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/lmb.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 extern char vdso32_start, vdso32_end;
54 static void *vdso32_kbase = &vdso32_start;
55 static unsigned int vdso32_pages;
56 static struct page **vdso32_pagelist;
57 unsigned long vdso32_sigtramp;
58 unsigned long vdso32_rt_sigtramp;
61 extern char vdso64_start, vdso64_end;
62 static void *vdso64_kbase = &vdso64_start;
63 static unsigned int vdso64_pages;
64 static struct page **vdso64_pagelist;
65 unsigned long vdso64_rt_sigtramp;
66 #endif /* CONFIG_PPC64 */
68 static int vdso_ready;
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 __page_aligned_data;
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
105 "__kernel_clock_gettime", NULL
109 "__kernel_clock_getres", NULL
113 "__kernel_get_tbfreq", NULL
118 * Some infos carried around for each of them during parsing at
123 Elf32_Ehdr *hdr; /* ptr to ELF */
124 Elf32_Sym *dynsym; /* ptr to .dynsym section */
125 unsigned long dynsymsize; /* size of .dynsym section */
126 char *dynstr; /* ptr to .dynstr section */
127 unsigned long text; /* offset of .text section in .so */
134 unsigned long dynsymsize;
141 static void dump_one_vdso_page(struct page *pg, struct page *upg)
143 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
146 if (upg && !IS_ERR(upg) /* && pg != upg*/) {
147 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
155 static void dump_vdso_pages(struct vm_area_struct * vma)
159 if (!vma || test_thread_flag(TIF_32BIT)) {
160 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
161 for (i=0; i<vdso32_pages; i++) {
162 struct page *pg = virt_to_page(vdso32_kbase +
164 struct page *upg = (vma && vma->vm_mm) ?
165 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
167 dump_one_vdso_page(pg, upg);
170 if (!vma || !test_thread_flag(TIF_32BIT)) {
171 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
172 for (i=0; i<vdso64_pages; i++) {
173 struct page *pg = virt_to_page(vdso64_kbase +
175 struct page *upg = (vma && vma->vm_mm) ?
176 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
178 dump_one_vdso_page(pg, upg);
185 * This is called from binfmt_elf, we create the special vma for the
186 * vDSO and insert it into the mm struct tree
188 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
190 struct mm_struct *mm = current->mm;
191 struct page **vdso_pagelist;
192 unsigned long vdso_pages;
193 unsigned long vdso_base;
200 if (test_thread_flag(TIF_32BIT)) {
201 vdso_pagelist = vdso32_pagelist;
202 vdso_pages = vdso32_pages;
203 vdso_base = VDSO32_MBASE;
205 vdso_pagelist = vdso64_pagelist;
206 vdso_pages = vdso64_pages;
208 * On 64bit we don't have a preferred map address. This
209 * allows get_unmapped_area to find an area near other mmaps
210 * and most likely share a SLB entry.
215 vdso_pagelist = vdso32_pagelist;
216 vdso_pages = vdso32_pages;
217 vdso_base = VDSO32_MBASE;
220 current->mm->context.vdso_base = 0;
222 /* vDSO has a problem and was disabled, just don't "enable" it for the
227 /* Add a page to the vdso size for the data page */
231 * pick a base address for the vDSO in process space. We try to put it
232 * at vdso_base which is the "natural" base for it, but we might fail
233 * and end up putting it elsewhere.
235 down_write(&mm->mmap_sem);
236 vdso_base = get_unmapped_area(NULL, vdso_base,
237 vdso_pages << PAGE_SHIFT, 0, 0);
238 if (IS_ERR_VALUE(vdso_base)) {
244 * our vma flags don't have VM_WRITE so by default, the process isn't
245 * allowed to write those pages.
246 * gdb can break that with ptrace interface, and thus trigger COW on
247 * those pages but it's then your responsibility to never do that on
248 * the "data" page of the vDSO or you'll stop getting kernel updates
249 * and your nice userland gettimeofday will be totally dead.
