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/prom.h>
29#include <asm/machdep.h>
30#include <asm/cputable.h>
31#include <asm/sections.h>
32#include <asm/firmware.h>
33#include <asm/vdso.h>
34#include <asm/vdso_datapage.h>
35#include <asm/setup.h>
36
37/* The alignment of the vDSO */
38#define VDSO_ALIGNMENT (1 << 16)
39
40extern char vdso32_start, vdso32_end;
41extern char vdso64_start, vdso64_end;
42
43/*
44 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
45 * Once the early boot kernel code no longer needs to muck around
46 * with it, it will become dynamically allocated
47 */
48static union {
49 struct vdso_arch_data data;
50 u8 page[PAGE_SIZE];
51} vdso_data_store __page_aligned_data;
52struct vdso_arch_data *vdso_data = &vdso_data_store.data;
53
54enum vvar_pages {
55 VVAR_DATA_PAGE_OFFSET,
56 VVAR_TIMENS_PAGE_OFFSET,
57 VVAR_NR_PAGES,
58};
59
60static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma,
61 unsigned long text_size)
62{
63 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
64
65 if (new_size != text_size)
66 return -EINVAL;
67
68 current->mm->context.vdso = (void __user *)new_vma->vm_start;
69
70 return 0;
71}
72
73static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
74{
75 return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start);
76}
77
78static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
79{
80 return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start);
81}
82
83static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
84 struct vm_area_struct *vma, struct vm_fault *vmf);
85
86static struct vm_special_mapping vvar_spec __ro_after_init = {
87 .name = "[vvar]",
88 .fault = vvar_fault,
89};
90
91static struct vm_special_mapping vdso32_spec __ro_after_init = {
92 .name = "[vdso]",
93 .mremap = vdso32_mremap,
94};
95
96static struct vm_special_mapping vdso64_spec __ro_after_init = {
97 .name = "[vdso]",
98 .mremap = vdso64_mremap,
99};
100
101#ifdef CONFIG_TIME_NS
102struct vdso_data *arch_get_vdso_data(void *vvar_page)
103{
104 return ((struct vdso_arch_data *)vvar_page)->data;
105}
106
107/*
108 * The vvar mapping contains data for a specific time namespace, so when a task
109 * changes namespace we must unmap its vvar data for the old namespace.
110 * Subsequent faults will map in data for the new namespace.
111 *
112 * For more details see timens_setup_vdso_data().
113 */
114int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
115{
116 struct mm_struct *mm = task->mm;
117 struct vm_area_struct *vma;
118
119 mmap_read_lock(mm);
120
121 for (vma = mm->mmap; vma; vma = vma->vm_next) {
122 unsigned long size = vma->vm_end - vma->vm_start;
123
124 if (vma_is_special_mapping(vma, &vvar_spec))
125 zap_page_range(vma, vma->vm_start, size);
126 }
127
128 mmap_read_unlock(mm);
129 return 0;
130}
131
132static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
133{
134 if (likely(vma->vm_mm == current->mm))
135 return current->nsproxy->time_ns->vvar_page;
136
137 /*
138 * VM_PFNMAP | VM_IO protect .fault() handler from being called
139 * through interfaces like /proc/$pid/mem or
140 * process_vm_{readv,writev}() as long as there's no .access()
141 * in special_mapping_vmops.
142 * For more details check_vma_flags() and __access_remote_vm()
143 */
144 WARN(1, "vvar_page accessed remotely");
145
146 return NULL;
147}
148#else
149static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
150{
151 return NULL;
152}
153#endif
154
155static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
156 struct vm_area_struct *vma, struct vm_fault *vmf)
157{
158 struct page *timens_page = find_timens_vvar_page(vma);
159 unsigned long pfn;
160
161 switch (vmf->pgoff) {
162 case VVAR_DATA_PAGE_OFFSET:
163 if (timens_page)
164 pfn = page_to_pfn(timens_page);
165 else
166 pfn = virt_to_pfn(vdso_data);
167 break;
168#ifdef CONFIG_TIME_NS
169 case VVAR_TIMENS_PAGE_OFFSET:
170 /*
171 * If a task belongs to a time namespace then a namespace
172 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
173 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
174 * offset.
