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, int uses_interp)
189 struct mm_struct *mm = current->mm;
190 struct page **vdso_pagelist;
191 unsigned long vdso_pages;
192 unsigned long vdso_base;
199 if (test_thread_flag(TIF_32BIT)) {
200 vdso_pagelist = vdso32_pagelist;
201 vdso_pages = vdso32_pages;
202 vdso_base = VDSO32_MBASE;
204 vdso_pagelist = vdso64_pagelist;
205 vdso_pages = vdso64_pages;
207 * On 64bit we don't have a preferred map address. This
208 * allows get_unmapped_area to find an area near other mmaps
209 * and most likely share a SLB entry.
214 vdso_pagelist = vdso32_pagelist;
215 vdso_pages = vdso32_pages;
216 vdso_base = VDSO32_MBASE;
219 current->mm->context.vdso_base = 0;
221 /* vDSO has a problem and was disabled, just don't "enable" it for the
226 /* Add a page to the vdso size for the data page */
230 * pick a base address for the vDSO in process space. We try to put it
231 * at vdso_base which is the "natural" base for it, but we might fail
232 * and end up putting it elsewhere.
234 down_write(&mm->mmap_sem);
235 vdso_base = get_unmapped_area(NULL, vdso_base,
236 vdso_pages << PAGE_SHIFT, 0, 0);
237 if (IS_ERR_VALUE(vdso_base)) {
243 * our vma flags don't have VM_WRITE so by default, the process isn't
244 * allowed to write those pages.
245 * gdb can break that with ptrace interface, and thus trigger COW on
246 * those pages but it's then your responsibility to never do that on
247 * the "data" page of the vDSO or you'll stop getting kernel updates
248 * and your nice userland gettimeofday will be totally dead.
249 * It's fine to use that for setting breakpoints in the vDSO code
252 * Make sure the vDSO gets into every core dump.
253 * Dumping its contents makes post-mortem fully interpretable later
254 * without matching up the same kernel and hardware config to see
255 * what PC values meant.
257 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
259 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
265 /* Put vDSO base into mm struct */
266 current->mm->context.vdso_base = vdso_base;
268 up_write(&mm->mmap_sem);
272 up_write(&mm->mmap_sem);
276 const char *arch_vma_name(struct vm_area_struct *vma)
278 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
285 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
292 /* Grab section headers and strings so we can tell who is who */
293 sechdrs = (void *)ehdr + ehdr->e_shoff;
294 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
296 /* Find the section they want */
297 for (i = 1; i < ehdr->e_shnum; i++) {
298 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
300 *size = sechdrs[i].sh_size;
301 return (void *)ehdr + sechdrs[i].sh_offset;
308 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
312 char name[MAX_SYMNAME], *c;
314 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
315 if (lib->dynsym[i].st_name == 0)
317 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
319 c = strchr(name, '@');
322 if (strcmp(symname, name) == 0)
323 return &lib->dynsym[i];
328 /* Note that we assume the section is .text and the symbol is relative to
331 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
334 Elf32_Sym *sym = find_symbol32(lib, symname);
337 printk(KERN_WARNING "vDSO32: function %s not found !\n",
341 return sym->st_value - VDSO32_LBASE;
344 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
345 struct lib64_elfinfo *v64,
346 const char *orig, const char *fix)
348 Elf32_Sym *sym32_gen, *sym32_fix;
350 sym32_gen = find_symbol32(v32, orig);
351 if (sym32_gen == NULL) {
352 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
356 sym32_gen->st_name = 0;
359 sym32_fix = find_symbol32(v32, fix);
360 if (sym32_fix == NULL) {
361 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
364 sym32_gen->st_value = sym32_fix->st_value;
365 sym32_gen->st_size = sym32_fix->st_size;
366 sym32_gen->st_info = sym32_fix->st_info;
367 sym32_gen->st_other = sym32_fix->st_other;
368 sym32_gen->st_shndx = sym32_fix->st_shndx;
376 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
383 /* Grab section headers and strings so we can tell who is who */
384 sechdrs = (void *)ehdr + ehdr->e_shoff;
385 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
387 /* Find the section they want */
388 for (i = 1; i < ehdr->e_shnum; i++) {
389 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
391 *size = sechdrs[i].sh_size;
392 return (void *)ehdr + sechdrs[i].sh_offset;
400 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
404 char name[MAX_SYMNAME], *c;
406 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
407 if (lib->dynsym[i].st_name == 0)
409 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
411 c = strchr(name, '@');
414 if (strcmp(symname, name) == 0)
415 return &lib->dynsym[i];
