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/config.h>
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/smp_lock.h>
19 #include <linux/stddef.h>
20 #include <linux/unistd.h>
21 #include <linux/slab.h>
22 #include <linux/user.h>
23 #include <linux/elf.h>
24 #include <linux/security.h>
25 #include <linux/bootmem.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>
37 #include <asm/vdso_datapage.h>
42 #define DBG(fmt...) printk(fmt)
47 /* Max supported size for symbol names */
48 #define MAX_SYMNAME 64
50 extern char vdso32_start, vdso32_end;
51 static void *vdso32_kbase = &vdso32_start;
52 unsigned int vdso32_pages;
53 unsigned long vdso32_sigtramp;
54 unsigned long vdso32_rt_sigtramp;
57 extern char vdso64_start, vdso64_end;
58 static void *vdso64_kbase = &vdso64_start;
59 unsigned int vdso64_pages;
60 unsigned long vdso64_rt_sigtramp;
61 #endif /* CONFIG_PPC64 */
64 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
65 * Once the early boot kernel code no longer needs to muck around
66 * with it, it will become dynamically allocated
69 struct vdso_data data;
71 } vdso_data_store __attribute__((__section__(".data.page_aligned")));
72 struct vdso_data *vdso_data = &vdso_data_store.data;
74 /* Format of the patch table */
77 unsigned long ftr_mask, ftr_value;
82 /* Table of functions to patch based on the CPU type/revision
84 * Currently, we only change sync_dicache to do nothing on processors
85 * with a coherent icache
87 static struct vdso_patch_def vdso_patches[] = {
89 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
90 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
94 "__kernel_gettimeofday", NULL
99 * Some infos carried around for each of them during parsing at
104 Elf32_Ehdr *hdr; /* ptr to ELF */
105 Elf32_Sym *dynsym; /* ptr to .dynsym section */
106 unsigned long dynsymsize; /* size of .dynsym section */
107 char *dynstr; /* ptr to .dynstr section */
108 unsigned long text; /* offset of .text section in .so */
115 unsigned long dynsymsize;
122 static void dump_one_vdso_page(struct page *pg, struct page *upg)
124 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
127 if (upg/* && pg != upg*/) {
128 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
136 static void dump_vdso_pages(struct vm_area_struct * vma)
140 if (!vma || test_thread_flag(TIF_32BIT)) {
141 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
142 for (i=0; i<vdso32_pages; i++) {
143 struct page *pg = virt_to_page(vdso32_kbase +
145 struct page *upg = (vma && vma->vm_mm) ?
146 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
148 dump_one_vdso_page(pg, upg);
151 if (!vma || !test_thread_flag(TIF_32BIT)) {
152 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
153 for (i=0; i<vdso64_pages; i++) {
154 struct page *pg = virt_to_page(vdso64_kbase +
156 struct page *upg = (vma && vma->vm_mm) ?
157 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
159 dump_one_vdso_page(pg, upg);
166 * Keep a dummy vma_close for now, it will prevent VMA merging.
168 static void vdso_vma_close(struct vm_area_struct * vma)
173 * Our nopage() function, maps in the actual vDSO kernel pages, they will
174 * be mapped read-only by do_no_page(), and eventually COW'ed, either
175 * right away for an initial write access, or by do_wp_page().
177 static struct page * vdso_vma_nopage(struct vm_area_struct * vma,
178 unsigned long address, int *type)
180 unsigned long offset = address - vma->vm_start;
183 void *vbase = test_thread_flag(TIF_32BIT) ?
184 vdso32_kbase : vdso64_kbase;
186 void *vbase = vdso32_kbase;
189 DBG("vdso_vma_nopage(current: %s, address: %016lx, off: %lx)\n",
190 current->comm, address, offset);
192 if (address < vma->vm_start || address > vma->vm_end)
193 return NOPAGE_SIGBUS;
196 * Last page is systemcfg.
