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 if (vma_is_special_mapping(vma, &vvar_spec))
124 zap_vma_pages(vma);
125 }
126 mmap_read_unlock(mm);
127
128 return 0;
129}
130#endif
131
132static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
133 struct vm_area_struct *vma, struct vm_fault *vmf)
134{
135 struct page *timens_page = find_timens_vvar_page(vma);
136 unsigned long pfn;
137
138 switch (vmf->pgoff) {
139 case VVAR_DATA_PAGE_OFFSET:
140 if (timens_page)
141 pfn = page_to_pfn(timens_page);
142 else
143 pfn = virt_to_pfn(vdso_data);
144 break;
145#ifdef CONFIG_TIME_NS
146 case VVAR_TIMENS_PAGE_OFFSET:
147 /*
148 * If a task belongs to a time namespace then a namespace
149 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
150 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
151 * offset.
152 * See also the comment near timens_setup_vdso_data().
153 */
154 if (!timens_page)
155 return VM_FAULT_SIGBUS;
156 pfn = virt_to_pfn(vdso_data);
157 break;
158#endif /* CONFIG_TIME_NS */
159 default:
160 return VM_FAULT_SIGBUS;
161 }
162
163 return vmf_insert_pfn(vma, vmf->address, pfn);
164}
165
166/*
167 * This is called from binfmt_elf, we create the special vma for the
168 * vDSO and insert it into the mm struct tree
169 */
170static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
171{
172 unsigned long vdso_size, vdso_base, mappings_size;
173 struct vm_special_mapping *vdso_spec;
174 unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE;
175 struct mm_struct *mm = current->mm;
176 struct vm_area_struct *vma;
177
178 if (is_32bit_task()) {
179 vdso_spec = &vdso32_spec;
180 vdso_size = &vdso32_end - &vdso32_start;
181 } else {
182 vdso_spec = &vdso64_spec;
183 vdso_size = &vdso64_end - &vdso64_start;
184 }
185
186 mappings_size = vdso_size + vvar_size;
187 mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
188
189 /*
190 * Pick a base address for the vDSO in process space.
191 * Add enough to the size so that the result can be aligned.
192 */
193 vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0);
194 if (IS_ERR_VALUE(vdso_base))
195 return vdso_base;
196
197 /* Add required alignment. */
198 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
199
200 /*
201 * Put vDSO base into mm struct. We need to do this before calling
202 * install_special_mapping or the perf counter mmap tracking code
203 * will fail to recognise it as a vDSO.
204 */
205 mm->context.vdso = (void __user *)vdso_base + vvar_size;
206
207 vma = _install_special_mapping(mm, vdso_base, vvar_size,
208 VM_READ | VM_MAYREAD | VM_IO |
209 VM_DONTDUMP | VM_PFNMAP, &vvar_spec);
210 if (IS_ERR(vma))
211 return PTR_ERR(vma);
212
213 /*
214 * our vma flags don't have VM_WRITE so by default, the process isn't
215 * allowed to write those pages.
216 * gdb can break that with ptrace interface, and thus trigger COW on
217 * those pages but it's then your responsibility to never do that on
218 * the "data" page of the vDSO or you'll stop getting kernel updates
219 * and your nice userland gettimeofday will be totally dead.
220 * It's fine to use that for setting breakpoints in the vDSO code
221 * pages though.
