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/stddef.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/user.h>
21 #include <linux/elf.h>
22 #include <linux/security.h>
23 #include <linux/bootmem.h>
24 #include <linux/lmb.h>
26 #include <asm/pgtable.h>
27 #include <asm/system.h>
28 #include <asm/processor.h>
30 #include <asm/mmu_context.h>
32 #include <asm/machdep.h>
33 #include <asm/cputable.h>
34 #include <asm/sections.h>
35 #include <asm/firmware.h>
37 #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 static unsigned int vdso32_pages;
55 static struct page **vdso32_pagelist;
56 unsigned long vdso32_sigtramp;
57 unsigned long vdso32_rt_sigtramp;
60 extern char vdso64_start, vdso64_end;
61 static void *vdso64_kbase = &vdso64_start;
62 static unsigned int vdso64_pages;
63 static struct page **vdso64_pagelist;
64 unsigned long vdso64_rt_sigtramp;
65 #endif /* CONFIG_PPC64 */
67 static int vdso_ready;
70 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
71 * Once the early boot kernel code no longer needs to muck around
72 * with it, it will become dynamically allocated
75 struct vdso_data data;
77 } vdso_data_store __attribute__((__section__(".data.page_aligned")));
78 struct vdso_data *vdso_data = &vdso_data_store.data;
80 /* Format of the patch table */
83 unsigned long ftr_mask, ftr_value;
88 /* Table of functions to patch based on the CPU type/revision
90 * Currently, we only change sync_dicache to do nothing on processors
91 * with a coherent icache
93 static struct vdso_patch_def vdso_patches[] = {
95 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
96 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
100 "__kernel_gettimeofday", NULL
104 "__kernel_clock_gettime", NULL
108 "__kernel_clock_getres", NULL
112 "__kernel_get_tbfreq", NULL
117 * Some infos carried around for each of them during parsing at
122 Elf32_Ehdr *hdr; /* ptr to ELF */
123 Elf32_Sym *dynsym; /* ptr to .dynsym section */
124 unsigned long dynsymsize; /* size of .dynsym section */
125 char *dynstr; /* ptr to .dynstr section */
126 unsigned long text; /* offset of .text section in .so */
133 unsigned long dynsymsize;
140 static void dump_one_vdso_page(struct page *pg, struct page *upg)
142 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
145 if (upg && !IS_ERR(upg) /* && pg != upg*/) {
146 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
154 static void dump_vdso_pages(struct vm_area_struct * vma)
158 if (!vma || test_thread_flag(TIF_32BIT)) {
159 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
160 for (i=0; i<vdso32_pages; i++) {
161 struct page *pg = virt_to_page(vdso32_kbase +
163 struct page *upg = (vma && vma->vm_mm) ?
164 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
166 dump_one_vdso_page(pg, upg);
169 if (!vma || !test_thread_flag(TIF_32BIT)) {
170 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
171 for (i=0; i<vdso64_pages; i++) {
172 struct page *pg = virt_to_page(vdso64_kbase +
174 struct page *upg = (vma && vma->vm_mm) ?
175 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
177 dump_one_vdso_page(pg, upg);
184 * This is called from binfmt_elf, we create the special vma for the
185 * vDSO and insert it into the mm struct tree
187 int arch_setup_additional_pages(struct linux_binprm *bprm,
188 int executable_stack)
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;
207 vdso_base = VDSO64_MBASE;
210 vdso_pagelist = vdso32_pagelist;
211 vdso_pages = vdso32_pages;
212 vdso_base = VDSO32_MBASE;
215 current->mm->context.vdso_base = 0;
217 /* vDSO has a problem and was disabled, just don't "enable" it for the
222 /* Add a page to the vdso size for the data page */
226 * pick a base address for the vDSO in process space. We try to put it
227 * at vdso_base which is the "natural" base for it, but we might fail
228 * and end up putting it elsewhere.
