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/smp_lock.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/slab.h>
21 #include <linux/user.h>
22 #include <linux/elf.h>
23 #include <linux/security.h>
24 #include <linux/bootmem.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 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;
230 if (test_thread_flag(TIF_32BIT)) {
231 vdso_pages = vdso32_pages;
232 vdso_base = VDSO32_MBASE;
234 vdso_pages = vdso64_pages;
235 vdso_base = VDSO64_MBASE;
238 vdso_pages = vdso32_pages;
239 vdso_base = VDSO32_MBASE;
242 current->mm->context.vdso_base = 0;
244 /* vDSO has a problem and was disabled, just don't "enable" it for the
249 /* Add a page to the vdso size for the data page */
253 * pick a base address for the vDSO in process space. We try to put it
254 * at vdso_base which is the "natural" base for it, but we might fail
255 * and end up putting it elsewhere.
257 down_write(&mm->mmap_sem);
258 vdso_base = get_unmapped_area(NULL, vdso_base,
259 vdso_pages << PAGE_SHIFT, 0, 0);
260 if (IS_ERR_VALUE(vdso_base)) {
266 /* Allocate a VMA structure and fill it up */
267 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
273 vma->vm_start = 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;
288 * Make sure the vDSO gets into every core dump.
289 * Dumping its contents makes post-mortem fully interpretable later
290 * without matching up the same kernel and hardware config to see
291 * what PC values meant.
293 vma->vm_flags |= VM_ALWAYSDUMP;
294 vma->vm_flags |= mm->def_flags;
295 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
296 vma->vm_ops = &vdso_vmops;
299 rc = insert_vm_struct(mm, vma);
303 /* Put vDSO base into mm struct and account for memory usage */
304 current->mm->context.vdso_base = vdso_base;
305 mm->total_vm += (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
306 up_write(&mm->mmap_sem);
310 kmem_cache_free(vm_area_cachep, vma);
312 up_write(&mm->mmap_sem);
316 const char *arch_vma_name(struct vm_area_struct *vma)
318 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
325 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
332 /* Grab section headers and strings so we can tell who is who */
333 sechdrs = (void *)ehdr + ehdr->e_shoff;
334 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
336 /* Find the section they want */
337 for (i = 1; i < ehdr->e_shnum; i++) {
338 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
340 *size = sechdrs[i].sh_size;
341 return (void *)ehdr + sechdrs[i].sh_offset;
348 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
352 char name[MAX_SYMNAME], *c;
354 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
355 if (lib->dynsym[i].st_name == 0)
357 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
359 c = strchr(name, '@');
362 if (strcmp(symname, name) == 0)
363 return &lib->dynsym[i];
368 /* Note that we assume the section is .text and the symbol is relative to
371 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
374 Elf32_Sym *sym = find_symbol32(lib, symname);
377 printk(KERN_WARNING "vDSO32: function %s not found !\n",
381 return sym->st_value - VDSO32_LBASE;
384 static int vdso_do_func_patch32(struct lib32_elfinfo *v32,
385 struct lib64_elfinfo *v64,
386 const char *orig, const char *fix)
388 Elf32_Sym *sym32_gen, *sym32_fix;
390 sym32_gen = find_symbol32(v32, orig);
391 if (sym32_gen == NULL) {
392 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
396 sym32_gen->st_name = 0;
399 sym32_fix = find_symbol32(v32, fix);
400 if (sym32_fix == NULL) {
401 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
404 sym32_gen->st_value = sym32_fix->st_value;
405 sym32_gen->st_size = sym32_fix->st_size;
406 sym32_gen->st_info = sym32_fix->st_info;
407 sym32_gen->st_other = sym32_fix->st_other;
408 sym32_gen->st_shndx = sym32_fix->st_shndx;
416 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
423 /* Grab section headers and strings so we can tell who is who */
424 sechdrs = (void *)ehdr + ehdr->e_shoff;
425 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
427 /* Find the section they want */
428 for (i = 1; i < ehdr->e_shnum; i++) {
429 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
431 *size = sechdrs[i].sh_size;
432 return (void *)ehdr + sechdrs[i].sh_offset;
440 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
444 char name[MAX_SYMNAME], *c;
446 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
