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"
99 CPU_FTR_USE_RTC
, CPU_FTR_USE_RTC
,
100 "__kernel_gettimeofday", NULL
103 CPU_FTR_USE_RTC
, CPU_FTR_USE_RTC
,
104 "__kernel_clock_gettime", NULL
107 CPU_FTR_USE_RTC
, CPU_FTR_USE_RTC
,
108 "__kernel_clock_getres", NULL
111 CPU_FTR_USE_RTC
, CPU_FTR_USE_RTC
,
112 "__kernel_get_tbfreq", NULL
115 CPU_FTR_USE_RTC
, CPU_FTR_USE_RTC
,
116 "__kernel_time", NULL
122 * Some infos carried around for each of them during parsing at
127 Elf32_Ehdr
*hdr
; /* ptr to ELF */
128 Elf32_Sym
*dynsym
; /* ptr to .dynsym section */
129 unsigned long dynsymsize
; /* size of .dynsym section */
130 char *dynstr
; /* ptr to .dynstr section */
131 unsigned long text
; /* offset of .text section in .so */
138 unsigned long dynsymsize
;
145 * This is called from binfmt_elf, we create the special vma for the
146 * vDSO and insert it into the mm struct tree
148 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int uses_interp
)
150 struct mm_struct
*mm
= current
->mm
;
151 struct page
**vdso_pagelist
;
152 unsigned long vdso_pages
;
153 unsigned long vdso_base
;
160 if (is_32bit_task()) {
161 vdso_pagelist
= vdso32_pagelist
;
162 vdso_pages
= vdso32_pages
;
163 vdso_base
= VDSO32_MBASE
;
165 vdso_pagelist
= vdso64_pagelist
;
166 vdso_pages
= vdso64_pages
;
168 * On 64bit we don't have a preferred map address. This
169 * allows get_unmapped_area to find an area near other mmaps
170 * and most likely share a SLB entry.
175 vdso_pagelist
= vdso32_pagelist
;
176 vdso_pages
= vdso32_pages
;
177 vdso_base
= VDSO32_MBASE
;
180 current
->mm
->context
.vdso_base
= 0;
182 /* vDSO has a problem and was disabled, just don't "enable" it for the
187 /* Add a page to the vdso size for the data page */
191 * pick a base address for the vDSO in process space. We try to put it
192 * at vdso_base which is the "natural" base for it, but we might fail
193 * and end up putting it elsewhere.
194 * Add enough to the size so that the result can be aligned.
196 if (down_write_killable(&mm
->mmap_sem
))
198 vdso_base
= get_unmapped_area(NULL
, vdso_base
,
199 (vdso_pages
<< PAGE_SHIFT
) +
200 ((VDSO_ALIGNMENT
- 1) & PAGE_MASK
),
202 if (IS_ERR_VALUE(vdso_base
)) {
207 /* Add required alignment. */
208 vdso_base
= ALIGN(vdso_base
, VDSO_ALIGNMENT
);
211 * Put vDSO base into mm struct. We need to do this before calling
212 * install_special_mapping or the perf counter mmap tracking code
213 * will fail to recognise it as a vDSO (since arch_vma_name fails).
215 current
->mm
->context
.vdso_base
= vdso_base
;
218 * our vma flags don't have VM_WRITE so by default, the process isn't
219 * allowed to write those pages.
220 * gdb can break that with ptrace interface, and thus trigger COW on
221 * those pages but it's then your responsibility to never do that on
222 * the "data" page of the vDSO or you'll stop getting kernel updates
223 * and your nice userland gettimeofday will be totally dead.
