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/module.h>
13 #include <linux/errno.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
17 #include <linux/smp.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/slab.h>
21 #include <linux/user.h>
22 #include <linux/elf.h>
23 #include <linux/security.h>
24 #include <linux/bootmem.h>
25 #include <linux/memblock.h>
27 #include <asm/pgtable.h>
28 #include <asm/system.h>
29 #include <asm/processor.h>
31 #include <asm/mmu_context.h>
33 #include <asm/machdep.h>
34 #include <asm/cputable.h>
35 #include <asm/sections.h>
36 #include <asm/firmware.h>
38 #include <asm/vdso_datapage.h>
45 #define DBG(fmt...) printk(fmt)
50 /* Max supported size for symbol names */
51 #define MAX_SYMNAME 64
53 /* The alignment of the vDSO */
54 #define VDSO_ALIGNMENT (1 << 16)
56 extern char vdso32_start, vdso32_end;
57 static void *vdso32_kbase = &vdso32_start;
58 static unsigned int vdso32_pages;
59 static struct page **vdso32_pagelist;
60 unsigned long vdso32_sigtramp;
61 unsigned long vdso32_rt_sigtramp;
64 extern char vdso64_start, vdso64_end;
65 static void *vdso64_kbase = &vdso64_start;
66 static unsigned int vdso64_pages;
67 static struct page **vdso64_pagelist;
68 unsigned long vdso64_rt_sigtramp;
69 #endif /* CONFIG_PPC64 */
71 static int vdso_ready;
74 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
75 * Once the early boot kernel code no longer needs to muck around
76 * with it, it will become dynamically allocated
79 struct vdso_data data;
81 } vdso_data_store __page_aligned_data;
82 struct vdso_data *vdso_data = &vdso_data_store.data;
84 /* Format of the patch table */
87 unsigned long ftr_mask, ftr_value;
92 /* Table of functions to patch based on the CPU type/revision
94 * Currently, we only change sync_dicache to do nothing on processors
95 * with a coherent icache
97 static struct vdso_patch_def vdso_patches[] = {
99 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
100 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
104 "__kernel_gettimeofday", NULL
108 "__kernel_clock_gettime", NULL
112 "__kernel_clock_getres", NULL
116 "__kernel_get_tbfreq", NULL
121 * Some infos carried around for each of them during parsing at
126 Elf32_Ehdr *hdr; /* ptr to ELF */
127 Elf32_Sym *dynsym; /* ptr to .dynsym section */
128 unsigned long dynsymsize; /* size of .dynsym section */
129 char *dynstr; /* ptr to .dynstr section */
130 unsigned long text; /* offset of .text section in .so */
137 unsigned long dynsymsize;
144 static void dump_one_vdso_page(struct page *pg, struct page *upg)
146 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
149 if (upg && !IS_ERR(upg) /* && pg != upg*/) {
150 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
158 static void dump_vdso_pages(struct vm_area_struct * vma)
162 if (!vma || is_32bit_task()) {
163 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
164 for (i=0; i<vdso32_pages; i++) {
165 struct page *pg = virt_to_page(vdso32_kbase +
167 struct page *upg = (vma && vma->vm_mm) ?
168 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
170 dump_one_vdso_page(pg, upg);
173 if (!vma || !is_32bit_task()) {
174 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
175 for (i=0; i<vdso64_pages; i++) {
176 struct page *pg = virt_to_page(vdso64_kbase +
178 struct page *upg = (vma && vma->vm_mm) ?
179 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
181 dump_one_vdso_page(pg, upg);
188 * This is called from binfmt_elf, we create the special vma for the
189 * vDSO and insert it into the mm struct tree
191 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
193 struct mm_struct *mm = current->mm;
194 struct page **vdso_pagelist;
195 unsigned long vdso_pages;
196 unsigned long vdso_base;
203 if (is_32bit_task()) {
204 vdso_pagelist = vdso32_pagelist;
205 vdso_pages = vdso32_pages;
206 vdso_base = VDSO32_MBASE;
208 vdso_pagelist = vdso64_pagelist;
209 vdso_pages = vdso64_pages;
211 * On 64bit we don't have a preferred map address. This
212 * allows get_unmapped_area to find an area near other mmaps
213 * and most likely share a SLB entry.
