2 * linux/arch/ppc64/kernel/vdso.c
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #include <linux/config.h>
14 #include <linux/module.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
19 #include <linux/smp.h>
20 #include <linux/smp_lock.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/elf.h>
26 #include <linux/security.h>
27 #include <linux/bootmem.h>
29 #include <asm/pgtable.h>
30 #include <asm/system.h>
31 #include <asm/processor.h>
33 #include <asm/mmu_context.h>
35 #include <asm/machdep.h>
36 #include <asm/cputable.h>
37 #include <asm/sections.h>
39 #include <asm/vdso_datapage.h>
44 #define DBG(fmt...) printk(fmt)
49 /* Max supported size for symbol names */
50 #define MAX_SYMNAME 64
52 extern char vdso32_start, vdso32_end;
53 static void *vdso32_kbase = &vdso32_start;
54 unsigned int vdso32_pages;
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 unsigned int vdso64_pages;
62 unsigned long vdso64_rt_sigtramp;
63 #endif /* CONFIG_PPC64 */
66 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
67 * Once the early boot kernel code no longer needs to muck around
68 * with it, it will become dynamically allocated
71 struct vdso_data data;
73 } vdso_data_store __attribute__((__section__(".data.page_aligned")));
74 struct vdso_data *vdso_data = &vdso_data_store.data;
76 /* Format of the patch table */
79 unsigned long ftr_mask, ftr_value;
84 /* Table of functions to patch based on the CPU type/revision
86 * Currently, we only change sync_dicache to do nothing on processors
87 * with a coherent icache
89 static struct vdso_patch_def vdso_patches[] = {
91 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
92 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
96 "__kernel_gettimeofday", NULL
101 * Some infos carried around for each of them during parsing at
106 Elf32_Ehdr *hdr; /* ptr to ELF */
107 Elf32_Sym *dynsym; /* ptr to .dynsym section */
108 unsigned long dynsymsize; /* size of .dynsym section */
109 char *dynstr; /* ptr to .dynstr section */
110 unsigned long text; /* offset of .text section in .so */
117 unsigned long dynsymsize;
124 static void dump_one_vdso_page(struct page *pg, struct page *upg)
126 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
129 if (upg/* && pg != upg*/) {
130 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
138 static void dump_vdso_pages(struct vm_area_struct * vma)
142 if (!vma || test_thread_flag(TIF_32BIT)) {
143 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
144 for (i=0; i<vdso32_pages; i++) {
145 struct page *pg = virt_to_page(vdso32_kbase +
147 struct page *upg = (vma && vma->vm_mm) ?
148 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
150 dump_one_vdso_page(pg, upg);
153 if (!vma || !test_thread_flag(TIF_32BIT)) {
154 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
155 for (i=0; i<vdso64_pages; i++) {
156 struct page *pg = virt_to_page(vdso64_kbase +
158 struct page *upg = (vma && vma->vm_mm) ?
159 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
161 dump_one_vdso_page(pg, upg);
168 * Keep a dummy vma_close for now, it will prevent VMA merging.
170 static void vdso_vma_close(struct vm_area_struct * vma)
175 * Our nopage() function, maps in the actual vDSO kernel pages, they will
176 * be mapped read-only by do_no_page(), and eventually COW'ed, either
177 * right away for an initial write access, or by do_wp_page().
179 static struct page * vdso_vma_nopage(struct vm_area_struct * vma,
180 unsigned long address, int *type)
182 unsigned long offset = address - vma->vm_start;
185 void *vbase = (vma->vm_mm->task_size > TASK_SIZE_USER32) ?
186 vdso64_kbase : vdso32_kbase;
188 void *vbase = vdso32_kbase;
191 DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
192 current->comm, address, offset);
194 if (address < vma->vm_start || address > vma->vm_end)
195 return NOPAGE_SIGBUS;
198 * Last page is systemcfg.
200 if ((vma->vm_end - address) <= PAGE_SIZE)
201 pg = virt_to_page(vdso_data);
203 pg = virt_to_page(vbase + offset);
206 DBG(" ->page count: %d\n", page_count(pg));
211 static struct vm_operations_struct vdso_vmops = {
212 .close = vdso_vma_close,
213 .nopage = vdso_vma_nopage,
217 * This is called from binfmt_elf, we create the special vma for the
218 * vDSO and insert it into the mm struct tree
220 int arch_setup_additional_pages(struct linux_binprm *bprm,
221 int executable_stack)
223 struct mm_struct *mm = current->mm;
224 struct vm_area_struct *vma;
225 unsigned long vdso_pages;
226 unsigned long vdso_base;
229 if (test_thread_flag(TIF_32BIT)) {
230 vdso_pages = vdso32_pages;
231 vdso_base = VDSO32_MBASE;
233 vdso_pages = vdso64_pages;
234 vdso_base = VDSO64_MBASE;
237 vdso_pages = vdso32_pages;
238 vdso_base = VDSO32_MBASE;
241 current->thread.vdso_base = 0;
243 /* vDSO has a problem and was disabled, just don't "enable" it for the
249 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
253 memset(vma, 0, sizeof(*vma));
255 /* Add a page to the vdso size for the data page */
259 * pick a base address for the vDSO in process space. We try to put it
260 * at vdso_base which is the "natural" base for it, but we might fail
261 * and end up putting it elsewhere.