224 * It's fine to use that for setting breakpoints in the vDSO code
227 rc
= install_special_mapping(mm
, vdso_base
, vdso_pages
<< PAGE_SHIFT
,
229 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
,
232 current
->mm
->context
.vdso_base
= 0;
236 up_write(&mm
->mmap_sem
);
240 up_write(&mm
->mmap_sem
);
244 const char *arch_vma_name(struct vm_area_struct
*vma
)
246 if (vma
->vm_mm
&& vma
->vm_start
== vma
->vm_mm
->context
.vdso_base
)
254 static void * __init
find_section32(Elf32_Ehdr
*ehdr
, const char *secname
,
261 /* Grab section headers and strings so we can tell who is who */
262 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
263 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
265 /* Find the section they want */
266 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
267 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
269 *size
= sechdrs
[i
].sh_size
;
270 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
277 static Elf32_Sym
* __init
find_symbol32(struct lib32_elfinfo
*lib
,
281 char name
[MAX_SYMNAME
], *c
;
283 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf32_Sym
)); i
++) {
284 if (lib
->dynsym
[i
].st_name
== 0)
286 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
288 c
= strchr(name
, '@');
291 if (strcmp(symname
, name
) == 0)
292 return &lib
->dynsym
[i
];
297 /* Note that we assume the section is .text and the symbol is relative to
300 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
303 Elf32_Sym
*sym
= find_symbol32(lib
, symname
);
306 printk(KERN_WARNING
"vDSO32: function %s not found !\n",
310 return sym
->st_value
- VDSO32_LBASE
;
313 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
314 struct lib64_elfinfo
*v64
,
315 const char *orig
, const char *fix
)
317 Elf32_Sym
*sym32_gen
, *sym32_fix
;
319 sym32_gen
= find_symbol32(v32
, orig
);
320 if (sym32_gen
== NULL
) {
321 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", orig
);
325 sym32_gen
->st_name
= 0;
328 sym32_fix
= find_symbol32(v32
, fix
);
329 if (sym32_fix
== NULL
) {
330 printk(KERN_ERR
"vDSO32: Can't find symbol %s !\n", fix
);
333 sym32_gen
->st_value
= sym32_fix
->st_value
;
334 sym32_gen
->st_size
= sym32_fix
->st_size
;
335 sym32_gen
->st_info
= sym32_fix
->st_info
;
336 sym32_gen
->st_other
= sym32_fix
->st_other
;
337 sym32_gen
->st_shndx
= sym32_fix
->st_shndx
;
341 #else /* !CONFIG_VDSO32 */
342 static unsigned long __init
find_function32(struct lib32_elfinfo
*lib
,
348 static int __init
vdso_do_func_patch32(struct lib32_elfinfo
*v32
,
349 struct lib64_elfinfo
*v64
,
350 const char *orig
, const char *fix
)
354 #endif /* CONFIG_VDSO32 */
359 static void * __init
find_section64(Elf64_Ehdr
*ehdr
, const char *secname
,
366 /* Grab section headers and strings so we can tell who is who */
367 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
368 secnames
= (void *)ehdr
+ sechdrs
[ehdr
->e_shstrndx
].sh_offset
;
370 /* Find the section they want */
371 for (i
= 1; i
< ehdr
->e_shnum
; i
++) {
372 if (strcmp(secnames
+sechdrs
[i
].sh_name
, secname
) == 0) {
374 *size
= sechdrs
[i
].sh_size
;
375 return (void *)ehdr
+ sechdrs
[i
].sh_offset
;
383 static Elf64_Sym
* __init
find_symbol64(struct lib64_elfinfo
*lib
,
387 char name
[MAX_SYMNAME
], *c
;
389 for (i
= 0; i
< (lib
->dynsymsize
/ sizeof(Elf64_Sym
)); i
++) {
390 if (lib
->dynsym
[i
].st_name
== 0)
392 strlcpy(name
, lib
->dynstr
+ lib
