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 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
105 "__kernel_gettimeofday", NULL
108 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
109 "__kernel_clock_gettime", NULL
112 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
113 "__kernel_clock_getres", NULL
116 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
117 "__kernel_get_tbfreq", NULL
120 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
121 "__kernel_time", NULL
127 * Some infos carried around for each of them during parsing at
132 Elf32_Ehdr *hdr; /* ptr to ELF */
133 Elf32_Sym *dynsym; /* ptr to .dynsym section */
134 unsigned long dynsymsize; /* size of .dynsym section */
135 char *dynstr; /* ptr to .dynstr section */
136 unsigned long text; /* offset of .text section in .so */
143 unsigned long dynsymsize;
150 * This is called from binfmt_elf, we create the special vma for the
151 * vDSO and insert it into the mm struct tree
153 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
155 struct mm_struct *mm = current->mm;
156 struct page **vdso_pagelist;
157 unsigned long vdso_pages;
158 unsigned long vdso_base;
165 if (is_32bit_task()) {
166 vdso_pagelist = vdso32_pagelist;
167 vdso_pages = vdso32_pages;
168 vdso_base = VDSO32_MBASE;
170 vdso_pagelist = vdso64_pagelist;
171 vdso_pages = vdso64_pages;
173 * On 64bit we don't have a preferred map address. This
174 * allows get_unmapped_area to find an area near other mmaps
175 * and most likely share a SLB entry.
180 vdso_pagelist = vdso32_pagelist;
181 vdso_pages = vdso32_pages;
182 vdso_base = VDSO32_MBASE;
185 current->mm->context.vdso_base = 0;
187 /* vDSO has a problem and was disabled, just don't "enable" it for the
192 /* Add a page to the vdso size for the data page */
196 * pick a base address for the vDSO in process space. We try to put it
197 * at vdso_base which is the "natural" base for it, but we might fail
198 * and end up putting it elsewhere.
199 * Add enough to the size so that the result can be aligned.
201 if (down_write_killable(&mm->mmap_sem))
203 vdso_base = get_unmapped_area(NULL, vdso_base,
204 (vdso_pages << PAGE_SHIFT) +
205 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
207 if (IS_ERR_VALUE(vdso_base)) {
212 /* Add required alignment. */
213 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
216 * Put vDSO base into mm struct. We need to do this before calling
217 * install_special_mapping or the perf counter mmap tracking code
218 * will fail to recognise it as a vDSO (since arch_vma_name fails).
220 current->mm->context.vdso_base = vdso_base;
223 * our vma flags don't have VM_WRITE so by default, the process isn't
224 * allowed to write those pages.
225 * gdb can break that with ptrace interface, and thus trigger COW on
226 * those pages but it's then your responsibility to never do that on
227 * the "data" page of the vDSO or you'll stop getting kernel updates
228 * and your nice userland gettimeofday will be totally dead.
229 * It's fine to use that for setting breakpoints in the vDSO code
232 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
234 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
237 current->mm->context.vdso_base = 0;
241 up_write(&mm->mmap_sem);
245 up_write(&mm->mmap_sem);
249 const char *arch_vma_name(struct vm_area_struct *vma)
251 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
259 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
266 /* Grab section headers and strings so we can tell who is who */
267 sechdrs = (void *)ehdr + ehdr->e_shoff;
268 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
270 /* Find the section they want */
271 for (i = 1; i < ehdr->e_shnum; i++) {
272 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
274 *size = sechdrs[i].sh_size;
275 return (void *)ehdr + sechdrs[i].sh_offset;
282 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
286 char name[MAX_SYMNAME], *c;
288 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
289 if (lib->dynsym[i].st_name == 0)
291 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
293 c = strchr(name, '@');
296 if (strcmp(symname, name) == 0)
297 return &lib->dynsym[i];
302 /* Note that we assume the section is .text and the symbol is relative to
305 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
308 Elf32_Sym *sym = find_symbol32(lib, symname);
311 printk(KERN_WARNING "vDSO32: function %s not found !\n",
315 return sym->st_value - VDSO32_LBASE;
318 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
319 struct lib64_elfinfo *v64,
320 const char *orig, const char *fix)
322 Elf32_Sym *sym32_gen, *sym32_fix;
324 sym32_gen = find_symbol32(v32, orig);
325 if (sym32_gen == NULL) {
326 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
330 sym32_gen->st_name = 0;
333 sym32_fix = find_symbol32(v32, fix);
334 if (sym32_fix == NULL) {
335 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
338 sym32_gen->st_value = sym32_fix->st_value;
339 sym32_gen->st_size = sym32_fix->st_size;
340 sym32_gen->st_info = sym32_fix->st_info;
341 sym32_gen->st_other = sym32_fix->st_other;
342 sym32_gen->st_shndx = sym32_fix->st_shndx;
346 #else /* !CONFIG_VDSO32 */
347 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
353 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
354 struct lib64_elfinfo *v64,
355 const char *orig, const char *fix)
359 #endif /* CONFIG_VDSO32 */
364 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