222 */
223 vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
224 VM_READ | VM_EXEC | VM_MAYREAD |
225 VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
226 if (IS_ERR(vma))
227 do_munmap(mm, vdso_base, vvar_size, NULL);
228
229 return PTR_ERR_OR_ZERO(vma);
230}
231
232int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
233{
234 struct mm_struct *mm = current->mm;
235 int rc;
236
237 mm->context.vdso = NULL;
238
239 if (mmap_write_lock_killable(mm))
240 return -EINTR;
241
242 rc = __arch_setup_additional_pages(bprm, uses_interp);
243 if (rc)
244 mm->context.vdso = NULL;
245
246 mmap_write_unlock(mm);
247 return rc;
248}
249
250#define VDSO_DO_FIXUPS(type, value, bits, sec) do { \
251 void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start); \
252 void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end); \
253 \
254 do_##type##_fixups((value), __start, __end); \
255} while (0)
256
257static void __init vdso_fixup_features(void)
258{
259#ifdef CONFIG_PPC64
260 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
261 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
262 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
263 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
264#endif /* CONFIG_PPC64 */
265
266#ifdef CONFIG_VDSO32
267 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
268 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
269#ifdef CONFIG_PPC64
270 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
271#endif /* CONFIG_PPC64 */
272 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
273#endif
274}
275
276/*
277 * Called from setup_arch to initialize the bitmap of available
278 * syscalls in the systemcfg page
279 */
280static void __init vdso_setup_syscall_map(void)
281{
282 unsigned int i;
283
284 for (i = 0; i < NR_syscalls; i++) {
285 if (sys_call_table[i] != (void *)&sys_ni_syscall)
286 vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
287 if (IS_ENABLED(CONFIG_COMPAT) &&
288 compat_sys_call_table[i] != (void *)&sys_ni_syscall)
289 vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
290 }
291}
292
293#ifdef CONFIG_PPC64
294int vdso_getcpu_init(void)
295{
296 unsigned long cpu, node, val;
297
298 /*
299 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
300 * in the next 16 bits. The VDSO uses this to implement getcpu().
301 */
302 cpu = get_cpu();
303 WARN_ON_ONCE(cpu > 0xffff);
304
305 node = cpu_to_node(cpu);
306 WARN_ON_ONCE(node > 0xffff);
307
308 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
309 mtspr(SPRN_SPRG_VDSO_WRITE, val);
310 get_paca()->sprg_vdso = val;
311
312 put_cpu();
313
314 return 0;
315}
316/* We need to call this before SMP init */
317early_initcall(vdso_getcpu_init);
318#endif
319
320static struct page ** __init vdso_setup_pages(void *start, void *end)
321{
322 int i;
323 struct page **pagelist;
324 int pages = (end - start) >> PAGE_SHIFT;
325
326 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
327 if (!pagelist)
328 panic("%s: Cannot allocate page list for VDSO", __func__);
329
330 for (i = 0; i < pages; i++)
331 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
332
333 return pagelist;
334}
335
336static int __init vdso_init(void)
337{
338#ifdef CONFIG_PPC64
339 /*
340 * Fill up the "systemcfg" stuff for backward compatibility
341 */
342 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
343 vdso_data->version.major = SYSTEMCFG_MAJOR;
344 vdso_data->version.minor = SYSTEMCFG_MINOR;
345 vdso_data->processor = mfspr(SPRN_PVR);
346 /*
347 * Fake the old platform number for pSeries and add
348 * in LPAR bit if necessary
349 */
350 vdso_data->platform = 0x100;
351 if (firmware_has_feature(FW_FEATURE_LPAR))
352 vdso_data->platform |= 1;
353 vdso_data->physicalMemorySize = memblock_phys_mem_size();
354 vdso_data->dcache_size = ppc64_caches.l1d.size;
355 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
356 vdso_data->icache_size = ppc64_caches.l1i.size;
357 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
358 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