->dynsym
[i
].st_name
,
394 c
= strchr(name
, '@');
397 if (strcmp(symname
, name
) == 0)
398 return &lib
->dynsym
[i
];
403 /* Note that we assume the section is .text and the symbol is relative to
406 static unsigned long __init
find_function64(struct lib64_elfinfo
*lib
,
409 Elf64_Sym
*sym
= find_symbol64(lib
, symname
);
412 printk(KERN_WARNING
"vDSO64: function %s not found !\n",
416 #ifdef VDS64_HAS_DESCRIPTORS
417 return *((u64
*)(vdso64_kbase
+ sym
->st_value
- VDSO64_LBASE
)) -
420 return sym
->st_value
- VDSO64_LBASE
;
424 static int __init
vdso_do_func_patch64(struct lib32_elfinfo
*v32
,
425 struct lib64_elfinfo
*v64
,
426 const char *orig
, const char *fix
)
428 Elf64_Sym
*sym64_gen
, *sym64_fix
;
430 sym64_gen
= find_symbol64(v64
, orig
);
431 if (sym64_gen
== NULL
) {
432 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", orig
);
436 sym64_gen
->st_name
= 0;
439 sym64_fix
= find_symbol64(v64
, fix
);
440 if (sym64_fix
== NULL
) {
441 printk(KERN_ERR
"vDSO64: Can't find symbol %s !\n", fix
);
444 sym64_gen
->st_value
= sym64_fix
->st_value
;
445 sym64_gen
->st_size
= sym64_fix
->st_size
;
446 sym64_gen
->st_info
= sym64_fix
->st_info
;
447 sym64_gen
->st_other
= sym64_fix
->st_other
;
448 sym64_gen
->st_shndx
= sym64_fix
->st_shndx
;
453 #endif /* CONFIG_PPC64 */
456 static __init
int vdso_do_find_sections(struct lib32_elfinfo
*v32
,
457 struct lib64_elfinfo
*v64
)
462 * Locate symbol tables & text section
466 v32
->dynsym
= find_section32(v32
->hdr
, ".dynsym", &v32
->dynsymsize
);
467 v32
->dynstr
= find_section32(v32
->hdr
, ".dynstr", NULL
);
468 if (v32
->dynsym
== NULL
|| v32
->dynstr
== NULL
) {
469 printk(KERN_ERR
"vDSO32: required symbol section not found\n");
472 sect
= find_section32(v32
->hdr
, ".text", NULL
);
474 printk(KERN_ERR
"vDSO32: the .text section was not found\n");
477 v32
->text
= sect
- vdso32_kbase
;
481 v64
->dynsym
= find_section64(v64
->hdr
, ".dynsym", &v64
->dynsymsize
);
482 v64
->dynstr
= find_section64(v64
->hdr
, ".dynstr", NULL
);
483 if (v64
->dynsym
== NULL
|| v64
->dynstr
== NULL
) {
484 printk(KERN_ERR
"vDSO64: required symbol section not found\n");
487 sect
= find_section64(v64
->hdr
, ".text", NULL
);
489 printk(KERN_ERR
"vDSO64: the .text section was not found\n");
492 v64
->text
= sect
- vdso64_kbase
;
493 #endif /* CONFIG_PPC64 */
498 static __init
void vdso_setup_trampolines(struct lib32_elfinfo
*v32
,
499 struct lib64_elfinfo
*v64
)
502 * Find signal trampolines
506 vdso64_rt_sigtramp
= find_function64(v64
, "__kernel_sigtramp_rt64");
508 vdso32_sigtramp
= find_function32(v32
, "__kernel_sigtramp32");
509 vdso32_rt_sigtramp
= find_function32(v32
, "__kernel_sigtramp_rt32");
512 static __init
int vdso_fixup_datapage(struct lib32_elfinfo
*v32
,
513 struct lib64_elfinfo
*v64
)
521 sym64
= find_symbol64(v64
, "__kernel_datapage_offset");
523 printk(KERN_ERR
"vDSO64: Can't find symbol "
524 "__kernel_datapage_offset !\n");
527 *((int *)(vdso64_kbase
+ sym64
->st_value
- VDSO64_LBASE
)) =
528 (vdso64_pages
<< PAGE_SHIFT
) -
529 (sym64
->st_value
- VDSO64_LBASE
);
530 #endif /* CONFIG_PPC64 */
533 sym32
= find_symbol32(v32
, "__kernel_datapage_offset");
535 printk(KERN_ERR
"vDSO32: Can't find symbol "
536 "__kernel_datapage_offset !\n");
539 *((int *)(vdso32_kbase
+ (sym32
->st_value
- VDSO32_LBASE
))) =
540 (vdso32_pages
<< PAGE_SHIFT
) -
541 (sym32
->st_value
- VDSO32_LBASE
);