198 if ((vma->vm_end - address) <= PAGE_SIZE)
199 pg = virt_to_page(vdso_data);
201 pg = virt_to_page(vbase + offset);
204 DBG(" ->page count: %d\n", page_count(pg));
209 static struct vm_operations_struct vdso_vmops = {
210 .close = vdso_vma_close,
211 .nopage = vdso_vma_nopage,
215 * This is called from binfmt_elf, we create the special vma for the
216 * vDSO and insert it into the mm struct tree
218 int arch_setup_additional_pages(struct linux_binprm *bprm,
219 int executable_stack)
221 struct mm_struct *mm = current->mm;
222 struct vm_area_struct *vma;
223 unsigned long vdso_pages;
224 unsigned long vdso_base;
227 if (test_thread_flag(TIF_32BIT)) {
228 vdso_pages = vdso32_pages;
229 vdso_base = VDSO32_MBASE;
231 vdso_pages = vdso64_pages;
232 vdso_base = VDSO64_MBASE;
235 vdso_pages = vdso32_pages;
236 vdso_base = VDSO32_MBASE;
239 current->thread.vdso_base = 0;
241 /* vDSO has a problem and was disabled, just don't "enable" it for the
247 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
251 memset(vma, 0, sizeof(*vma));
253 /* Add a page to the vdso size for the data page */
257 * pick a base address for the vDSO in process space. We try to put it
258 * at vdso_base which is the "natural" base for it, but we might fail
259 * and end up putting it elsewhere.
261 vdso_base = get_unmapped_area(NULL, vdso_base,
262 vdso_pages << PAGE_SHIFT, 0, 0);
263 if (vdso_base & ~PAGE_MASK) {
264 kmem_cache_free(vm_area_cachep, vma);
265 return (int)vdso_base;
268 current->thread.vdso_base = vdso_base;
271 vma->vm_start = current->thread.vdso_base;
272 vma->vm_end = vma->vm_start + (vdso_pages << PAGE_SHIFT);
275 * our vma flags don't have VM_WRITE so by default, the process isn't
276 * allowed to write those pages.
277 * gdb can break that with ptrace interface, and thus trigger COW on
278 * those pages but it's then your responsibility to never do that on
279 * the "data" page of the vDSO or you'll stop getting kernel updates
280 * and your nice userland gettimeofday will be totally dead.
281 * It's fine to use that for setting breakpoints in the vDSO code
284 vma->vm_flags = VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
285 vma->vm_flags |= mm->def_flags;
286 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
287 vma->vm_ops = &vdso_vmops;
289 down_write(&mm->mmap_sem);
290 if (insert_vm_struct(mm, vma)) {
291 up_write(&mm->mmap_sem);
292 kmem_cache_free(vm_area_cachep, vma);
295 mm->total_vm += (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
296 up_write(&mm->mmap_sem);
301 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
308 /* Grab section headers and strings so we can tell who is who */
309 sechdrs = (void *)ehdr + ehdr->e_shoff;
310 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
312 /* Find the section they want */
313 for (i = 1; i < ehdr->e_shnum; i++) {
314 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
316 *size = sechdrs[i].sh_size;
317 return (void *)ehdr + sechdrs[i].sh_offset;
324 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
328 char name[MAX_SYMNAME], *c;
330 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
331 if (lib->dynsym[i].st_name == 0)
333 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
335 c = strchr(name, '@');
338 if (strcmp(symname, name) == 0)
339 return &lib->dynsym[i];
344 /* Note that we assume the section is .text and the symbol is relative to
347 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
350 Elf32_Sym *sym = find_symbol32(lib, symname);
353 printk(KERN_WARNING "vDSO32: function %s not found !\n",
357 return sym->st_value - VDSO32_LBASE;
360 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
361 struct lib64_elfinfo *v64,
362 const char *orig, const char *fix)
364 Elf32_Sym *sym32_gen, *sym32_fix;
366 sym32_gen = find_symbol32(v32, orig);
367 if (sym32_gen == NULL) {
368 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
372 sym32_gen->st_name = 0;
375 sym32_fix = find_symbol32(v32, fix);
376 if (sym32_fix == NULL) {
377 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
380 sym32_gen->st_value = sym32_fix->st_value;
381 sym32_gen->st_size = sym32_fix->st_size;
382 sym32_gen->st_info = sym32_fix->st_info;
383 sym32_gen->st_other = sym32_fix->st_other;
384 sym32_gen->st_shndx = sym32_fix->st_shndx;