218 vdso_pagelist = vdso32_pagelist;
219 vdso_pages = vdso32_pages;
220 vdso_base = VDSO32_MBASE;
223 current->mm->context.vdso_base = 0;
225 /* vDSO has a problem and was disabled, just don't "enable" it for the
230 /* Add a page to the vdso size for the data page */
234 * pick a base address for the vDSO in process space. We try to put it
235 * at vdso_base which is the "natural" base for it, but we might fail
236 * and end up putting it elsewhere.
237 * Add enough to the size so that the result can be aligned.
239 down_write(&mm->mmap_sem);
240 vdso_base = get_unmapped_area(NULL, vdso_base,
241 (vdso_pages << PAGE_SHIFT) +
242 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
244 if (IS_ERR_VALUE(vdso_base)) {
249 /* Add required alignment. */
250 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
253 * Put vDSO base into mm struct. We need to do this before calling
254 * install_special_mapping or the perf counter mmap tracking code
255 * will fail to recognise it as a vDSO (since arch_vma_name fails).
257 current->mm->context.vdso_base = vdso_base;
260 * our vma flags don't have VM_WRITE so by default, the process isn't
261 * allowed to write those pages.
262 * gdb can break that with ptrace interface, and thus trigger COW on
263 * those pages but it's then your responsibility to never do that on
264 * the "data" page of the vDSO or you'll stop getting kernel updates
265 * and your nice userland gettimeofday will be totally dead.
266 * It's fine to use that for setting breakpoints in the vDSO code
269 * Make sure the vDSO gets into every core dump.
270 * Dumping its contents makes post-mortem fully interpretable later
271 * without matching up the same kernel and hardware config to see
272 * what PC values meant.
274 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
276 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
280 current->mm->context.vdso_base = 0;
284 up_write(&mm->mmap_sem);
288 up_write(&mm->mmap_sem);
292 const char *arch_vma_name(struct vm_area_struct *vma)
294 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
301 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
308 /* Grab section headers and strings so we can tell who is who */
309 sechdrs = (void *)ehdr + ehdr->e_shoff;
310 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
312 /* Find the section they want */
313 for (i = 1; i < ehdr->e_shnum; i++) {
314 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
316 *size = sechdrs[i].sh_size;
317 return (void *)ehdr + sechdrs[i].sh_offset;
324 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
328 char name[MAX_SYMNAME], *c;
330 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
331 if (lib->dynsym[i].st_name == 0)
333 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
335 c = strchr(name, '@');
338 if (strcmp(symname, name) == 0)
339 return &lib->dynsym[i];
344 /* Note that we assume the section is .text and the symbol is relative to
347 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
350 Elf32_Sym *sym = find_symbol32(lib, symname);
353 printk(KERN_WARNING "vDSO32: function %s not found !\n",
357 return sym->st_value - VDSO32_LBASE;
360 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
361 struct lib64_elfinfo *v64,
362 const char *orig, const char *fix)
364 Elf32_Sym *sym32_gen, *sym32_fix;
366 sym32_gen = find_symbol32(v32, orig);
367 if (sym32_gen == NULL) {
368 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
372 sym32_gen->st_name = 0;
375 sym32_fix = find_symbol32(v32, fix);
376 if (sym32_fix == NULL) {
377 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
380 sym32_gen->st_value = sym32_fix->st_value;
381 sym32_gen->st_size = sym32_fix->st_size;
382 sym32_gen->st_info = sym32_fix->st_info;
383 sym32_gen->st_other = sym32_fix->st_other;
384 sym32_gen->st_shndx = sym32_fix->st_shndx;
392 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
399 /* Grab section headers and strings so we can tell who is who */
400 sechdrs = (void *)ehdr + ehdr->e_shoff;
401 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
403 /* Find the section they want */
404 for (i = 1; i < ehdr->e_shnum; i++) {
405 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
407 *size = sechdrs[i].sh_size;
408 return (void *)ehdr + sechdrs[i].sh_offset;
416 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
420 char name[MAX_SYMNAME], *c;
422 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
423 if (lib->dynsym[i].st_name == 0)
425 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
427 c = strchr(name, '@');
430 if (strcmp(symname, name) == 0)
431 return &lib->dynsym[i];
436 /* Note that we assume the section is .text and the symbol is relative to