175 * See also the comment near timens_setup_vdso_data().
176 */
177 if (!timens_page)
178 return VM_FAULT_SIGBUS;
179 pfn = virt_to_pfn(vdso_data);
180 break;
181#endif /* CONFIG_TIME_NS */
182 default:
183 return VM_FAULT_SIGBUS;
184 }
185
186 return vmf_insert_pfn(vma, vmf->address, pfn);
187}
188
189/*
190 * This is called from binfmt_elf, we create the special vma for the
191 * vDSO and insert it into the mm struct tree
192 */
193static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
194{
195 unsigned long vdso_size, vdso_base, mappings_size;
196 struct vm_special_mapping *vdso_spec;
197 unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE;
198 struct mm_struct *mm = current->mm;
199 struct vm_area_struct *vma;
200
201 if (is_32bit_task()) {
202 vdso_spec = &vdso32_spec;
203 vdso_size = &vdso32_end - &vdso32_start;
204 vdso_base = VDSO32_MBASE;
205 } else {
206 vdso_spec = &vdso64_spec;
207 vdso_size = &vdso64_end - &vdso64_start;
208 /*
209 * On 64bit we don't have a preferred map address. This
210 * allows get_unmapped_area to find an area near other mmaps
211 * and most likely share a SLB entry.
212 */
213 vdso_base = 0;
214 }
215
216 mappings_size = vdso_size + vvar_size;
217 mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK;
218
219 /*
220 * pick a base address for the vDSO in process space. We try to put it
221 * at vdso_base which is the "natural" base for it, but we might fail
222 * and end up putting it elsewhere.
223 * Add enough to the size so that the result can be aligned.
224 */
225 vdso_base = get_unmapped_area(NULL, vdso_base, mappings_size, 0, 0);
226 if (IS_ERR_VALUE(vdso_base))
227 return vdso_base;
228
229 /* Add required alignment. */
230 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
231
232 /*
233 * Put vDSO base into mm struct. We need to do this before calling
234 * install_special_mapping or the perf counter mmap tracking code
235 * will fail to recognise it as a vDSO.
236 */
237 mm->context.vdso = (void __user *)vdso_base + vvar_size;
238
239 vma = _install_special_mapping(mm, vdso_base, vvar_size,
240 VM_READ | VM_MAYREAD | VM_IO |
241 VM_DONTDUMP | VM_PFNMAP, &vvar_spec);
242 if (IS_ERR(vma))
243 return PTR_ERR(vma);
244
245 /*
246 * our vma flags don't have VM_WRITE so by default, the process isn't
247 * allowed to write those pages.
248 * gdb can break that with ptrace interface, and thus trigger COW on
249 * those pages but it's then your responsibility to never do that on
250 * the "data" page of the vDSO or you'll stop getting kernel updates
251 * and your nice userland gettimeofday will be totally dead.
252 * It's fine to use that for setting breakpoints in the vDSO code
253 * pages though.