230 down_write(&mm->mmap_sem);
231 vdso_base = get_unmapped_area(NULL, vdso_base,
232 vdso_pages << PAGE_SHIFT, 0, 0);
233 if (IS_ERR_VALUE(vdso_base)) {
239 * our vma flags don't have VM_WRITE so by default, the process isn't
240 * allowed to write those pages.
241 * gdb can break that with ptrace interface, and thus trigger COW on
242 * those pages but it's then your responsibility to never do that on
243 * the "data" page of the vDSO or you'll stop getting kernel updates
244 * and your nice userland gettimeofday will be totally dead.
245 * It's fine to use that for setting breakpoints in the vDSO code
248 * Make sure the vDSO gets into every core dump.
249 * Dumping its contents makes post-mortem fully interpretable later
250 * without matching up the same kernel and hardware config to see
251 * what PC values meant.
253 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
255 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
261 /* Put vDSO base into mm struct */
262 current->mm->context.vdso_base = vdso_base;
264 up_write(&mm->mmap_sem);
268 up_write(&mm->mmap_sem);
272 const char *arch_vma_name(struct vm_area_struct *vma)
274 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
281 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
288 /* Grab section headers and strings so we can tell who is who */
289 sechdrs = (void *)ehdr + ehdr->e_shoff;
290 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
292 /* Find the section they want */
293 for (i = 1; i < ehdr->e_shnum; i++) {
294 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
296 *size = sechdrs[i].sh_size;
297 return (void *)ehdr + sechdrs[i].sh_offset;
304 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
308 char name[MAX_SYMNAME], *c;
310 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
311 if (lib->dynsym[i].st_name == 0)
313 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
315 c = strchr(name, '@');
318 if (strcmp(symname, name) == 0)
319 return &lib->dynsym[i];
324 /* Note that we assume the section is .text and the symbol is relative to
327 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
330 Elf32_Sym *sym = find_symbol32(lib, symname);
333 printk(KERN_WARNING "vDSO32: function %s not found !\n",
337 return sym->st_value - VDSO32_LBASE;
340 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
341 struct lib64_elfinfo *v64,
342 const char *orig, const char *fix)
344 Elf32_Sym *sym32_gen, *sym32_fix;
346 sym32_gen = find_symbol32(v32, orig);
347 if (sym32_gen == NULL) {
348 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
352 sym32_gen->st_name = 0;
355 sym32_fix = find_symbol32(v32, fix);
356 if (sym32_fix == NULL) {
357 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
360 sym32_gen->st_value = sym32_fix->st_value;
361 sym32_gen->st_size = sym32_fix->st_size;
362 sym32_gen->st_info = sym32_fix->st_info;
363 sym32_gen->st_other = sym32_fix->st_other;
364 sym32_gen->st_shndx = sym32_fix->st_shndx;
372 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
379 /* Grab section headers and strings so we can tell who is who */
380 sechdrs = (void *)ehdr + ehdr->e_shoff;
381 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
383 /* Find the section they want */
384 for (i = 1; i < ehdr->e_shnum; i++) {
385 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
387 *size = sechdrs[i].sh_size;
388 return (void *)ehdr + sechdrs[i].sh_offset;
396 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
400 char name[MAX_SYMNAME], *c;
402 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
403 if (lib->dynsym[i].st_name == 0)
405 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
407 c = strchr(name, '@');
410 if (strcmp(symname, name) == 0)
411 return &lib->dynsym[i];
416 /* Note that we assume the section is .text and the symbol is relative to
419 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
422 Elf64_Sym *sym = find_symbol64(lib, symname);
425 printk(KERN_WARNING "vDSO64: function %s not found !\n",
429 #ifdef VDS64_HAS_DESCRIPTORS
430 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
433 return sym->st_value - VDSO64_LBASE;
437 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
