1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
8 #include <linux/errno.h>
9 #include <linux/sched.h>
10 #include <linux/kernel.h>
12 #include <linux/smp.h>
13 #include <linux/stddef.h>
14 #include <linux/unistd.h>
15 #include <linux/slab.h>
16 #include <linux/user.h>
17 #include <linux/elf.h>
18 #include <linux/security.h>
19 #include <linux/memblock.h>
21 #include <asm/pgtable.h>
22 #include <asm/processor.h>
24 #include <asm/mmu_context.h>
26 #include <asm/machdep.h>
27 #include <asm/cputable.h>
28 #include <asm/sections.h>
29 #include <asm/firmware.h>
31 #include <asm/vdso_datapage.h>
32 #include <asm/setup.h>
37 #define DBG(fmt...) printk(fmt)
42 /* Max supported size for symbol names */
43 #define MAX_SYMNAME 64
45 /* The alignment of the vDSO */
46 #define VDSO_ALIGNMENT (1 << 16)
48 static unsigned int vdso32_pages
;
49 static void *vdso32_kbase
;
50 static struct page
**vdso32_pagelist
;
51 unsigned long vdso32_sigtramp
;
52 unsigned long vdso32_rt_sigtramp
;
55 extern char vdso32_start
, vdso32_end
;
59 extern char vdso64_start
, vdso64_end
;
60 static void *vdso64_kbase
= &vdso64_start
;
61 static unsigned int vdso64_pages
;
62 static struct page
**vdso64_pagelist
;
63 unsigned long vdso64_rt_sigtramp
;
64 #endif /* CONFIG_PPC64 */
66 static int vdso_ready
;
69 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
70 * Once the early boot kernel code no longer needs to muck around
71 * with it, it will become dynamically allocated
74 struct vdso_data data
;
76 } vdso_data_store __page_aligned_data
;
77 struct vdso_data
*vdso_data
= &vdso_data_store
.data
;
79 /* Format of the patch table */
82 unsigned long ftr_mask
, ftr_value
;
87 /* Table of functions to patch based on the CPU type/revision
89 * Currently, we only change sync_dicache to do nothing on processors
90 * with a coherent icache
92 static struct vdso_patch_def vdso_patches
[] = {
94 CPU_FTR_COHERENT_ICACHE
, CPU_FTR_COHERENT_ICACHE
,
95 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
100 * Some infos carried around for each of them during parsing at
105 Elf32_Ehdr
*hdr
; /* ptr to ELF */
106 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
107 unsigned long dynsymsize
; /* size of .dynsym section */
108 char *dynstr
; /* ptr to .dynstr section */
109 unsigned long text
; /* offset of .text section in .so */
116 unsigned long dynsymsize
;
123 * This is called from binfmt_elf, we create the special vma for the
124 * vDSO and insert it into the mm struct tree
126 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int uses_interp
)
128 struct mm_struct
*mm
= current
->mm
;
129 struct page
**vdso_pagelist
;
130 unsigned long vdso_pages
;
131 unsigned long vdso_base
;
138 if (is_32bit_task()) {
139 vdso_pagelist
= vdso32_pagelist
;
140 vdso_pages
= vdso32_pages
;
141 vdso_base
= VDSO32_MBASE
;
143 vdso_pagelist
= vdso64_pagelist
;
144 vdso_pages
= vdso64_pages
;
146 * On 64bit we don't have a preferred map address. This
147 * allows get_unmapped_area to find an area near other mmaps
148 * and most likely share a SLB entry.
153 vdso_pagelist
= vdso32_pagelist
;
154 vdso_pages
= vdso32_pages
;
155 vdso_base
= VDSO32_MBASE
;
158 current
->mm
->context
.vdso_base
= 0;
160 /* vDSO has a problem and was disabled, just don't "enable" it for the
165 /* Add a page to the vdso size for the data page */
169 * pick a base address for the vDSO in process space. We try to put it
170 * at vdso_base which is the "natural" base for it, but we might fail
171 * and end up putting it elsewhere.
172 * Add enough to the size so that the result can be aligned.