254 */
255 vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size,
256 VM_READ | VM_EXEC | VM_MAYREAD |
257 VM_MAYWRITE | VM_MAYEXEC, vdso_spec);
258 if (IS_ERR(vma))
259 do_munmap(mm, vdso_base, vvar_size, NULL);
260
261 return PTR_ERR_OR_ZERO(vma);
262}
263
264int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
265{
266 struct mm_struct *mm = current->mm;
267 int rc;
268
269 mm->context.vdso = NULL;
270
271 if (mmap_write_lock_killable(mm))
272 return -EINTR;
273
274 rc = __arch_setup_additional_pages(bprm, uses_interp);
275 if (rc)
276 mm->context.vdso = NULL;
277
278 mmap_write_unlock(mm);
279 return rc;
280}
281
282#define VDSO_DO_FIXUPS(type, value, bits, sec) do { \
283 void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start); \
284 void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end); \
285 \
286 do_##type##_fixups((value), __start, __end); \
287} while (0)
288
289static void __init vdso_fixup_features(void)
290{
291#ifdef CONFIG_PPC64
292 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup);
293 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup);
294 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup);
295 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup);
296#endif /* CONFIG_PPC64 */
297
298#ifdef CONFIG_VDSO32
299 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup);
300 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup);
301#ifdef CONFIG_PPC64
302 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup);
303#endif /* CONFIG_PPC64 */
304 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup);
305#endif
306}
307
308/*
309 * Called from setup_arch to initialize the bitmap of available
310 * syscalls in the systemcfg page
311 */
312static void __init vdso_setup_syscall_map(void)
313{
314 unsigned int i;
315
316 for (i = 0; i < NR_syscalls; i++) {
317 if (sys_call_table[i] != (unsigned long)&sys_ni_syscall)
318 vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
319 if (IS_ENABLED(CONFIG_COMPAT) &&
320 compat_sys_call_table[i] != (unsigned long)&sys_ni_syscall)
321 vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f);
322 }
323}
324
325#ifdef CONFIG_PPC64
326int vdso_getcpu_init(void)
327{
328 unsigned long cpu, node, val;
329
330 /*
331 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
332 * in the next 16 bits. The VDSO uses this to implement getcpu().
333 */
334 cpu = get_cpu();
335 WARN_ON_ONCE(cpu > 0xffff);
336
337 node = cpu_to_node(cpu);
338 WARN_ON_ONCE(node > 0xffff);
339
340 val = (cpu & 0xffff) | ((node & 0xffff) << 16);
341 mtspr(SPRN_SPRG_VDSO_WRITE, val);
342 get_paca()->sprg_vdso = val;
343
344 put_cpu();
345
346 return 0;
347}
348/* We need to call this before SMP init */
349early_initcall(vdso_getcpu_init);
350#endif
351
352static struct page ** __init vdso_setup_pages(void *start, void *end)
353{
354 int i;
355 struct page **pagelist;
356 int pages = (end - start) >> PAGE_SHIFT;
357
358 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
359 if (!pagelist)
360 panic("%s: Cannot allocate page list for VDSO", __func__);
361
362 for (i = 0; i < pages; i++)
363 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
364
365 return pagelist;
366}
367
368static int __init vdso_init(void)
369{
370#ifdef CONFIG_PPC64
371 /*
372 * Fill up the "systemcfg" stuff for backward compatibility
373 */
374 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
375 vdso_data->version.major = SYSTEMCFG_MAJOR;
376 vdso_data->version.minor = SYSTEMCFG_MINOR;
377 vdso_data->processor = mfspr(SPRN_PVR);
378 /*
379 * Fake the old platform number for pSeries and add
380 * in LPAR bit if necessary
381 */
382 vdso_data->platform = 0x100;
383 if (firmware_has_feature(FW_FEATURE_LPAR))
384 vdso_data->platform |= 1;
385 vdso_data->physicalMemorySize = memblock_phys_mem_size();
386 vdso_data->dcache_size = ppc64_caches.l1d.size;
387 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
388 vdso_data->icache_size = ppc64_caches.l1i.size;
389 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
390 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
391 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
392 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
393 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
394#endif /* CONFIG_PPC64 */
395
396 vdso_setup_syscall_map();
397
398 vdso_fixup_features();
399
400 if (IS_ENABLED(CONFIG_VDSO32))
401 vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end);
402
403 if (IS_ENABLED(CONFIG_PPC64))
404 vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end);
405
406 smp_wmb();
407
408 return 0;
409}
410arch_initcall(vdso_init);
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
21#include <asm/pgtable.h>
22#include <asm/processor.h>
23#include <asm/mmu.h>
24#include <asm/mmu_context.h>
25#include <asm/prom.h>
26#include <asm/machdep.h>
27#include <asm/cputable.h>
28#include <asm/sections.h>
29#include <asm/firmware.h>
30#include <asm/vdso.h>
31#include <asm/vdso_datapage.h>
32#include <asm/setup.h>
33
34#undef DEBUG
35
36#ifdef DEBUG
37#define DBG(fmt...) printk(fmt)
38#else
39#define DBG(fmt...)
