3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
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/cpu_has_feature.h>
26 #include <asm/pgtable.h>
27 #include <asm/processor.h>
29 #include <asm/mmu_context.h>
31 #include <asm/machdep.h>
32 #include <asm/cputable.h>
33 #include <asm/sections.h>
34 #include <asm/firmware.h>
36 #include <asm/vdso_datapage.h>
37 #include <asm/setup.h>
42 #define DBG(fmt...) printk(fmt)
47 /* Max supported size for symbol names */
48 #define MAX_SYMNAME 64
50 /* The alignment of the vDSO */
51 #define VDSO_ALIGNMENT (1 << 16)
53 static unsigned int vdso32_pages;
54 static void *vdso32_kbase;
55 static struct page **vdso32_pagelist;
56 unsigned long vdso32_sigtramp;
57 unsigned long vdso32_rt_sigtramp;
60 extern char vdso32_start, vdso32_end;
64 extern char vdso64_start, vdso64_end;
65 static void *vdso64_kbase = &vdso64_start;
66 static unsigned int vdso64_pages;
67 static struct page **vdso64_pagelist;
68 unsigned long vdso64_rt_sigtramp;
69 #endif /* CONFIG_PPC64 */
71 static int vdso_ready;
74 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
75 * Once the early boot kernel code no longer needs to muck around
76 * with it, it will become dynamically allocated
79 struct vdso_data data;
81 } vdso_data_store __page_aligned_data;
82 struct vdso_data *vdso_data = &vdso_data_store.data;
84 /* Format of the patch table */
87 unsigned long ftr_mask, ftr_value;
92 /* Table of functions to patch based on the CPU type/revision
94 * Currently, we only change sync_dicache to do nothing on processors
95 * with a coherent icache
97 static struct vdso_patch_def vdso_patches[] = {
99 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
100 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
104 "__kernel_gettimeofday", NULL
108 "__kernel_clock_gettime", NULL
112 "__kernel_clock_getres", NULL
116 "__kernel_get_tbfreq", NULL
120 "__kernel_time", NULL
125 * Some infos carried around for each of them during parsing at
130 Elf32_Ehdr *hdr; /* ptr to ELF */
131 Elf32_Sym *dynsym; /* ptr to .dynsym section */
132 unsigned long dynsymsize; /* size of .dynsym section */
133 char *dynstr; /* ptr to .dynstr section */
134 unsigned long text; /* offset of .text section in .so */
141 unsigned long dynsymsize;
148 * This is called from binfmt_elf, we create the special vma for the
149 * vDSO and insert it into the mm struct tree
151 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
153 struct mm_struct *mm = current->mm;
154 struct page **vdso_pagelist;
155 unsigned long vdso_pages;
156 unsigned long vdso_base;
163 if (is_32bit_task()) {
164 vdso_pagelist = vdso32_pagelist;
165 vdso_pages = vdso32_pages;
166 vdso_base = VDSO32_MBASE;
168 vdso_pagelist = vdso64_pagelist;
169 vdso_pages = vdso64_pages;
171 * On 64bit we don't have a preferred map address. This
172 * allows get_unmapped_area to find an area near other mmaps
173 * and most likely share a SLB entry.
178 vdso_pagelist = vdso32_pagelist;
179 vdso_pages = vdso32_pages;
180 vdso_base = VDSO32_MBASE;
183 current->mm->context.vdso_base = 0;
185 /* vDSO has a problem and was disabled, just don't "enable" it for the
190 /* Add a page to the vdso size for the data page */
194 * pick a base address for the vDSO in process space. We try to put it
195 * at vdso_base which is the "natural" base for it, but we might fail
196 * and end up putting it elsewhere.
197 * Add enough to the size so that the result can be aligned.
199 if (down_write_killable(&mm->mmap_sem))
201 vdso_base = get_unmapped_area(NULL, vdso_base,
202 (vdso_pages << PAGE_SHIFT) +
203 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
205 if (IS_ERR_VALUE(vdso_base)) {
210 /* Add required alignment. */
211 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
214 * Put vDSO base into mm struct. We need to do this before calling
215 * install_special_mapping or the perf counter mmap tracking code
216 * will fail to recognise it as a vDSO (since arch_vma_name fails).
218 current->mm->context.vdso_base = vdso_base;
221 * our vma flags don't have VM_WRITE so by default, the process isn't
222 * allowed to write those pages.
223 * gdb can break that with ptrace interface, and thus trigger COW on
224 * those pages but it's then your responsibility to never do that on
225 * the "data" page of the vDSO or you'll stop getting kernel updates
226 * and your nice userland gettimeofday will be totally dead.
