2 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
3 * <benh@kernel.crashing.org>
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; either version
8 * 2 of the License, or (at your option) any later version.
11 #include <linux/module.h>
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/bootmem.h>
25 #include <asm/pgtable.h>
26 #include <asm/system.h>
27 #include <asm/processor.h>
29 #include <asm/mmu_context.h>
31 #include <asm/machdep.h>
32 #include <asm/cputable.h>
33 #include <asm/sections.h>
34 #include <asm/firmware.h>
36 #include <asm/vdso_datapage.h>
43 #define DBG(fmt...) printk(fmt)
48 /* Max supported size for symbol names */
49 #define MAX_SYMNAME 64
51 extern char vdso32_start, vdso32_end;
52 static void *vdso32_kbase = &vdso32_start;
53 static unsigned int vdso32_pages;
54 static struct page **vdso32_pagelist;
55 unsigned long vdso32_sigtramp;
56 unsigned long vdso32_rt_sigtramp;
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 __attribute__((__section__(".data.page_aligned")));
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 "__kernel_gettimeofday", NULL
104 * Some infos carried around for each of them during parsing at
109 Elf32_Ehdr *hdr; /* ptr to ELF */
110 Elf32_Sym *dynsym; /* ptr to .dynsym section */
111 unsigned long dynsymsize; /* size of .dynsym section */
112 char *dynstr; /* ptr to .dynstr section */
113 unsigned long text; /* offset of .text section in .so */
120 unsigned long dynsymsize;
127 static void dump_one_vdso_page(struct page *pg, struct page *upg)
129 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
132 if (upg/* && pg != upg*/) {
133 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
141 static void dump_vdso_pages(struct vm_area_struct * vma)
145 if (!vma || test_thread_flag(TIF_32BIT)) {
146 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
147 for (i=0; i<vdso32_pages; i++) {
148 struct page *pg = virt_to_page(vdso32_kbase +
150 struct page *upg = (vma && vma->vm_mm) ?
151 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
153 dump_one_vdso_page(pg, upg);
156 if (!vma || !test_thread_flag(TIF_32BIT)) {
157 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
158 for (i=0; i<vdso64_pages; i++) {
159 struct page *pg = virt_to_page(vdso64_kbase +
161 struct page *upg = (vma && vma->vm_mm) ?
162 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
164 dump_one_vdso_page(pg, upg);
171 * This is called from binfmt_elf, we create the special vma for the
172 * vDSO and insert it into the mm struct tree
174 int arch_setup_additional_pages(struct linux_binprm *bprm,
175 int executable_stack)
177 struct mm_struct *mm = current->mm;
178 struct page **vdso_pagelist;
179 unsigned long vdso_pages;
180 unsigned long vdso_base;
187 if (test_thread_flag(TIF_32BIT)) {
188 vdso_pagelist = vdso32_pagelist;
189 vdso_pages = vdso32_pages;
190 vdso_base = VDSO32_MBASE;
192 vdso_pagelist = vdso64_pagelist;
193 vdso_pages = vdso64_pages;
194 vdso_base = VDSO64_MBASE;
197 vdso_pagelist = vdso32_pagelist;
198 vdso_pages = vdso32_pages;
199 vdso_base = VDSO32_MBASE;
202 current->mm->context.vdso_base = 0;
204 /* vDSO has a problem and was disabled, just don't "enable" it for the
209 /* Add a page to the vdso size for the data page */
213 * pick a base address for the vDSO in process space. We try to put it
214 * at vdso_base which is the "natural" base for it, but we might fail
215 * and end up putting it elsewhere.
217 down_write(&mm->mmap_sem);
218 vdso_base = get_unmapped_area(NULL, vdso_base,
219 vdso_pages << PAGE_SHIFT, 0, 0);
220 if (IS_ERR_VALUE(vdso_base)) {
226 * our vma flags don't have VM_WRITE so by default, the process isn't
227 * allowed to write those pages.
228 * gdb can break that with ptrace interface, and thus trigger COW on
229 * those pages but it's then your responsibility to never do that on
230 * the "data" page of the vDSO or you'll stop getting kernel updates
231 * and your nice userland gettimeofday will be totally dead.
232 * It's fine to use that for setting breakpoints in the vDSO code
235 * Make sure the vDSO gets into every core dump.
