3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
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/memblock.h>
25 #include <asm/pgtable.h>
26 #include <asm/processor.h>
28 #include <asm/mmu_context.h>
30 #include <asm/machdep.h>
31 #include <asm/cputable.h>
32 #include <asm/sections.h>
33 #include <asm/firmware.h>
35 #include <asm/vdso_datapage.h>
36 #include <asm/setup.h>
41 #define DBG(fmt...) printk(fmt)
46 /* Max supported size for symbol names */
47 #define MAX_SYMNAME 64
49 /* The alignment of the vDSO */
50 #define VDSO_ALIGNMENT (1 << 16)
52 static unsigned int vdso32_pages;
53 static void *vdso32_kbase;
54 static struct page **vdso32_pagelist;
55 unsigned long vdso32_sigtramp;
56 unsigned long vdso32_rt_sigtramp;
59 extern char vdso32_start, vdso32_end;
63 extern char vdso64_start, vdso64_end;
64 static void *vdso64_kbase = &vdso64_start;
65 static unsigned int vdso64_pages;
66 static struct page **vdso64_pagelist;
67 unsigned long vdso64_rt_sigtramp;
68 #endif /* CONFIG_PPC64 */
70 static int vdso_ready;
73 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
74 * Once the early boot kernel code no longer needs to muck around
75 * with it, it will become dynamically allocated
78 struct vdso_data data;
80 } vdso_data_store __page_aligned_data;
81 struct vdso_data *vdso_data = &vdso_data_store.data;
83 /* Format of the patch table */
86 unsigned long ftr_mask, ftr_value;
91 /* Table of functions to patch based on the CPU type/revision
93 * Currently, we only change sync_dicache to do nothing on processors
94 * with a coherent icache
96 static struct vdso_patch_def vdso_patches[] = {
98 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
99 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
103 "__kernel_gettimeofday", NULL
107 "__kernel_clock_gettime", NULL
111 "__kernel_clock_getres", NULL
115 "__kernel_get_tbfreq", NULL
119 "__kernel_time", NULL
124 * Some infos carried around for each of them during parsing at
129 Elf32_Ehdr *hdr; /* ptr to ELF */
130 Elf32_Sym *dynsym; /* ptr to .dynsym section */
131 unsigned long dynsymsize; /* size of .dynsym section */
132 char *dynstr; /* ptr to .dynstr section */
133 unsigned long text; /* offset of .text section in .so */
140 unsigned long dynsymsize;
147 * This is called from binfmt_elf, we create the special vma for the
148 * vDSO and insert it into the mm struct tree
150 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
152 struct mm_struct *mm = current->mm;
153 struct page **vdso_pagelist;
154 unsigned long vdso_pages;
155 unsigned long vdso_base;
162 if (is_32bit_task()) {
163 vdso_pagelist = vdso32_pagelist;
164 vdso_pages = vdso32_pages;
165 vdso_base = VDSO32_MBASE;
167 vdso_pagelist = vdso64_pagelist;
168 vdso_pages = vdso64_pages;
170 * On 64bit we don't have a preferred map address. This
171 * allows get_unmapped_area to find an area near other mmaps
172 * and most likely share a SLB entry.
177 vdso_pagelist = vdso32_pagelist;
178 vdso_pages = vdso32_pages;
179 vdso_base = VDSO32_MBASE;
182 current->mm->context.vdso_base = 0;
184 /* vDSO has a problem and was disabled, just don't "enable" it for the
189 /* Add a page to the vdso size for the data page */
193 * pick a base address for the vDSO in process space. We try to put it
194 * at vdso_base which is the "natural" base for it, but we might fail
195 * and end up putting it elsewhere.
196 * Add enough to the size so that the result can be aligned.
198 down_write(&mm->mmap_sem);
199 vdso_base = get_unmapped_area(NULL, vdso_base,
200 (vdso_pages << PAGE_SHIFT) +
201 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
203 if (IS_ERR_VALUE(vdso_base)) {
208 /* Add required alignment. */
209 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
212 * Put vDSO base into mm struct. We need to do this before calling
213 * install_special_mapping or the perf counter mmap tracking code
214 * will fail to recognise it as a vDSO (since arch_vma_name fails).
