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>
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)
222 struct mm_struct *mm = current->mm;
223 struct vm_area_struct *vma;
224 unsigned long vdso_pages;
225 unsigned long vdso_base;
228 if (test_thread_flag(TIF_32BIT)) {
229 vdso_pages = vdso32_pages;
230 vdso_base = VDSO32_MBASE;
232 vdso_pages = vdso64_pages;
233 vdso_base = VDSO64_MBASE;
236 vdso_pages = vdso32_pages;
237 vdso_base = VDSO32_MBASE;
240 current->thread.vdso_base = 0;
242 /* vDSO has a problem and was disabled, just don't "enable" it for the
248 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
252 memset(vma, 0, sizeof(*vma));
254 /* Add a page to the vdso size for the data page */
258 * pick a base address for the vDSO in process space. We try to put it
259 * at vdso_base which is the "natural" base for it, but we might fail
260 * and end up putting it elsewhere.
262 vdso_base = get_unmapped_area(NULL, vdso_base,
263 vdso_pages << PAGE_SHIFT, 0, 0);
264 if (vdso_base & ~PAGE_MASK) {
265 kmem_cache_free(vm_area_cachep, vma);
266 return (int)vdso_base;
269 current->thread.vdso_base = vdso_base;
272 vma->vm_start = current->thread.vdso_base;
273 vma->vm_end = vma->vm_start + (vdso_pages << PAGE_SHIFT);
276 * our vma flags don't have VM_WRITE so by default, the process isn't
277 * allowed to write those pages.
278 * gdb can break that with ptrace interface, and thus trigger COW on
279 * those pages but it's then your responsibility to never do that on
280 * the "data" page of the vDSO or you'll stop getting kernel updates
281 * and your nice userland gettimeofday will be totally dead.
282 * It's fine to use that for setting breakpoints in the vDSO code
285 vma->vm_flags = VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
286 vma->vm_flags |= mm->def_flags;
287 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
288 vma->vm_ops = &vdso_vmops;
290 down_write(&mm->mmap_sem);
291 if (insert_vm_struct(mm, vma)) {
292 up_write(&mm->mmap_sem);
293 kmem_cache_free(vm_area_cachep, vma);
296 mm->total_vm += (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
297 up_write(&mm->mmap_sem);
302 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
309 /* Grab section headers and strings so we can tell who is who */
310 sechdrs = (void *)ehdr + ehdr->e_shoff;
311 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
313 /* Find the section they want */
314 for (i = 1; i < ehdr->e_shnum; i++) {
315 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
317 *size = sechdrs[i].sh_size;
318 return (void *)ehdr + sechdrs[i].sh_offset;
325 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
329 char name[MAX_SYMNAME], *c;
331 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
332 if (lib->dynsym[i].st_name == 0)
334 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
336 c = strchr(name, '@');
339 if (strcmp(symname, name) == 0)
340 return &lib->dynsym[i];
345 /* Note that we assume the section is .text and the symbol is relative to
348 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
351 Elf32_Sym *sym = find_symbol32(lib, symname);
354 printk(KERN_WARNING "vDSO32: function %s not found !\n",
358 return sym->st_value - VDSO32_LBASE;
361 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
362 struct lib64_elfinfo *v64,
363 const char *orig, const char *fix)
365 Elf32_Sym *sym32_gen, *sym32_fix;
367 sym32_gen = find_symbol32(v32, orig);
368 if (sym32_gen == NULL) {
369 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
373 sym32_gen->st_name = 0;
376 sym32_fix = find_symbol32(v32, fix);
377 if (sym32_fix == NULL) {
378 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
381 sym32_gen->st_value = sym32_fix->st_value;
382 sym32_gen->st_size = sym32_fix->st_size;
383 sym32_gen->st_info = sym32_fix->st_info;
384 sym32_gen->st_other = sym32_fix->st_other;
385 sym32_gen->st_shndx = sym32_fix->st_shndx;
393 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
400 /* Grab section headers and strings so we can tell who is who */
401 sechdrs = (void *)ehdr + ehdr->e_shoff;
402 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
404 /* Find the section they want */
405 for (i = 1; i < ehdr->e_shnum; i++) {
406 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
408 *size = sechdrs[i].sh_size;
409 return (void *)ehdr + sechdrs[i].sh_offset;
417 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
421 char name[MAX_SYMNAME], *c;
423 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
424 if (lib->dynsym[i].st_name == 0)
426 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
428 c = strchr(name, '@');
431 if (strcmp(symname, name) == 0)
432 return &lib->dynsym[i];
437 /* Note that we assume the section is .text and the symbol is relative to
440 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
443 Elf64_Sym *sym = find_symbol64(lib, symname);
446 printk(KERN_WARNING "vDSO64: function %s not found !\n",
450 #ifdef VDS64_HAS_DESCRIPTORS
451 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
454 return sym->st_value - VDSO64_LBASE;
458 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
459 struct lib64_elfinfo *v64,
460 const char *orig, const char *fix)
462 Elf64_Sym *sym64_gen, *sym64_fix;
464 sym64_gen = find_symbol64(v64, orig);
465 if (sym64_gen == NULL) {
466 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
470 sym64_gen->st_name = 0;
473 sym64_fix = find_symbol64(v64, fix);
474 if (sym64_fix == NULL) {
475 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
478 sym64_gen->st_value = sym64_fix->st_value;
479 sym64_gen->st_size = sym64_fix->st_size;
480 sym64_gen->st_info = sym64_fix->st_info;
481 sym64_gen->st_other = sym64_fix->st_other;
