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/smp_lock.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>
26 #include <asm/pgtable.h>
27 #include <asm/system.h>
28 #include <asm/processor.h>
30 #include <asm/mmu_context.h>
32 #include <asm/machdep.h>
33 #include <asm/cputable.h>
34 #include <asm/sections.h>
35 #include <asm/firmware.h>
37 #include <asm/vdso_datapage.h>
44 #define DBG(fmt...) printk(fmt)
49 /* Max supported size for symbol names */
50 #define MAX_SYMNAME 64
52 #define VDSO32_MAXPAGES (((0x3000 + PAGE_MASK) >> PAGE_SHIFT) + 2)
53 #define VDSO64_MAXPAGES (((0x3000 + PAGE_MASK) >> PAGE_SHIFT) + 2)
55 extern char vdso32_start, vdso32_end;
56 static void *vdso32_kbase = &vdso32_start;
57 unsigned int vdso32_pages;
58 static struct page *vdso32_pagelist[VDSO32_MAXPAGES];
59 unsigned long vdso32_sigtramp;
60 unsigned long vdso32_rt_sigtramp;
63 extern char vdso64_start, vdso64_end;
64 static void *vdso64_kbase = &vdso64_start;
65 unsigned int vdso64_pages;
66 static struct page *vdso64_pagelist[VDSO64_MAXPAGES];
67 unsigned long vdso64_rt_sigtramp;
68 #endif /* CONFIG_PPC64 */
71 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
72 * Once the early boot kernel code no longer needs to muck around
73 * with it, it will become dynamically allocated
76 struct vdso_data data;
78 } vdso_data_store __attribute__((__section__(".data.page_aligned")));
79 struct vdso_data *vdso_data = &vdso_data_store.data;
81 /* Format of the patch table */
84 unsigned long ftr_mask, ftr_value;
89 /* Table of functions to patch based on the CPU type/revision
91 * Currently, we only change sync_dicache to do nothing on processors
92 * with a coherent icache
94 static struct vdso_patch_def vdso_patches[] = {
96 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
97 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
101 "__kernel_gettimeofday", NULL
106 * Some infos carried around for each of them during parsing at
111 Elf32_Ehdr *hdr; /* ptr to ELF */
112 Elf32_Sym *dynsym; /* ptr to .dynsym section */
113 unsigned long dynsymsize; /* size of .dynsym section */
114 char *dynstr; /* ptr to .dynstr section */
115 unsigned long text; /* offset of .text section in .so */
122 unsigned long dynsymsize;
129 static void dump_one_vdso_page(struct page *pg, struct page *upg)
131 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
134 if (upg/* && pg != upg*/) {
135 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
143 static void dump_vdso_pages(struct vm_area_struct * vma)
147 if (!vma || test_thread_flag(TIF_32BIT)) {
148 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
149 for (i=0; i<vdso32_pages; i++) {
150 struct page *pg = virt_to_page(vdso32_kbase +
152 struct page *upg = (vma && vma->vm_mm) ?
153 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
155 dump_one_vdso_page(pg, upg);
158 if (!vma || !test_thread_flag(TIF_32BIT)) {
159 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
160 for (i=0; i<vdso64_pages; i++) {
161 struct page *pg = virt_to_page(vdso64_kbase +
163 struct page *upg = (vma && vma->vm_mm) ?
164 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
166 dump_one_vdso_page(pg, upg);
173 * This is called from binfmt_elf, we create the special vma for the
174 * vDSO and insert it into the mm struct tree
176 int arch_setup_additional_pages(struct linux_binprm *bprm,
177 int executable_stack)
179 struct mm_struct *mm = current->mm;
180 struct page **vdso_pagelist;
181 unsigned long vdso_pages;
182 unsigned long vdso_base;
186 if (test_thread_flag(TIF_32BIT)) {
187 vdso_pagelist = vdso32_pagelist;
188 vdso_pages = vdso32_pages;
189 vdso_base = VDSO32_MBASE;
191 vdso_pagelist = vdso64_pagelist;
192 vdso_pages = vdso64_pages;
193 vdso_base = VDSO64_MBASE;
196 vdso_pagelist = vdso32_pagelist;
197 vdso_pages = vdso32_pages;
198 vdso_base = VDSO32_MBASE;
201 current->mm->context.vdso_base = 0;
203 /* vDSO has a problem and was disabled, just don't "enable" it for the
208 /* Add a page to the vdso size for the data page */
212 * pick a base address for the vDSO in process space. We try to put it
213 * at vdso_base which is the "natural" base for it, but we might fail
214 * and end up putting it elsewhere.