359 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
360 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
361 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
362#endif /* CONFIG_PPC64 */
363
364 vdso_setup_syscall_map();
365
366 vdso_fixup_features();
367
368 if (IS_ENABLED(CONFIG_VDSO32))
369 vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
370
371 if (IS_ENABLED(CONFIG_PPC64))
372 vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
373
374 smp_wmb();
375
376 return 0;
377}
378arch_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/bootmem.h>
24#include <linux/memblock.h>
25
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
53extern char vdso32_start, vdso32_end;
54static void *vdso32_kbase = &vdso32_start;
55static unsigned int vdso32_pages;
56static struct page **vdso32_pagelist;
57unsigned long vdso32_sigtramp;
58unsigned long vdso32_rt_sigtramp;
59
60#ifdef CONFIG_PPC64
61extern char vdso64_start, vdso64_end;
62static void *vdso64_kbase = &vdso64_start;
63static unsigned int vdso64_pages;
64static struct page **vdso64_pagelist;
65unsigned long vdso64_rt_sigtramp;
66#endif /* CONFIG_PPC64 */
67
68static int vdso_ready;
69
70/*
71 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
72 * Once the early boot kernel code no longer needs to muck around
73 * with it, it will become dynamically allocated
74 */
75static union {
76 struct vdso_data data;
77 u8 page[PAGE_SIZE];
78} vdso_data_store __page_aligned_data;
79struct vdso_data *vdso_data = &vdso_data_store.data;
80
81/* Format of the patch table */
82struct vdso_patch_def
83{
84 unsigned long ftr_mask, ftr_value;
85 const char *gen_name;
86 const char *fix_name;
87};
88
89/* Table of functions to patch based on the CPU type/revision
90 *
91 * Currently, we only change sync_dicache to do nothing on processors
92 * with a coherent icache
93 */
94static struct vdso_patch_def vdso_patches[] = {
95 {
96 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
97 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
98 },
99 {
100 CPU_FTR_USE_TB, 0,
101 "__kernel_gettimeofday", NULL
102 },
103 {
104 CPU_FTR_USE_TB, 0,
105 "__kernel_clock_gettime", NULL
106 },
107 {
108 CPU_FTR_USE_TB, 0,
109 "__kernel_clock_getres", NULL
110 },
111 {
112 CPU_FTR_USE_TB, 0,
113 "__kernel_get_tbfreq", NULL
114 },
115 {
116 CPU_FTR_USE_TB, 0,
117 "__kernel_time", NULL
118 },
119};
120
121/*
122 * Some infos carried around for each of them during parsing at
123 * boot time.
124 */
125struct lib32_elfinfo
126{
127 Elf32_Ehdr *hdr; /* ptr to ELF */
128 Elf32_Sym *dynsym; /* ptr to .dynsym section */
129 unsigned long dynsymsize; /* size of .dynsym section */
130 char *dynstr; /* ptr to .dynstr section */
131 unsigned long text; /* offset of .text section in .so */
132};
133
134struct lib64_elfinfo
135{
136 Elf64_Ehdr *hdr;
137 Elf64_Sym *dynsym;
138 unsigned long dynsymsize;
139 char *dynstr;
140 unsigned long text;
141};
142
143
144#ifdef __DEBUG
145static void dump_one_vdso_page(struct page *pg, struct page *upg)
146{
147 printk("kpg: %p (c:%d,f:%08lx)", __va(page_to_pfn(pg) << PAGE_SHIFT),
148 page_count(pg),
149 pg->flags);
150 if (upg && !IS_ERR(upg) /* && pg != upg*/) {
151 printk(" upg: %p (c:%d,f:%08lx)", __va(page_to_pfn(upg)
152 << PAGE_SHIFT),
153 page_count(upg),
154 upg->flags);
155 }
156 printk("\n");
157}
158
159static void dump_vdso_pages(struct vm_area_struct * vma)
160{
161 int i;
162
163 if (!vma || is_32bit_task()) {
164 printk("vDSO32 @ %016lx:\n", (unsigned long)vdso32_kbase);
165 for (i=0; i<vdso32_pages; i++) {
166 struct page *pg = virt_to_page(vdso32_kbase +
167 i*PAGE_SIZE);
168 struct page *upg = (vma && vma->vm_mm) ?
169 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
170 : NULL;
171 dump_one_vdso_page(pg, upg);
172 }
173 }
174 if (!vma || !is_32bit_task()) {
175 printk("vDSO64 @ %016lx:\n", (unsigned long)vdso64_kbase);
176 for (i=0; i<vdso64_pages; i++) {
177 struct page *pg = virt_to_page(vdso64_kbase +
178 i*PAGE_SIZE);
179 struct page *upg = (vma && vma->vm_mm) ?