174 if (down_write_killable(&mm
->mmap_sem
))
176 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
177 (vdso_pages
<< PAGE_SHIFT
) +
178 ((VDSO_ALIGNMENT
- 1) & PAGE_MASK
),
180 if (IS_ERR_VALUE(vdso_base
)) {
185 /* Add required alignment. */
186 vdso_base
= ALIGN(vdso_base
, VDSO_ALIGNMENT
);
189 * Put vDSO base into mm struct. We need to do this before calling
190 * install_special_mapping or the perf counter mmap tracking code
191 * will fail to recognise it as a vDSO (since arch_vma_name fails).
193 current
->mm
->context
.vdso_base
= vdso_base
;
196 * our vma flags don't have VM_WRITE so by default, the process isn't
197 * allowed to write those pages.
198 * gdb can break that with ptrace interface, and thus trigger COW on
199 * those pages but it's then your responsibility to never do that on
200 * the "data" page of the vDSO or you'll stop getting kernel updates
201 * and your nice userland gettimeofday will be totally dead.
202 * It's fine to use that for setting breakpoints in the vDSO code
205 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
207 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
,
210 current
->mm
->context
.vdso_base
= 0;
214 up_write(&mm
->mmap_sem
);
218 up_write(&mm
->mmap_sem
);
222 const char *arch_vma_name(struct vm_area_struct
*vma
)
224 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
232 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
239 /* Grab section headers and strings so we can tell who is who */
240 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
241 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
243 /* Find the section they want */
244 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
245 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
247 *size
= sechdrs
[i
].sh_size
;
248 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
255 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
259 char name
[MAX_SYMNAME
], *c
;
261 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
262 if (lib
->dynsym
[i
].st_name
== 0)
264 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
266 c
= strchr(name
, '@');
269 if (strcmp(symname
, name
) == 0)
270 return &lib
->dynsym
[i
];
275 /* Note that we assume the section is .text and the symbol is relative to
278 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
281 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
284 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
288 return sym
->st_value
- VDSO32_LBASE
;
291 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
292 struct lib64_elfinfo
*v64
,
293 const char *orig
, const char *fix
)
295 Elf32_Sym
*sym32_gen
, *sym32_fix
;
297 sym32_gen
= find_symbol32(v32
, orig
);
298 if (sym32_gen
== NULL
) {
299 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
303 sym32_gen
->st_name
= 0;
306 sym32_fix
= find_symbol32(v32
, fix
);
307 if (sym32_fix
== NULL
) {
308 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
311 sym32_gen
->st_value
= sym32_fix
->st_value
;
312 sym32_gen
->st_size
= sym32_fix
->st_size
;
313 sym32_gen
->st_info
= sym32_fix
->st_info
;
314 sym32_gen
->st_other
= sym32_fix
->st_other
;
315 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
319 #else /* !CONFIG_VDSO32 */
320 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
326 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
327 struct lib64_elfinfo
*v64
,
328 const char *orig
, const char *fix
)
332 #endif /* CONFIG_VDSO32 */
337 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
344 /* Grab section headers and strings so we can tell who is who */
345 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
346 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
348 /* Find the section they want */
349 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
350 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
352 *size
= sechdrs
[i
].sh_size
;
353 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
361 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
365 char name
[MAX_SYMNAME
], *c
;
367 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
368 if (lib
->dynsym
[i
].st_name
== 0)
370 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
372 c
= strchr(name
, '@');
375 if (strcmp(symname
, name
) == 0)
376 return &lib
->dynsym
[i
];
381 /* Note that we assume the section is .text and the symbol is relative to
384 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
387 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
390 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
394 #ifdef VDS64_HAS_DESCRIPTORS
395 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
398 return sym
->st_value
- VDSO64_LBASE
;
402 static int __init
vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
403 struct lib64_elfinfo
*v64
,
404 const char *orig
, const char *fix
)
406 Elf64_Sym
*sym64_gen
, *sym64_fix
;
408 sym64_gen
= find_symbol64(v64
, orig
);
409 if (sym64_gen
== NULL
) {
410 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
414 sym64_gen
->st_name
= 0;
417 sym64_fix
= find_symbol64(v64
, fix
);
418 if (sym64_fix
== NULL
) {
419 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
422 sym64_gen
->st_value
= sym64_fix
->st_value
;
423 sym64_gen
->st_size
= sym64_fix
->st_size
;
424 sym64_gen
->st_info
= sym64_fix
->st_info
;
425 sym64_gen
->st_other
= sym64_fix
->st_other
;