216 current->mm->context.vdso_base = vdso_base;
219 * our vma flags don't have VM_WRITE so by default, the process isn't
220 * allowed to write those pages.
221 * gdb can break that with ptrace interface, and thus trigger COW on
222 * those pages but it's then your responsibility to never do that on
223 * the "data" page of the vDSO or you'll stop getting kernel updates
224 * and your nice userland gettimeofday will be totally dead.
225 * It's fine to use that for setting breakpoints in the vDSO code
228 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
230 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
233 current->mm->context.vdso_base = 0;
237 up_write(&mm->mmap_sem);
241 up_write(&mm->mmap_sem);
245 const char *arch_vma_name(struct vm_area_struct *vma)
247 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
255 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
262 /* Grab section headers and strings so we can tell who is who */
263 sechdrs = (void *)ehdr + ehdr->e_shoff;
264 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
266 /* Find the section they want */
267 for (i = 1; i < ehdr->e_shnum; i++) {
268 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
270 *size = sechdrs[i].sh_size;
271 return (void *)ehdr + sechdrs[i].sh_offset;
278 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
282 char name[MAX_SYMNAME], *c;
284 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
285 if (lib->dynsym[i].st_name == 0)
287 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
289 c = strchr(name, '@');
292 if (strcmp(symname, name) == 0)
293 return &lib->dynsym[i];
298 /* Note that we assume the section is .text and the symbol is relative to
301 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
304 Elf32_Sym *sym = find_symbol32(lib, symname);
307 printk(KERN_WARNING "vDSO32: function %s not found !\n",
311 return sym->st_value - VDSO32_LBASE;
314 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
315 struct lib64_elfinfo *v64,
316 const char *orig, const char *fix)
318 Elf32_Sym *sym32_gen, *sym32_fix;
320 sym32_gen = find_symbol32(v32, orig);
321 if (sym32_gen == NULL) {
322 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
326 sym32_gen->st_name = 0;
329 sym32_fix = find_symbol32(v32, fix);
330 if (sym32_fix == NULL) {
331 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
334 sym32_gen->st_value = sym32_fix->st_value;
335 sym32_gen->st_size = sym32_fix->st_size;
336 sym32_gen->st_info = sym32_fix->st_info;
337 sym32_gen->st_other = sym32_fix->st_other;
338 sym32_gen->st_shndx = sym32_fix->st_shndx;
342 #else /* !CONFIG_VDSO32 */
343 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
349 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
350 struct lib64_elfinfo *v64,
351 const char *orig, const char *fix)
355 #endif /* CONFIG_VDSO32 */
360 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
367 /* Grab section headers and strings so we can tell who is who */
368 sechdrs = (void *)ehdr + ehdr->e_shoff;
369 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
371 /* Find the section they want */
372 for (i = 1; i < ehdr->e_shnum; i++) {
373 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
375 *size = sechdrs[i].sh_size;
376 return (void *)ehdr + sechdrs[i].sh_offset;
384 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
388 char name[MAX_SYMNAME], *c;
390 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
391 if (lib->dynsym[i].st_name == 0)
393 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
395 c = strchr(name, '@');
398 if (strcmp(symname, name) == 0)
399 return &lib->dynsym[i];
404 /* Note that we assume the section is .text and the symbol is relative to
407 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
410 Elf64_Sym *sym = find_symbol64(lib, symname);
413 printk(KERN_WARNING "vDSO64: function %s not found !\n",
417 #ifdef VDS64_HAS_DESCRIPTORS
418 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
421 return sym->st_value - VDSO64_LBASE;
425 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
426 struct lib64_elfinfo *v64,
427 const char *orig, const char *fix)
429 Elf64_Sym *sym64_gen, *sym64_fix;
431 sym64_gen = find_symbol64(v64, orig);