392 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
399 /* Grab section headers and strings so we can tell who is who */
400 sechdrs = (void *)ehdr + ehdr->e_shoff;
401 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
403 /* Find the section they want */
404 for (i = 1; i < ehdr->e_shnum; i++) {
405 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
407 *size = sechdrs[i].sh_size;
408 return (void *)ehdr + sechdrs[i].sh_offset;
416 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
420 char name[MAX_SYMNAME], *c;
422 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
423 if (lib->dynsym[i].st_name == 0)
425 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
427 c = strchr(name, '@');
430 if (strcmp(symname, name) == 0)
431 return &lib->dynsym[i];
436 /* Note that we assume the section is .text and the symbol is relative to
439 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
442 Elf64_Sym *sym = find_symbol64(lib, symname);
445 printk(KERN_WARNING "vDSO64: function %s not found !\n",
449 #ifdef VDS64_HAS_DESCRIPTORS
450 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
453 return sym->st_value - VDSO64_LBASE;
457 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
458 struct lib64_elfinfo *v64,
459 const char *orig, const char *fix)
461 Elf64_Sym *sym64_gen, *sym64_fix;
463 sym64_gen = find_symbol64(v64, orig);
464 if (sym64_gen == NULL) {
465 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
469 sym64_gen->st_name = 0;
472 sym64_fix = find_symbol64(v64, fix);
473 if (sym64_fix == NULL) {
474 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
477 sym64_gen->st_value = sym64_fix->st_value;
478 sym64_gen->st_size = sym64_fix->st_size;
479 sym64_gen->st_info = sym64_fix->st_info;
480 sym64_gen->st_other = sym64_fix->st_other;
481 sym64_gen->st_shndx = sym64_fix->st_shndx;
486 #endif /* CONFIG_PPC64 */
489 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
490 struct lib64_elfinfo *v64)
495 * Locate symbol tables & text section
498 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
499 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
500 if (v32->dynsym == NULL || v32->dynstr == NULL) {
501 printk(KERN_ERR "vDSO32: required symbol section not found\n");
504 sect = find_section32(v32->hdr, ".text", NULL);
506 printk(KERN_ERR "vDSO32: the .text section was not found\n");
509 v32->text = sect - vdso32_kbase;
512 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
513 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
514 if (v64->dynsym == NULL || v64->dynstr == NULL) {
515 printk(KERN_ERR "vDSO64: required symbol section not found\n");
518 sect = find_section64(v64->hdr, ".text", NULL);
520 printk(KERN_ERR "vDSO64: the .text section was not found\n");
523 v64->text = sect - vdso64_kbase;
524 #endif /* CONFIG_PPC64 */
529 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
530 struct lib64_elfinfo *v64)
533 * Find signal trampolines
537 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
539 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
540 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
543 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
544 struct lib64_elfinfo *v64)
550 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
552 printk(KERN_ERR "vDSO64: Can't find symbol "
553 "__kernel_datapage_offset !\n");
556 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
557 (vdso64_pages << PAGE_SHIFT) -
558 (sym64->st_value - VDSO64_LBASE);
559 #endif /* CONFIG_PPC64 */
561 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
563 printk(KERN_ERR "vDSO32: Can't find symbol "
564 "__kernel_datapage_offset !\n");
567 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
568 (vdso32_pages << PAGE_SHIFT) -
569 (sym32->st_value - VDSO32_LBASE);
574 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
575 struct lib64_elfinfo *v64)
579 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
580 struct vdso_patch_def *patch = &vdso_patches[i];
581 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
586 DBG("replacing %s with %s...\n", patch->gen_name,
587 patch->fix_name ? "NONE" : patch->fix_name);
590 * Patch the 32 bits and 64 bits symbols. Note that we do not
591 * patch the "." symbol on 64 bits.