250 * It's fine to use that for setting breakpoints in the vDSO code
253 * Make sure the vDSO gets into every core dump.
254 * Dumping its contents makes post-mortem fully interpretable later
255 * without matching up the same kernel and hardware config to see
256 * what PC values meant.
258 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
260 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
266 /* Put vDSO base into mm struct */
267 current->mm->context.vdso_base = vdso_base;
269 up_write(&mm->mmap_sem);
273 up_write(&mm->mmap_sem);
277 const char *arch_vma_name(struct vm_area_struct *vma)
279 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
286 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
293 /* Grab section headers and strings so we can tell who is who */
294 sechdrs = (void *)ehdr + ehdr->e_shoff;
295 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
297 /* Find the section they want */
298 for (i = 1; i < ehdr->e_shnum; i++) {
299 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
301 *size = sechdrs[i].sh_size;
302 return (void *)ehdr + sechdrs[i].sh_offset;
309 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
313 char name[MAX_SYMNAME], *c;
315 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
316 if (lib->dynsym[i].st_name == 0)
318 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
320 c = strchr(name, '@');
323 if (strcmp(symname, name) == 0)
324 return &lib->dynsym[i];
329 /* Note that we assume the section is .text and the symbol is relative to
332 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
335 Elf32_Sym *sym = find_symbol32(lib, symname);
338 printk(KERN_WARNING "vDSO32: function %s not found !\n",
342 return sym->st_value - VDSO32_LBASE;
345 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
346 struct lib64_elfinfo *v64,
347 const char *orig, const char *fix)
349 Elf32_Sym *sym32_gen, *sym32_fix;
351 sym32_gen = find_symbol32(v32, orig);
352 if (sym32_gen == NULL) {
353 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
357 sym32_gen->st_name = 0;
360 sym32_fix = find_symbol32(v32, fix);
361 if (sym32_fix == NULL) {
362 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
365 sym32_gen->st_value = sym32_fix->st_value;
366 sym32_gen->st_size = sym32_fix->st_size;
367 sym32_gen->st_info = sym32_fix->st_info;
368 sym32_gen->st_other = sym32_fix->st_other;
369 sym32_gen->st_shndx = sym32_fix->st_shndx;
377 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
384 /* Grab section headers and strings so we can tell who is who */
385 sechdrs = (void *)ehdr + ehdr->e_shoff;
386 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
388 /* Find the section they want */
389 for (i = 1; i < ehdr->e_shnum; i++) {
390 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
392 *size = sechdrs[i].sh_size;
393 return (void *)ehdr + sechdrs[i].sh_offset;
401 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
405 char name[MAX_SYMNAME], *c;
407 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
408 if (lib->dynsym[i].st_name == 0)
410 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
412 c = strchr(name, '@');
415 if (strcmp(symname, name) == 0)
416 return &lib->dynsym[i];
421 /* Note that we assume the section is .text and the symbol is relative to
424 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
427 Elf64_Sym *sym = find_symbol64(lib, symname);
430 printk(KERN_WARNING "vDSO64: function %s not found !\n",
434 #ifdef VDS64_HAS_DESCRIPTORS
435 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
438 return sym->st_value - VDSO64_LBASE;
442 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
443 struct lib64_elfinfo *v64,
444 const char *orig, const char *fix)
446 Elf64_Sym *sym64_gen, *sym64_fix;
448 sym64_gen = find_symbol64(v64, orig);
449 if (sym64_gen == NULL) {
450 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
454 sym64_gen->st_name = 0;
457 sym64_fix = find_symbol64(v64, fix);
458 if (sym64_fix == NULL) {
459 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
462 sym64_gen->st_value = sym64_fix->st_value;