40#endif
41
42/* Max supported size for symbol names */
43#define MAX_SYMNAME 64
44
45/* The alignment of the vDSO */
46#define VDSO_ALIGNMENT (1 << 16)
47
48static unsigned int vdso32_pages;
49static void *vdso32_kbase;
50static struct page **vdso32_pagelist;
51unsigned long vdso32_sigtramp;
52unsigned long vdso32_rt_sigtramp;
53
54#ifdef CONFIG_VDSO32
55extern char vdso32_start, vdso32_end;
56#endif
57
58#ifdef CONFIG_PPC64
59extern char vdso64_start, vdso64_end;
60static void *vdso64_kbase = &vdso64_start;
61static unsigned int vdso64_pages;
62static struct page **vdso64_pagelist;
63unsigned long vdso64_rt_sigtramp;
64#endif /* CONFIG_PPC64 */
65
66static int vdso_ready;
67
68/*
69 * The vdso data page (aka. systemcfg for old ppc64 fans) is here.
70 * Once the early boot kernel code no longer needs to muck around
71 * with it, it will become dynamically allocated
72 */
73static union {
74 struct vdso_data data;
75 u8 page[PAGE_SIZE];
76} vdso_data_store __page_aligned_data;
77struct vdso_data *vdso_data = &vdso_data_store.data;
78
79/* Format of the patch table */
80struct vdso_patch_def
81{
82 unsigned long ftr_mask, ftr_value;
83 const char *gen_name;
84 const char *fix_name;
85};
86
87/* Table of functions to patch based on the CPU type/revision
88 *
89 * Currently, we only change sync_dicache to do nothing on processors
90 * with a coherent icache
91 */
92static struct vdso_patch_def vdso_patches[] = {
93 {
94 CPU_FTR_COHERENT_ICACHE, CPU_FTR_COHERENT_ICACHE,
95 "__kernel_sync_dicache", "__kernel_sync_dicache_p5"
96 },
97};
98
99/*
100 * Some infos carried around for each of them during parsing at
101 * boot time.
102 */
103struct lib32_elfinfo
104{
105 Elf32_Ehdr *hdr; /* ptr to ELF */
106 Elf32_Sym *dynsym; /* ptr to .dynsym section */
107 unsigned long dynsymsize; /* size of .dynsym section */
108 char *dynstr; /* ptr to .dynstr section */
109 unsigned long text; /* offset of .text section in .so */
110};
111
112struct lib64_elfinfo
113{
114 Elf64_Ehdr *hdr;
115 Elf64_Sym *dynsym;
116 unsigned long dynsymsize;
117 char *dynstr;
118 unsigned long text;
119};
120
121
122/*
123 * This is called from binfmt_elf, we create the special vma for the
124 * vDSO and insert it into the mm struct tree
125 */
126int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
127{
128 struct mm_struct *mm = current->mm;
129 struct page **vdso_pagelist;
130 unsigned long vdso_pages;
131 unsigned long vdso_base;
132 int rc;
133
134 if (!vdso_ready)
135 return 0;
136
137#ifdef CONFIG_PPC64
138 if (is_32bit_task()) {
139 vdso_pagelist = vdso32_pagelist;
140 vdso_pages = vdso32_pages;
141 vdso_base = VDSO32_MBASE;
142 } else {
143 vdso_pagelist = vdso64_pagelist;
144 vdso_pages = vdso64_pages;
145 /*
146 * On 64bit we don't have a preferred map address. This
147 * allows get_unmapped_area to find an area near other mmaps
148 * and most likely share a SLB entry.