447 if (lib->dynsym[i].st_name == 0)
449 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
451 c = strchr(name, '@');
454 if (strcmp(symname, name) == 0)
455 return &lib->dynsym[i];
460 /* Note that we assume the section is .text and the symbol is relative to
463 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
466 Elf64_Sym *sym = find_symbol64(lib, symname);
469 printk(KERN_WARNING "vDSO64: function %s not found !\n",
473 #ifdef VDS64_HAS_DESCRIPTORS
474 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
477 return sym->st_value - VDSO64_LBASE;
481 static int vdso_do_func_patch64(struct lib32_elfinfo *v32,
482 struct lib64_elfinfo *v64,
483 const char *orig, const char *fix)
485 Elf64_Sym *sym64_gen, *sym64_fix;
487 sym64_gen = find_symbol64(v64, orig);
488 if (sym64_gen == NULL) {
489 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
493 sym64_gen->st_name = 0;
496 sym64_fix = find_symbol64(v64, fix);
497 if (sym64_fix == NULL) {
498 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
501 sym64_gen->st_value = sym64_fix->st_value;
502 sym64_gen->st_size = sym64_fix->st_size;
503 sym64_gen->st_info = sym64_fix->st_info;
504 sym64_gen->st_other = sym64_fix->st_other;
505 sym64_gen->st_shndx = sym64_fix->st_shndx;
510 #endif /* CONFIG_PPC64 */
513 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
514 struct lib64_elfinfo *v64)
519 * Locate symbol tables & text section
522 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
523 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
524 if (v32->dynsym == NULL || v32->dynstr == NULL) {
525 printk(KERN_ERR "vDSO32: required symbol section not found\n");
528 sect = find_section32(v32->hdr, ".text", NULL);
530 printk(KERN_ERR "vDSO32: the .text section was not found\n");
533 v32->text = sect - vdso32_kbase;
536 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
537 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
538 if (v64->dynsym == NULL || v64->dynstr == NULL) {
539 printk(KERN_ERR "vDSO64: required symbol section not found\n");
542 sect = find_section64(v64->hdr, ".text", NULL);
544 printk(KERN_ERR "vDSO64: the .text section was not found\n");
547 v64->text = sect - vdso64_kbase;
548 #endif /* CONFIG_PPC64 */
553 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
554 struct lib64_elfinfo *v64)
557 * Find signal trampolines
561 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
563 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
564 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
567 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
568 struct lib64_elfinfo *v64)
574 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
576 printk(KERN_ERR "vDSO64: Can't find symbol "
577 "__kernel_datapage_offset !\n");
580 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
581 (vdso64_pages << PAGE_SHIFT) -
582 (sym64->st_value - VDSO64_LBASE);
583 #endif /* CONFIG_PPC64 */
585 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
587 printk(KERN_ERR "vDSO32: Can't find symbol "
588 "__kernel_datapage_offset !\n");
591 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
592 (vdso32_pages << PAGE_SHIFT) -
593 (sym32->st_value - VDSO32_LBASE);
599 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
600 struct lib64_elfinfo *v64)
603 unsigned long size32;
607 unsigned long size64;
609 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
611 do_feature_fixups(cur_cpu_spec->cpu_features,
612 start64, start64 + size64);
614 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
616 do_feature_fixups(powerpc_firmware_features,
617 start64, start64 + size64);
618 #endif /* CONFIG_PPC64 */
620 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
622 do_feature_fixups(cur_cpu_spec->cpu_features,
623 start32, start32 + size32);
626 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
628 do_feature_fixups(powerpc_firmware_features,
629 start32, start32 + size32);
630 #endif /* CONFIG_PPC64 */
635 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
636 struct lib64_elfinfo *v64)
640 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
641 struct vdso_patch_def *patch = &vdso_patches[i];
642 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
647 DBG("replacing %s with %s...\n", patch->gen_name,
648 patch->fix_name ? "NONE" : patch->fix_name);
651 * Patch the 32 bits and 64 bits symbols. Note that we do not
652 * patch the "." symbol on 64 bits.