420 /* Note that we assume the section is .text and the symbol is relative to
423 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
426 Elf64_Sym *sym = find_symbol64(lib, symname);
429 printk(KERN_WARNING "vDSO64: function %s not found !\n",
433 #ifdef VDS64_HAS_DESCRIPTORS
434 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
437 return sym->st_value - VDSO64_LBASE;
441 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
442 struct lib64_elfinfo *v64,
443 const char *orig, const char *fix)
445 Elf64_Sym *sym64_gen, *sym64_fix;
447 sym64_gen = find_symbol64(v64, orig);
448 if (sym64_gen == NULL) {
449 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
453 sym64_gen->st_name = 0;
456 sym64_fix = find_symbol64(v64, fix);
457 if (sym64_fix == NULL) {
458 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
461 sym64_gen->st_value = sym64_fix->st_value;
462 sym64_gen->st_size = sym64_fix->st_size;
463 sym64_gen->st_info = sym64_fix->st_info;
464 sym64_gen->st_other = sym64_fix->st_other;
465 sym64_gen->st_shndx = sym64_fix->st_shndx;
470 #endif /* CONFIG_PPC64 */
473 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
474 struct lib64_elfinfo *v64)
479 * Locate symbol tables & text section
482 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
483 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
484 if (v32->dynsym == NULL || v32->dynstr == NULL) {
485 printk(KERN_ERR "vDSO32: required symbol section not found\n");
488 sect = find_section32(v32->hdr, ".text", NULL);
490 printk(KERN_ERR "vDSO32: the .text section was not found\n");
493 v32->text = sect - vdso32_kbase;
496 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
497 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
498 if (v64->dynsym == NULL || v64->dynstr == NULL) {
499 printk(KERN_ERR "vDSO64: required symbol section not found\n");
502 sect = find_section64(v64->hdr, ".text", NULL);
504 printk(KERN_ERR "vDSO64: the .text section was not found\n");
507 v64->text = sect - vdso64_kbase;
508 #endif /* CONFIG_PPC64 */
513 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
514 struct lib64_elfinfo *v64)
517 * Find signal trampolines
521 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
523 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
524 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
527 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
528 struct lib64_elfinfo *v64)
534 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
536 printk(KERN_ERR "vDSO64: Can't find symbol "
537 "__kernel_datapage_offset !\n");
540 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
541 (vdso64_pages << PAGE_SHIFT) -
542 (sym64->st_value - VDSO64_LBASE);
543 #endif /* CONFIG_PPC64 */
545 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
547 printk(KERN_ERR "vDSO32: Can't find symbol "
548 "__kernel_datapage_offset !\n");
551 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
552 (vdso32_pages << PAGE_SHIFT) -
553 (sym32->st_value - VDSO32_LBASE);
559 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
560 struct lib64_elfinfo *v64)
563 unsigned long size32;
567 unsigned long size64;
569 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
571 do_feature_fixups(cur_cpu_spec->cpu_features,
572 start64, start64 + size64);
574 start64 = find_section64(v64->hdr, "__mmu_ftr_fixup", &size64);
576 do_feature_fixups(cur_cpu_spec->mmu_features,
577 start64, start64 + size64);
579 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
581 do_feature_fixups(powerpc_firmware_features,
582 start64, start64 + size64);
584 start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
586 do_lwsync_fixups(cur_cpu_spec->cpu_features,
587 start64, start64 + size64);
588 #endif /* CONFIG_PPC64 */
590 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
592 do_feature_fixups(cur_cpu_spec->cpu_features,
593 start32, start32 + size32);
595 start32 = find_section32(v32->hdr, "__mmu_ftr_fixup", &size32);
597 do_feature_fixups(cur_cpu_spec->mmu_features,
598 start32, start32 + size32);
601 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
603 do_feature_fixups(powerpc_firmware_features,
604 start32, start32 + size32);
605 #endif /* CONFIG_PPC64 */
607 start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
609 do_lwsync_fixups(cur_cpu_spec->cpu_features,
610 start32, start32 + size32);
615 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
616 struct lib64_elfinfo *v64)
620 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
621 struct vdso_patch_def *patch = &vdso_patches[i];
622 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
627 DBG("replacing %s with %s...\n", patch->gen_name,
628 patch->fix_name ? "NONE" : patch->fix_name);
631 * Patch the 32 bits and 64 bits symbols. Note that we do not
632 * patch the "." symbol on 64 bits.