426 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
431 #endif /* CONFIG_PPC64 */
434 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
435 struct lib64_elfinfo
*v64
)
440 * Locate symbol tables & text section
444 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
445 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
446 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
447 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
450 sect
= find_section32(v32
->hdr
, ".text", NULL
);
452 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
455 v32
->text
= sect
- vdso32_kbase
;
459 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
460 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
461 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
462 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
465 sect
= find_section64(v64
->hdr
, ".text", NULL
);
467 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
470 v64
->text
= sect
- vdso64_kbase
;
471 #endif /* CONFIG_PPC64 */
476 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
477 struct lib64_elfinfo
*v64
)
480 * Find signal trampolines
484 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
486 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
487 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
490 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
491 struct lib64_elfinfo
*v64
)
499 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
501 printk(KERN_ERR
"vDSO64: Can't find symbol "
502 "__kernel_datapage_offset !\n");
505 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
506 (vdso64_pages
<< PAGE_SHIFT
) -
507 (sym64
->st_value
- VDSO64_LBASE
);
508 #endif /* CONFIG_PPC64 */
511 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
513 printk(KERN_ERR
"vDSO32: Can't find symbol "
514 "__kernel_datapage_offset !\n");
517 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
518 (vdso32_pages
<< PAGE_SHIFT
) -
519 (sym32
->st_value
- VDSO32_LBASE
);
526 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
527 struct lib64_elfinfo
*v64
)
533 start
= find_section64(v64
->hdr
, "__ftr_fixup", &size
);
535 do_feature_fixups(cur_cpu_spec
->cpu_features
,
536 start
, start
+ size
);
538 start
= find_section64(v64
->hdr
, "__mmu_ftr_fixup", &size
);
540 do_feature_fixups(cur_cpu_spec
->mmu_features
,
541 start
, start
+ size
);
543 start
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size
);
545 do_feature_fixups(powerpc_firmware_features
,
546 start
, start
+ size
);
548 start
= find_section64(v64
->hdr
, "__lwsync_fixup", &size
);
550 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
551 start
, start
+ size
);
552 #endif /* CONFIG_PPC64 */
555 start
= find_section32(v32
->hdr
, "__ftr_fixup", &size
);
557 do_feature_fixups(cur_cpu_spec
->cpu_features
,
558 start
, start
+ size
);
560 start
= find_section32(v32
->hdr
, "__mmu_ftr_fixup", &size
);
562 do_feature_fixups(cur_cpu_spec
->mmu_features
,
563 start
, start
+ size
);
566 start
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size
);
568 do_feature_fixups(powerpc_firmware_features
,
569 start
, start
+ size
);
570 #endif /* CONFIG_PPC64 */
572 start
= find_section32(v32
->hdr
, "__lwsync_fixup", &size
);
574 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
575 start
, start
+ size
);
581 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
582 struct lib64_elfinfo
*v64
)
586 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
587 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
588 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
593 DBG("replacing %s with %s...\n", patch
->gen_name
,
594 patch
->fix_name
? "NONE" : patch
->fix_name
);
597 * Patch the 32 bits and 64 bits symbols. Note that we do not
598 * patch the "." symbol on 64 bits.
599 * It would be easy to do, but doesn't seem to be necessary,
600 * patching the OPD symbol is enough.
602 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
605 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
607 #endif /* CONFIG_PPC64 */
614 static __init
int vdso_setup(void)
616 struct lib32_elfinfo v32
;
617 struct lib64_elfinfo v64
;
619 v32
.hdr
= vdso32_kbase
;
621 v64
.hdr
= vdso64_kbase
;
623 if (vdso_do_find_sections(&v32
, &v64
))
626 if (vdso_fixup_datapage(&v32
, &v64
))
629 if (vdso_fixup_features(&v32
, &v64
))
632 if (vdso_fixup_alt_funcs(&v32
, &v64
))
635 vdso_setup_trampolines(&v32
, &v64
);
641 * Called from setup_arch to initialize the bitmap of available
642 * syscalls in the systemcfg page
644 static void __init
vdso_setup_syscall_map(void)
647 extern unsigned long *sys_call_table
;
649 extern unsigned long *compat_sys_call_table
;
651 extern unsigned long sys_ni_syscall
;
654 for (i
= 0; i
< NR_syscalls
; i
++) {
656 if (sys_call_table
[i
] != sys_ni_syscall
)
657 vdso_data
->syscall_map_64
[i
>> 5] |=
658 0x80000000UL
>> (i
& 0x1f);
659 if (compat_sys_call_table
[i
] != sys_ni_syscall
)
660 vdso_data
->syscall_map_32
[i
>> 5] |=
661 0x80000000UL
>> (i
& 0x1f);
662 #else /* CONFIG_PPC64 */
663 if (sys_call_table
[i
] != sys_ni_syscall
)
664 vdso_data
->syscall_map_32
[i
>> 5] |=
665 0x80000000UL
>> (i
& 0x1f);
666 #endif /* CONFIG_PPC64 */
671 int vdso_getcpu_init(void)
673 unsigned long cpu
, node
, val
;
676 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
677 * in the next 16 bits. The VDSO uses this to implement getcpu().