463 sym64_gen->st_size = sym64_fix->st_size;
464 sym64_gen->st_info = sym64_fix->st_info;
465 sym64_gen->st_other = sym64_fix->st_other;
466 sym64_gen->st_shndx = sym64_fix->st_shndx;
471 #endif /* CONFIG_PPC64 */
474 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
475 struct lib64_elfinfo *v64)
480 * Locate symbol tables & text section
483 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
484 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
485 if (v32->dynsym == NULL || v32->dynstr == NULL) {
486 printk(KERN_ERR "vDSO32: required symbol section not found\n");
489 sect = find_section32(v32->hdr, ".text", NULL);
491 printk(KERN_ERR "vDSO32: the .text section was not found\n");
494 v32->text = sect - vdso32_kbase;
497 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
498 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
499 if (v64->dynsym == NULL || v64->dynstr == NULL) {
500 printk(KERN_ERR "vDSO64: required symbol section not found\n");
503 sect = find_section64(v64->hdr, ".text", NULL);
505 printk(KERN_ERR "vDSO64: the .text section was not found\n");
508 v64->text = sect - vdso64_kbase;
509 #endif /* CONFIG_PPC64 */
514 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
515 struct lib64_elfinfo *v64)
518 * Find signal trampolines
522 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
524 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
525 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
528 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
529 struct lib64_elfinfo *v64)
535 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
537 printk(KERN_ERR "vDSO64: Can't find symbol "
538 "__kernel_datapage_offset !\n");
541 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
542 (vdso64_pages << PAGE_SHIFT) -
543 (sym64->st_value - VDSO64_LBASE);
544 #endif /* CONFIG_PPC64 */
546 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
548 printk(KERN_ERR "vDSO32: Can't find symbol "
549 "__kernel_datapage_offset !\n");
552 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
553 (vdso32_pages << PAGE_SHIFT) -
554 (sym32->st_value - VDSO32_LBASE);
560 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
561 struct lib64_elfinfo *v64)
564 unsigned long size32;
568 unsigned long size64;
570 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
572 do_feature_fixups(cur_cpu_spec->cpu_features,
573 start64, start64 + size64);
575 start64 = find_section64(v64->hdr, "__mmu_ftr_fixup", &size64);
577 do_feature_fixups(cur_cpu_spec->mmu_features,
578 start64, start64 + size64);
580 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
582 do_feature_fixups(powerpc_firmware_features,
583 start64, start64 + size64);
585 start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
587 do_lwsync_fixups(cur_cpu_spec->cpu_features,
588 start64, start64 + size64);
589 #endif /* CONFIG_PPC64 */
591 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
593 do_feature_fixups(cur_cpu_spec->cpu_features,
594 start32, start32 + size32);
596 start32 = find_section32(v32->hdr, "__mmu_ftr_fixup", &size32);
598 do_feature_fixups(cur_cpu_spec->mmu_features,
599 start32, start32 + size32);
602 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
604 do_feature_fixups(powerpc_firmware_features,
605 start32, start32 + size32);
606 #endif /* CONFIG_PPC64 */
608 start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
610 do_lwsync_fixups(cur_cpu_spec->cpu_features,
611 start32, start32 + size32);
616 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
617 struct lib64_elfinfo *v64)
621 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
622 struct vdso_patch_def *patch = &vdso_patches[i];
623 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
628 DBG("replacing %s with %s...\n", patch->gen_name,
629 patch->fix_name ? "NONE" : patch->fix_name);
632 * Patch the 32 bits and 64 bits symbols. Note that we do not
633 * patch the "." symbol on 64 bits.
634 * It would be easy to do, but doesn't seem to be necessary,
635 * patching the OPD symbol is enough.