263 vdso_base = get_unmapped_area(NULL, vdso_base,
264 vdso_pages << PAGE_SHIFT, 0, 0);
265 if (vdso_base & ~PAGE_MASK) {
266 kmem_cache_free(vm_area_cachep, vma);
267 return (int)vdso_base;
270 current->thread.vdso_base = vdso_base;
273 vma->vm_start = current->thread.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;
291 down_write(&mm->mmap_sem);
292 if (insert_vm_struct(mm, vma)) {
293 up_write(&mm->mmap_sem);
294 kmem_cache_free(vm_area_cachep, vma);
297 mm->total_vm += (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
298 up_write(&mm->mmap_sem);
303 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
310 /* Grab section headers and strings so we can tell who is who */
311 sechdrs = (void *)ehdr + ehdr->e_shoff;
312 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
314 /* Find the section they want */
315 for (i = 1; i < ehdr->e_shnum; i++) {
316 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
318 *size = sechdrs[i].sh_size;
319 return (void *)ehdr + sechdrs[i].sh_offset;
326 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
330 char name[MAX_SYMNAME], *c;
332 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
333 if (lib->dynsym[i].st_name == 0)
335 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
337 c = strchr(name, '@');
340 if (strcmp(symname, name) == 0)
341 return &lib->dynsym[i];
346 /* Note that we assume the section is .text and the symbol is relative to
349 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
352 Elf32_Sym *sym = find_symbol32(lib, symname);
355 printk(KERN_WARNING "vDSO32: function %s not found !\n",
359 return sym->st_value - VDSO32_LBASE;
362 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
363 struct lib64_elfinfo *v64,
364 const char *orig, const char *fix)
366 Elf32_Sym *sym32_gen, *sym32_fix;
368 sym32_gen = find_symbol32(v32, orig);
369 if (sym32_gen == NULL) {
370 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
374 sym32_gen->st_name = 0;
377 sym32_fix = find_symbol32(v32, fix);
378 if (sym32_fix == NULL) {
379 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
382 sym32_gen->st_value = sym32_fix->st_value;
383 sym32_gen->st_size = sym32_fix->st_size;
384 sym32_gen->st_info = sym32_fix->st_info;
385 sym32_gen->st_other = sym32_fix->st_other;
386 sym32_gen->st_shndx = sym32_fix->st_shndx;
394 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
401 /* Grab section headers and strings so we can tell who is who */
402 sechdrs = (void *)ehdr + ehdr->e_shoff;
403 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
405 /* Find the section they want */
406 for (i = 1; i < ehdr->e_shnum; i++) {
407 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
409 *size = sechdrs[i].sh_size;
410 return (void *)ehdr + sechdrs[i].sh_offset;
418 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
422 char name[MAX_SYMNAME], *c;
424 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
425 if (lib->dynsym[i].st_name == 0)
427 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
429 c = strchr(name, '@');
432 if (strcmp(symname, name) == 0)
433 return &lib->dynsym[i];
438 /* Note that we assume the section is .text and the symbol is relative to
441 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
444 Elf64_Sym *sym = find_symbol64(lib, symname);
447 printk(KERN_WARNING "vDSO64: function %s not found !\n",
451 #ifdef VDS64_HAS_DESCRIPTORS
452 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
455 return sym->st_value - VDSO64_LBASE;
459 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
460 struct lib64_elfinfo *v64,
461 const char *orig, const char *fix)
463 Elf64_Sym *sym64_gen, *sym64_fix;
465 sym64_gen = find_symbol64(v64, orig);
466 if (sym64_gen == NULL) {
467 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
471 sym64_gen->st_name = 0;
474 sym64_fix = find_symbol64(v64, fix);
475 if (sym64_fix == NULL) {
476 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
479 sym64_gen->st_value = sym64_fix->st_value;
480 sym64_gen->st_size = sym64_fix->st_size;
481 sym64_gen->st_info = sym64_fix->st_info;
482 sym64_gen->st_other = sym64_fix->st_other;
483 sym64_gen->st_shndx = sym64_fix->st_shndx;
488 #endif /* CONFIG_PPC64 */
491 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
492 struct lib64_elfinfo *v64)
497 * Locate symbol tables & text section
500 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
501 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
502 if (v32->dynsym == NULL || v32->dynstr == NULL) {
503 printk(KERN_ERR "vDSO32: required symbol section not found\n");
506 sect = find_section32(v32->hdr, ".text", NULL);
508 printk(KERN_ERR "vDSO32: the .text section was not found\n");
511 v32->text = sect - vdso32_kbase;
514 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