439 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
442 Elf64_Sym *sym = find_symbol64(lib, symname);
445 printk(KERN_WARNING "vDSO64: function %s not found !\n",
449 #ifdef VDS64_HAS_DESCRIPTORS
450 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
453 return sym->st_value - VDSO64_LBASE;
457 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
458 struct lib64_elfinfo *v64,
459 const char *orig, const char *fix)
461 Elf64_Sym *sym64_gen, *sym64_fix;
463 sym64_gen = find_symbol64(v64, orig);
464 if (sym64_gen == NULL) {
465 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
469 sym64_gen->st_name = 0;
472 sym64_fix = find_symbol64(v64, fix);
473 if (sym64_fix == NULL) {
474 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
477 sym64_gen->st_value = sym64_fix->st_value;
478 sym64_gen->st_size = sym64_fix->st_size;
479 sym64_gen->st_info = sym64_fix->st_info;
480 sym64_gen->st_other = sym64_fix->st_other;
481 sym64_gen->st_shndx = sym64_fix->st_shndx;
486 #endif /* CONFIG_PPC64 */
489 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
490 struct lib64_elfinfo *v64)
495 * Locate symbol tables & text section
498 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
499 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
500 if (v32->dynsym == NULL || v32->dynstr == NULL) {
501 printk(KERN_ERR "vDSO32: required symbol section not found\n");
504 sect = find_section32(v32->hdr, ".text", NULL);
506 printk(KERN_ERR "vDSO32: the .text section was not found\n");
509 v32->text = sect - vdso32_kbase;
512 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
513 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
514 if (v64->dynsym == NULL || v64->dynstr == NULL) {
515 printk(KERN_ERR "vDSO64: required symbol section not found\n");
518 sect = find_section64(v64->hdr, ".text", NULL);
520 printk(KERN_ERR "vDSO64: the .text section was not found\n");
523 v64->text = sect - vdso64_kbase;
524 #endif /* CONFIG_PPC64 */
529 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
530 struct lib64_elfinfo *v64)
533 * Find signal trampolines
537 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
539 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
540 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
543 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
544 struct lib64_elfinfo *v64)
550 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
552 printk(KERN_ERR "vDSO64: Can't find symbol "
553 "__kernel_datapage_offset !\n");
556 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
557 (vdso64_pages << PAGE_SHIFT) -
558 (sym64->st_value - VDSO64_LBASE);
559 #endif /* CONFIG_PPC64 */
561 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
563 printk(KERN_ERR "vDSO32: Can't find symbol "
564 "__kernel_datapage_offset !\n");
567 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
568 (vdso32_pages << PAGE_SHIFT) -
569 (sym32->st_value - VDSO32_LBASE);
575 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
576 struct lib64_elfinfo *v64)
579 unsigned long size32;
583 unsigned long size64;
585 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
587 do_feature_fixups(cur_cpu_spec->cpu_features,
588 start64, start64 + size64);
590 start64 = find_section64(v64->hdr, "__mmu_ftr_fixup", &size64);
592 do_feature_fixups(cur_cpu_spec->mmu_features,
593 start64, start64 + size64);
595 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
597 do_feature_fixups(powerpc_firmware_features,
598 start64, start64 + size64);
600 start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
602 do_lwsync_fixups(cur_cpu_spec->cpu_features,
603 start64, start64 + size64);
604 #endif /* CONFIG_PPC64 */
606 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
608 do_feature_fixups(cur_cpu_spec->cpu_features,
609 start32, start32 + size32);
611 start32 = find_section32(v32->hdr, "__mmu_ftr_fixup", &size32);
613 do_feature_fixups(cur_cpu_spec->mmu_features,
614 start32, start32 + size32);
617 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
619 do_feature_fixups(powerpc_firmware_features,
620 start32, start32 + size32);
621 #endif /* CONFIG_PPC64 */
623 start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
625 do_lwsync_fixups(cur_cpu_spec->cpu_features,
626 start32, start32 + size32);
631 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
632 struct lib64_elfinfo *v64)
636 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
637 struct vdso_patch_def *patch = &vdso_patches[i];
638 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
643 DBG("replacing %s with %s...\n", patch->gen_name,
644 patch->fix_name ? "NONE" : patch->fix_name);
647 * Patch the 32 bits and 64 bits symbols. Note that we do not
648 * patch the "." symbol on 64 bits.
