
);
576 #endif /* CONFIG_PPC64 */
579 start
= find_section32(v32
->hdr
, "__ftr_fixup", &size
);
581 do_feature_fixups(cur_cpu_spec
->cpu_features
,
582 start
, start
+ size
);
584 start
= find_section32(v32
->hdr
, "__mmu_ftr_fixup", &size
);
586 do_feature_fixups(cur_cpu_spec
->mmu_features
,
587 start
, start
+ size
);
590 start
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size
);
592 do_feature_fixups(powerpc_firmware_features
,
593 start
, start
+ size
);
594 #endif /* CONFIG_PPC64 */
596 start
= find_section32(v32
->hdr
, "__lwsync_fixup", &size
);
598 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
599 start
, start
+ size
);
605 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
606 struct lib64_elfinfo
*v64
)
610 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
611 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
612 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
617 DBG("replacing %s with %s...\n", patch
->gen_name
,
618 patch
->fix_name
? "NONE" : patch
->fix_name
);
621 * Patch the 32 bits and 64 bits symbols. Note that we do not
622 * patch the "." symbol on 64 bits.
623 * It would be easy to do, but doesn't seem to be necessary,
624 * patching the OPD symbol is enough.
626 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
629 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
631 #endif /* CONFIG_PPC64 */
638 static __init
int vdso_setup(void)
640 struct lib32_elfinfo v32
;
641 struct lib64_elfinfo v64
;
643 v32
.hdr
= vdso32_kbase
;
645 v64
.hdr
= vdso64_kbase
;
647 if (vdso_do_find_sections(&v32
, &v64
))
650 if (vdso_fixup_datapage(&v32
, &v64
))
653 if (vdso_fixup_features(&v32
, &v64
))
656 if (vdso_fixup_alt_funcs(&v32
, &v64
))
659 vdso_setup_trampolines(&v32
, &v64
);
665 * Called from setup_arch to initialize the bitmap of available
666 * syscalls in the systemcfg page
668 static void __init
vdso_setup_syscall_map(void)
671 extern unsigned long *sys_call_table
;
672 extern unsigned long sys_ni_syscall
;
675 for (i
= 0; i
< NR_syscalls
; i
++) {
677 if (sys_call_table
[i
*2] != sys_ni_syscall
)
678 vdso_data
->syscall_map_64
[i
>> 5] |=
679 0x80000000UL
>> (i
& 0x1f);
680 if (sys_call_table
[i
*2+1] != sys_ni_syscall
)
681 vdso_data
->syscall_map_32
[i
>> 5] |=
682 0x80000000UL
>> (i
& 0x1f);
683 #else /* CONFIG_PPC64 */
684 if (sys_call_table
[i
] != sys_ni_syscall
)
685 vdso_data
->syscall_map_32
[i
>> 5] |=
686 0x80000000UL
>> (i
& 0x1f);
687 #endif /* CONFIG_PPC64 */
692 int vdso_getcpu_init(void)
694 unsigned long cpu
, node
, val
;
697 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
698 * in the next 16 bits. The VDSO uses this to implement getcpu().
701 WARN_ON_ONCE(cpu
> 0xffff);
703 node
= cpu_to_node(cpu
);
704 WARN_ON_ONCE(node
> 0xffff);
706 val
= (cpu
& 0xfff) | ((node
& 0xffff) << 16);
707 mtspr(SPRN_SPRG_VDSO_WRITE
, val
);
708 get_paca()->sprg_vdso
= val
;
714 /* We need to call this before SMP init */
715 early_initcall(vdso_getcpu_init
);
718 static int __init
vdso_init(void)
724 * Fill up the "systemcfg" stuff for backward compatibility
726 strcpy((char *)vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
727 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
728 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
729 vdso_data
->processor
= mfspr(SPRN_PVR
);
731 * Fake the old platform number for pSeries and add
732 * in LPAR bit if necessary
734 vdso_data
->platform
= 0x100;
735 if (firmware_has_feature(FW_FEATURE_LPAR
))
736 vdso_data
->platform
|= 1;
737 vdso_data
->physicalMemorySize
= memblock_phys_mem_size();
738 vdso_data
->dcache_size
= ppc64_caches
.dsize
;
739 vdso_data
->dcache_line_size
= ppc64_caches
.dline_size
;
740 vdso_data
->icache_size
= ppc64_caches
.isize
;
741 vdso_data
->icache_line_size
= ppc64_caches
.iline_size
;
743 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
744 vdso_data
->dcache_block_size
= ppc64_caches
.dline_size
;
745 vdso_data
->icache_block_size
= ppc64_caches
.iline_size
;
746 vdso_data
->dcache_log_block_size
= ppc64_caches
.log_dline_size
;
747 vdso_data
->icache_log_block_size
= ppc64_caches
.log_iline_size
;
750 * Calculate the size of the 64 bits vDSO
752 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
753 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
755 vdso_data
->dcache_block_size
= L1_CACHE_BYTES
;
756 vdso_data
->dcache_log_block_size
= L1_CACHE_SHIFT
;
757 vdso_data
->icache_block_size
= L1_CACHE_BYTES
;
758 vdso_data
->icache_log_block_size
= L1_CACHE_SHIFT
;
759 #endif /* CONFIG_PPC64 */
763 vdso32_kbase
= &vdso32_start
;
766 * Calculate the size of the 32 bits vDSO
768 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
769 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
774 * Setup the syscall map in the vDOS
776 vdso_setup_syscall_map();
779 * Initialize the vDSO images in memory, that is do necessary
780 * fixups of vDSO symbols, locate trampolines, etc...
783 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
792 /* Make sure pages are in the correct state */
793 vdso32_pagelist
= kzalloc(sizeof(struct page
*) * (vdso32_pages
+ 2),
795 BUG_ON(vdso32_pagelist
== NULL
);
796 for (i
= 0; i
< vdso32_pages
; i
++) {
797 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
798 ClearPageReserved(pg
);
800 vdso32_pagelist
[i
] = pg
;
802 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
803 vdso32_pagelist
[i
] = NULL
;
807 vdso64_pagelist
= kzalloc(sizeof(struct page
*) * (vdso64_pages
+ 2),
809 BUG_ON(vdso64_pagelist
== NULL
);
810 for (i
= 0; i
< vdso64_pages
; i
++) {
811 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
812 ClearPageReserved(pg
);
814 vdso64_pagelist
[i
] = pg
;
816 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
817 vdso64_pagelist
[i
] = NULL
;
818 #endif /* CONFIG_PPC64 */
820 get_page(virt_to_page(vdso_data
));
827 arch_initcall(vdso_init
);