432 if (sym64_gen == NULL) {
433 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
437 sym64_gen->st_name = 0;
440 sym64_fix = find_symbol64(v64, fix);
441 if (sym64_fix == NULL) {
442 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
445 sym64_gen->st_value = sym64_fix->st_value;
446 sym64_gen->st_size = sym64_fix->st_size;
447 sym64_gen->st_info = sym64_fix->st_info;
448 sym64_gen->st_other = sym64_fix->st_other;
449 sym64_gen->st_shndx = sym64_fix->st_shndx;
454 #endif /* CONFIG_PPC64 */
457 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
458 struct lib64_elfinfo *v64)
463 * Locate symbol tables & text section
467 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
468 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
469 if (v32->dynsym == NULL || v32->dynstr == NULL) {
470 printk(KERN_ERR "vDSO32: required symbol section not found\n");
473 sect = find_section32(v32->hdr, ".text", NULL);
475 printk(KERN_ERR "vDSO32: the .text section was not found\n");
478 v32->text = sect - vdso32_kbase;
482 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
483 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
484 if (v64->dynsym == NULL || v64->dynstr == NULL) {
485 printk(KERN_ERR "vDSO64: required symbol section not found\n");
488 sect = find_section64(v64->hdr, ".text", NULL);
490 printk(KERN_ERR "vDSO64: the .text section was not found\n");
493 v64->text = sect - vdso64_kbase;
494 #endif /* CONFIG_PPC64 */
499 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
500 struct lib64_elfinfo *v64)
503 * Find signal trampolines
507 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
509 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
510 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
513 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
514 struct lib64_elfinfo *v64)
522 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
524 printk(KERN_ERR "vDSO64: Can't find symbol "
525 "__kernel_datapage_offset !\n");
528 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
529 (vdso64_pages << PAGE_SHIFT) -
530 (sym64->st_value - VDSO64_LBASE);
531 #endif /* CONFIG_PPC64 */
534 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
536 printk(KERN_ERR "vDSO32: Can't find symbol "
537 "__kernel_datapage_offset !\n");
540 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
541 (vdso32_pages << PAGE_SHIFT) -
542 (sym32->st_value - VDSO32_LBASE);
549 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
550 struct lib64_elfinfo *v64)
556 start = find_section64(v64->hdr, "__ftr_fixup", &size);
558 do_feature_fixups(cur_cpu_spec->cpu_features,
559 start, start + size);
561 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
563 do_feature_fixups(cur_cpu_spec->mmu_features,
564 start, start + size);
566 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
568 do_feature_fixups(powerpc_firmware_features,
569 start, start + size);
571 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
573 do_lwsync_fixups(cur_cpu_spec->cpu_features,
574 start, start + size);
575 #endif /* CONFIG_PPC64 */
578 start = find_section32(v32->hdr, "__ftr_fixup", &size);
580 do_feature_fixups(cur_cpu_spec->cpu_features,
581 start, start + size);
583 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
585 do_feature_fixups(cur_cpu_spec->mmu_features,
586 start, start + size);
589 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
591 do_feature_fixups(powerpc_firmware_features,
592 start, start + size);
593 #endif /* CONFIG_PPC64 */
595 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
597 do_lwsync_fixups(cur_cpu_spec->cpu_features,
598 start, start + size);
604 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
605 struct lib64_elfinfo *v64)
609 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
610 struct vdso_patch_def *patch = &vdso_patches[i];
611 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
616 DBG("replacing %s with %s...\n", patch->gen_name,
617 patch->fix_name ? "NONE" : patch->fix_name);
620 * Patch the 32 bits and 64 bits symbols. Note that we do not
621 * patch the "." symbol on 64 bits.
622 * It would be easy to do, but doesn't seem to be necessary,
623 * patching the OPD symbol is enough.