592 * It would be easy to do, but doesn't seem to be necessary,
593 * patching the OPD symbol is enough.
595 vdso_do_func_patch32(v32, v64, patch->gen_name,
598 vdso_do_func_patch64(v32, v64, patch->gen_name,
600 #endif /* CONFIG_PPC64 */
607 static __init int vdso_setup(void)
609 struct lib32_elfinfo v32;
610 struct lib64_elfinfo v64;
612 v32.hdr = vdso32_kbase;
614 v64.hdr = vdso64_kbase;
616 if (vdso_do_find_sections(&v32, &v64))
619 if (vdso_fixup_datapage(&v32, &v64))
622 if (vdso_fixup_alt_funcs(&v32, &v64))
625 vdso_setup_trampolines(&v32, &v64);
631 * Called from setup_arch to initialize the bitmap of available
632 * syscalls in the systemcfg page
634 static void __init vdso_setup_syscall_map(void)
637 extern unsigned long *sys_call_table;
638 extern unsigned long sys_ni_syscall;
641 for (i = 0; i < __NR_syscalls; i++) {
643 if (sys_call_table[i*2] != sys_ni_syscall)
644 vdso_data->syscall_map_64[i >> 5] |=
645 0x80000000UL >> (i & 0x1f);
646 if (sys_call_table[i*2+1] != sys_ni_syscall)
647 vdso_data->syscall_map_32[i >> 5] |=
648 0x80000000UL >> (i & 0x1f);
649 #else /* CONFIG_PPC64 */
650 if (sys_call_table[i] != sys_ni_syscall)
651 vdso_data->syscall_map_32[i >> 5] |=
652 0x80000000UL >> (i & 0x1f);
653 #endif /* CONFIG_PPC64 */
658 void __init vdso_init(void)
664 * Fill up the "systemcfg" stuff for backward compatiblity
666 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
667 vdso_data->version.major = SYSTEMCFG_MAJOR;
668 vdso_data->version.minor = SYSTEMCFG_MINOR;
669 vdso_data->processor = mfspr(SPRN_PVR);
670 vdso_data->platform = _machine;
671 vdso_data->physicalMemorySize = lmb_phys_mem_size();
672 vdso_data->dcache_size = ppc64_caches.dsize;
673 vdso_data->dcache_line_size = ppc64_caches.dline_size;
674 vdso_data->icache_size = ppc64_caches.isize;
675 vdso_data->icache_line_size = ppc64_caches.iline_size;
678 * Calculate the size of the 64 bits vDSO
680 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
681 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
682 #endif /* CONFIG_PPC64 */
686 * Calculate the size of the 32 bits vDSO
688 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
689 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
693 * Setup the syscall map in the vDOS
695 vdso_setup_syscall_map();
697 * Initialize the vDSO images in memory, that is do necessary
698 * fixups of vDSO symbols, locate trampolines, etc...
701 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
709 /* Make sure pages are in the correct state */
710 for (i = 0; i < vdso32_pages; i++) {
711 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
712 ClearPageReserved(pg);
717 for (i = 0; i < vdso64_pages; i++) {
718 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
719 ClearPageReserved(pg);
722 #endif /* CONFIG_PPC64 */
724 get_page(virt_to_page(vdso_data));
727 int in_gate_area_no_task(unsigned long addr)
732 int in_gate_area(struct task_struct *task, unsigned long addr)
737 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)