548 static __init
int vdso_fixup_features(struct lib32_elfinfo
*v32
,
549 struct lib64_elfinfo
*v64
)
555 start
= find_section64(v64
->hdr
, "__ftr_fixup", &size
);
557 do_feature_fixups(cur_cpu_spec
->cpu_features
,
558 start
, start
+ size
);
560 start
= find_section64(v64
->hdr
, "__mmu_ftr_fixup", &size
);
562 do_feature_fixups(cur_cpu_spec
->mmu_features
,
563 start
, start
+ size
);
565 start
= find_section64(v64
->hdr
, "__fw_ftr_fixup", &size
);
567 do_feature_fixups(powerpc_firmware_features
,
568 start
, start
+ size
);
570 start
= find_section64(v64
->hdr
, "__lwsync_fixup", &size
);
572 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
573 start
, start
+ size
);
574 #endif /* CONFIG_PPC64 */
577 start
= find_section32(v32
->hdr
, "__ftr_fixup", &size
);
579 do_feature_fixups(cur_cpu_spec
->cpu_features
,
580 start
, start
+ size
);
582 start
= find_section32(v32
->hdr
, "__mmu_ftr_fixup", &size
);
584 do_feature_fixups(cur_cpu_spec
->mmu_features
,
585 start
, start
+ size
);
588 start
= find_section32(v32
->hdr
, "__fw_ftr_fixup", &size
);
590 do_feature_fixups(powerpc_firmware_features
,
591 start
, start
+ size
);
592 #endif /* CONFIG_PPC64 */
594 start
= find_section32(v32
->hdr
, "__lwsync_fixup", &size
);
596 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
597 start
, start
+ size
);
603 static __init
int vdso_fixup_alt_funcs(struct lib32_elfinfo
*v32
,
604 struct lib64_elfinfo
*v64
)
608 for (i
= 0; i
< ARRAY_SIZE(vdso_patches
); i
++) {
609 struct vdso_patch_def
*patch
= &vdso_patches
[i
];
610 int match
= (cur_cpu_spec
->cpu_features
& patch
->ftr_mask
)
615 DBG("replacing %s with %s...\n", patch
->gen_name
,
616 patch
->fix_name
? "NONE" : patch
->fix_name
);
619 * Patch the 32 bits and 64 bits symbols. Note that we do not
620 * patch the "." symbol on 64 bits.
621 * It would be easy to do, but doesn't seem to be necessary,
622 * patching the OPD symbol is enough.
624 vdso_do_func_patch32(v32
, v64
, patch
->gen_name
,
627 vdso_do_func_patch64(v32
, v64
, patch
->gen_name
,
629 #endif /* CONFIG_PPC64 */
636 static __init
int vdso_setup(void)
638 struct lib32_elfinfo v32
;
639 struct lib64_elfinfo v64
;
641 v32
.hdr
= vdso32_kbase
;
643 v64
.hdr
= vdso64_kbase
;
645 if (vdso_do_find_sections(&v32
, &v64
))
648 if (vdso_fixup_datapage(&v32
, &v64
))
651 if (vdso_fixup_features(&v32
, &v64
))
654 if (vdso_fixup_alt_funcs(&v32
, &v64
))
657 vdso_setup_trampolines(&v32
, &v64
);
663 * Called from setup_arch to initialize the bitmap of available
664 * syscalls in the systemcfg page
666 static void __init
vdso_setup_syscall_map(void)
669 extern unsigned long *sys_call_table
;
671 extern unsigned long *compat_sys_call_table
;
673 extern unsigned long sys_ni_syscall
;
676 for (i
= 0; i
< NR_syscalls
; i
++) {
678 if (sys_call_table
[i
] != sys_ni_syscall
)
679 vdso_data
->syscall_map_64
[i
>> 5] |=
680 0x80000000UL
>> (i
& 0x1f);
681 if (compat_sys_call_table
[i
] != sys_ni_syscall
)
682 vdso_data
->syscall_map_32
[i
>> 5] |=
683 0x80000000UL
>> (i
& 0x1f);
684 #else /* CONFIG_PPC64 */
685 if (sys_call_table
[i
] != sys_ni_syscall
)
686 vdso_data
->syscall_map_32
[i
>> 5] |=
687 0x80000000UL
>> (i
& 0x1f);
688 #endif /* CONFIG_PPC64 */
693 int vdso_getcpu_init(void)
695 unsigned long cpu
, node
, val
;
698 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
699 * in the next 16 bits. The VDSO uses this to implement getcpu().