637 vdso_do_func_patch32(v32, v64, patch->gen_name,
640 vdso_do_func_patch64(v32, v64, patch->gen_name,
642 #endif /* CONFIG_PPC64 */
649 static __init int vdso_setup(void)
651 struct lib32_elfinfo v32;
652 struct lib64_elfinfo v64;
654 v32.hdr = vdso32_kbase;
656 v64.hdr = vdso64_kbase;
658 if (vdso_do_find_sections(&v32, &v64))
661 if (vdso_fixup_datapage(&v32, &v64))
664 if (vdso_fixup_features(&v32, &v64))
667 if (vdso_fixup_alt_funcs(&v32, &v64))
670 vdso_setup_trampolines(&v32, &v64);
676 * Called from setup_arch to initialize the bitmap of available
677 * syscalls in the systemcfg page
679 static void __init vdso_setup_syscall_map(void)
682 extern unsigned long *sys_call_table;
683 extern unsigned long sys_ni_syscall;
686 for (i = 0; i < __NR_syscalls; i++) {
688 if (sys_call_table[i*2] != sys_ni_syscall)
689 vdso_data->syscall_map_64[i >> 5] |=
690 0x80000000UL >> (i & 0x1f);
691 if (sys_call_table[i*2+1] != sys_ni_syscall)
692 vdso_data->syscall_map_32[i >> 5] |=
693 0x80000000UL >> (i & 0x1f);
694 #else /* CONFIG_PPC64 */
695 if (sys_call_table[i] != sys_ni_syscall)
696 vdso_data->syscall_map_32[i >> 5] |=
697 0x80000000UL >> (i & 0x1f);
698 #endif /* CONFIG_PPC64 */
703 static int __init vdso_init(void)
709 * Fill up the "systemcfg" stuff for backward compatiblity
711 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
712 vdso_data->version.major = SYSTEMCFG_MAJOR;
713 vdso_data->version.minor = SYSTEMCFG_MINOR;
714 vdso_data->processor = mfspr(SPRN_PVR);
716 * Fake the old platform number for pSeries and iSeries and add
717 * in LPAR bit if necessary
719 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
720 if (firmware_has_feature(FW_FEATURE_LPAR))
721 vdso_data->platform |= 1;
722 vdso_data->physicalMemorySize = lmb_phys_mem_size();
723 vdso_data->dcache_size = ppc64_caches.dsize;
724 vdso_data->dcache_line_size = ppc64_caches.dline_size;
725 vdso_data->icache_size = ppc64_caches.isize;
726 vdso_data->icache_line_size = ppc64_caches.iline_size;
728 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
729 vdso_data->dcache_block_size = ppc64_caches.dline_size;
730 vdso_data->icache_block_size = ppc64_caches.iline_size;
731 vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
732 vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
735 * Calculate the size of the 64 bits vDSO
737 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
738 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
740 vdso_data->dcache_block_size = L1_CACHE_BYTES;
741 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
742 vdso_data->icache_block_size = L1_CACHE_BYTES;
743 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
744 #endif /* CONFIG_PPC64 */
748 * Calculate the size of the 32 bits vDSO
750 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
751 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
755 * Setup the syscall map in the vDOS
757 vdso_setup_syscall_map();
760 * Initialize the vDSO images in memory, that is do necessary
761 * fixups of vDSO symbols, locate trampolines, etc...
764 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
772 /* Make sure pages are in the correct state */
773 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
775 BUG_ON(vdso32_pagelist == NULL);
776 for (i = 0; i < vdso32_pages; i++) {
777 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
778 ClearPageReserved(pg);
780 vdso32_pagelist[i] = pg;
782 vdso32_pagelist[i++] = virt_to_page(vdso_data);
783 vdso32_pagelist[i] = NULL;
786 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
788 BUG_ON(vdso64_pagelist == NULL);
789 for (i = 0; i < vdso64_pages; i++) {
790 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
791 ClearPageReserved(pg);
793 vdso64_pagelist[i] = pg;
795 vdso64_pagelist[i++] = virt_to_page(vdso_data);
796 vdso64_pagelist[i] = NULL;
797 #endif /* CONFIG_PPC64 */
799 get_page(virt_to_page(vdso_data));
806 arch_initcall(vdso_init);
808 int in_gate_area_no_task(unsigned long addr)
813 int in_gate_area(struct task_struct *task, unsigned long addr)
818 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)