633 * It would be easy to do, but doesn't seem to be necessary,
634 * patching the OPD symbol is enough.
636 vdso_do_func_patch32(v32, v64, patch->gen_name,
639 vdso_do_func_patch64(v32, v64, patch->gen_name,
641 #endif /* CONFIG_PPC64 */
648 static __init int vdso_setup(void)
650 struct lib32_elfinfo v32;
651 struct lib64_elfinfo v64;
653 v32.hdr = vdso32_kbase;
655 v64.hdr = vdso64_kbase;
657 if (vdso_do_find_sections(&v32, &v64))
660 if (vdso_fixup_datapage(&v32, &v64))
663 if (vdso_fixup_features(&v32, &v64))
666 if (vdso_fixup_alt_funcs(&v32, &v64))
669 vdso_setup_trampolines(&v32, &v64);
675 * Called from setup_arch to initialize the bitmap of available
676 * syscalls in the systemcfg page
678 static void __init vdso_setup_syscall_map(void)
681 extern unsigned long *sys_call_table;
682 extern unsigned long sys_ni_syscall;
685 for (i = 0; i < __NR_syscalls; i++) {
687 if (sys_call_table[i*2] != sys_ni_syscall)
688 vdso_data->syscall_map_64[i >> 5] |=
689 0x80000000UL >> (i & 0x1f);
690 if (sys_call_table[i*2+1] != sys_ni_syscall)
691 vdso_data->syscall_map_32[i >> 5] |=
692 0x80000000UL >> (i & 0x1f);
693 #else /* CONFIG_PPC64 */
694 if (sys_call_table[i] != sys_ni_syscall)
695 vdso_data->syscall_map_32[i >> 5] |=
696 0x80000000UL >> (i & 0x1f);
697 #endif /* CONFIG_PPC64 */
702 static int __init vdso_init(void)
708 * Fill up the "systemcfg" stuff for backward compatiblity
710 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
711 vdso_data->version.major = SYSTEMCFG_MAJOR;
712 vdso_data->version.minor = SYSTEMCFG_MINOR;
713 vdso_data->processor = mfspr(SPRN_PVR);
715 * Fake the old platform number for pSeries and iSeries and add
716 * in LPAR bit if necessary
718 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
719 if (firmware_has_feature(FW_FEATURE_LPAR))
720 vdso_data->platform |= 1;
721 vdso_data->physicalMemorySize = lmb_phys_mem_size();
722 vdso_data->dcache_size = ppc64_caches.dsize;
723 vdso_data->dcache_line_size = ppc64_caches.dline_size;
724 vdso_data->icache_size = ppc64_caches.isize;
725 vdso_data->icache_line_size = ppc64_caches.iline_size;
727 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
728 vdso_data->dcache_block_size = ppc64_caches.dline_size;
729 vdso_data->icache_block_size = ppc64_caches.iline_size;
730 vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
731 vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
734 * Calculate the size of the 64 bits vDSO
736 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
737 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
739 vdso_data->dcache_block_size = L1_CACHE_BYTES;
740 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
741 vdso_data->icache_block_size = L1_CACHE_BYTES;
742 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
743 #endif /* CONFIG_PPC64 */
747 * Calculate the size of the 32 bits vDSO
749 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
750 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
754 * Setup the syscall map in the vDOS
756 vdso_setup_syscall_map();
759 * Initialize the vDSO images in memory, that is do necessary
760 * fixups of vDSO symbols, locate trampolines, etc...
763 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
771 /* Make sure pages are in the correct state */
772 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
774 BUG_ON(vdso32_pagelist == NULL);
775 for (i = 0; i < vdso32_pages; i++) {
776 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
777 ClearPageReserved(pg);
779 vdso32_pagelist[i] = pg;
781 vdso32_pagelist[i++] = virt_to_page(vdso_data);
782 vdso32_pagelist[i] = NULL;
785 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
787 BUG_ON(vdso64_pagelist == NULL);
788 for (i = 0; i < vdso64_pages; i++) {
789 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
790 ClearPageReserved(pg);
792 vdso64_pagelist[i] = pg;
794 vdso64_pagelist[i++] = virt_to_page(vdso_data);
795 vdso64_pagelist[i] = NULL;
796 #endif /* CONFIG_PPC64 */
798 get_page(virt_to_page(vdso_data));
805 arch_initcall(vdso_init);
807 int in_gate_area_no_task(unsigned long addr)
812 int in_gate_area(struct task_struct *task, unsigned long addr)
817 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)