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/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 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
104 "__kernel_gettimeofday", NULL
107 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
108 "__kernel_clock_gettime", NULL
111 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
112 "__kernel_clock_getres", NULL
115 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
116 "__kernel_get_tbfreq", NULL
119 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
120 "__kernel_time", NULL
126 * Some infos carried around for each of them during parsing at
131 Elf32_Ehdr *hdr; /* ptr to ELF */
132 Elf32_Sym *dynsym; /* ptr to .dynsym section */
133 unsigned long dynsymsize; /* size of .dynsym section */
134 char *dynstr; /* ptr to .dynstr section */
135 unsigned long text; /* offset of .text section in .so */
142 unsigned long dynsymsize;
149 * This is called from binfmt_elf, we create the special vma for the
150 * vDSO and insert it into the mm struct tree
152 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
154 struct mm_struct *mm = current->mm;
155 struct page **vdso_pagelist;
156 unsigned long vdso_pages;
157 unsigned long vdso_base;
164 if (is_32bit_task()) {
165 vdso_pagelist = vdso32_pagelist;
166 vdso_pages = vdso32_pages;
167 vdso_base = VDSO32_MBASE;
169 vdso_pagelist = vdso64_pagelist;
170 vdso_pages = vdso64_pages;
172 * On 64bit we don't have a preferred map address. This
173 * allows get_unmapped_area to find an area near other mmaps
174 * and most likely share a SLB entry.
179 vdso_pagelist = vdso32_pagelist;
180 vdso_pages = vdso32_pages;
181 vdso_base = VDSO32_MBASE;
184 current->mm->context.vdso_base = 0;
186 /* vDSO has a problem and was disabled, just don't "enable" it for the
191 /* Add a page to the vdso size for the data page */
195 * pick a base address for the vDSO in process space. We try to put it
196 * at vdso_base which is the "natural" base for it, but we might fail
197 * and end up putting it elsewhere.
198 * Add enough to the size so that the result can be aligned.
200 if (down_write_killable(&mm->mmap_sem))
202 vdso_base = get_unmapped_area(NULL, vdso_base,
203 (vdso_pages << PAGE_SHIFT) +
204 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
206 if (IS_ERR_VALUE(vdso_base)) {
211 /* Add required alignment. */
212 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
215 * Put vDSO base into mm struct. We need to do this before calling
216 * install_special_mapping or the perf counter mmap tracking code
217 * will fail to recognise it as a vDSO (since arch_vma_name fails).
219 current->mm->context.vdso_base = vdso_base;
222 * our vma flags don't have VM_WRITE so by default, the process isn't
223 * allowed to write those pages.
224 * gdb can break that with ptrace interface, and thus trigger COW on
225 * those pages but it's then your responsibility to never do that on
226 * the "data" page of the vDSO or you'll stop getting kernel updates
227 * and your nice userland gettimeofday will be totally dead.
228 * It's fine to use that for setting breakpoints in the vDSO code
231 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
233 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
236 current->mm->context.vdso_base = 0;
240 up_write(&mm->mmap_sem);
244 up_write(&mm->mmap_sem);
248 const char *arch_vma_name(struct vm_area_struct *vma)
250 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
258 static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
265 /* Grab section headers and strings so we can tell who is who */
266 sechdrs = (void *)ehdr + ehdr->e_shoff;
267 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
269 /* Find the section they want */
270 for (i = 1; i < ehdr->e_shnum; i++) {
271 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
273 *size = sechdrs[i].sh_size;
274 return (void *)ehdr + sechdrs[i].sh_offset;
281 static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
285 char name[MAX_SYMNAME], *c;
287 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
288 if (lib->dynsym[i].st_name == 0)
290 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
292 c = strchr(name, '@');
295 if (strcmp(symname, name) == 0)
296 return &lib->dynsym[i];
301 /* Note that we assume the section is .text and the symbol is relative to
304 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
307 Elf32_Sym *sym = find_symbol32(lib, symname);
310 printk(KERN_WARNING "vDSO32: function %s not found !\n",
314 return sym->st_value - VDSO32_LBASE;
317 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
318 struct lib64_elfinfo *v64,
319 const char *orig, const char *fix)
321 Elf32_Sym *sym32_gen, *sym32_fix;
323 sym32_gen = find_symbol32(v32, orig);
324 if (sym32_gen == NULL) {
325 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
329 sym32_gen->st_name = 0;
332 sym32_fix = find_symbol32(v32, fix);
333 if (sym32_fix == NULL) {
334 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
337 sym32_gen->st_value = sym32_fix->st_value;
338 sym32_gen->st_size = sym32_fix->st_size;
339 sym32_gen->st_info = sym32_fix->st_info;
340 sym32_gen->st_other = sym32_fix->st_other;
341 sym32_gen->st_shndx = sym32_fix->st_shndx;
345 #else /* !CONFIG_VDSO32 */
346 static unsigned long __init find_function32(struct lib32_elfinfo *lib,
352 static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
353 struct lib64_elfinfo *v64,
354 const char *orig, const char *fix)
358 #endif /* CONFIG_VDSO32 */