371 /* Grab section headers and strings so we can tell who is who */
372 sechdrs = (void *)ehdr + ehdr->e_shoff;
373 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
375 /* Find the section they want */
376 for (i = 1; i < ehdr->e_shnum; i++) {
377 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
379 *size = sechdrs[i].sh_size;
380 return (void *)ehdr + sechdrs[i].sh_offset;
388 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
392 char name[MAX_SYMNAME], *c;
394 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
395 if (lib->dynsym[i].st_name == 0)
397 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
399 c = strchr(name, '@');
402 if (strcmp(symname, name) == 0)
403 return &lib->dynsym[i];
408 /* Note that we assume the section is .text and the symbol is relative to
411 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
414 Elf64_Sym *sym = find_symbol64(lib, symname);
417 printk(KERN_WARNING "vDSO64: function %s not found !\n",
421 #ifdef VDS64_HAS_DESCRIPTORS
422 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
425 return sym->st_value - VDSO64_LBASE;
429 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
430 struct lib64_elfinfo *v64,
431 const char *orig, const char *fix)
433 Elf64_Sym *sym64_gen, *sym64_fix;
435 sym64_gen = find_symbol64(v64, orig);
436 if (sym64_gen == NULL) {
437 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
441 sym64_gen->st_name = 0;
444 sym64_fix = find_symbol64(v64, fix);
445 if (sym64_fix == NULL) {
446 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
449 sym64_gen->st_value = sym64_fix->st_value;
450 sym64_gen->st_size = sym64_fix->st_size;
451 sym64_gen->st_info = sym64_fix->st_info;
452 sym64_gen->st_other = sym64_fix->st_other;
453 sym64_gen->st_shndx = sym64_fix->st_shndx;
458 #endif /* CONFIG_PPC64 */
461 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
462 struct lib64_elfinfo *v64)
467 * Locate symbol tables & text section
471 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
472 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
473 if (v32->dynsym == NULL || v32->dynstr == NULL) {
474 printk(KERN_ERR "vDSO32: required symbol section not found\n");
477 sect = find_section32(v32->hdr, ".text", NULL);
479 printk(KERN_ERR "vDSO32: the .text section was not found\n");
482 v32->text = sect - vdso32_kbase;
486 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
487 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
488 if (v64->dynsym == NULL || v64->dynstr == NULL) {
489 printk(KERN_ERR "vDSO64: required symbol section not found\n");
492 sect = find_section64(v64->hdr, ".text", NULL);
494 printk(KERN_ERR "vDSO64: the .text section was not found\n");
497 v64->text = sect - vdso64_kbase;
498 #endif /* CONFIG_PPC64 */
503 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
504 struct lib64_elfinfo *v64)
507 * Find signal trampolines
511 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
513 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
514 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
517 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
518 struct lib64_elfinfo *v64)
526 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
528 printk(KERN_ERR "vDSO64: Can't find symbol "
529 "__kernel_datapage_offset !\n");
532 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
533 (vdso64_pages << PAGE_SHIFT) -
534 (sym64->st_value - VDSO64_LBASE);
535 #endif /* CONFIG_PPC64 */
538 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
540 printk(KERN_ERR "vDSO32: Can't find symbol "
541 "__kernel_datapage_offset !\n");
544 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
545 (vdso32_pages << PAGE_SHIFT) -
546 (sym32->st_value - VDSO32_LBASE);
553 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
554 struct lib64_elfinfo *v64)
560 start = find_section64(v64->hdr, "__ftr_fixup", &size);
562 do_feature_fixups(cur_cpu_spec->cpu_features,
563 start, start + size);
565 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
567 do_feature_fixups(cur_cpu_spec->mmu_features,
568 start, start + size);
570 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
572 do_feature_fixups(powerpc_firmware_features,
573 start, start + size);
575 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
577 do_lwsync_fixups(cur_cpu_spec->cpu_features,
578 start, start + size);
579 #endif /* CONFIG_PPC64 */
582 start = find_section32(v32->hdr, "__ftr_fixup", &size);
584 do_feature_fixups(cur_cpu_spec->cpu_features,
585 start, start + size);
587 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
589 do_feature_fixups(cur_cpu_spec->mmu_features,
590 start, start + size);
593 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
595 do_feature_fixups(powerpc_firmware_features,
596 start, start + size);
597 #endif /* CONFIG_PPC64 */
599 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
601 do_lwsync_fixups(cur_cpu_spec->cpu_features,
602 start, start + size);
608 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
609 struct lib64_elfinfo *v64)
613 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
614 struct vdso_patch_def *patch = &vdso_patches[i];
615 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
620 DBG("replacing %s with %s...\n", patch->gen_name,
621 patch->fix_name ? "NONE" : patch->fix_name);
624 * Patch the 32 bits and 64 bits symbols. Note that we do not
625 * patch the "." symbol on 64 bits.