438 struct lib64_elfinfo *v64,
439 const char *orig, const char *fix)
441 Elf64_Sym *sym64_gen, *sym64_fix;
443 sym64_gen = find_symbol64(v64, orig);
444 if (sym64_gen == NULL) {
445 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
449 sym64_gen->st_name = 0;
452 sym64_fix = find_symbol64(v64, fix);
453 if (sym64_fix == NULL) {
454 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
457 sym64_gen->st_value = sym64_fix->st_value;
458 sym64_gen->st_size = sym64_fix->st_size;
459 sym64_gen->st_info = sym64_fix->st_info;
460 sym64_gen->st_other = sym64_fix->st_other;
461 sym64_gen->st_shndx = sym64_fix->st_shndx;
466 #endif /* CONFIG_PPC64 */
469 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
470 struct lib64_elfinfo *v64)
475 * Locate symbol tables & text section
478 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
479 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
480 if (v32->dynsym == NULL || v32->dynstr == NULL) {
481 printk(KERN_ERR "vDSO32: required symbol section not found\n");
484 sect = find_section32(v32->hdr, ".text", NULL);
486 printk(KERN_ERR "vDSO32: the .text section was not found\n");
489 v32->text = sect - vdso32_kbase;
492 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
493 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
494 if (v64->dynsym == NULL || v64->dynstr == NULL) {
495 printk(KERN_ERR "vDSO64: required symbol section not found\n");
498 sect = find_section64(v64->hdr, ".text", NULL);
500 printk(KERN_ERR "vDSO64: the .text section was not found\n");
503 v64->text = sect - vdso64_kbase;
504 #endif /* CONFIG_PPC64 */
509 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
510 struct lib64_elfinfo *v64)
513 * Find signal trampolines
517 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
519 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
520 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
523 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
524 struct lib64_elfinfo *v64)
530 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
532 printk(KERN_ERR "vDSO64: Can't find symbol "
533 "__kernel_datapage_offset !\n");
536 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
537 (vdso64_pages << PAGE_SHIFT) -
538 (sym64->st_value - VDSO64_LBASE);
539 #endif /* CONFIG_PPC64 */
541 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
543 printk(KERN_ERR "vDSO32: Can't find symbol "
544 "__kernel_datapage_offset !\n");
547 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
548 (vdso32_pages << PAGE_SHIFT) -
549 (sym32->st_value - VDSO32_LBASE);
555 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
556 struct lib64_elfinfo *v64)
559 unsigned long size32;
563 unsigned long size64;
565 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
567 do_feature_fixups(cur_cpu_spec->cpu_features,
568 start64, start64 + size64);
570 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
572 do_feature_fixups(powerpc_firmware_features,
573 start64, start64 + size64);
575 start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
577 do_lwsync_fixups(cur_cpu_spec->cpu_features,
578 start64, start64 + size64);
579 #endif /* CONFIG_PPC64 */
581 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
583 do_feature_fixups(cur_cpu_spec->cpu_features,
584 start32, start32 + size32);
587 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
589 do_feature_fixups(powerpc_firmware_features,
590 start32, start32 + size32);
591 #endif /* CONFIG_PPC64 */
593 start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
595 do_lwsync_fixups(cur_cpu_spec->cpu_features,
596 start32, start32 + size32);
601 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
602 struct lib64_elfinfo *v64)
606 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
607 struct vdso_patch_def *patch = &vdso_patches[i];
608 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
613 DBG("replacing %s with %s...\n", patch->gen_name,
614 patch->fix_name ? "NONE" : patch->fix_name);
617 * Patch the 32 bits and 64 bits symbols. Note that we do not
618 * patch the "." symbol on 64 bits.
619 * It would be easy to do, but doesn't seem to be necessary,
620 * patching the OPD symbol is enough.