363 static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
370 /* Grab section headers and strings so we can tell who is who */
371 sechdrs = (void *)ehdr + ehdr->e_shoff;
372 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
374 /* Find the section they want */
375 for (i = 1; i < ehdr->e_shnum; i++) {
376 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
378 *size = sechdrs[i].sh_size;
379 return (void *)ehdr + sechdrs[i].sh_offset;
387 static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
391 char name[MAX_SYMNAME], *c;
393 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
394 if (lib->dynsym[i].st_name == 0)
396 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
398 c = strchr(name, '@');
401 if (strcmp(symname, name) == 0)
402 return &lib->dynsym[i];
407 /* Note that we assume the section is .text and the symbol is relative to
410 static unsigned long __init find_function64(struct lib64_elfinfo *lib,
413 Elf64_Sym *sym = find_symbol64(lib, symname);
416 printk(KERN_WARNING "vDSO64: function %s not found !\n",
420 #ifdef VDS64_HAS_DESCRIPTORS
421 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
424 return sym->st_value - VDSO64_LBASE;
428 static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
429 struct lib64_elfinfo *v64,
430 const char *orig, const char *fix)
432 Elf64_Sym *sym64_gen, *sym64_fix;
434 sym64_gen = find_symbol64(v64, orig);
435 if (sym64_gen == NULL) {
436 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
440 sym64_gen->st_name = 0;
443 sym64_fix = find_symbol64(v64, fix);
444 if (sym64_fix == NULL) {
445 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
448 sym64_gen->st_value = sym64_fix->st_value;
449 sym64_gen->st_size = sym64_fix->st_size;
450 sym64_gen->st_info = sym64_fix->st_info;
451 sym64_gen->st_other = sym64_fix->st_other;
452 sym64_gen->st_shndx = sym64_fix->st_shndx;
457 #endif /* CONFIG_PPC64 */
460 static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
461 struct lib64_elfinfo *v64)
466 * Locate symbol tables & text section
470 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
471 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
472 if (v32->dynsym == NULL || v32->dynstr == NULL) {
473 printk(KERN_ERR "vDSO32: required symbol section not found\n");
476 sect = find_section32(v32->hdr, ".text", NULL);
478 printk(KERN_ERR "vDSO32: the .text section was not found\n");
481 v32->text = sect - vdso32_kbase;
485 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
486 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
487 if (v64->dynsym == NULL || v64->dynstr == NULL) {
488 printk(KERN_ERR "vDSO64: required symbol section not found\n");
491 sect = find_section64(v64->hdr, ".text", NULL);
493 printk(KERN_ERR "vDSO64: the .text section was not found\n");
496 v64->text = sect - vdso64_kbase;
497 #endif /* CONFIG_PPC64 */
502 static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
503 struct lib64_elfinfo *v64)
506 * Find signal trampolines
510 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
512 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
513 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
516 static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
517 struct lib64_elfinfo *v64)
525 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
527 printk(KERN_ERR "vDSO64: Can't find symbol "
528 "__kernel_datapage_offset !\n");
531 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
532 (vdso64_pages << PAGE_SHIFT) -
533 (sym64->st_value - VDSO64_LBASE);
534 #endif /* CONFIG_PPC64 */
537 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
539 printk(KERN_ERR "vDSO32: Can't find symbol "
540 "__kernel_datapage_offset !\n");
543 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
544 (vdso32_pages << PAGE_SHIFT) -
545 (sym32->st_value - VDSO32_LBASE);
552 static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
553 struct lib64_elfinfo *v64)
559 start = find_section64(v64->hdr, "__ftr_fixup", &size);
561 do_feature_fixups(cur_cpu_spec->cpu_features,
562 start, start + size);
564 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
566 do_feature_fixups(cur_cpu_spec->mmu_features,
567 start, start + size);
569 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
571 do_feature_fixups(powerpc_firmware_features,
572 start, start + size);
574 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
576 do_lwsync_fixups(cur_cpu_spec->cpu_features,
577 start, start + size);
578 #endif /* CONFIG_PPC64 */
581 start = find_section32(v32->hdr, "__ftr_fixup", &size);
583 do_feature_fixups(cur_cpu_spec->cpu_features,
584 start, start + size);
586 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
588 do_feature_fixups(cur_cpu_spec->mmu_features,
589 start, start + size);
592 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
594 do_feature_fixups(powerpc_firmware_features,
595 start, start + size);
596 #endif /* CONFIG_PPC64 */
598 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
600 do_lwsync_fixups(cur_cpu_spec->cpu_features,
601 start, start + size);
607 static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
608 struct lib64_elfinfo *v64)
612 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
613 struct vdso_patch_def *patch = &vdso_patches[i];
614 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
619 DBG("replacing %s with %s...\n", patch->gen_name,
620 patch->fix_name ? "NONE" : patch->fix_name);
623 * Patch the 32 bits and 64 bits symbols. Note that we do not
624 * patch the "." symbol on 64 bits.