236 * Dumping its contents makes post-mortem fully interpretable later
237 * without matching up the same kernel and hardware config to see
238 * what PC values meant.
240 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
242 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
248 /* Put vDSO base into mm struct */
249 current->mm->context.vdso_base = vdso_base;
251 up_write(&mm->mmap_sem);
255 up_write(&mm->mmap_sem);
259 const char *arch_vma_name(struct vm_area_struct *vma)
261 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
268 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
275 /* Grab section headers and strings so we can tell who is who */
276 sechdrs = (void *)ehdr + ehdr->e_shoff;
277 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
279 /* Find the section they want */
280 for (i = 1; i < ehdr->e_shnum; i++) {
281 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
283 *size = sechdrs[i].sh_size;
284 return (void *)ehdr + sechdrs[i].sh_offset;
291 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
295 char name[MAX_SYMNAME], *c;
297 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
298 if (lib->dynsym[i].st_name == 0)
300 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
302 c = strchr(name, '@');
305 if (strcmp(symname, name) == 0)
306 return &lib->dynsym[i];
311 /* Note that we assume the section is .text and the symbol is relative to
314 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
317 Elf32_Sym *sym = find_symbol32(lib, symname);
320 printk(KERN_WARNING "vDSO32: function %s not found !\n",
324 return sym->st_value - VDSO32_LBASE;
327 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
328 struct lib64_elfinfo *v64,
329 const char *orig, const char *fix)
331 Elf32_Sym *sym32_gen, *sym32_fix;
333 sym32_gen = find_symbol32(v32, orig);
334 if (sym32_gen == NULL) {
335 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
339 sym32_gen->st_name = 0;
342 sym32_fix = find_symbol32(v32, fix);
343 if (sym32_fix == NULL) {
344 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
347 sym32_gen->st_value = sym32_fix->st_value;
348 sym32_gen->st_size = sym32_fix->st_size;
349 sym32_gen->st_info = sym32_fix->st_info;
350 sym32_gen->st_other = sym32_fix->st_other;
351 sym32_gen->st_shndx = sym32_fix->st_shndx;
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 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
465 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
466 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
467 if (v32->dynsym == NULL || v32->dynstr == NULL) {
468 printk(KERN_ERR "vDSO32: required symbol section not found\n");
471 sect = find_section32(v32->hdr, ".text", NULL);
473 printk(KERN_ERR "vDSO32: the .text section was not found\n");
476 v32->text = sect - vdso32_kbase;
479 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
480 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
481 if (v64->dynsym == NULL || v64->dynstr == NULL) {
482 printk(KERN_ERR "vDSO64: required symbol section not found\n");
485 sect = find_section64(v64->hdr, ".text", NULL);
487 printk(KERN_ERR "vDSO64: the .text section was not found\n");
490 v64->text = sect - vdso64_kbase;
491 #endif /* CONFIG_PPC64 */
496 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
497 struct lib64_elfinfo *v64)
500 * Find signal trampolines
504 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
506 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
507 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
510 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
511 struct lib64_elfinfo *v64)
517 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
519 printk(KERN_ERR "vDSO64: Can't find symbol "
520 "__kernel_datapage_offset !\n");
523 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
524 (vdso64_pages << PAGE_SHIFT) -
525 (sym64->st_value - VDSO64_LBASE);
526 #endif /* CONFIG_PPC64 */
528 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
530 printk(KERN_ERR "vDSO32: Can't find symbol "
531 "__kernel_datapage_offset !\n");
534 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
535 (vdso32_pages << PAGE_SHIFT) -
536 (sym32->st_value - VDSO32_LBASE);
542 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
543 struct lib64_elfinfo *v64)
546 unsigned long size32;
550 unsigned long size64;
552 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
554 do_feature_fixups(cur_cpu_spec->cpu_features,
555 start64, start64 + size64);
557 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
559 do_feature_fixups(powerpc_firmware_features,
560 start64, start64 + size64);
561 #endif /* CONFIG_PPC64 */
563 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
565 do_feature_fixups(cur_cpu_spec->cpu_features,
566 start32, start32 + size32);
569 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
571 do_feature_fixups(powerpc_firmware_features,
572 start32, start32 + size32);
573 #endif /* CONFIG_PPC64 */
578 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
579 struct lib64_elfinfo *v64)
583 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
584 struct vdso_patch_def *patch = &vdso_patches[i];
585 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
590 DBG("replacing %s with %s...\n", patch->gen_name,
591 patch->fix_name ? "NONE" : patch->fix_name);
594 * Patch the 32 bits and 64 bits symbols. Note that we do not
595 * patch the "." symbol on 64 bits.