149 */
150 vdso_base = 0;
151 }
152#else
153 vdso_pagelist = vdso32_pagelist;
154 vdso_pages = vdso32_pages;
155 vdso_base = VDSO32_MBASE;
156#endif
157
158 current->mm->context.vdso_base = 0;
159
160 /* vDSO has a problem and was disabled, just don't "enable" it for the
161 * process
162 */
163 if (vdso_pages == 0)
164 return 0;
165 /* Add a page to the vdso size for the data page */
166 vdso_pages ++;
167
168 /*
169 * pick a base address for the vDSO in process space. We try to put it
170 * at vdso_base which is the "natural" base for it, but we might fail
171 * and end up putting it elsewhere.
172 * Add enough to the size so that the result can be aligned.
173 */
174 if (down_write_killable(&mm->mmap_sem))
175 return -EINTR;
176 vdso_base = get_unmapped_area(NULL, vdso_base,
177 (vdso_pages << PAGE_SHIFT) +
178 ((VDSO_ALIGNMENT - 1) & PAGE_MASK),
179 0, 0);
180 if (IS_ERR_VALUE(vdso_base)) {
181 rc = vdso_base;
182 goto fail_mmapsem;
183 }
184
185 /* Add required alignment. */
186 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT);
187
188 /*
189 * Put vDSO base into mm struct. We need to do this before calling
190 * install_special_mapping or the perf counter mmap tracking code
191 * will fail to recognise it as a vDSO (since arch_vma_name fails).
192 */
193 current->mm->context.vdso_base = vdso_base;
194
195 /*
196 * our vma flags don't have VM_WRITE so by default, the process isn't
197 * allowed to write those pages.
198 * gdb can break that with ptrace interface, and thus trigger COW on
199 * those pages but it's then your responsibility to never do that on
200 * the "data" page of the vDSO or you'll stop getting kernel updates
201 * and your nice userland gettimeofday will be totally dead.
202 * It's fine to use that for setting breakpoints in the vDSO code
203 * pages though.
204 */
205 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
206 VM_READ|VM_EXEC|
207 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
208 vdso_pagelist);
209 if (rc) {
210 current->mm->context.vdso_base = 0;
211 goto fail_mmapsem;
212 }
213
214 up_write(&mm->mmap_sem);
215 return 0;
216
217 fail_mmapsem:
218 up_write(&mm->mmap_sem);
219 return rc;
220}
221
222const char *arch_vma_name(struct vm_area_struct *vma)
223{
224 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
225 return "[vdso]";
226 return NULL;
227}
228
229
230
231#ifdef CONFIG_VDSO32
232static void * __init find_section32(Elf32_Ehdr *ehdr, const char *secname,
233 unsigned long *size)
234{
235 Elf32_Shdr *sechdrs;
236 unsigned int i;
237 char *secnames;
238
239 /* Grab section headers and strings so we can tell who is who */
240 sechdrs = (void *)ehdr + ehdr->e_shoff;
241 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
242
243 /* Find the section they want */
244 for (i = 1; i < ehdr->e_shnum; i++) {
245 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
246 if (size)
247 *size = sechdrs[i].sh_size;
248 return (void *)ehdr + sechdrs[i].sh_offset;
249 }
250 }
251 *size = 0;
252 return NULL;
253}
254
255static Elf32_Sym * __init find_symbol32(struct lib32_elfinfo *lib,
256 const char *symname)
257{
258 unsigned int i;
259 char name[MAX_SYMNAME], *c;
260
261 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