626 * It would be easy to do, but doesn't seem to be necessary,
627 * patching the OPD symbol is enough.
629 vdso_do_func_patch32(v32, v64, patch->gen_name,
632 vdso_do_func_patch64(v32, v64, patch->gen_name,
634 #endif /* CONFIG_PPC64 */
641 static __init int vdso_setup(void)
643 struct lib32_elfinfo v32;
644 struct lib64_elfinfo v64;
646 v32.hdr = vdso32_kbase;
648 v64.hdr = vdso64_kbase;
650 if (vdso_do_find_sections(&v32, &v64))
653 if (vdso_fixup_datapage(&v32, &v64))
656 if (vdso_fixup_features(&v32, &v64))
659 if (vdso_fixup_alt_funcs(&v32, &v64))
662 vdso_setup_trampolines(&v32, &v64);
668 * Called from setup_arch to initialize the bitmap of available
669 * syscalls in the systemcfg page
671 static void __init vdso_setup_syscall_map(void)
674 extern unsigned long *sys_call_table;
675 extern unsigned long sys_ni_syscall;
678 for (i = 0; i < NR_syscalls; i++) {
680 if (sys_call_table[i*2] != sys_ni_syscall)
681 vdso_data->syscall_map_64[i >> 5] |=
682 0x80000000UL >> (i & 0x1f);
683 if (sys_call_table[i*2+1] != sys_ni_syscall)
684 vdso_data->syscall_map_32[i >> 5] |=
685 0x80000000UL >> (i & 0x1f);
686 #else /* CONFIG_PPC64 */
687 if (sys_call_table[i] != sys_ni_syscall)
688 vdso_data->syscall_map_32[i >> 5] |=
689 0x80000000UL >> (i & 0x1f);
690 #endif /* CONFIG_PPC64 */
695 int vdso_getcpu_init(void)
697 unsigned long cpu, node, val;
700 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
701 * in the next 16 bits. The VDSO uses this to implement getcpu().
704 WARN_ON_ONCE(cpu > 0xffff);
706 node = cpu_to_node(cpu);
707 WARN_ON_ONCE(node > 0xffff);
709 val = (cpu & 0xfff) | ((node & 0xffff) << 16);
710 mtspr(SPRN_SPRG_VDSO_WRITE, val);
711 get_paca()->sprg_vdso = val;
717 /* We need to call this before SMP init */
718 early_initcall(vdso_getcpu_init);
721 static int __init vdso_init(void)
727 * Fill up the "systemcfg" stuff for backward compatibility
729 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
730 vdso_data->version.major = SYSTEMCFG_MAJOR;
731 vdso_data->version.minor = SYSTEMCFG_MINOR;
732 vdso_data->processor = mfspr(SPRN_PVR);
734 * Fake the old platform number for pSeries and add
735 * in LPAR bit if necessary
737 vdso_data->platform = 0x100;
738 if (firmware_has_feature(FW_FEATURE_LPAR))
739 vdso_data->platform |= 1;
740 vdso_data->physicalMemorySize = memblock_phys_mem_size();
741 vdso_data->dcache_size = ppc64_caches.l1d.size;
742 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
743 vdso_data->icache_size = ppc64_caches.l1i.size;
744 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
745 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
746 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
747 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
748 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
751 * Calculate the size of the 64 bits vDSO
753 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
754 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
756 vdso_data->dcache_block_size = L1_CACHE_BYTES;
757 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
758 vdso_data->icache_block_size = L1_CACHE_BYTES;
759 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
760 #endif /* CONFIG_PPC64 */
764 vdso32_kbase = &vdso32_start;
767 * Calculate the size of the 32 bits vDSO
769 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
770 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
775 * Setup the syscall map in the vDOS
777 vdso_setup_syscall_map();
780 * Initialize the vDSO images in memory, that is do necessary
781 * fixups of vDSO symbols, locate trampolines, etc...
784 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
793 /* Make sure pages are in the correct state */
794 vdso32_pagelist = kcalloc(vdso32_pages + 2, sizeof(struct page *),
796 BUG_ON(vdso32_pagelist == NULL);
797 for (i = 0; i < vdso32_pages; i++) {
798 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
799 ClearPageReserved(pg);
801 vdso32_pagelist[i] = pg;
803 vdso32_pagelist[i++] = virt_to_page(vdso_data);
804 vdso32_pagelist[i] = NULL;
808 vdso64_pagelist = kcalloc(vdso64_pages + 2, sizeof(struct page *),
810 BUG_ON(vdso64_pagelist == NULL);
811 for (i = 0; i < vdso64_pages; i++) {
812 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
813 ClearPageReserved(pg);
815 vdso64_pagelist[i] = pg;
817 vdso64_pagelist[i++] = virt_to_page(vdso_data);
818 vdso64_pagelist[i] = NULL;
819 #endif /* CONFIG_PPC64 */
821 get_page(virt_to_page(vdso_data));
828 arch_initcall(vdso_init);