227 * It's fine to use that for setting breakpoints in the vDSO code
230 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
232 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
235 current->mm->context.vdso_base = 0;
239 up_write(&mm->mmap_sem);
243 up_write(&mm->mmap_sem);
247 const char *arch_vma_name(struct vm_area_struct *vma)
249 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
257 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
264 /* Grab section headers and strings so we can tell who is who */
265 sechdrs = (void *)ehdr + ehdr->e_shoff;
266 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
268 /* Find the section they want */
269 for (i = 1; i < ehdr->e_shnum; i++) {
270 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
272 *size = sechdrs[i].sh_size;
273 return (void *)ehdr + sechdrs[i].sh_offset;
280 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
284 char name[MAX_SYMNAME], *c;
286 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
287 if (lib->dynsym[i].st_name == 0)
289 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
291 c = strchr(name, '@');
294 if (strcmp(symname, name) == 0)
295 return &lib->dynsym[i];
300 /* Note that we assume the section is .text and the symbol is relative to
303 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
306 Elf32_Sym *sym = find_symbol32(lib, symname);
309 printk(KERN_WARNING "vDSO32: function %s not found !\n",
313 return sym->st_value - VDSO32_LBASE;
316 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
317 struct lib64_elfinfo *v64,
318 const char *orig, const char *fix)
320 Elf32_Sym *sym32_gen, *sym32_fix;
322 sym32_gen = find_symbol32(v32, orig);
323 if (sym32_gen == NULL) {
324 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
328 sym32_gen->st_name = 0;
331 sym32_fix = find_symbol32(v32, fix);
332 if (sym32_fix == NULL) {
333 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
336 sym32_gen->st_value = sym32_fix->st_value;
337 sym32_gen->st_size = sym32_fix->st_size;
338 sym32_gen->st_info = sym32_fix->st_info;
339 sym32_gen->st_other = sym32_fix->st_other;
340 sym32_gen->st_shndx = sym32_fix->st_shndx;
344 #else /* !CONFIG_VDSO32 */
345 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
351 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
352 struct lib64_elfinfo *v64,
353 const char *orig, const char *fix)
357 #endif /* CONFIG_VDSO32 */
362 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
369 /* Grab section headers and strings so we can tell who is who */
370 sechdrs = (void *)ehdr + ehdr->e_shoff;
371 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
373 /* Find the section they want */
374 for (i = 1; i < ehdr->e_shnum; i++) {
375 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
377 *size = sechdrs[i].sh_size;
378 return (void *)ehdr + sechdrs[i].sh_offset;
386 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
390 char name[MAX_SYMNAME], *c;
392 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
393 if (lib->dynsym[i].st_name == 0)
395 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
397 c = strchr(name, '@');
400 if (strcmp(symname, name) == 0)
401 return &lib->dynsym[i];
406 /* Note that we assume the section is .text and the symbol is relative to
409 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
412 Elf64_Sym *sym = find_symbol64(lib, symname);
415 printk(KERN_WARNING "vDSO64: function %s not found !\n",
419 #ifdef VDS64_HAS_DESCRIPTORS
420 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
423 return sym->st_value - VDSO64_LBASE;
427 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
428 struct lib64_elfinfo *v64,
429 const char *orig, const char *fix)
431 Elf64_Sym *sym64_gen, *sym64_fix;
433 sym64_gen = find_symbol64(v64, orig);
434 if (sym64_gen == NULL) {
435 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
439 sym64_gen->st_name = 0;
442 sym64_fix = find_symbol64(v64, fix);
443 if (sym64_fix == NULL) {
444 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
447 sym64_gen->st_value = sym64_fix->st_value;
448 sym64_gen->st_size = sym64_fix->st_size;
449 sym64_gen->st_info = sym64_fix->st_info;
450 sym64_gen->st_other = sym64_fix->st_other;
451 sym64_gen->st_shndx = sym64_fix->st_shndx;
456 #endif /* CONFIG_PPC64 */
459 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
460 struct lib64_elfinfo *v64)
465 * Locate symbol tables & text section
469 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
470 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
471 if (v32->dynsym == NULL || v32->dynstr == NULL) {
472 printk(KERN_ERR "vDSO32: required symbol section not found\n");
475 sect = find_section32(v32->hdr, ".text", NULL);
477 printk(KERN_ERR "vDSO32: the .text section was not found\n");
480 v32->text = sect - vdso32_kbase;
484 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
485 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
486 if (v64->dynsym == NULL || v64->dynstr == NULL) {
487 printk(KERN_ERR "vDSO64: required symbol section not found\n");
490 sect = find_section64(v64->hdr, ".text", NULL);
492 printk(KERN_ERR "vDSO64: the .text section was not found\n");
495 v64->text = sect - vdso64_kbase;
496 #endif /* CONFIG_PPC64 */
501 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
502 struct lib64_elfinfo *v64)
505 * Find signal trampolines