216 down_write(&mm->mmap_sem);
217 vdso_base = get_unmapped_area(NULL, vdso_base,
218 vdso_pages << PAGE_SHIFT, 0, 0);
219 if (IS_ERR_VALUE(vdso_base)) {
225 * our vma flags don't have VM_WRITE so by default, the process isn't
226 * allowed to write those pages.
227 * gdb can break that with ptrace interface, and thus trigger COW on
228 * those pages but it's then your responsibility to never do that on
229 * the "data" page of the vDSO or you'll stop getting kernel updates
230 * and your nice userland gettimeofday will be totally dead.
231 * It's fine to use that for setting breakpoints in the vDSO code
234 * Make sure the vDSO gets into every core dump.
235 * Dumping its contents makes post-mortem fully interpretable later
236 * without matching up the same kernel and hardware config to see
237 * what PC values meant.
239 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
241 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
247 /* Put vDSO base into mm struct */
248 current->mm->context.vdso_base = vdso_base;
250 up_write(&mm->mmap_sem);
254 up_write(&mm->mmap_sem);
258 const char *arch_vma_name(struct vm_area_struct *vma)
260 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
267 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
274 /* Grab section headers and strings so we can tell who is who */
275 sechdrs = (void *)ehdr + ehdr->e_shoff;
276 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
278 /* Find the section they want */
279 for (i = 1; i < ehdr->e_shnum; i++) {
280 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
282 *size = sechdrs[i].sh_size;
283 return (void *)ehdr + sechdrs[i].sh_offset;
290 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
294 char name[MAX_SYMNAME], *c;
296 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
297 if (lib->dynsym[i].st_name == 0)
299 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
301 c = strchr(name, '@');
304 if (strcmp(symname, name) == 0)
305 return &lib->dynsym[i];
310 /* Note that we assume the section is .text and the symbol is relative to
313 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
316 Elf32_Sym *sym = find_symbol32(lib, symname);
319 printk(KERN_WARNING "vDSO32: function %s not found !\n",
323 return sym->st_value - VDSO32_LBASE;
326 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
327 struct lib64_elfinfo *v64,
328 const char *orig, const char *fix)
330 Elf32_Sym *sym32_gen, *sym32_fix;
332 sym32_gen = find_symbol32(v32, orig);
333 if (sym32_gen == NULL) {
334 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
338 sym32_gen->st_name = 0;
341 sym32_fix = find_symbol32(v32, fix);
342 if (sym32_fix == NULL) {
343 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
346 sym32_gen->st_value = sym32_fix->st_value;
347 sym32_gen->st_size = sym32_fix->st_size;
348 sym32_gen->st_info = sym32_fix->st_info;
349 sym32_gen->st_other = sym32_fix->st_other;
350 sym32_gen->st_shndx = sym32_fix->st_shndx;
358 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
365 /* Grab section headers and strings so we can tell who is who */
366 sechdrs = (void *)ehdr + ehdr->e_shoff;
367 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
369 /* Find the section they want */
370 for (i = 1; i < ehdr->e_shnum; i++) {
371 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
373 *size = sechdrs[i].sh_size;
374 return (void *)ehdr + sechdrs[i].sh_offset;
382 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
386 char name[MAX_SYMNAME], *c;
388 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
389 if (lib->dynsym[i].st_name == 0)
391 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
393 c = strchr(name, '@');
396 if (strcmp(symname, name) == 0)
397 return &lib->dynsym[i];
402 /* Note that we assume the section is .text and the symbol is relative to
405 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
408 Elf64_Sym *sym = find_symbol64(lib, symname);
411 printk(KERN_WARNING "vDSO64: function %s not found !\n",
415 #ifdef VDS64_HAS_DESCRIPTORS
