Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2
3/*
4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
5 * <benh@kernel.crashing.org>
6 */
7
8#include <linux/errno.h>
9#include <linux/sched.h>
10#include <linux/kernel.h>
11#include <linux/mm.h>
12#include <linux/smp.h>
13#include <linux/stddef.h>
14#include <linux/unistd.h>
15#include <linux/slab.h>
16#include <linux/user.h>
17#include <linux/elf.h>
18#include <linux/security.h>
19#include <linux/memblock.h>
20#include <linux/syscalls.h>
21#include <linux/time_namespace.h>
22#include <vdso/datapage.h>
23
24#include <asm/syscall.h>
25#include <asm/processor.h>
26#include <asm/mmu.h>
27#include <asm/mmu_context.h>
28#include <asm/machdep.h>
29#include <asm/cputable.h>
30#include <asm/sections.h>
31#include <asm/firmware.h>
32#include <asm/vdso.h>
33#include <asm/vdso_datapage.h>
34#include <asm/setup.h>
35
36/* The alignment of the vDSO */
37#define VDSO_ALIGNMENT (1 << 16)
38
39extern char vdso32_start, vdso32_end;
40extern char vdso64_start, vdso64_end;
41
42long sys_ni_syscall(void);
43
44/*
45 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
46 * Once the early boot kernel code no longer needs to muck around
47 * with it, it will become dynamically allocated
48 */
49static union {
50 struct vdso_arch_data data;
51 u8 page[PAGE_SIZE];
52} vdso_data_store __page_aligned_data;
53struct vdso_arch_data *vdso_data = &vdso_data_store.data;
54
55enum vvar_pages {
56 VVAR_DATA_PAGE_OFFSET,
57 VVAR_TIMENS_PAGE_OFFSET,
58 VVAR_NR_PAGES,
59};
60
61static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma,
62 unsigned long text_size)
63{
64 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
65
66 if (new_size != text_size)
67 return -EINVAL;
68
69 current->mm->context.vdso = (void __user *)new_vma->vm_start;
70
71 return 0;
72}
73
74static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
75{
76 return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start);
77}
78
79static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
80{
81 return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start);
82}
83
84static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
85 struct vm_area_struct *vma, struct vm_fault *vmf);
86
87static struct vm_special_mapping vvar_spec __ro_after_init = {
88 .name = "[vvar]",
89 .fault = vvar_fault,
90};
91
92static struct vm_special_mapping vdso32_spec __ro_after_init = {
93 .name = "[vdso]",
94 .mremap = vdso32_mremap,
95};
96
97static struct vm_special_mapping vdso64_spec __ro_after_init = {
98 .name = "[vdso]",
99 .mremap = vdso64_mremap,
100};
101
102#ifdef CONFIG_TIME_NS
103struct vdso_data *arch_get_vdso_data(void *vvar_page)
104{
105 return ((struct vdso_arch_data *)vvar_page)->data;
106}
107
108/*
109 * The vvar mapping contains data for a specific time namespace, so when a task
110 * changes namespace we must unmap its vvar data for the old namespace.
111 * Subsequent faults will map in data for the new namespace.
112 *
113 * For more details see timens_setup_vdso_data().
114 */
115int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
116{
117 struct mm_struct *mm = task->mm;
118 VMA_ITERATOR(vmi, mm, 0);
119 struct vm_area_struct *vma;
120
121 mmap_read_lock(mm);
122 for_each_vma(vmi, vma) {
123 unsigned long size = vma->vm_end - vma->vm_start;
124
125 if (vma_is_special_mapping(vma, &vvar_spec))
126 zap_page_range(vma, vma->vm_start, size);
127 }
128 mmap_read_unlock(mm);
129
130 return 0;
131}
132#endif
133
134static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
135 struct vm_area_struct *vma, struct vm_fault *vmf)
136{
137 struct page *timens_page = find_timens_vvar_page(vma);
138 unsigned long pfn;
139
140 switch (vmf->pgoff) {
141 case VVAR_DATA_PAGE_OFFSET:
142 if (timens_page)
143 pfn = page_to_pfn(timens_page);
144 else
145 pfn = virt_to_pfn(vdso_data);
146 break;
147#ifdef CONFIG_TIME_NS
148 case VVAR_TIMENS_PAGE_OFFSET:
149 /*
150 * If a task belongs to a time namespace then a namespace
151 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
152 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
153 * offset.