649 * It would be easy to do, but doesn't seem to be necessary,
650 * patching the OPD symbol is enough.
652 vdso_do_func_patch32(v32, v64, patch->gen_name,
655 vdso_do_func_patch64(v32, v64, patch->gen_name,
657 #endif /* CONFIG_PPC64 */
664 static __init int vdso_setup(void)
666 struct lib32_elfinfo v32;
667 struct lib64_elfinfo v64;
669 v32.hdr = vdso32_kbase;
671 v64.hdr = vdso64_kbase;
673 if (vdso_do_find_sections(&v32, &v64))
676 if (vdso_fixup_datapage(&v32, &v64))
679 if (vdso_fixup_features(&v32, &v64))
682 if (vdso_fixup_alt_funcs(&v32, &v64))
685 vdso_setup_trampolines(&v32, &v64);
691 * Called from setup_arch to initialize the bitmap of available
692 * syscalls in the systemcfg page
694 static void __init vdso_setup_syscall_map(void)
697 extern unsigned long *sys_call_table;
698 extern unsigned long sys_ni_syscall;
701 for (i = 0; i < __NR_syscalls; i++) {
703 if (sys_call_table[i*2] != sys_ni_syscall)
704 vdso_data->syscall_map_64[i >> 5] |=
705 0x80000000UL >> (i & 0x1f);
706 if (sys_call_table[i*2+1] != sys_ni_syscall)
707 vdso_data->syscall_map_32[i >> 5] |=
708 0x80000000UL >> (i & 0x1f);
709 #else /* CONFIG_PPC64 */
710 if (sys_call_table[i] != sys_ni_syscall)
711 vdso_data->syscall_map_32[i >> 5] |=
712 0x80000000UL >> (i & 0x1f);
713 #endif /* CONFIG_PPC64 */
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 iSeries and add
732 * in LPAR bit if necessary
734 vdso_data->platform = machine_is(iseries) ? 0x200 : 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 * Calculate the size of the 32 bits vDSO
765 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
766 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
770 * Setup the syscall map in the vDOS
772 vdso_setup_syscall_map();
775 * Initialize the vDSO images in memory, that is do necessary
776 * fixups of vDSO symbols, locate trampolines, etc...
779 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
787 /* Make sure pages are in the correct state */
788 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
790 BUG_ON(vdso32_pagelist == NULL);
791 for (i = 0; i < vdso32_pages; i++) {
792 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
793 ClearPageReserved(pg);
795 vdso32_pagelist[i] = pg;
797 vdso32_pagelist[i++] = virt_to_page(vdso_data);
798 vdso32_pagelist[i] = NULL;
801 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
803 BUG_ON(vdso64_pagelist == NULL);
804 for (i = 0; i < vdso64_pages; i++) {
805 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
806 ClearPageReserved(pg);
808 vdso64_pagelist[i] = pg;
810 vdso64_pagelist[i++] = virt_to_page(vdso_data);
811 vdso64_pagelist[i] = NULL;
812 #endif /* CONFIG_PPC64 */
814 get_page(virt_to_page(vdso_data));
821 arch_initcall(vdso_init);
823 int in_gate_area_no_mm(unsigned long addr)
828 int in_gate_area(struct mm_struct *mm, unsigned long addr)
833 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)