180 follow_page(vma, vma->vm_start + i*PAGE_SIZE, 0)
181 : NULL;
182 dump_one_vdso_page(pg, upg);
183 }
184 }
185}
186#endif /* DEBUG */
187
188/*
189 * This is called from binfmt_elf, we create the special vma for the
190 * vDSO and insert it into the mm struct tree
191 */
192int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
193{
194 struct mm_struct *mm = current->mm;
195 struct page **vdso_pagelist;
196 unsigned long vdso_pages;
197 unsigned long vdso_base;
198 int rc;
199
200 if (!vdso_ready)
201 return 0;
202
203#ifdef CONFIG_PPC64
204 if (is_32bit_task()) {
205 vdso_pagelist = vdso32_pagelist;
206 vdso_pages = vdso32_pages;
207 vdso_base = VDSO32_MBASE;
208 } else {
209 vdso_pagelist = vdso64_pagelist;
210 vdso_pages = vdso64_pages;
211 /*
212 * On 64bit we don't have a preferred map address. This
213 * allows get_unmapped_area to find an area near other mmaps
214 * and most likely share a SLB entry.
215 */
216 vdso_base = 0;
217 }
218#else
219 vdso_pagelist = vdso32_pagelist;
220 vdso_pages = vdso32_pages;
221 vdso_base = VDSO32_MBASE;
222#endif
223
224 current->mm->context.vdso_base = 0;
225
226 /* vDSO has a problem and was disabled, just don't "enable" it for the
227 * process
228 */
229 if (vdso_pages == 0)
230 return 0;
231 /* Add a page to the vdso size for the data page */
232 vdso_pages ++;
233
234 /*
235 * pick a base address for the vDSO in process space. We try to put it
236 * at vdso_base which is the "natural" base for it, but we might fail
237 * and end up putting it elsewhere.
238 * Add enough to the size so that the result can be aligned.
239 */
240 down_write(&mm->mmap_sem);
241 vdso_base = get_unmapped_area(NULL, vdso_base,
242 (vdso_pages << PAGE_SHIFT) +
243 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
244 0, 0);
245 if (IS_ERR_VALUE(vdso_base)) {
246 rc = vdso_base;
247 goto fail_mmapsem;
248 }
249
250 /* Add required alignment. */
251 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
252
253 /*
254 * Put vDSO base into mm struct. We need to do this before calling
255 * install_special_mapping or the perf counter mmap tracking code
256 * will fail to recognise it as a vDSO (since arch_vma_name fails).
257 */
258 current->mm->context.vdso_base = vdso_base;
259
260 /*
261 * our vma flags don't have VM_WRITE so by default, the process isn't
262 * allowed to write those pages.
263 * gdb can break that with ptrace interface, and thus trigger COW on
264 * those pages but it's then your responsibility to never do that on
265 * the "data" page of the vDSO or you'll stop getting kernel updates
266 * and your nice userland gettimeofday will be totally dead.
267 * It's fine to use that for setting breakpoints in the vDSO code
268 * pages though.
269 */
270 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
271 VM_READ|VM_EXEC|
272 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
273 vdso_pagelist);
274 if (rc) {
275 current->mm->context.vdso_base = 0;
276 goto fail_mmapsem;
277 }
278
279 up_write(&mm->mmap_sem);
280 return 0;
281
282 fail_mmapsem:
283 up_write(&mm->mmap_sem);
284 return rc;
285}
286
287const char *arch_vma_name(struct vm_area_struct *vma)
288{
289 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
290 return "[vdso]";
291 return NULL;
292}
293
294
295
296static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
297 unsigned long *size)
298{
299 Elf32_Shdr *sechdrs;
300 unsigned int i;
301 char *secnames;
302
303 /* Grab section headers and strings so we can tell who is who */
304 sechdrs = (void *)ehdr + ehdr->e_shoff;
305 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
306
307 /* Find the section they want */
308 for (i = 1; i < ehdr->e_shnum; i++) {