482 sym64_gen->st_shndx = sym64_fix->st_shndx;
487 #endif /* CONFIG_PPC64 */
490 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
491 struct lib64_elfinfo *v64)
496 * Locate symbol tables & text section
499 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
500 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
501 if (v32->dynsym == NULL || v32->dynstr == NULL) {
502 printk(KERN_ERR "vDSO32: required symbol section not found\n");
505 sect = find_section32(v32->hdr, ".text", NULL);
507 printk(KERN_ERR "vDSO32: the .text section was not found\n");
510 v32->text = sect - vdso32_kbase;
513 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
514 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
515 if (v64->dynsym == NULL || v64->dynstr == NULL) {
516 printk(KERN_ERR "vDSO64: required symbol section not found\n");
519 sect = find_section64(v64->hdr, ".text", NULL);
521 printk(KERN_ERR "vDSO64: the .text section was not found\n");
524 v64->text = sect - vdso64_kbase;
525 #endif /* CONFIG_PPC64 */
530 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
531 struct lib64_elfinfo *v64)
534 * Find signal trampolines
538 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
540 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
541 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
544 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
545 struct lib64_elfinfo *v64)
551 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
553 printk(KERN_ERR "vDSO64: Can't find symbol "
554 "__kernel_datapage_offset !\n");
557 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
558 (vdso64_pages << PAGE_SHIFT) -
559 (sym64->st_value - VDSO64_LBASE);
560 #endif /* CONFIG_PPC64 */
562 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
564 printk(KERN_ERR "vDSO32: Can't find symbol "
565 "__kernel_datapage_offset !\n");
568 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
569 (vdso32_pages << PAGE_SHIFT) -
570 (sym32->st_value - VDSO32_LBASE);
575 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
576 struct lib64_elfinfo *v64)
580 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
581 struct vdso_patch_def *patch = &vdso_patches[i];
582 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
587 DBG("replacing %s with %s...\n", patch->gen_name,
588 patch->fix_name ? "NONE" : patch->fix_name);
591 * Patch the 32 bits and 64 bits symbols. Note that we do not
592 * patch the "." symbol on 64 bits.
593 * It would be easy to do, but doesn't seem to be necessary,
594 * patching the OPD symbol is enough.
596 vdso_do_func_patch32(v32, v64, patch->gen_name,
599 vdso_do_func_patch64(v32, v64, patch->gen_name,
601 #endif /* CONFIG_PPC64 */
608 static __init int vdso_setup(void)
610 struct lib32_elfinfo v32;
611 struct lib64_elfinfo v64;
613 v32.hdr = vdso32_kbase;
615 v64.hdr = vdso64_kbase;
617 if (vdso_do_find_sections(&v32, &v64))
620 if (vdso_fixup_datapage(&v32, &v64))
623 if (vdso_fixup_alt_funcs(&v32, &v64))
626 vdso_setup_trampolines(&v32, &v64);
632 * Called from setup_arch to initialize the bitmap of available
633 * syscalls in the systemcfg page
635 static void __init vdso_setup_syscall_map(void)
638 extern unsigned long *sys_call_table;
639 extern unsigned long sys_ni_syscall;
642 for (i = 0; i < __NR_syscalls; i++) {
644 if (sys_call_table[i*2] != sys_ni_syscall)
645 vdso_data->syscall_map_64[i >> 5] |=
646 0x80000000UL >> (i & 0x1f);
647 if (sys_call_table[i*2+1] != sys_ni_syscall)
648 vdso_data->syscall_map_32[i >> 5] |=
649 0x80000000UL >> (i & 0x1f);
650 #else /* CONFIG_PPC64 */
651 if (sys_call_table[i] != sys_ni_syscall)
652 vdso_data->syscall_map_32[i >> 5] |=
653 0x80000000UL >> (i & 0x1f);
654 #endif /* CONFIG_PPC64 */
659 void __init vdso_init(void)
665 * Fill up the "systemcfg" stuff for backward compatiblity
667 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
668 vdso_data->version.major = SYSTEMCFG_MAJOR;
669 vdso_data->version.minor = SYSTEMCFG_MINOR;
670 vdso_data->processor = mfspr(SPRN_PVR);
672 * Fake the old platform number for pSeries and iSeries and add
673 * in LPAR bit if necessary
675 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
676 if (firmware_has_feature(FW_FEATURE_LPAR))
677 vdso_data->platform |= 1;
678 vdso_data->physicalMemorySize = lmb_phys_mem_size();
679 vdso_data->dcache_size = ppc64_caches.dsize;
680 vdso_data->dcache_line_size = ppc64_caches.dline_size;
681 vdso_data->icache_size = ppc64_caches.isize;
682 vdso_data->icache_line_size = ppc64_caches.iline_size;
685 * Calculate the size of the 64 bits vDSO
687 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
688 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
689 #endif /* CONFIG_PPC64 */
693 * Calculate the size of the 32 bits vDSO
695 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
696 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
700 * Setup the syscall map in the vDOS
702 vdso_setup_syscall_map();
704 * Initialize the vDSO images in memory, that is do necessary
705 * fixups of vDSO symbols, locate trampolines, etc...
708 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
716 /* Make sure pages are in the correct state */
717 for (i = 0; i < vdso32_pages; i++) {
718 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
719 ClearPageReserved(pg);
724 for (i = 0; i < vdso64_pages; i++) {
725 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
726 ClearPageReserved(pg);
729 #endif /* CONFIG_PPC64 */
731 get_page(virt_to_page(vdso_data));
734 int in_gate_area_no_task(unsigned long addr)
739 int in_gate_area(struct task_struct *task, unsigned long addr)
744 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)