622 vdso_do_func_patch32(v32, v64, patch->gen_name,
625 vdso_do_func_patch64(v32, v64, patch->gen_name,
627 #endif /* CONFIG_PPC64 */
634 static __init int vdso_setup(void)
636 struct lib32_elfinfo v32;
637 struct lib64_elfinfo v64;
639 v32.hdr = vdso32_kbase;
641 v64.hdr = vdso64_kbase;
643 if (vdso_do_find_sections(&v32, &v64))
646 if (vdso_fixup_datapage(&v32, &v64))
649 if (vdso_fixup_features(&v32, &v64))
652 if (vdso_fixup_alt_funcs(&v32, &v64))
655 vdso_setup_trampolines(&v32, &v64);
661 * Called from setup_arch to initialize the bitmap of available
662 * syscalls in the systemcfg page
664 static void __init vdso_setup_syscall_map(void)
667 extern unsigned long *sys_call_table;
668 extern unsigned long sys_ni_syscall;
671 for (i = 0; i < __NR_syscalls; i++) {
673 if (sys_call_table[i*2] != sys_ni_syscall)
674 vdso_data->syscall_map_64[i >> 5] |=
675 0x80000000UL >> (i & 0x1f);
676 if (sys_call_table[i*2+1] != sys_ni_syscall)
677 vdso_data->syscall_map_32[i >> 5] |=
678 0x80000000UL >> (i & 0x1f);
679 #else /* CONFIG_PPC64 */
680 if (sys_call_table[i] != sys_ni_syscall)
681 vdso_data->syscall_map_32[i >> 5] |=
682 0x80000000UL >> (i & 0x1f);
683 #endif /* CONFIG_PPC64 */
688 static int __init vdso_init(void)
694 * Fill up the "systemcfg" stuff for backward compatiblity
696 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
697 vdso_data->version.major = SYSTEMCFG_MAJOR;
698 vdso_data->version.minor = SYSTEMCFG_MINOR;
699 vdso_data->processor = mfspr(SPRN_PVR);
701 * Fake the old platform number for pSeries and iSeries and add
702 * in LPAR bit if necessary
704 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
705 if (firmware_has_feature(FW_FEATURE_LPAR))
706 vdso_data->platform |= 1;
707 vdso_data->physicalMemorySize = lmb_phys_mem_size();
708 vdso_data->dcache_size = ppc64_caches.dsize;
709 vdso_data->dcache_line_size = ppc64_caches.dline_size;
710 vdso_data->icache_size = ppc64_caches.isize;
711 vdso_data->icache_line_size = ppc64_caches.iline_size;
713 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
714 vdso_data->dcache_block_size = ppc64_caches.dline_size;
715 vdso_data->icache_block_size = ppc64_caches.iline_size;
716 vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
717 vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
720 * Calculate the size of the 64 bits vDSO
722 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
723 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
725 vdso_data->dcache_block_size = L1_CACHE_BYTES;
726 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
727 vdso_data->icache_block_size = L1_CACHE_BYTES;
728 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
729 #endif /* CONFIG_PPC64 */
733 * Calculate the size of the 32 bits vDSO
735 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
736 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
740 * Setup the syscall map in the vDOS
742 vdso_setup_syscall_map();
745 * Initialize the vDSO images in memory, that is do necessary
746 * fixups of vDSO symbols, locate trampolines, etc...
749 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
757 /* Make sure pages are in the correct state */
758 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
760 BUG_ON(vdso32_pagelist == NULL);
761 for (i = 0; i < vdso32_pages; i++) {
762 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
763 ClearPageReserved(pg);
765 vdso32_pagelist[i] = pg;
767 vdso32_pagelist[i++] = virt_to_page(vdso_data);
768 vdso32_pagelist[i] = NULL;
771 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
773 BUG_ON(vdso64_pagelist == NULL);
774 for (i = 0; i < vdso64_pages; i++) {
775 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
776 ClearPageReserved(pg);
778 vdso64_pagelist[i] = pg;
780 vdso64_pagelist[i++] = virt_to_page(vdso_data);
781 vdso64_pagelist[i] = NULL;
782 #endif /* CONFIG_PPC64 */
784 get_page(virt_to_page(vdso_data));
791 arch_initcall(vdso_init);
793 int in_gate_area_no_task(unsigned long addr)
798 int in_gate_area(struct task_struct *task, unsigned long addr)
803 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)