309 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
310 if (size)
311 *size = sechdrs[i].sh_size;
312 return (void *)ehdr + sechdrs[i].sh_offset;
313 }
314 }
315 *size = 0;
316 return NULL;
317}
318
319static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
320 const char *symname)
321{
322 unsigned int i;
323 char name[MAX_SYMNAME], *c;
324
325 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
326 if (lib->dynsym[i].st_name == 0)
327 continue;
328 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
329 MAX_SYMNAME);
330 c = strchr(name, '@');
331 if (c)
332 *c = 0;
333 if (strcmp(symname, name) == 0)
334 return &lib->dynsym[i];
335 }
336 return NULL;
337}
338
339/* Note that we assume the section is .text and the symbol is relative to
340 * the library base
341 */
342static unsigned long __init find_function32(struct lib32_elfinfo *lib,
343 const char *symname)
344{
345 Elf32_Sym *sym = find_symbol32(lib, symname);
346
347 if (sym == NULL) {
348 printk(KERN_WARNING "vDSO32: function %s not found !\n",
349 symname);
350 return 0;
351 }
352 return sym->st_value - VDSO32_LBASE;
353}
354
355static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
356 struct lib64_elfinfo *v64,
357 const char *orig, const char *fix)
358{
359 Elf32_Sym *sym32_gen, *sym32_fix;
360
361 sym32_gen = find_symbol32(v32, orig);
362 if (sym32_gen == NULL) {
363 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
364 return -1;
365 }
366 if (fix == NULL) {
367 sym32_gen->st_name = 0;
368 return 0;
369 }
370 sym32_fix = find_symbol32(v32, fix);
371 if (sym32_fix == NULL) {
372 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
373 return -1;
374 }
375 sym32_gen->st_value = sym32_fix->st_value;
376 sym32_gen->st_size = sym32_fix->st_size;
377 sym32_gen->st_info = sym32_fix->st_info;
378 sym32_gen->st_other = sym32_fix->st_other;
379 sym32_gen->st_shndx = sym32_fix->st_shndx;
380
381 return 0;
382}
383
384
385#ifdef CONFIG_PPC64
386
387static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
388 unsigned long *size)
389{
390 Elf64_Shdr *sechdrs;
391 unsigned int i;
392 char *secnames;
393
394 /* Grab section headers and strings so we can tell who is who */
395 sechdrs = (void *)ehdr + ehdr->e_shoff;
396 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
397
398 /* Find the section they want */
399 for (i = 1; i < ehdr->e_shnum; i++) {
400 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
401 if (size)
402 *size = sechdrs[i].sh_size;
403 return (void *)ehdr + sechdrs[i].sh_offset;
404 }
405 }
406 if (size)
407 *size = 0;
408 return NULL;
409}
410
411static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
412 const char *symname)
413{
414 unsigned int i;
415 char name[MAX_SYMNAME], *c;
416
417 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
418 if (lib->dynsym[i].st_name == 0)
419 continue;
420 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
421 MAX_SYMNAME);
422 c = strchr(name, '@');
423 if (c)
424 *c = 0;
425 if (strcmp(symname, name) == 0)
426 return &lib->dynsym[i];
427 }
428 return NULL;
429}
430
431/* Note that we assume the section is .text and the symbol is relative to
432 * the library base
433 */
434static unsigned long __init find_function64(struct lib64_elfinfo *lib,
435 const char *symname)
436{
437 Elf64_Sym *sym = find_symbol64(lib, symname);
438
439 if (sym == NULL) {
440 printk(KERN_WARNING "vDSO64: function %s not found !\n",
441 symname);
442 return 0;
443 }
444#ifdef VDS64_HAS_DESCRIPTORS
445 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
446 VDSO64_LBASE;
447#else
448 return sym->st_value - VDSO64_LBASE;
449#endif
450}
451