154 * See also the comment near timens_setup_vdso_data().
155 */
156 if (!timens_page)
157 return VM_FAULT_SIGBUS;
158 pfn = virt_to_pfn(vdso_data);
159 break;
160#endif /* CONFIG_TIME_NS */
161 default:
162 return VM_FAULT_SIGBUS;
163 }
164
165 return vmf_insert_pfn(vma, vmf->address, pfn);
166}
167
168/*
169 * This is called from binfmt_elf, we create the special vma for the
170 * vDSO and insert it into the mm struct tree
171 */
172static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
173{
174 unsigned long vdso_size, vdso_base, mappings_size;
175 struct vm_special_mapping *vdso_spec;
176 unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE;
177 struct mm_struct *mm = current->mm;
178 struct vm_area_struct *vma;
179
180 if (is_32bit_task()) {
181 vdso_spec = &vdso32_spec;
182 vdso_size = &vdso32_end - &vdso32_start;
183 } else {
184 vdso_spec = &vdso64_spec;
185 vdso_size = &vdso64_end - &vdso64_start;
186 }
187
188 mappings_size = vdso_size + vvar_size;
189 mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
190
191 /*
192 * Pick a base address for the vDSO in process space.
193 * Add enough to the size so that the result can be aligned.
194 */
195 vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0);
196 if (IS_ERR_VALUE(vdso_base))
197 return vdso_base;
198
199 /* Add required alignment. */
200 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
201
202 /*
203 * Put vDSO base into mm struct. We need to do this before calling
204 * install_special_mapping or the perf counter mmap tracking code
205 * will fail to recognise it as a vDSO.
206 */
207 mm->context.vdso = (void __user *)vdso_base + vvar_size;
208
209 vma = _install_special_mapping(mm, vdso_base, vvar_size,
210 VM_READ | VM_MAYREAD | VM_IO |
211 VM_DONTDUMP | VM_PFNMAP, &vvar_spec);
212 if (IS_ERR(vma))
213 return PTR_ERR(vma);
214
215 /*
216 * our vma flags don't have VM_WRITE so by default, the process isn't
217 * allowed to write those pages.
218 * gdb can break that with ptrace interface, and thus trigger COW on
219 * those pages but it's then your responsibility to never do that on
220 * the "data" page of the vDSO or you'll stop getting kernel updates
221 * and your nice userland gettimeofday will be totally dead.
222 * It's fine to use that for setting breakpoints in the vDSO code
223 * pages though.
224 */
225 vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
226 VM_READ | VM_EXEC | VM_MAYREAD |
227 VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
228 if (IS_ERR(vma))
229 do_munmap(mm, vdso_base, vvar_size, NULL);
230
231 return PTR_ERR_OR_ZERO(vma);
232}
233
234int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
235{
236 struct mm_struct *mm = current->mm;
237 int rc;
238
239 mm->context.vdso = NULL;
240
241 if (mmap_write_lock_killable(mm))
242 return -EINTR;
243
244 rc = __arch_setup_additional_pages(bprm, uses_interp);
245 if (rc)
246 mm->context.vdso = NULL;
247
248 mmap_write_unlock(mm);
249 return rc;
250}
251
252#define VDSO_DO_FIXUPS(type, value, bits, sec) do { \
253 void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start); \
254 void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end); \
255 \
256 do_##type##_fixups((value), __start, __end); \
257} while (0)
258
259static void __init vdso_fixup_features(void)
260{
261#ifdef CONFIG_PPC64
262 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
263 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
264 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
265 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
266#endif /* CONFIG_PPC64 */
267
268#ifdef CONFIG_VDSO32
269 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
270 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
271#ifdef CONFIG_PPC64
272 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
273#endif /* CONFIG_PPC64 */
274 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
275#endif
276}
277
278/*
279 * Called from setup_arch to initialize the bitmap of available
280 * syscalls in the systemcfg page
281 */
282static void __init vdso_setup_syscall_map(void)
283{
284 unsigned int i;
285
286 for (i = 0; i < NR_syscalls; i++) {
287 if (sys_call_table[i] != (void *)&sys_ni_syscall)
288 vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
289 if (IS_ENABLED(CONFIG_COMPAT) &&
290 compat_sys_call_table[i] != (void *)&sys_ni_syscall)
291 vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
292 }
293}
294
295#ifdef CONFIG_PPC64
296int vdso_getcpu_init(void)
297{
298 unsigned long cpu, node, val;
299
300 /*
301 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
302 * in the next 16 bits. The VDSO uses this to implement getcpu().