653 * It would be easy to do, but doesn't seem to be necessary,
654 * patching the OPD symbol is enough.
656 vdso_do_func_patch32(v32, v64, patch->gen_name,
659 vdso_do_func_patch64(v32, v64, patch->gen_name,
661 #endif /* CONFIG_PPC64 */
668 static __init int vdso_setup(void)
670 struct lib32_elfinfo v32;
671 struct lib64_elfinfo v64;
673 v32.hdr = vdso32_kbase;
675 v64.hdr = vdso64_kbase;
677 if (vdso_do_find_sections(&v32, &v64))
680 if (vdso_fixup_datapage(&v32, &v64))
683 if (vdso_fixup_features(&v32, &v64))
686 if (vdso_fixup_alt_funcs(&v32, &v64))
689 vdso_setup_trampolines(&v32, &v64);
695 * Called from setup_arch to initialize the bitmap of available
696 * syscalls in the systemcfg page
698 static void __init vdso_setup_syscall_map(void)
701 extern unsigned long *sys_call_table;
702 extern unsigned long sys_ni_syscall;
705 for (i = 0; i < __NR_syscalls; i++) {
707 if (sys_call_table[i*2] != sys_ni_syscall)
708 vdso_data->syscall_map_64[i >> 5] |=
709 0x80000000UL >> (i & 0x1f);
710 if (sys_call_table[i*2+1] != sys_ni_syscall)
711 vdso_data->syscall_map_32[i >> 5] |=
712 0x80000000UL >> (i & 0x1f);
713 #else /* CONFIG_PPC64 */
714 if (sys_call_table[i] != sys_ni_syscall)
715 vdso_data->syscall_map_32[i >> 5] |=
716 0x80000000UL >> (i & 0x1f);
717 #endif /* CONFIG_PPC64 */
722 void __init vdso_init(void)
728 * Fill up the "systemcfg" stuff for backward compatiblity
730 strcpy(vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
731 vdso_data->version.major = SYSTEMCFG_MAJOR;
732 vdso_data->version.minor = SYSTEMCFG_MINOR;
733 vdso_data->processor = mfspr(SPRN_PVR);
735 * Fake the old platform number for pSeries and iSeries and add
736 * in LPAR bit if necessary
738 vdso_data->platform = machine_is(iseries) ? 0x200 : 0x100;
739 if (firmware_has_feature(FW_FEATURE_LPAR))
740 vdso_data->platform |= 1;
741 vdso_data->physicalMemorySize = lmb_phys_mem_size();
742 vdso_data->dcache_size = ppc64_caches.dsize;
743 vdso_data->dcache_line_size = ppc64_caches.dline_size;
744 vdso_data->icache_size = ppc64_caches.isize;
745 vdso_data->icache_line_size = ppc64_caches.iline_size;
748 * Calculate the size of the 64 bits vDSO
750 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
751 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
752 #endif /* CONFIG_PPC64 */
756 * Calculate the size of the 32 bits vDSO
758 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
759 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
763 * Setup the syscall map in the vDOS
765 vdso_setup_syscall_map();
768 * Initialize the vDSO images in memory, that is do necessary
769 * fixups of vDSO symbols, locate trampolines, etc...
772 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
780 /* Make sure pages are in the correct state */
781 for (i = 0; i < vdso32_pages; i++) {
782 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
783 ClearPageReserved(pg);
788 for (i = 0; i < vdso64_pages; i++) {
789 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
790 ClearPageReserved(pg);
793 #endif /* CONFIG_PPC64 */
795 get_page(virt_to_page(vdso_data));
798 int in_gate_area_no_task(unsigned long addr)
803 int in_gate_area(struct task_struct *task, unsigned long addr)
808 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)