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>
25 #include <linux/vs_memory.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>
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 unsigned int vdso32_pages;
54 unsigned long vdso32_sigtramp;
55 unsigned long vdso32_rt_sigtramp;
58 extern char vdso64_start, vdso64_end;
59 static void *vdso64_kbase = &vdso64_start;
60 unsigned int vdso64_pages;
61 unsigned long vdso64_rt_sigtramp;
62 #endif /* CONFIG_PPC64 */
65 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
66 * Once the early boot kernel code no longer needs to muck around
67 * with it, it will become dynamically allocated
70 struct vdso_data data;
72 } vdso_data_store __attribute__((__section__(".data.page_aligned")));
73 struct vdso_data *vdso_data = &vdso_data_store.data;
75 /* Format of the patch table */
78 unsigned long ftr_mask, ftr_value;
83 /* Table of functions to patch based on the CPU type/revision
85 * Currently, we only change sync_dicache to do nothing on processors
86 * with a coherent icache
88 static struct vdso_patch_def vdso_patches[] = {
90 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
91 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
95 "__kernel_gettimeofday", NULL
100 * Some infos carried around for each of them during parsing at
105 Elf32_Ehdr *hdr; /* ptr to ELF */
106 Elf32_Sym *dynsym; /* ptr to .dynsym section */
107 unsigned long dynsymsize; /* size of .dynsym section */
108 char *dynstr; /* ptr to .dynstr section */
109 unsigned long text; /* offset of .text section in .so */
116 unsigned long dynsymsize;
123 static void dump_one_vdso_page(struct page *pg, struct page *upg)
125 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
128 if (upg/* && pg != upg*/) {
129 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
137 static void dump_vdso_pages(struct vm_area_struct * vma)
141 if (!vma || test_thread_flag(TIF_32BIT)) {
142 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
143 for (i=0; i<vdso32_pages; i++) {
144 struct page *pg = virt_to_page(vdso32_kbase +
146 struct page *upg = (vma && vma->vm_mm) ?
147 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
149 dump_one_vdso_page(pg, upg);
152 if (!vma || !test_thread_flag(TIF_32BIT)) {
153 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
154 for (i=0; i<vdso64_pages; i++) {
155 struct page *pg = virt_to_page(vdso64_kbase +
157 struct page *upg = (vma && vma->vm_mm) ?
158 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
160 dump_one_vdso_page(pg, upg);
167 * Keep a dummy vma_close for now, it will prevent VMA merging.
169 static void vdso_vma_close(struct vm_area_struct * vma)
174 * Our nopage() function, maps in the actual vDSO kernel pages, they will
175 * be mapped read-only by do_no_page(), and eventually COW'ed, either
176 * right away for an initial write access, or by do_wp_page().
178 static struct page * vdso_vma_nopage(struct vm_area_struct * vma,
179 unsigned long address, int *type)
181 unsigned long offset = address - vma->vm_start;
184 void *vbase = (vma->vm_mm->task_size > TASK_SIZE_USER32) ?
185 vdso64_kbase : vdso32_kbase;
187 void *vbase = vdso32_kbase;
190 DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
191 current->comm, address, offset);
193 if (address < vma->vm_start || address > vma->vm_end)
194 return NOPAGE_SIGBUS;
197 * Last page is systemcfg.
199 if ((vma->vm_end - address) <= PAGE_SIZE)
200 pg = virt_to_page(vdso_data);
202 pg = virt_to_page(vbase + offset);
205 DBG(" ->page count: %d\n", page_count(pg));
210 static struct vm_operations_struct vdso_vmops = {
211 .close = vdso_vma_close,
212 .nopage = vdso_vma_nopage,
216 * This is called from binfmt_elf, we create the special vma for the
217 * vDSO and insert it into the mm struct tree
219 int arch_setup_additional_pages(struct linux_binprm *bprm,
220 int executable_stack, unsigned long start_code,
221 unsigned long interp_map_address)
223 struct mm_struct *mm = current->mm;
224 struct vm_area_struct *vma;
225 unsigned long vdso_pages;
226 unsigned long vdso_base;
230 if (test_thread_flag(TIF_32BIT)) {
231 vdso_pages = vdso32_pages;
232 vdso_base = VDSO32_MBASE;
234 vdso_pages = vdso64_pages;
235 vdso_base = VDSO64_MBASE;
238 vdso_pages = vdso32_pages;
239 vdso_base = VDSO32_MBASE;
242 current->mm->context.vdso_base = 0;
244 /* vDSO has a problem and was disabled, just don't "enable" it for the
249 /* Add a page to the vdso size for the data page */
253 * pick a base address for the vDSO in process space. We try to put it
254 * at vdso_base which is the "natural" base for it, but we might fail
255 * and end up putting it elsewhere.
257 down_write(&mm->mmap_sem);
258 vdso_base = get_unmapped_area(NULL, vdso_base,
259 vdso_pages << PAGE_SHIFT, 0, 0);
260 if (IS_ERR_VALUE(vdso_base)) {
266 /* Allocate a VMA structure and fill it up */
267 vma = kmem_cache_zalloc(vm_area_cachep, SLAB_KERNEL);
273 vma->vm_start = vdso_base;
274 vma->vm_end = vma->vm_start + (vdso_pages << PAGE_SHIFT);
277 * our vma flags don't have VM_WRITE so by default, the process isn't
278 * allowed to write those pages.