262 if (lib->dynsym[i].st_name == 0)
263 continue;
264 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
265 MAX_SYMNAME);
266 c = strchr(name, '@');
267 if (c)
268 *c = 0;
269 if (strcmp(symname, name) == 0)
270 return &lib->dynsym[i];
271 }
272 return NULL;
273}
274
275/* Note that we assume the section is .text and the symbol is relative to
276 * the library base
277 */
278static unsigned long __init find_function32(struct lib32_elfinfo *lib,
279 const char *symname)
280{
281 Elf32_Sym *sym = find_symbol32(lib, symname);
282
283 if (sym == NULL) {
284 printk(KERN_WARNING "vDSO32: function %s not found !\n",
285 symname);
286 return 0;
287 }
288 return sym->st_value - VDSO32_LBASE;
289}
290
291static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
292 struct lib64_elfinfo *v64,
293 const char *orig, const char *fix)
294{
295 Elf32_Sym *sym32_gen, *sym32_fix;
296
297 sym32_gen = find_symbol32(v32, orig);
298 if (sym32_gen == NULL) {
299 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", orig);
300 return -1;
301 }
302 if (fix == NULL) {
303 sym32_gen->st_name = 0;
304 return 0;
305 }
306 sym32_fix = find_symbol32(v32, fix);
307 if (sym32_fix == NULL) {
308 printk(KERN_ERR "vDSO32: Can't find symbol %s !\n", fix);
309 return -1;
310 }
311 sym32_gen->st_value = sym32_fix->st_value;
312 sym32_gen->st_size = sym32_fix->st_size;
313 sym32_gen->st_info = sym32_fix->st_info;
314 sym32_gen->st_other = sym32_fix->st_other;
315 sym32_gen->st_shndx = sym32_fix->st_shndx;
316
317 return 0;
318}
319#else /* !CONFIG_VDSO32 */
320static unsigned long __init find_function32(struct lib32_elfinfo *lib,
321 const char *symname)
322{
323 return 0;
324}
325
326static int __init vdso_do_func_patch32(struct lib32_elfinfo *v32,
327 struct lib64_elfinfo *v64,
328 const char *orig, const char *fix)
329{
330 return 0;
331}
332#endif /* CONFIG_VDSO32 */
333
334
335#ifdef CONFIG_PPC64
336
337static void * __init find_section64(Elf64_Ehdr *ehdr, const char *secname,
338 unsigned long *size)
339{
340 Elf64_Shdr *sechdrs;
341 unsigned int i;
342 char *secnames;
343
344 /* Grab section headers and strings so we can tell who is who */
345 sechdrs = (void *)ehdr + ehdr->e_shoff;
346 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
347
348 /* Find the section they want */
349 for (i = 1; i < ehdr->e_shnum; i++) {
350 if (strcmp(secnames+sechdrs[i].sh_name, secname) == 0) {
351 if (size)
352 *size = sechdrs[i].sh_size;
353 return (void *)ehdr + sechdrs[i].sh_offset;
354 }
355 }
356 if (size)
357 *size = 0;
358 return NULL;
359}
360
361static Elf64_Sym * __init find_symbol64(struct lib64_elfinfo *lib,
362 const char *symname)
363{
364 unsigned int i;
365 char name[MAX_SYMNAME], *c;
366
367 for (i = 0; i < (lib->dynsymsize / sizeof(Elf64_Sym)); i++) {
368 if (lib->dynsym[i].st_name == 0)
369 continue;
370 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
371 MAX_SYMNAME);
372 c = strchr(name, '@');
373 if (c)
374 *c = 0;
375 if (strcmp(symname, name) == 0)
376 return &lib->dynsym[i];
377 }
378 return NULL;
379}
380
381/* Note that we assume the section is .text and the symbol is relative to
382 * the library base
383 */
384static unsigned long __init find_function64(struct lib64_elfinfo *lib,