303 */
304 cpu = get_cpu();
305 WARN_ON_ONCE(cpu > 0xffff);
306
307 node = cpu_to_node(cpu);
308 WARN_ON_ONCE(node > 0xffff);
309
310 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
311 mtspr(SPRN_SPRG_VDSO_WRITE, val);
312 get_paca()->sprg_vdso = val;
313
314 put_cpu();
315
316 return 0;
317}
318/* We need to call this before SMP init */
319early_initcall(vdso_getcpu_init);
320#endif
321
322static struct page ** __init vdso_setup_pages(void *start, void *end)
323{
324 int i;
325 struct page **pagelist;
326 int pages = (end - start) >> PAGE_SHIFT;
327
328 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
329 if (!pagelist)
330 panic("%s: Cannot allocate page list for VDSO", __func__);
331
332 for (i = 0; i < pages; i++)
333 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
334
335 return pagelist;
336}
337
338static int __init vdso_init(void)
339{
340#ifdef CONFIG_PPC64
341 /*
342 * Fill up the "systemcfg" stuff for backward compatibility
343 */
344 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
345 vdso_data->version.major = SYSTEMCFG_MAJOR;
346 vdso_data->version.minor = SYSTEMCFG_MINOR;
347 vdso_data->processor = mfspr(SPRN_PVR);
348 /*
349 * Fake the old platform number for pSeries and add
350 * in LPAR bit if necessary
351 */
352 vdso_data->platform = 0x100;
353 if (firmware_has_feature(FW_FEATURE_LPAR))
354 vdso_data->platform |= 1;
355 vdso_data->physicalMemorySize = memblock_phys_mem_size();
356 vdso_data->dcache_size = ppc64_caches.l1d.size;
357 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
358 vdso_data->icache_size = ppc64_caches.l1i.size;
359 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
360 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
361 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
362 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
363 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
364#endif /* CONFIG_PPC64 */
365
366 vdso_setup_syscall_map();
367
368 vdso_fixup_features();
369
370 if (IS_ENABLED(CONFIG_VDSO32))
371 vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
372
373 if (IS_ENABLED(CONFIG_PPC64))
374 vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
375
376 smp_wmb();
377
378 return 0;
379}
380arch_initcall(vdso_init);
1
2/*
3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
5 *
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.
10 */
11
12#include <linux/errno.h>
13#include <linux/sched.h>
14#include <linux/kernel.h>
15#include <linux/mm.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>
24
25#include <asm/cpu_has_feature.h>
26#include <asm/pgtable.h>
27#include <asm/processor.h>
28#include <asm/mmu.h>
29#include <asm/mmu_context.h>
30#include <asm/prom.h>
31#include <asm/machdep.h>
32#include <asm/cputable.h>
33#include <asm/sections.h>
34#include <asm/firmware.h>
35#include <asm/vdso.h>
36#include <asm/vdso_datapage.h>
37#include <asm/setup.h>
38
39#undef DEBUG
40
41#ifdef DEBUG
42#define DBG(fmt...) printk(fmt)
43#else
44#define DBG(fmt...)