416 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
419 return sym->st_value - VDSO64_LBASE;
423 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
424 struct lib64_elfinfo *v64,
425 const char *orig, const char *fix)
427 Elf64_Sym *sym64_gen, *sym64_fix;
429 sym64_gen = find_symbol64(v64, orig);
430 if (sym64_gen == NULL) {
431 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
435 sym64_gen->st_name = 0;
438 sym64_fix = find_symbol64(v64, fix);
439 if (sym64_fix == NULL) {
440 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
443 sym64_gen->st_value = sym64_fix->st_value;
444 sym64_gen->st_size = sym64_fix->st_size;
445 sym64_gen->st_info = sym64_fix->st_info;
446 sym64_gen->st_other = sym64_fix->st_other;
447 sym64_gen->st_shndx = sym64_fix->st_shndx;
452 #endif /* CONFIG_PPC64 */
455 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
456 struct lib64_elfinfo *v64)
461 * Locate symbol tables & text section
464 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
465 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
466 if (v32->dynsym == NULL || v32->dynstr == NULL) {
467 printk(KERN_ERR "vDSO32: required symbol section not found\n");
470 sect = find_section32(v32->hdr, ".text", NULL);
472 printk(KERN_ERR "vDSO32: the .text section was not found\n");
475 v32->text = sect - vdso32_kbase;
478 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
479 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
480 if (v64->dynsym == NULL || v64->dynstr == NULL) {
481 printk(KERN_ERR "vDSO64: required symbol section not found\n");
484 sect = find_section64(v64->hdr, ".text", NULL);
486 printk(KERN_ERR "vDSO64: the .text section was not found\n");
489 v64->text = sect - vdso64_kbase;
490 #endif /* CONFIG_PPC64 */
495 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
496 struct lib64_elfinfo *v64)
499 * Find signal trampolines
503 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
505 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
506 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
509 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
510 struct lib64_elfinfo *v64)
516 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
518 printk(KERN_ERR "vDSO64: Can't find symbol "
519 "__kernel_datapage_offset !\n");
522 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
523 (vdso64_pages << PAGE_SHIFT) -
524 (sym64->st_value - VDSO64_LBASE);
525 #endif /* CONFIG_PPC64 */
527 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
529 printk(KERN_ERR "vDSO32: Can't find symbol "
530 "__kernel_datapage_offset !\n");
533 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
534 (vdso32_pages << PAGE_SHIFT) -
535 (sym32->st_value - VDSO32_LBASE);
541 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
542 struct lib64_elfinfo *v64)
545 unsigned long size32;
549 unsigned long size64;
551 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
553 do_feature_fixups(cur_cpu_spec->cpu_features,
554 start64, start64 + size64);
556 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
558 do_feature_fixups(powerpc_firmware_features,
559 start64, start64 + size64);
560 #endif /* CONFIG_PPC64 */
562 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
564 do_feature_fixups(cur_cpu_spec->cpu_features,
565 start32, start32 + size32);
568 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
570 do_feature_fixups(powerpc_firmware_features,
571 start32, start32 + size32);
572 #endif /* CONFIG_PPC64 */
577 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
578 struct lib64_elfinfo *v64)
582 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
583 struct vdso_patch_def *patch = &vdso_patches[i];
584 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
589 DBG("replacing %s with %s...\n", patch->gen_name,
590 patch->fix_name ? "NONE" : patch->fix_name);
593 * Patch the 32 bits and 64 bits symbols. Note that we do not
594 * patch the "." symbol on 64 bits.
595 * It would be easy to do, but doesn't seem to be necessary,
596 * patching the OPD symbol is enough.