452static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
453 struct lib64_elfinfo *v64,
454 const char *orig, const char *fix)
455{
456 Elf64_Sym *sym64_gen, *sym64_fix;
457
458 sym64_gen = find_symbol64(v64, orig);
459 if (sym64_gen == NULL) {
460 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
461 return -1;
462 }
463 if (fix == NULL) {
464 sym64_gen->st_name = 0;
465 return 0;
466 }
467 sym64_fix = find_symbol64(v64, fix);
468 if (sym64_fix == NULL) {
469 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
470 return -1;
471 }
472 sym64_gen->st_value = sym64_fix->st_value;
473 sym64_gen->st_size = sym64_fix->st_size;
474 sym64_gen->st_info = sym64_fix->st_info;
475 sym64_gen->st_other = sym64_fix->st_other;
476 sym64_gen->st_shndx = sym64_fix->st_shndx;
477
478 return 0;
479}
480
481#endif /* CONFIG_PPC64 */
482
483
484static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
485 struct lib64_elfinfo *v64)
486{
487 void *sect;
488
489 /*
490 * Locate symbol tables & text section
491 */
492
493 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
494 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
495 if (v32->dynsym == NULL || v32->dynstr == NULL) {
496 printk(KERN_ERR "vDSO32: required symbol section not found\n");
497 return -1;
498 }
499 sect = find_section32(v32->hdr, ".text", NULL);
500 if (sect == NULL) {
501 printk(KERN_ERR "vDSO32: the .text section was not found\n");
502 return -1;
503 }
504 v32->text = sect - vdso32_kbase;
505
506#ifdef CONFIG_PPC64
507 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
508 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
509 if (v64->dynsym == NULL || v64->dynstr == NULL) {
510 printk(KERN_ERR "vDSO64: required symbol section not found\n");
511 return -1;
512 }
513 sect = find_section64(v64->hdr, ".text", NULL);
514 if (sect == NULL) {
515 printk(KERN_ERR "vDSO64: the .text section was not found\n");
516 return -1;
517 }
518 v64->text = sect - vdso64_kbase;
519#endif /* CONFIG_PPC64 */
520
521 return 0;
522}
523
524static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
525 struct lib64_elfinfo *v64)
526{
527 /*
528 * Find signal trampolines
529 */
530
531#ifdef CONFIG_PPC64
532 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
533#endif
534 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
535 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
536}
537
538static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
539 struct lib64_elfinfo *v64)
540{
541 Elf32_Sym *sym32;
542#ifdef CONFIG_PPC64
543 Elf64_Sym *sym64;
544
545 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
546 if (sym64 == NULL) {
547 printk(KERN_ERR "vDSO64: Can't find symbol "
548 "__kernel_datapage_offset !\n");
549 return -1;
550 }
551 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
552 (vdso64_pages << PAGE_SHIFT) -
553 (sym64->st_value - VDSO64_LBASE);
554#endif /* CONFIG_PPC64 */
555
556 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
557 if (sym32 == NULL) {
558 printk(KERN_ERR "vDSO32: Can't find symbol "
559 "__kernel_datapage_offset !\n");
560 return -1;
561 }
562 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
563 (vdso32_pages << PAGE_SHIFT) -
564 (sym32->st_value - VDSO32_LBASE);
565
566 return 0;
567}
568
569
570static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
571 struct lib64_elfinfo *v64)
572{
573 void *start32;
574 unsigned long size32;
575
576#ifdef CONFIG_PPC64
577 void *start64;
578 unsigned long size64;
579
580 start64 = find_section64(v64->hdr, "__ftr_fixup", &size64);
581 if (start64)
582 do_feature_fixups(cur_cpu_spec->cpu_features,
583 start64, start64 + size64);
584
585 start64 = find_section64(v64->hdr, "__mmu_ftr_fixup", &size64);