45#endif
46
47/* Max supported size for symbol names */
48#define MAX_SYMNAME 64
49
50/* The alignment of the vDSO */
51#define VDSO_ALIGNMENT (1 << 16)
52
53static unsigned int vdso32_pages;
54static void *vdso32_kbase;
55static struct page **vdso32_pagelist;
56unsigned long vdso32_sigtramp;
57unsigned long vdso32_rt_sigtramp;
58
59#ifdef CONFIG_VDSO32
60extern char vdso32_start, vdso32_end;
61#endif
62
63#ifdef CONFIG_PPC64
64extern char vdso64_start, vdso64_end;
65static void *vdso64_kbase = &vdso64_start;
66static unsigned int vdso64_pages;
67static struct page **vdso64_pagelist;
68unsigned long vdso64_rt_sigtramp;
69#endif /* CONFIG_PPC64 */
70
71static int vdso_ready;
72
73/*
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
77 */
78static union {
79 struct vdso_data data;
80 u8 page[PAGE_SIZE];
81} vdso_data_store __page_aligned_data;
82struct vdso_data *vdso_data = &vdso_data_store.data;
83
84/* Format of the patch table */
85struct vdso_patch_def
86{
87 unsigned long ftr_mask, ftr_value;
88 const char *gen_name;
89 const char *fix_name;
90};
91
92/* Table of functions to patch based on the CPU type/revision
93 *
94 * Currently, we only change sync_dicache to do nothing on processors
95 * with a coherent icache
96 */
97static struct vdso_patch_def vdso_patches[] = {
98 {
99 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
100 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
101 },
102#ifdef CONFIG_PPC32
103 {
104 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
105 "__kernel_gettimeofday", NULL
106 },
107 {
108 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
109 "__kernel_clock_gettime", NULL
110 },
111 {
112 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
113 "__kernel_clock_getres", NULL
114 },
115 {
116 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
117 "__kernel_get_tbfreq", NULL
118 },
119 {
120 CPU_FTR_USE_RTC, CPU_FTR_USE_RTC,
121 "__kernel_time", NULL
122 },
123#endif
124};
125
126/*
127 * Some infos carried around for each of them during parsing at
128 * boot time.
129 */
130struct lib32_elfinfo
131{
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 */
137};
138
139struct lib64_elfinfo
140{
141 Elf64_Ehdr *hdr;
142 Elf64_Sym *dynsym;
143 unsigned long dynsymsize;
144 char *dynstr;
145 unsigned long text;
146};
147
148
149/*
150 * This is called from binfmt_elf, we create the special vma for the
151 * vDSO and insert it into the mm struct tree
152 */
153int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
154{
155 struct mm_struct *mm = current->mm;
156 struct page **vdso_pagelist;
157 unsigned long vdso_pages;
158 unsigned long vdso_base;
159 int rc;
160
161 if (!vdso_ready)
162 return 0;
163
164#ifdef CONFIG_PPC64
165 if (is_32bit_task()) {
166 vdso_pagelist = vdso32_pagelist;
167 vdso_pages = vdso32_pages;
168 vdso_base = VDSO32_MBASE;
169 } else {
170 vdso_pagelist = vdso64_pagelist;
171 vdso_pages = vdso64_pages;
172 /*
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.
176 */
177 vdso_base = 0;
178 }
179#else
180 vdso_pagelist = vdso32_pagelist;
181 vdso_pages = vdso32_pages;
182 vdso_base = VDSO32_MBASE;
183#endif
184
185 current->mm->context.vdso_base = 0;
186
187 /* vDSO has a problem and was disabled, just don't "enable" it for the
188 * process
189 */
190 if (vdso_pages == 0)
191 return 0;
192 /* Add a page to the vdso size for the data page */
193 vdso_pages ++;
194
195 /*
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.
200 */
201 if (down_write_killable(&mm->mmap_sem))
202 return -EINTR;
203 vdso_base = get_unmapped_area(NULL, vdso_base,
204 (vdso_pages << PAGE_SHIFT) +
205 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
206 0, 0);
207 if (IS_ERR_VALUE(vdso_base)) {
208 rc = vdso_base;
209 goto fail_mmapsem;
210 }
211
212 /* Add required alignment. */
213 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
214
215 /*
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).
219 */
220 current->mm->context.vdso_base = vdso_base;
221
222 /*
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
230 * pages though.