279 * gdb can break that with ptrace interface, and thus trigger COW on
280 * those pages but it's then your responsibility to never do that on
281 * the "data" page of the vDSO or you'll stop getting kernel updates
282 * and your nice userland gettimeofday will be totally dead.
283 * It's fine to use that for setting breakpoints in the vDSO code
286 vma->vm_flags = VM_READ|VM_EXEC|VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC;
287 vma->vm_flags |= mm->def_flags;
288 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
289 vma->vm_ops = &vdso_vmops;
292 rc = insert_vm_struct(mm, vma);
296 /* Put vDSO base into mm struct and account for memory usage */
297 current->mm->context.vdso_base = vdso_base;
298 vx_vmpages_add(mm, (vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
299 up_write(&mm->mmap_sem);
303 kmem_cache_free(vm_area_cachep, vma);
305 up_write(&mm->mmap_sem);
309 const char *arch_vma_name(struct vm_area_struct *vma)
311 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
318 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
325 /* Grab section headers and strings so we can tell who is who */
326 sechdrs = (void *)ehdr + ehdr->e_shoff;
327 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
329 /* Find the section they want */
330 for (i = 1; i < ehdr->e_shnum; i++) {
331 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
333 *size = sechdrs[i].sh_size;
334 return (void *)ehdr + sechdrs[i].sh_offset;
341 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
345 char name[MAX_SYMNAME], *c;
347 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
348 if (lib->dynsym[i].st_name == 0)
350 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
352 c = strchr(name, '@');
355 if (strcmp(symname, name) == 0)
356 return &lib->dynsym[i];
361 /* Note that we assume the section is .text and the symbol is relative to
364 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
367 Elf32_Sym *sym = find_symbol32(lib, symname);
370 printk(KERN_WARNING "vDSO32: function %s not found !\n",
374 return sym->st_value - VDSO32_LBASE;
377 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
378 struct lib64_elfinfo *v64,
379 const char *orig, const char *fix)
381 Elf32_Sym *sym32_gen, *sym32_fix;
383 sym32_gen = find_symbol32(v32, orig);
384 if (sym32_gen == NULL) {
385 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
389 sym32_gen->st_name = 0;
392 sym32_fix = find_symbol32(v32, fix);
393 if (sym32_fix == NULL) {
394 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
397 sym32_gen->st_value = sym32_fix->st_value;
398 sym32_gen->st_size = sym32_fix->st_size;
399 sym32_gen->st_info = sym32_fix->st_info;
400 sym32_gen->st_other = sym32_fix->st_other;
401 sym32_gen->st_shndx = sym32_fix->st_shndx;
409 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
416 /* Grab section headers and strings so we can tell who is who */
417 sechdrs = (void *)ehdr + ehdr->e_shoff;
418 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
420 /* Find the section they want */
421 for (i = 1; i < ehdr->e_shnum; i++) {
422 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
424 *size = sechdrs[i].sh_size;
425 return (void *)ehdr + sechdrs[i].sh_offset;
433 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
437 char name[MAX_SYMNAME], *c;
439 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
440 if (lib->dynsym[i].st_name == 0)
442 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
444 c = strchr(name, '@');
447 if (strcmp(symname, name) == 0)
448 return &lib->dynsym[i];
453 /* Note that we assume the section is .text and the symbol is relative to
456 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
459 Elf64_Sym *sym = find_symbol64(lib, symname);
462 printk(KERN_WARNING "vDSO64: function %s not found !\n",
466 #ifdef VDS64_HAS_DESCRIPTORS
467 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
470 return sym->st_value - VDSO64_LBASE;
474 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
475 struct lib64_elfinfo *v64,
476 const char *orig, const char *fix)
478 Elf64_Sym *sym64_gen, *sym64_fix;
480 sym64_gen = find_symbol64(v64, orig);
481 if (sym64_gen == NULL) {
482 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
486 sym64_gen->st_name = 0;
489 sym64_fix = find_symbol64(v64, fix);
490 if (sym64_fix == NULL) {
491 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
494 sym64_gen->st_value = sym64_fix->st_value;
495 sym64_gen->st_size = sym64_fix->st_size;
496 sym64_gen->st_info = sym64_fix->st_info;
497 sym64_gen->st_other = sym64_fix->st_other;
498 sym64_gen->st_shndx = sym64_fix->st_shndx;
503 #endif /* CONFIG_PPC64 */
506 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
507 struct lib64_elfinfo *v64)
512 * Locate symbol tables & text section
515 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
516 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
517 if (v32->dynsym == NULL || v32->dynstr == NULL) {
518 printk(KERN_ERR "vDSO32: required symbol section not found\n");
521 sect = find_section32(v32->hdr, ".text", NULL);
523 printk(KERN_ERR "vDSO32: the .text section was not found\n");