586 if (start64)
587 do_feature_fixups(cur_cpu_spec->mmu_features,
588 start64, start64 + size64);
589
590 start64 = find_section64(v64->hdr, "__fw_ftr_fixup", &size64);
591 if (start64)
592 do_feature_fixups(powerpc_firmware_features,
593 start64, start64 + size64);
594
595 start64 = find_section64(v64->hdr, "__lwsync_fixup", &size64);
596 if (start64)
597 do_lwsync_fixups(cur_cpu_spec->cpu_features,
598 start64, start64 + size64);
599#endif /* CONFIG_PPC64 */
600
601 start32 = find_section32(v32->hdr, "__ftr_fixup", &size32);
602 if (start32)
603 do_feature_fixups(cur_cpu_spec->cpu_features,
604 start32, start32 + size32);
605
606 start32 = find_section32(v32->hdr, "__mmu_ftr_fixup", &size32);
607 if (start32)
608 do_feature_fixups(cur_cpu_spec->mmu_features,
609 start32, start32 + size32);
610
611#ifdef CONFIG_PPC64
612 start32 = find_section32(v32->hdr, "__fw_ftr_fixup", &size32);
613 if (start32)
614 do_feature_fixups(powerpc_firmware_features,
615 start32, start32 + size32);
616#endif /* CONFIG_PPC64 */
617
618 start32 = find_section32(v32->hdr, "__lwsync_fixup", &size32);
619 if (start32)
620 do_lwsync_fixups(cur_cpu_spec->cpu_features,
621 start32, start32 + size32);
622
623 return 0;
624}
625
626static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
627 struct lib64_elfinfo *v64)
628{
629 int i;
630
631 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
632 struct vdso_patch_def *patch = &vdso_patches[i];
633 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
634 == patch->ftr_value;
635 if (!match)
636 continue;
637
638 DBG("replacing %s with %s...\n", patch->gen_name,
639 patch->fix_name ? "NONE" : patch->fix_name);
640
641 /*
642 * Patch the 32 bits and 64 bits symbols. Note that we do not
643 * patch the "." symbol on 64 bits.
644 * It would be easy to do, but doesn't seem to be necessary,
645 * patching the OPD symbol is enough.
646 */
647 vdso_do_func_patch32(v32, v64, patch->gen_name,
648 patch->fix_name);
649#ifdef CONFIG_PPC64
650 vdso_do_func_patch64(v32, v64, patch->gen_name,
651 patch->fix_name);
652#endif /* CONFIG_PPC64 */
653 }
654
655 return 0;
656}
657
658
659static __init int vdso_setup(void)
660{
661 struct lib32_elfinfo v32;
662 struct lib64_elfinfo v64;
663
664 v32.hdr = vdso32_kbase;
665#ifdef CONFIG_PPC64
666 v64.hdr = vdso64_kbase;
667#endif
668 if (vdso_do_find_sections(&v32, &v64))
669 return -1;
670
671 if (vdso_fixup_datapage(&v32, &v64))
672 return -1;
673
674 if (vdso_fixup_features(&v32, &v64))
675 return -1;
676
677 if (vdso_fixup_alt_funcs(&v32, &v64))
678 return -1;
679
680 vdso_setup_trampolines(&v32, &v64);
681
682 return 0;
683}
684
685/*
686 * Called from setup_arch to initialize the bitmap of available
687 * syscalls in the systemcfg page
688 */
689static void __init vdso_setup_syscall_map(void)
690{
691 unsigned int i;
692 extern unsigned long *sys_call_table;
693 extern unsigned long sys_ni_syscall;
694
695
696 for (i = 0; i < __NR_syscalls; i++) {
697#ifdef CONFIG_PPC64
698 if (sys_call_table[i*2] != sys_ni_syscall)
699 vdso_data->syscall_map_64[i >> 5] |=
700 0x80000000UL >> (i & 0x1f);
701 if (sys_call_table[i*2+1] != sys_ni_syscall)
702 vdso_data->syscall_map_32[i >> 5] |=
703 0x80000000UL >> (i & 0x1f);
704#else /* CONFIG_PPC64 */
705 if (sys_call_table[i] != sys_ni_syscall)
706 vdso_data->syscall_map_32[i >> 5] |=
707 0x80000000UL >> (i & 0x1f);
708#endif /* CONFIG_PPC64 */
709 }
710}
711
712#ifdef CONFIG_PPC64
713int vdso_getcpu_init(void)
714{
715 unsigned long cpu, node, val;
716
717 /*
718 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
719 * in the next 16 bits. The VDSO uses this to implement getcpu().