231 */
232 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
233 VM_READ|VM_EXEC|
234 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
235 vdso_pagelist);
236 if (rc) {
237 current->mm->context.vdso_base = 0;
238 goto fail_mmapsem;
239 }
240
241 up_write(&mm->mmap_sem);
242 return 0;
243
244 fail_mmapsem:
245 up_write(&mm->mmap_sem);
246 return rc;
247}
248
249const char *arch_vma_name(struct vm_area_struct *vma)
250{
251 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
252 return "[vdso]";
253 return NULL;
254}
255
256
257
258#ifdef CONFIG_VDSO32
259static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
260 unsigned long *size)
261{
262 Elf32_Shdr *sechdrs;
263 unsigned int i;
264 char *secnames;
265
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;
269
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) {
273 if (size)
274 *size = sechdrs[i].sh_size;
275 return (void *)ehdr + sechdrs[i].sh_offset;
276 }
277 }
278 *size = 0;
279 return NULL;
280}
281
282static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
283 const char *symname)
284{
285 unsigned int i;
286 char name[MAX_SYMNAME], *c;
287
288 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
289 if (lib->dynsym[i].st_name == 0)
290 continue;
291 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
292 MAX_SYMNAME);
293 c = strchr(name, '@');
294 if (c)
295 *c = 0;
296 if (strcmp(symname, name) == 0)
297 return &lib->dynsym[i];
298 }
299 return NULL;
300}
301
302/* Note that we assume the section is .text and the symbol is relative to
303 * the library base
304 */
305static unsigned long __init find_function32(struct lib32_elfinfo *lib,
306 const char *symname)
307{
308 Elf32_Sym *sym = find_symbol32(lib, symname);
309
310 if (sym == NULL) {
311 printk(KERN_WARNING "vDSO32: function %s not found !\n",
312 symname);
313 return 0;
314 }
315 return sym->st_value - VDSO32_LBASE;
316}
317
318static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
319 struct lib64_elfinfo *v64,
320 const char *orig, const char *fix)
321{
322 Elf32_Sym *sym32_gen, *sym32_fix;
323
324 sym32_gen = find_symbol32(v32, orig);
325 if (sym32_gen == NULL) {
326 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
327 return -1;
328 }
329 if (fix == NULL) {
330 sym32_gen->st_name = 0;
331 return 0;
332 }
333 sym32_fix = find_symbol32(v32, fix);
334 if (sym32_fix == NULL) {
335 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
336 return -1;
337 }
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;
343
344 return 0;
345}
346#else /* !CONFIG_VDSO32 */
347static unsigned long __init find_function32(struct lib32_elfinfo *lib,
348 const char *symname)
349{
350 return 0;
351}
352
353static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
354 struct lib64_elfinfo *v64,
355 const char *orig, const char *fix)
356{
357 return 0;
358}
359#endif /* CONFIG_VDSO32 */
360
361
362#ifdef CONFIG_PPC64
363
364static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
365 unsigned long *size)
366{
367 Elf64_Shdr *sechdrs;
368 unsigned int i;
369 char *secnames;
370
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;
374
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) {
378 if (size)
379 *size = sechdrs[i].sh_size;
380 return (void *)ehdr + sechdrs[i].sh_offset;
381 }
382 }
383 if (size)
384 *size = 0;
385 return NULL;
386}
387
388static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
389 const char *symname)
390{
391 unsigned int i;
392 char name[MAX_SYMNAME], *c;
393
394 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
395 if (lib->dynsym[i].st_name == 0)
396 continue;
397 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
398 MAX_SYMNAME);
399 c = strchr(name, '@');
400 if (c)
401 *c = 0;
402 if (strcmp(symname, name) == 0)
403 return &lib->dynsym[i];
404 }
405 return NULL;
406}
407
408/* Note that we assume the section is .text and the symbol is relative to
409 * the library base
410 */
411static unsigned long __init find_function64(struct lib64_elfinfo *lib,
412 const char *symname)
413{
414 Elf64_Sym *sym = find_symbol64(lib, symname);
415
416 if (sym == NULL) {
417 printk(KERN_WARNING "vDSO64: function %s not found !\n",
418 symname);
419 return 0;
420 }
421#ifdef VDS64_HAS_DESCRIPTORS
422 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
423 VDSO64_LBASE;
424#else
425 return sym->st_value - VDSO64_LBASE;
426#endif
427}
428
429static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
430 struct lib64_elfinfo *v64,
431 const char *orig, const char *fix)
432{
433 Elf64_Sym *sym64_gen, *sym64_fix;
434