598 vdso_do_func_patch32(v32, v64, patch->gen_name,
601 vdso_do_func_patch64(v32, v64, patch->gen_name,
603 #endif /* CONFIG_PPC64 */
610 static __init int vdso_setup(void)
612 struct lib32_elfinfo v32;
613 struct lib64_elfinfo v64;
615 v32.hdr = vdso32_kbase;
617 v64.hdr = vdso64_kbase;
619 if (vdso_do_find_sections(&v32, &v64))
622 if (vdso_fixup_datapage(&v32, &v64))
625 if (vdso_fixup_features(&v32, &v64))
628 if (vdso_fixup_alt_funcs(&v32, &v64))
631 vdso_setup_trampolines(&v32, &v64);
637 * Called from setup_arch to initialize the bitmap of available
638 * syscalls in the systemcfg page
640 static void __init vdso_setup_syscall_map(void)
643 extern unsigned long *sys_call_table;
644 extern unsigned long sys_ni_syscall;
647 for (i = 0; i < __NR_syscalls; i++) {
649 if (sys_call_table[i*2] != sys_ni_syscall)
650 vdso_data->syscall_map_64[i >> 5] |=
651 0x80000000UL >> (i & 0x1f);
652 if (sys_call_table[i*2+1] != sys_ni_syscall)
653 vdso_data->syscall_map_32[i >> 5] |=
654 0x80000000UL >> (i & 0x1f);
655 #else /* CONFIG_PPC64 */
656 if (sys_call_table[i] != sys_ni_syscall)
657 vdso_data->syscall_map_32[i >> 5] |=
658 0x80000000UL >> (i & 0x1f);
659 #endif /* CONFIG_PPC64 */
664 void __init vdso_init(void)
670 * Fill up the "systemcfg" stuff for backward compatiblity
672 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
673 vdso_data->version.major = SYSTEMCFG_MAJOR;
674 vdso_data->version.minor = SYSTEMCFG_MINOR;
675 vdso_data->processor = mfspr(SPRN_PVR);
677 * Fake the old platform number for pSeries and iSeries and add
678 * in LPAR bit if necessary
680 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
681 if (firmware_has_feature(FW_FEATURE_LPAR))
682 vdso_data->platform |= 1;
683 vdso_data->physicalMemorySize = lmb_phys_mem_size();
684 vdso_data->dcache_size = ppc64_caches.dsize;
685 vdso_data->dcache_line_size = ppc64_caches.dline_size;
686 vdso_data->icache_size = ppc64_caches.isize;
687 vdso_data->icache_line_size = ppc64_caches.iline_size;
690 * Calculate the size of the 64 bits vDSO
692 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
693 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
694 #endif /* CONFIG_PPC64 */
698 * Calculate the size of the 32 bits vDSO
700 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
701 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
705 * Setup the syscall map in the vDOS
707 vdso_setup_syscall_map();
710 * Initialize the vDSO images in memory, that is do necessary
711 * fixups of vDSO symbols, locate trampolines, etc...
714 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
722 /* Make sure pages are in the correct state */
723 BUG_ON(vdso32_pages + 2 > VDSO32_MAXPAGES);
724 for (i = 0; i < vdso32_pages; i++) {
725 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
726 ClearPageReserved(pg);
728 vdso32_pagelist[i] = pg;
730 vdso32_pagelist[i++] = virt_to_page(vdso_data);
731 vdso32_pagelist[i] = NULL;
734 BUG_ON(vdso64_pages + 2 > VDSO64_MAXPAGES);
735 for (i = 0; i < vdso64_pages; i++) {
736 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
737 ClearPageReserved(pg);
739 vdso64_pagelist[i] = pg;
741 vdso64_pagelist[i++] = virt_to_page(vdso_data);
742 vdso64_pagelist[i] = NULL;
743 #endif /* CONFIG_PPC64 */
745 get_page(virt_to_page(vdso_data));
748 int in_gate_area_no_task(unsigned long addr)
753 int in_gate_area(struct task_struct *task, unsigned long addr)
758 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)