625 * It would be easy to do, but doesn't seem to be necessary,
626 * patching the OPD symbol is enough.
628 vdso_do_func_patch32(v32, v64, patch->gen_name,
631 vdso_do_func_patch64(v32, v64, patch->gen_name,
633 #endif /* CONFIG_PPC64 */
640 static __init int vdso_setup(void)
642 struct lib32_elfinfo v32;
643 struct lib64_elfinfo v64;
645 v32.hdr = vdso32_kbase;
647 v64.hdr = vdso64_kbase;
649 if (vdso_do_find_sections(&v32, &v64))
652 if (vdso_fixup_datapage(&v32, &v64))
655 if (vdso_fixup_features(&v32, &v64))
658 if (vdso_fixup_alt_funcs(&v32, &v64))
661 vdso_setup_trampolines(&v32, &v64);
667 * Called from setup_arch to initialize the bitmap of available
668 * syscalls in the systemcfg page
670 static void __init vdso_setup_syscall_map(void)
673 extern unsigned long *sys_call_table;
674 extern unsigned long sys_ni_syscall;
677 for (i = 0; i < NR_syscalls; i++) {
679 if (sys_call_table[i*2] != sys_ni_syscall)
680 vdso_data->syscall_map_64[i >> 5] |=
681 0x80000000UL >> (i & 0x1f);
682 if (sys_call_table[i*2+1] != sys_ni_syscall)
683 vdso_data->syscall_map_32[i >> 5] |=
684 0x80000000UL >> (i & 0x1f);
685 #else /* CONFIG_PPC64 */
686 if (sys_call_table[i] != sys_ni_syscall)
687 vdso_data->syscall_map_32[i >> 5] |=
688 0x80000000UL >> (i & 0x1f);
689 #endif /* CONFIG_PPC64 */
694 int vdso_getcpu_init(void)
696 unsigned long cpu, node, val;
699 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
700 * in the next 16 bits. The VDSO uses this to implement getcpu().
703 WARN_ON_ONCE(cpu > 0xffff);
705 node = cpu_to_node(cpu);
706 WARN_ON_ONCE(node > 0xffff);
708 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
709 mtspr(SPRN_SPRG_VDSO_WRITE, val);
710 get_paca()->sprg_vdso = val;
716 /* We need to call this before SMP init */
717 early_initcall(vdso_getcpu_init);
720 static int __init vdso_init(void)
726 * Fill up the "systemcfg" stuff for backward compatibility
728 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
729 vdso_data->version.major = SYSTEMCFG_MAJOR;
730 vdso_data->version.minor = SYSTEMCFG_MINOR;
731 vdso_data->processor = mfspr(SPRN_PVR);
733 * Fake the old platform number for pSeries and add
734 * in LPAR bit if necessary
736 vdso_data->platform = 0x100;
737 if (firmware_has_feature(FW_FEATURE_LPAR))
738 vdso_data->platform |= 1;
739 vdso_data->physicalMemorySize = memblock_phys_mem_size();
740 vdso_data->dcache_size = ppc64_caches.l1d.size;
741 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
742 vdso_data->icache_size = ppc64_caches.l1i.size;
743 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
744 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
745 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
746 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
747 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
750 * Calculate the size of the 64 bits vDSO
752 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
753 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
755 vdso_data->dcache_block_size = L1_CACHE_BYTES;
756 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
757 vdso_data->icache_block_size = L1_CACHE_BYTES;
758 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
759 #endif /* CONFIG_PPC64 */
763 vdso32_kbase = &vdso32_start;
766 * Calculate the size of the 32 bits vDSO
768 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
769 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
774 * Setup the syscall map in the vDOS
776 vdso_setup_syscall_map();
779 * Initialize the vDSO images in memory, that is do necessary
780 * fixups of vDSO symbols, locate trampolines, etc...
783 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
792 /* Make sure pages are in the correct state */
793 vdso32_pagelist = kcalloc(vdso32_pages + 2, sizeof(struct page *),
795 BUG_ON(vdso32_pagelist == NULL);
796 for (i = 0; i < vdso32_pages; i++) {
797 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
798 ClearPageReserved(pg);
800 vdso32_pagelist[i] = pg;
802 vdso32_pagelist[i++] = virt_to_page(vdso_data);
803 vdso32_pagelist[i] = NULL;
807 vdso64_pagelist = kcalloc(vdso64_pages + 2, sizeof(struct page *),
809 BUG_ON(vdso64_pagelist == NULL);
810 for (i = 0; i < vdso64_pages; i++) {
811 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
812 ClearPageReserved(pg);
814 vdso64_pagelist[i] = pg;
816 vdso64_pagelist[i++] = virt_to_page(vdso_data);
817 vdso64_pagelist[i] = NULL;
818 #endif /* CONFIG_PPC64 */
820 get_page(virt_to_page(vdso_data));
827 arch_initcall(vdso_init);