435 sym64_gen = find_symbol64(v64, orig);
436 if (sym64_gen == NULL) {
437 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
438 return -1;
439 }
440 if (fix == NULL) {
441 sym64_gen->st_name = 0;
442 return 0;
443 }
444 sym64_fix = find_symbol64(v64, fix);
445 if (sym64_fix == NULL) {
446 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
447 return -1;
448 }
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;
454
455 return 0;
456}
457
458#endif /* CONFIG_PPC64 */
459
460
461static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
462 struct lib64_elfinfo *v64)
463{
464 void *sect;
465
466 /*
467 * Locate symbol tables & text section
468 */
469
470#ifdef CONFIG_VDSO32
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");
475 return -1;
476 }
477 sect = find_section32(v32->hdr, ".text", NULL);
478 if (sect == NULL) {
479 printk(KERN_ERR "vDSO32: the .text section was not found\n");
480 return -1;
481 }
482 v32->text = sect - vdso32_kbase;
483#endif
484
485#ifdef CONFIG_PPC64
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");
490 return -1;
491 }
492 sect = find_section64(v64->hdr, ".text", NULL);
493 if (sect == NULL) {
494 printk(KERN_ERR "vDSO64: the .text section was not found\n");
495 return -1;
496 }
497 v64->text = sect - vdso64_kbase;
498#endif /* CONFIG_PPC64 */
499
500 return 0;
501}
502
503static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
504 struct lib64_elfinfo *v64)
505{
506 /*
507 * Find signal trampolines
508 */
509
510#ifdef CONFIG_PPC64
511 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
512#endif
513 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
514 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
515}
516
517static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
518 struct lib64_elfinfo *v64)
519{
520#ifdef CONFIG_VDSO32
521 Elf32_Sym *sym32;
522#endif
523#ifdef CONFIG_PPC64
524 Elf64_Sym *sym64;
525
526 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
527 if (sym64 == NULL) {
528 printk(KERN_ERR "vDSO64: Can't find symbol "
529 "__kernel_datapage_offset !\n");
530 return -1;
531 }
532 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
533 (vdso64_pages << PAGE_SHIFT) -
534 (sym64->st_value - VDSO64_LBASE);
535#endif /* CONFIG_PPC64 */
536
537#ifdef CONFIG_VDSO32
538 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
539 if (sym32 == NULL) {
540 printk(KERN_ERR "vDSO32: Can't find symbol "
541 "__kernel_datapage_offset !\n");
542 return -1;
543 }
544 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
545 (vdso32_pages << PAGE_SHIFT) -
546 (sym32->st_value - VDSO32_LBASE);
547#endif
548
549 return 0;
550}
551
552
553static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
554 struct lib64_elfinfo *v64)
555{
556 unsigned long size;
557 void *start;
558
559#ifdef CONFIG_PPC64
560 start = find_section64(v64->hdr, "__ftr_fixup", &size);
561 if (start)
562 do_feature_fixups(cur_cpu_spec->cpu_features,
563 start, start + size);
564
565 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
566 if (start)
567 do_feature_fixups(cur_cpu_spec->mmu_features,
568 start, start + size);
569
570 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
571 if (start)
572 do_feature_fixups(powerpc_firmware_features,
573 start, start + size);
574
575 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
576 if (start)
577 do_lwsync_fixups(cur_cpu_spec->cpu_features,
578 start, start + size);
579#endif /* CONFIG_PPC64 */
580
581#ifdef CONFIG_VDSO32
582 start = find_section32(v32->hdr, "__ftr_fixup", &size);
583 if (start)
584 do_feature_fixups(cur_cpu_spec->cpu_features,
585 start, start + size);
586
587 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
588 if (start)
589 do_feature_fixups(cur_cpu_spec->mmu_features,
590 start, start + size);
591
592#ifdef CONFIG_PPC64
593 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
594 if (start)
595 do_feature_fixups(powerpc_firmware_features,
596 start, start + size);
597#endif /* CONFIG_PPC64 */
598
599 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
600 if (start)
601 do_lwsync_fixups(cur_cpu_spec->cpu_features,
602 start, start + size);
603#endif
604
605 return 0;
606}
607
608static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
609 struct lib64_elfinfo *v64)
610{
611 int i;
612
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)
616 == patch->ftr_value;
617 if (!match)
618 continue;
619
620 DBG("replacing %s with %s...\n", patch->gen_name,
621 patch->fix_name ? "NONE" : patch->fix_name);
622
623 /*
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.