385 const char *symname)
386{
387 Elf64_Sym *sym = find_symbol64(lib, symname);
388
389 if (sym == NULL) {
390 printk(KERN_WARNING "vDSO64: function %s not found !\n",
391 symname);
392 return 0;
393 }
394#ifdef VDS64_HAS_DESCRIPTORS
395 return *((u64 *)(vdso64_kbase + sym->st_value - VDSO64_LBASE)) -
396 VDSO64_LBASE;
397#else
398 return sym->st_value - VDSO64_LBASE;
399#endif
400}
401
402static int __init vdso_do_func_patch64(struct lib32_elfinfo *v32,
403 struct lib64_elfinfo *v64,
404 const char *orig, const char *fix)
405{
406 Elf64_Sym *sym64_gen, *sym64_fix;
407
408 sym64_gen = find_symbol64(v64, orig);
409 if (sym64_gen == NULL) {
410 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", orig);
411 return -1;
412 }
413 if (fix == NULL) {
414 sym64_gen->st_name = 0;
415 return 0;
416 }
417 sym64_fix = find_symbol64(v64, fix);
418 if (sym64_fix == NULL) {
419 printk(KERN_ERR "vDSO64: Can't find symbol %s !\n", fix);
420 return -1;
421 }
422 sym64_gen->st_value = sym64_fix->st_value;
423 sym64_gen->st_size = sym64_fix->st_size;
424 sym64_gen->st_info = sym64_fix->st_info;
425 sym64_gen->st_other = sym64_fix->st_other;
426 sym64_gen->st_shndx = sym64_fix->st_shndx;
427
428 return 0;
429}
430
431#endif /* CONFIG_PPC64 */
432
433
434static __init int vdso_do_find_sections(struct lib32_elfinfo *v32,
435 struct lib64_elfinfo *v64)
436{
437 void *sect;
438
439 /*
440 * Locate symbol tables & text section
441 */
442
443#ifdef CONFIG_VDSO32
444 v32->dynsym = find_section32(v32->hdr, ".dynsym", &v32->dynsymsize);
445 v32->dynstr = find_section32(v32->hdr, ".dynstr", NULL);
446 if (v32->dynsym == NULL || v32->dynstr == NULL) {
447 printk(KERN_ERR "vDSO32: required symbol section not found\n");
448 return -1;
449 }
450 sect = find_section32(v32->hdr, ".text", NULL);
451 if (sect == NULL) {
452 printk(KERN_ERR "vDSO32: the .text section was not found\n");
453 return -1;
454 }
455 v32->text = sect - vdso32_kbase;
456#endif
457
458#ifdef CONFIG_PPC64
459 v64->dynsym = find_section64(v64->hdr, ".dynsym", &v64->dynsymsize);
460 v64->dynstr = find_section64(v64->hdr, ".dynstr", NULL);
461 if (v64->dynsym == NULL || v64->dynstr == NULL) {
462 printk(KERN_ERR "vDSO64: required symbol section not found\n");
463 return -1;
464 }
465 sect = find_section64(v64->hdr, ".text", NULL);
466 if (sect == NULL) {
467 printk(KERN_ERR "vDSO64: the .text section was not found\n");
468 return -1;
469 }
470 v64->text = sect - vdso64_kbase;
471#endif /* CONFIG_PPC64 */
472
473 return 0;
474}
475
476static __init void vdso_setup_trampolines(struct lib32_elfinfo *v32,
477 struct lib64_elfinfo *v64)
478{
479 /*
480 * Find signal trampolines
481 */
482
483#ifdef CONFIG_PPC64
484 vdso64_rt_sigtramp = find_function64(v64, "__kernel_sigtramp_rt64");
485#endif
486 vdso32_sigtramp = find_function32(v32, "__kernel_sigtramp32");
487 vdso32_rt_sigtramp = find_function32(v32, "__kernel_sigtramp_rt32");
488}
489
490static __init int vdso_fixup_datapage(struct lib32_elfinfo *v32,
491 struct lib64_elfinfo *v64)
492{
493#ifdef CONFIG_VDSO32
494 Elf32_Sym *sym32;
495#endif
496#ifdef CONFIG_PPC64
497 Elf64_Sym *sym64;
498
499 sym64 = find_symbol64(v64, "__kernel_datapage_offset");