625 vdso_do_func_patch32(v32, v64, patch->gen_name,
628 vdso_do_func_patch64(v32, v64, patch->gen_name,
630 #endif /* CONFIG_PPC64 */
637 static __init int vdso_setup(void)
639 struct lib32_elfinfo v32;
640 struct lib64_elfinfo v64;
642 v32.hdr = vdso32_kbase;
644 v64.hdr = vdso64_kbase;
646 if (vdso_do_find_sections(&v32, &v64))
649 if (vdso_fixup_datapage(&v32, &v64))
652 if (vdso_fixup_features(&v32, &v64))
655 if (vdso_fixup_alt_funcs(&v32, &v64))
658 vdso_setup_trampolines(&v32, &v64);
664 * Called from setup_arch to initialize the bitmap of available
665 * syscalls in the systemcfg page
667 static void __init vdso_setup_syscall_map(void)
670 extern unsigned long *sys_call_table;
671 extern unsigned long sys_ni_syscall;
674 for (i = 0; i < __NR_syscalls; i++) {
676 if (sys_call_table[i*2] != sys_ni_syscall)
677 vdso_data->syscall_map_64[i >> 5] |=
678 0x80000000UL >> (i & 0x1f);
679 if (sys_call_table[i*2+1] != sys_ni_syscall)
680 vdso_data->syscall_map_32[i >> 5] |=
681 0x80000000UL >> (i & 0x1f);
682 #else /* CONFIG_PPC64 */
683 if (sys_call_table[i] != sys_ni_syscall)
684 vdso_data->syscall_map_32[i >> 5] |=
685 0x80000000UL >> (i & 0x1f);
686 #endif /* CONFIG_PPC64 */
691 int vdso_getcpu_init(void)
693 unsigned long cpu, node, val;
696 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
697 * in the next 16 bits. The VDSO uses this to implement getcpu().
700 WARN_ON_ONCE(cpu > 0xffff);
702 node = cpu_to_node(cpu);
703 WARN_ON_ONCE(node > 0xffff);
705 val = (cpu & 0xfff) | ((node & 0xffff) << 16);
706 mtspr(SPRN_SPRG_VDSO_WRITE, val);
707 get_paca()->sprg_vdso = val;
713 /* We need to call this before SMP init */
714 early_initcall(vdso_getcpu_init);
717 static int __init vdso_init(void)
723 * Fill up the "systemcfg" stuff for backward compatibility
725 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
726 vdso_data->version.major = SYSTEMCFG_MAJOR;
727 vdso_data->version.minor = SYSTEMCFG_MINOR;
728 vdso_data->processor = mfspr(SPRN_PVR);
730 * Fake the old platform number for pSeries and add
731 * in LPAR bit if necessary
733 vdso_data->platform = 0x100;
734 if (firmware_has_feature(FW_FEATURE_LPAR))
735 vdso_data->platform |= 1;
736 vdso_data->physicalMemorySize = memblock_phys_mem_size();
737 vdso_data->dcache_size = ppc64_caches.dsize;
738 vdso_data->dcache_line_size = ppc64_caches.dline_size;
739 vdso_data->icache_size = ppc64_caches.isize;
740 vdso_data->icache_line_size = ppc64_caches.iline_size;
742 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
743 vdso_data->dcache_block_size = ppc64_caches.dline_size;
744 vdso_data->icache_block_size = ppc64_caches.iline_size;
745 vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
746 vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
749 * Calculate the size of the 64 bits vDSO
751 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
752 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
754 vdso_data->dcache_block_size = L1_CACHE_BYTES;
755 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
756 vdso_data->icache_block_size = L1_CACHE_BYTES;
757 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
758 #endif /* CONFIG_PPC64 */
762 vdso32_kbase = &vdso32_start;
765 * Calculate the size of the 32 bits vDSO
767 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
768 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
773 * Setup the syscall map in the vDOS
775 vdso_setup_syscall_map();
778 * Initialize the vDSO images in memory, that is do necessary
779 * fixups of vDSO symbols, locate trampolines, etc...
782 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
791 /* Make sure pages are in the correct state */
792 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
794 BUG_ON(vdso32_pagelist == NULL);
795 for (i = 0; i < vdso32_pages; i++) {
796 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
797 ClearPageReserved(pg);
799 vdso32_pagelist[i] = pg;
801 vdso32_pagelist[i++] = virt_to_page(vdso_data);
802 vdso32_pagelist[i] = NULL;
806 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
808 BUG_ON(vdso64_pagelist == NULL);
809 for (i = 0; i < vdso64_pages; i++) {
810 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
811 ClearPageReserved(pg);
813 vdso64_pagelist[i] = pg;
815 vdso64_pagelist[i++] = virt_to_page(vdso_data);
816 vdso64_pagelist[i] = NULL;
817 #endif /* CONFIG_PPC64 */
819 get_page(virt_to_page(vdso_data));
826 arch_initcall(vdso_init);