596 * It would be easy to do, but doesn't seem to be necessary,
597 * patching the OPD symbol is enough.
599 vdso_do_func_patch32(v32, v64, patch->gen_name,
602 vdso_do_func_patch64(v32, v64, patch->gen_name,
604 #endif /* CONFIG_PPC64 */
611 static __init int vdso_setup(void)
613 struct lib32_elfinfo v32;
614 struct lib64_elfinfo v64;
616 v32.hdr = vdso32_kbase;
618 v64.hdr = vdso64_kbase;
620 if (vdso_do_find_sections(&v32, &v64))
623 if (vdso_fixup_datapage(&v32, &v64))
626 if (vdso_fixup_features(&v32, &v64))
629 if (vdso_fixup_alt_funcs(&v32, &v64))
632 vdso_setup_trampolines(&v32, &v64);
638 * Called from setup_arch to initialize the bitmap of available
639 * syscalls in the systemcfg page
641 static void __init vdso_setup_syscall_map(void)
644 extern unsigned long *sys_call_table;
645 extern unsigned long sys_ni_syscall;
648 for (i = 0; i < __NR_syscalls; i++) {
650 if (sys_call_table[i*2] != sys_ni_syscall)
651 vdso_data->syscall_map_64[i >> 5] |=
652 0x80000000UL >> (i & 0x1f);
653 if (sys_call_table[i*2+1] != sys_ni_syscall)
654 vdso_data->syscall_map_32[i >> 5] |=
655 0x80000000UL >> (i & 0x1f);
656 #else /* CONFIG_PPC64 */
657 if (sys_call_table[i] != sys_ni_syscall)
658 vdso_data->syscall_map_32[i >> 5] |=
659 0x80000000UL >> (i & 0x1f);
660 #endif /* CONFIG_PPC64 */
665 static int __init vdso_init(void)
671 * Fill up the "systemcfg" stuff for backward compatiblity
673 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
674 vdso_data->version.major = SYSTEMCFG_MAJOR;
675 vdso_data->version.minor = SYSTEMCFG_MINOR;
676 vdso_data->processor = mfspr(SPRN_PVR);
678 * Fake the old platform number for pSeries and iSeries and add
679 * in LPAR bit if necessary
681 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
682 if (firmware_has_feature(FW_FEATURE_LPAR))
683 vdso_data->platform |= 1;
684 vdso_data->physicalMemorySize = lmb_phys_mem_size();
685 vdso_data->dcache_size = ppc64_caches.dsize;
686 vdso_data->dcache_line_size = ppc64_caches.dline_size;
687 vdso_data->icache_size = ppc64_caches.isize;
688 vdso_data->icache_line_size = ppc64_caches.iline_size;
691 * Calculate the size of the 64 bits vDSO
693 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
694 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
695 #endif /* CONFIG_PPC64 */
699 * Calculate the size of the 32 bits vDSO
701 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
702 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
706 * Setup the syscall map in the vDOS
708 vdso_setup_syscall_map();
711 * Initialize the vDSO images in memory, that is do necessary
712 * fixups of vDSO symbols, locate trampolines, etc...
715 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
723 /* Make sure pages are in the correct state */
724 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
726 BUG_ON(vdso32_pagelist == NULL);
727 for (i = 0; i < vdso32_pages; i++) {
728 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
729 ClearPageReserved(pg);
731 vdso32_pagelist[i] = pg;
733 vdso32_pagelist[i++] = virt_to_page(vdso_data);
734 vdso32_pagelist[i] = NULL;
737 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
739 BUG_ON(vdso64_pagelist == NULL);
740 for (i = 0; i < vdso64_pages; i++) {
741 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
742 ClearPageReserved(pg);
744 vdso64_pagelist[i] = pg;
746 vdso64_pagelist[i++] = virt_to_page(vdso_data);
747 vdso64_pagelist[i] = NULL;
748 #endif /* CONFIG_PPC64 */
750 get_page(virt_to_page(vdso_data));
757 arch_initcall(vdso_init);
759 int in_gate_area_no_task(unsigned long addr)
764 int in_gate_area(struct task_struct *task, unsigned long addr)
769 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)