680 WARN_ON_ONCE(cpu
> 0xffff);
682 node
= cpu_to_node(cpu
);
683 WARN_ON_ONCE(node
> 0xffff);
685 val
= (cpu
& 0xfff) | ((node
& 0xffff) << 16);
686 mtspr(SPRN_SPRG_VDSO_WRITE
, val
);
687 get_paca()->sprg_vdso
= val
;
693 /* We need to call this before SMP init */
694 early_initcall(vdso_getcpu_init
);
697 static int __init
vdso_init(void)
703 * Fill up the "systemcfg" stuff for backward compatibility
705 strcpy((char *)vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
706 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
707 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
708 vdso_data
->processor
= mfspr(SPRN_PVR
);
710 * Fake the old platform number for pSeries and add
711 * in LPAR bit if necessary
713 vdso_data
->platform
= 0x100;
714 if (firmware_has_feature(FW_FEATURE_LPAR
))
715 vdso_data
->platform
|= 1;
716 vdso_data
->physicalMemorySize
= memblock_phys_mem_size();
717 vdso_data
->dcache_size
= ppc64_caches
.l1d
.size
;
718 vdso_data
->dcache_line_size
= ppc64_caches
.l1d
.line_size
;
719 vdso_data
->icache_size
= ppc64_caches
.l1i
.size
;
720 vdso_data
->icache_line_size
= ppc64_caches
.l1i
.line_size
;
721 vdso_data
->dcache_block_size
= ppc64_caches
.l1d
.block_size
;
722 vdso_data
->icache_block_size
= ppc64_caches
.l1i
.block_size
;
723 vdso_data
->dcache_log_block_size
= ppc64_caches
.l1d
.log_block_size
;
724 vdso_data
->icache_log_block_size
= ppc64_caches
.l1i
.log_block_size
;
727 * Calculate the size of the 64 bits vDSO
729 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
730 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
732 vdso_data
->dcache_block_size
= L1_CACHE_BYTES
;
733 vdso_data
->dcache_log_block_size
= L1_CACHE_SHIFT
;
734 vdso_data
->icache_block_size
= L1_CACHE_BYTES
;
735 vdso_data
->icache_log_block_size
= L1_CACHE_SHIFT
;
736 #endif /* CONFIG_PPC64 */
740 vdso32_kbase
= &vdso32_start
;
743 * Calculate the size of the 32 bits vDSO
745 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
746 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
751 * Setup the syscall map in the vDOS
753 vdso_setup_syscall_map();
756 * Initialize the vDSO images in memory, that is do necessary
757 * fixups of vDSO symbols, locate trampolines, etc...
760 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
769 /* Make sure pages are in the correct state */
770 vdso32_pagelist
= kcalloc(vdso32_pages
+ 2, sizeof(struct page
*),
772 BUG_ON(vdso32_pagelist
== NULL
);
773 for (i
= 0; i
< vdso32_pages
; i
++) {
774 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
776 vdso32_pagelist
[i
] = pg
;
778 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
779 vdso32_pagelist
[i
] = NULL
;
783 vdso64_pagelist
= kcalloc(vdso64_pages
+ 2, sizeof(struct page
*),
785 BUG_ON(vdso64_pagelist
== NULL
);
786 for (i
= 0; i
< vdso64_pages
; i
++) {
787 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
789 vdso64_pagelist
[i
] = pg
;
791 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
792 vdso64_pagelist
[i
] = NULL
;
793 #endif /* CONFIG_PPC64 */
795 get_page(virt_to_page(vdso_data
));
802 arch_initcall(vdso_init
);