509 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
511 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
512 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
515 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
516 struct lib64_elfinfo *v64)
524 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
526 printk(KERN_ERR "vDSO64: Can't find symbol "
527 "__kernel_datapage_offset !\n");
530 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
531 (vdso64_pages << PAGE_SHIFT) -
532 (sym64->st_value - VDSO64_LBASE);
533 #endif /* CONFIG_PPC64 */
536 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
538 printk(KERN_ERR "vDSO32: Can't find symbol "
539 "__kernel_datapage_offset !\n");
542 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
543 (vdso32_pages << PAGE_SHIFT) -
544 (sym32->st_value - VDSO32_LBASE);
551 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
552 struct lib64_elfinfo *v64)
558 start = find_section64(v64->hdr, "__ftr_fixup", &size);
560 do_feature_fixups(cur_cpu_spec->cpu_features,
561 start, start + size);
563 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
565 do_feature_fixups(cur_cpu_spec->mmu_features,
566 start, start + size);
568 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
570 do_feature_fixups(powerpc_firmware_features,
571 start, start + size);
573 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
575 do_lwsync_fixups(cur_cpu_spec->cpu_features,
576 start, start + size);
577 #endif /* CONFIG_PPC64 */
580 start = find_section32(v32->hdr, "__ftr_fixup", &size);
582 do_feature_fixups(cur_cpu_spec->cpu_features,
583 start, start + size);
585 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
587 do_feature_fixups(cur_cpu_spec->mmu_features,
588 start, start + size);
591 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
593 do_feature_fixups(powerpc_firmware_features,
594 start, start + size);
595 #endif /* CONFIG_PPC64 */
597 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
599 do_lwsync_fixups(cur_cpu_spec->cpu_features,
600 start, start + size);
606 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
607 struct lib64_elfinfo *v64)
611 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
612 struct vdso_patch_def *patch = &vdso_patches[i];
613 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
618 DBG("replacing %s with %s...\n", patch->gen_name,
619 patch->fix_name ? "NONE" : patch->fix_name);
622 * Patch the 32 bits and 64 bits symbols. Note that we do not
623 * patch the "." symbol on 64 bits.
624 * It would be easy to do, but doesn't seem to be necessary,
625 * patching the OPD symbol is enough.
627 vdso_do_func_patch32(v32, v64, patch->gen_name,
630 vdso_do_func_patch64(v32, v64, patch->gen_name,
632 #endif /* CONFIG_PPC64 */
639 static __init int vdso_setup(void)
641 struct lib32_elfinfo v32;
642 struct lib64_elfinfo v64;
644 v32.hdr = vdso32_kbase;
646 v64.hdr = vdso64_kbase;
648 if (vdso_do_find_sections(&v32, &v64))
651 if (vdso_fixup_datapage(&v32, &v64))
654 if (vdso_fixup_features(&v32, &v64))
657 if (vdso_fixup_alt_funcs(&v32, &v64))
660 vdso_setup_trampolines(&v32, &v64);
666 * Called from setup_arch to initialize the bitmap of available
667 * syscalls in the systemcfg page
669 static void __init vdso_setup_syscall_map(void)
672 extern unsigned long *sys_call_table;
673 extern unsigned long sys_ni_syscall;
676 for (i = 0; i < NR_syscalls; i++) {
678 if (sys_call_table[i*2] != sys_ni_syscall)
679 vdso_data->syscall_map_64[i >> 5] |=
680 0x80000000UL >> (i & 0x1f);
681 if (sys_call_table[i*2+1] != sys_ni_syscall)
682 vdso_data->syscall_map_32[i >> 5] |=
683 0x80000000UL >> (i & 0x1f);
684 #else /* CONFIG_PPC64 */
685 if (sys_call_table[i] != sys_ni_syscall)
686 vdso_data->syscall_map_32[i >> 5] |=
687 0x80000000UL >> (i & 0x1f);
688 #endif /* CONFIG_PPC64 */
693 int vdso_getcpu_init(void)
695 unsigned long cpu, node, val;
698 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
699 * in the next 16 bits. The VDSO uses this to implement getcpu().
702 WARN_ON_ONCE(cpu > 0xffff);
704 node = cpu_to_node(cpu);
705 WARN_ON_ONCE(node > 0xffff);
707 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
708 mtspr(SPRN_SPRG_VDSO_WRITE, val);
709 get_paca()->sprg_vdso = val;
715 /* We need to call this before SMP init */
716 early_initcall(vdso_getcpu_init);
719 static int __init vdso_init(void)
725 * Fill up the "systemcfg" stuff for backward compatibility
727 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
728 vdso_data->version.major = SYSTEMCFG_MAJOR;
729 vdso_data->version.minor = SYSTEMCFG_MINOR;
730 vdso_data->processor = mfspr(SPRN_PVR);
732 * Fake the old platform number for pSeries and add
733 * in LPAR bit if necessary
735 vdso_data->platform = 0x100;
736 if (firmware_has_feature(FW_FEATURE_LPAR))
737 vdso_data->platform |= 1;
738 vdso_data->physicalMemorySize = memblock_phys_mem_size();
739 vdso_data->dcache_size = ppc64_caches.l1d.size;
740 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
741 vdso_data->icache_size = ppc64_caches.l1i.size;
742 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
743 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
744 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
745 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
746 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_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);