515 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
516 if (v64->dynsym == NULL || v64->dynstr == NULL) {
517 printk(KERN_ERR "vDSO64: required symbol section not found\n");
520 sect = find_section64(v64->hdr, ".text", NULL);
522 printk(KERN_ERR "vDSO64: the .text section was not found\n");
525 v64->text = sect - vdso64_kbase;
526 #endif /* CONFIG_PPC64 */
531 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
532 struct lib64_elfinfo *v64)
535 * Find signal trampolines
539 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
541 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
542 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
545 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
546 struct lib64_elfinfo *v64)
552 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
554 printk(KERN_ERR "vDSO64: Can't find symbol "
555 "__kernel_datapage_offset !\n");
558 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
559 (vdso64_pages << PAGE_SHIFT) -
560 (sym64->st_value - VDSO64_LBASE);
561 #endif /* CONFIG_PPC64 */
563 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
565 printk(KERN_ERR "vDSO32: Can't find symbol "
566 "__kernel_datapage_offset !\n");
569 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
570 (vdso32_pages << PAGE_SHIFT) -
571 (sym32->st_value - VDSO32_LBASE);
576 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
577 struct lib64_elfinfo *v64)
581 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
582 struct vdso_patch_def *patch = &vdso_patches[i];
583 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
588 DBG("replacing %s with %s...\n", patch->gen_name,
589 patch->fix_name ? "NONE" : patch->fix_name);
592 * Patch the 32 bits and 64 bits symbols. Note that we do not
593 * patch the "." symbol on 64 bits.
594 * It would be easy to do, but doesn't seem to be necessary,
595 * patching the OPD symbol is enough.
597 vdso_do_func_patch32(v32, v64, patch->gen_name,
600 vdso_do_func_patch64(v32, v64, patch->gen_name,
602 #endif /* CONFIG_PPC64 */
609 static __init int vdso_setup(void)
611 struct lib32_elfinfo v32;
612 struct lib64_elfinfo v64;
614 v32.hdr = vdso32_kbase;
616 v64.hdr = vdso64_kbase;
618 if (vdso_do_find_sections(&v32, &v64))
621 if (vdso_fixup_datapage(&v32, &v64))
624 if (vdso_fixup_alt_funcs(&v32, &v64))
627 vdso_setup_trampolines(&v32, &v64);
633 * Called from setup_arch to initialize the bitmap of available
634 * syscalls in the systemcfg page
636 static void __init vdso_setup_syscall_map(void)
639 extern unsigned long *sys_call_table;
640 extern unsigned long sys_ni_syscall;
643 for (i = 0; i < __NR_syscalls; i++) {
645 if (sys_call_table[i*2] != sys_ni_syscall)
646 vdso_data->syscall_map_64[i >> 5] |=
647 0x80000000UL >> (i & 0x1f);
648 if (sys_call_table[i*2+1] != sys_ni_syscall)
649 vdso_data->syscall_map_32[i >> 5] |=
650 0x80000000UL >> (i & 0x1f);
651 #else /* CONFIG_PPC64 */
652 if (sys_call_table[i] != sys_ni_syscall)
653 vdso_data->syscall_map_32[i >> 5] |=
654 0x80000000UL >> (i & 0x1f);
655 #endif /* CONFIG_PPC64 */
660 void __init vdso_init(void)
666 * Fill up the "systemcfg" stuff for backward compatiblity
668 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
669 vdso_data->version.major = SYSTEMCFG_MAJOR;
670 vdso_data->version.minor = SYSTEMCFG_MINOR;
671 vdso_data->processor = mfspr(SPRN_PVR);
672 vdso_data->platform = _machine;
673 vdso_data->physicalMemorySize = lmb_phys_mem_size();
674 vdso_data->dcache_size = ppc64_caches.dsize;
675 vdso_data->dcache_line_size = ppc64_caches.dline_size;
676 vdso_data->icache_size = ppc64_caches.isize;
677 vdso_data->icache_line_size = ppc64_caches.iline_size;
680 * Calculate the size of the 64 bits vDSO
682 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
683 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
684 #endif /* CONFIG_PPC64 */
688 * Calculate the size of the 32 bits vDSO
690 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
691 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
695 * Setup the syscall map in the vDOS
697 vdso_setup_syscall_map();
699 * Initialize the vDSO images in memory, that is do necessary
700 * fixups of vDSO symbols, locate trampolines, etc...
703 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
711 /* Make sure pages are in the correct state */
712 for (i = 0; i < vdso32_pages; i++) {
713 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
714 ClearPageReserved(pg);
719 for (i = 0; i < vdso64_pages; i++) {
720 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
721 ClearPageReserved(pg);
724 #endif /* CONFIG_PPC64 */
726 get_page(virt_to_page(vdso_data));
729 int in_gate_area_no_task(unsigned long addr)
734 int in_gate_area(struct task_struct *task, unsigned long addr)
739 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)