720 */
721 cpu = get_cpu();
722 WARN_ON_ONCE(cpu > 0xffff);
723
724 node = cpu_to_node(cpu);
725 WARN_ON_ONCE(node > 0xffff);
726
727 val = (cpu & 0xfff) | ((node & 0xffff) << 16);
728 mtspr(SPRN_SPRG_VDSO_WRITE, val);
729 get_paca()->sprg_vdso = val;
730
731 put_cpu();
732
733 return 0;
734}
735/* We need to call this before SMP init */
736early_initcall(vdso_getcpu_init);
737#endif
738
739static int __init vdso_init(void)
740{
741 int i;
742
743#ifdef CONFIG_PPC64
744 /*
745 * Fill up the "systemcfg" stuff for backward compatibility
746 */
747 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
748 vdso_data->version.major = SYSTEMCFG_MAJOR;
749 vdso_data->version.minor = SYSTEMCFG_MINOR;
750 vdso_data->processor = mfspr(SPRN_PVR);
751 /*
752 * Fake the old platform number for pSeries and add
753 * in LPAR bit if necessary
754 */
755 vdso_data->platform = 0x100;
756 if (firmware_has_feature(FW_FEATURE_LPAR))
757 vdso_data->platform |= 1;
758 vdso_data->physicalMemorySize = memblock_phys_mem_size();
759 vdso_data->dcache_size = ppc64_caches.dsize;
760 vdso_data->dcache_line_size = ppc64_caches.dline_size;
761 vdso_data->icache_size = ppc64_caches.isize;
762 vdso_data->icache_line_size = ppc64_caches.iline_size;
763
764 /* XXXOJN: Blocks should be added to ppc64_caches and used instead */
765 vdso_data->dcache_block_size = ppc64_caches.dline_size;
766 vdso_data->icache_block_size = ppc64_caches.iline_size;
767 vdso_data->dcache_log_block_size = ppc64_caches.log_dline_size;
768 vdso_data->icache_log_block_size = ppc64_caches.log_iline_size;
769
770 /*
771 * Calculate the size of the 64 bits vDSO
772 */
773 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
774 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
775#else
776 vdso_data->dcache_block_size = L1_CACHE_BYTES;
777 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
778 vdso_data->icache_block_size = L1_CACHE_BYTES;
779 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
780#endif /* CONFIG_PPC64 */
781
782
783 /*
784 * Calculate the size of the 32 bits vDSO
785 */
786 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
787 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
788
789
790 /*
791 * Setup the syscall map in the vDOS
792 */
793 vdso_setup_syscall_map();
794
795 /*
796 * Initialize the vDSO images in memory, that is do necessary
797 * fixups of vDSO symbols, locate trampolines, etc...
798 */
799 if (vdso_setup()) {
800 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
801 vdso32_pages = 0;
802#ifdef CONFIG_PPC64
803 vdso64_pages = 0;
804#endif
805 return 0;
806 }
807
808 /* Make sure pages are in the correct state */
809 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 2),
810 GFP_KERNEL);
811 BUG_ON(vdso32_pagelist == NULL);
812 for (i = 0; i < vdso32_pages; i++) {
813 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
814 ClearPageReserved(pg);
815 get_page(pg);
816 vdso32_pagelist[i] = pg;
817 }
818 vdso32_pagelist[i++] = virt_to_page(vdso_data);
819 vdso32_pagelist[i] = NULL;
820
821#ifdef CONFIG_PPC64
822 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 2),
823 GFP_KERNEL);
824 BUG_ON(vdso64_pagelist == NULL);
825 for (i = 0; i < vdso64_pages; i++) {
826 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
827 ClearPageReserved(pg);
828 get_page(pg);
829 vdso64_pagelist[i] = pg;
830 }
831 vdso64_pagelist[i++] = virt_to_page(vdso_data);
832 vdso64_pagelist[i] = NULL;
833#endif /* CONFIG_PPC64 */
834
835 get_page(virt_to_page(vdso_data));
836
837 smp_wmb();
838 vdso_ready = 1;
839
840 return 0;
841}
842arch_initcall(vdso_init);
843
844int in_gate_area_no_mm(unsigned long addr)
845{
846 return 0;
847}
848
849int in_gate_area(struct mm_struct *mm, unsigned long addr)
850{
851 return 0;
852}
853
854struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
855{
856 return NULL;
857}
858