628 */
629 vdso_do_func_patch32(v32, v64, patch->gen_name,
630 patch->fix_name);
631#ifdef CONFIG_PPC64
632 vdso_do_func_patch64(v32, v64, patch->gen_name,
633 patch->fix_name);
634#endif /* CONFIG_PPC64 */
635 }
636
637 return 0;
638}
639
640
641static __init int vdso_setup(void)
642{
643 struct lib32_elfinfo v32;
644 struct lib64_elfinfo v64;
645
646 v32.hdr = vdso32_kbase;
647#ifdef CONFIG_PPC64
648 v64.hdr = vdso64_kbase;
649#endif
650 if (vdso_do_find_sections(&v32, &v64))
651 return -1;
652
653 if (vdso_fixup_datapage(&v32, &v64))
654 return -1;
655
656 if (vdso_fixup_features(&v32, &v64))
657 return -1;
658
659 if (vdso_fixup_alt_funcs(&v32, &v64))
660 return -1;
661
662 vdso_setup_trampolines(&v32, &v64);
663
664 return 0;
665}
666
667/*
668 * Called from setup_arch to initialize the bitmap of available
669 * syscalls in the systemcfg page
670 */
671static void __init vdso_setup_syscall_map(void)
672{
673 unsigned int i;
674 extern unsigned long *sys_call_table;
675 extern unsigned long sys_ni_syscall;
676
677
678 for (i = 0; i < NR_syscalls; i++) {
679#ifdef CONFIG_PPC64
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 */
691 }
692}
693
694#ifdef CONFIG_PPC64
695int vdso_getcpu_init(void)
696{
697 unsigned long cpu, node, val;
698
699 /*
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().
702 */
703 cpu = get_cpu();
704 WARN_ON_ONCE(cpu > 0xffff);
705
706 node = cpu_to_node(cpu);
707 WARN_ON_ONCE(node > 0xffff);
708
709 val = (cpu & 0xfff) | ((node & 0xffff) << 16);
710 mtspr(SPRN_SPRG_VDSO_WRITE, val);
711 get_paca()->sprg_vdso = val;
712
713 put_cpu();
714
715 return 0;
716}
717/* We need to call this before SMP init */
718early_initcall(vdso_getcpu_init);
719#endif
720
721static int __init vdso_init(void)
722{
723 int i;
724
725#ifdef CONFIG_PPC64
726 /*
727 * Fill up the "systemcfg" stuff for backward compatibility
728 */
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);
733 /*
734 * Fake the old platform number for pSeries and add
735 * in LPAR bit if necessary
736 */
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;
749
750 /*
751 * Calculate the size of the 64 bits vDSO
752 */
753 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
754 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
755#else
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 */
761
762
763#ifdef CONFIG_VDSO32
764 vdso32_kbase = &vdso32_start;
765
766 /*
767 * Calculate the size of the 32 bits vDSO
768 */
769 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
770 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
771#endif
772
773
774 /*
775 * Setup the syscall map in the vDOS
776 */
777 vdso_setup_syscall_map();
778
779 /*
780 * Initialize the vDSO images in memory, that is do necessary
781 * fixups of vDSO symbols, locate trampolines, etc...
782 */
783 if (vdso_setup()) {
784 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
785 vdso32_pages = 0;
786#ifdef CONFIG_PPC64
787 vdso64_pages = 0;
788#endif
789 return 0;
790 }
791
792#ifdef CONFIG_VDSO32
793 /* Make sure pages are in the correct state */
794 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
795 GFP_KERNEL);
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);
800 get_page(pg);
801 vdso32_pagelist[i] = pg;
802 }
803 vdso32_pagelist[i++] = virt_to_page(vdso_data);
804 vdso32_pagelist[i] = NULL;
805#endif
806
807#ifdef CONFIG_PPC64
808 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
809 GFP_KERNEL);
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);
814 get_page(pg);
815 vdso64_pagelist[i] = pg;
816 }
817 vdso64_pagelist[i++] = virt_to_page(vdso_data);
818 vdso64_pagelist[i] = NULL;
819#endif /* CONFIG_PPC64 */
820
821 get_page(virt_to_page(vdso_data));
822
823 smp_wmb();
824 vdso_ready = 1;
825
826 return 0;
827}
828arch_initcall(vdso_init);