702 WARN_ON_ONCE(cpu
> 0xffff);
704 node
= cpu_to_node(cpu
);
705 WARN_ON_ONCE(node
> 0xffff);
707 val
= (cpu
& 0xfff) | ((node
& 0xffff) << 16);
708 mtspr(SPRN_SPRG_VDSO_WRITE
, val
);
709 get_paca()->sprg_vdso
= val
;
715 /* We need to call this before SMP init */
716 early_initcall(vdso_getcpu_init
);
719 static int __init
vdso_init(void)
725 * Fill up the "systemcfg" stuff for backward compatibility
727 strcpy((char *)vdso_data
->eye_catcher
, "SYSTEMCFG:PPC64");
728 vdso_data
->version
.major
= SYSTEMCFG_MAJOR
;
729 vdso_data
->version
.minor
= SYSTEMCFG_MINOR
;
730 vdso_data
->processor
= mfspr(SPRN_PVR
);
732 * Fake the old platform number for pSeries and add
733 * in LPAR bit if necessary
735 vdso_data
->platform
= 0x100;
736 if (firmware_has_feature(FW_FEATURE_LPAR
))
737 vdso_data
->platform
|= 1;
738 vdso_data
->physicalMemorySize
= memblock_phys_mem_size();
739 vdso_data
->dcache_size
= ppc64_caches
.l1d
.size
;
740 vdso_data
->dcache_line_size
= ppc64_caches
.l1d
.line_size
;
741 vdso_data
->icache_size
= ppc64_caches
.l1i
.size
;
742 vdso_data
->icache_line_size
= ppc64_caches
.l1i
.line_size
;
743 vdso_data
->dcache_block_size
= ppc64_caches
.l1d
.block_size
;
744 vdso_data
->icache_block_size
= ppc64_caches
.l1i
.block_size
;
745 vdso_data
->dcache_log_block_size
= ppc64_caches
.l1d
.log_block_size
;
746 vdso_data
->icache_log_block_size
= ppc64_caches
.l1i
.log_block_size
;
749 * Calculate the size of the 64 bits vDSO
751 vdso64_pages
= (&vdso64_end
- &vdso64_start
) >> PAGE_SHIFT
;
752 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase
, vdso64_pages
);
754 vdso_data
->dcache_block_size
= L1_CACHE_BYTES
;
755 vdso_data
->dcache_log_block_size
= L1_CACHE_SHIFT
;
756 vdso_data
->icache_block_size
= L1_CACHE_BYTES
;
757 vdso_data
->icache_log_block_size
= L1_CACHE_SHIFT
;
758 #endif /* CONFIG_PPC64 */
762 vdso32_kbase
= &vdso32_start
;
765 * Calculate the size of the 32 bits vDSO
767 vdso32_pages
= (&vdso32_end
- &vdso32_start
) >> PAGE_SHIFT
;
768 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase
, vdso32_pages
);
773 * Setup the syscall map in the vDOS
775 vdso_setup_syscall_map();
778 * Initialize the vDSO images in memory, that is do necessary
779 * fixups of vDSO symbols, locate trampolines, etc...
782 printk(KERN_ERR
"vDSO setup failure, not enabled !\n");
791 /* Make sure pages are in the correct state */
792 vdso32_pagelist
= kcalloc(vdso32_pages
+ 2, sizeof(struct page
*),
794 BUG_ON(vdso32_pagelist
== NULL
);
795 for (i
= 0; i
< vdso32_pages
; i
++) {
796 struct page
*pg
= virt_to_page(vdso32_kbase
+ i
*PAGE_SIZE
);
798 vdso32_pagelist
[i
] = pg
;
800 vdso32_pagelist
[i
++] = virt_to_page(vdso_data
);
801 vdso32_pagelist
[i
] = NULL
;
805 vdso64_pagelist
= kcalloc(vdso64_pages
+ 2, sizeof(struct page
*),
807 BUG_ON(vdso64_pagelist
== NULL
);
808 for (i
= 0; i
< vdso64_pages
; i
++) {
809 struct page
*pg
= virt_to_page(vdso64_kbase
+ i
*PAGE_SIZE
);
811 vdso64_pagelist
[i
] = pg
;
813 vdso64_pagelist
[i
++] = virt_to_page(vdso_data
);
814 vdso64_pagelist
[i
] = NULL
;
815 #endif /* CONFIG_PPC64 */
817 get_page(virt_to_page(vdso_data
));
824 arch_initcall(vdso_init
);