526 v32->text = sect - vdso32_kbase;
529 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
530 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
531 if (v64->dynsym == NULL || v64->dynstr == NULL) {
532 printk(KERN_ERR "vDSO64: required symbol section not found\n");
535 sect = find_section64(v64->hdr, ".text", NULL);
537 printk(KERN_ERR "vDSO64: the .text section was not found\n");
540 v64->text = sect - vdso64_kbase;
541 #endif /* CONFIG_PPC64 */
546 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
547 struct lib64_elfinfo *v64)
550 * Find signal trampolines
554 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
556 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
557 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
560 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
561 struct lib64_elfinfo *v64)
567 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
569 printk(KERN_ERR "vDSO64: Can't find symbol "
570 "__kernel_datapage_offset !\n");
573 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
574 (vdso64_pages << PAGE_SHIFT) -
575 (sym64->st_value - VDSO64_LBASE);
576 #endif /* CONFIG_PPC64 */
578 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
580 printk(KERN_ERR "vDSO32: Can't find symbol "
581 "__kernel_datapage_offset !\n");
584 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
585 (vdso32_pages << PAGE_SHIFT) -
586 (sym32->st_value - VDSO32_LBASE);
591 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
592 struct lib64_elfinfo *v64)
596 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
597 struct vdso_patch_def *patch = &vdso_patches[i];
598 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
603 DBG("replacing %s with %s...\n", patch->gen_name,
604 patch->fix_name ? "NONE" : patch->fix_name);
607 * Patch the 32 bits and 64 bits symbols. Note that we do not
608 * patch the "." symbol on 64 bits.
609 * It would be easy to do, but doesn't seem to be necessary,
610 * patching the OPD symbol is enough.
612 vdso_do_func_patch32(v32, v64, patch->gen_name,
615 vdso_do_func_patch64(v32, v64, patch->gen_name,
617 #endif /* CONFIG_PPC64 */
624 static __init int vdso_setup(void)
626 struct lib32_elfinfo v32;
627 struct lib64_elfinfo v64;
629 v32.hdr = vdso32_kbase;
631 v64.hdr = vdso64_kbase;
633 if (vdso_do_find_sections(&v32, &v64))
636 if (vdso_fixup_datapage(&v32, &v64))
639 if (vdso_fixup_alt_funcs(&v32, &v64))
642 vdso_setup_trampolines(&v32, &v64);
648 * Called from setup_arch to initialize the bitmap of available
649 * syscalls in the systemcfg page
651 static void __init vdso_setup_syscall_map(void)
654 extern unsigned long *sys_call_table;
655 extern unsigned long sys_ni_syscall;
658 for (i = 0; i < __NR_syscalls; i++) {
660 if (sys_call_table[i*2] != sys_ni_syscall)
661 vdso_data->syscall_map_64[i >> 5] |=
662 0x80000000UL >> (i & 0x1f);
663 if (sys_call_table[i*2+1] != sys_ni_syscall)
664 vdso_data->syscall_map_32[i >> 5] |=
665 0x80000000UL >> (i & 0x1f);
666 #else /* CONFIG_PPC64 */
667 if (sys_call_table[i] != sys_ni_syscall)
668 vdso_data->syscall_map_32[i >> 5] |=
669 0x80000000UL >> (i & 0x1f);
670 #endif /* CONFIG_PPC64 */
675 void __init vdso_init(void)
681 * Fill up the "systemcfg" stuff for backward compatiblity
683 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
684 vdso_data->version.major = SYSTEMCFG_MAJOR;
685 vdso_data->version.minor = SYSTEMCFG_MINOR;
686 vdso_data->processor = mfspr(SPRN_PVR);
688 * Fake the old platform number for pSeries and iSeries and add
689 * in LPAR bit if necessary
691 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
692 if (firmware_has_feature(FW_FEATURE_LPAR))
693 vdso_data->platform |= 1;
694 vdso_data->physicalMemorySize = lmb_phys_mem_size();
695 vdso_data->dcache_size = ppc64_caches.dsize;
696 vdso_data->dcache_line_size = ppc64_caches.dline_size;
697 vdso_data->icache_size = ppc64_caches.isize;
698 vdso_data->icache_line_size = ppc64_caches.iline_size;
701 * Calculate the size of the 64 bits vDSO
703 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
704 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
705 #endif /* CONFIG_PPC64 */
709 * Calculate the size of the 32 bits vDSO
711 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
712 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
716 * Setup the syscall map in the vDOS
718 vdso_setup_syscall_map();
720 * Initialize the vDSO images in memory, that is do necessary
721 * fixups of vDSO symbols, locate trampolines, etc...
724 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
732 /* Make sure pages are in the correct state */
733 for (i = 0; i < vdso32_pages; i++) {
734 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
735 ClearPageReserved(pg);
740 for (i = 0; i < vdso64_pages; i++) {
741 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
742 ClearPageReserved(pg);
745 #endif /* CONFIG_PPC64 */
747 get_page(virt_to_page(vdso_data));
750 int in_gate_area_no_task(unsigned long addr)
755 int in_gate_area(struct task_struct *task, unsigned long addr)
760 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)