500 if (sym64 == NULL) {
501 printk(KERN_ERR "vDSO64: Can't find symbol "
502 "__kernel_datapage_offset !\n");
503 return -1;
504 }
505 *((int *)(vdso64_kbase + sym64->st_value - VDSO64_LBASE)) =
506 (vdso64_pages << PAGE_SHIFT) -
507 (sym64->st_value - VDSO64_LBASE);
508#endif /* CONFIG_PPC64 */
509
510#ifdef CONFIG_VDSO32
511 sym32 = find_symbol32(v32, "__kernel_datapage_offset");
512 if (sym32 == NULL) {
513 printk(KERN_ERR "vDSO32: Can't find symbol "
514 "__kernel_datapage_offset !\n");
515 return -1;
516 }
517 *((int *)(vdso32_kbase + (sym32->st_value - VDSO32_LBASE))) =
518 (vdso32_pages << PAGE_SHIFT) -
519 (sym32->st_value - VDSO32_LBASE);
520#endif
521
522 return 0;
523}
524
525
526static __init int vdso_fixup_features(struct lib32_elfinfo *v32,
527 struct lib64_elfinfo *v64)
528{
529 unsigned long size;
530 void *start;
531
532#ifdef CONFIG_PPC64
533 start = find_section64(v64->hdr, "__ftr_fixup", &size);
534 if (start)
535 do_feature_fixups(cur_cpu_spec->cpu_features,
536 start, start + size);
537
538 start = find_section64(v64->hdr, "__mmu_ftr_fixup", &size);
539 if (start)
540 do_feature_fixups(cur_cpu_spec->mmu_features,
541 start, start + size);
542
543 start = find_section64(v64->hdr, "__fw_ftr_fixup", &size);
544 if (start)
545 do_feature_fixups(powerpc_firmware_features,
546 start, start + size);
547
548 start = find_section64(v64->hdr, "__lwsync_fixup", &size);
549 if (start)
550 do_lwsync_fixups(cur_cpu_spec->cpu_features,
551 start, start + size);
552#endif /* CONFIG_PPC64 */
553
554#ifdef CONFIG_VDSO32
555 start = find_section32(v32->hdr, "__ftr_fixup", &size);
556 if (start)
557 do_feature_fixups(cur_cpu_spec->cpu_features,
558 start, start + size);
559
560 start = find_section32(v32->hdr, "__mmu_ftr_fixup", &size);
561 if (start)
562 do_feature_fixups(cur_cpu_spec->mmu_features,
563 start, start + size);
564
565#ifdef CONFIG_PPC64
566 start = find_section32(v32->hdr, "__fw_ftr_fixup", &size);
567 if (start)
568 do_feature_fixups(powerpc_firmware_features,
569 start, start + size);
570#endif /* CONFIG_PPC64 */
571
572 start = find_section32(v32->hdr, "__lwsync_fixup", &size);
573 if (start)
574 do_lwsync_fixups(cur_cpu_spec->cpu_features,
575 start, start + size);
576#endif
577
578 return 0;
579}
580
581static __init int vdso_fixup_alt_funcs(struct lib32_elfinfo *v32,
582 struct lib64_elfinfo *v64)
583{
584 int i;
585
586 for (i = 0; i < ARRAY_SIZE(vdso_patches); i++) {
587 struct vdso_patch_def *patch = &vdso_patches[i];
588 int match = (cur_cpu_spec->cpu_features & patch->ftr_mask)
589 == patch->ftr_value;
590 if (!match)
591 continue;
592
593 DBG("replacing %s with %s...\n", patch->gen_name,
594 patch->fix_name ? "NONE" : patch->fix_name);
595
596 /*
597 * Patch the 32 bits and 64 bits symbols. Note that we do not
598 * patch the "." symbol on 64 bits.
599 * It would be easy to do, but doesn't seem to be necessary,
600 * patching the OPD symbol is enough.
601 */
602 vdso_do_func_patch32(v32, v64, patch->gen_name,
603 patch->fix_name);
604#ifdef CONFIG_PPC64
605 vdso_do_func_patch64(v32, v64, patch->gen_name,
606 patch->fix_name);
607#endif /* CONFIG_PPC64 */
608 }
609
610 return 0;
611}
612
613
614static __init int vdso_setup(void)
615{
616 struct lib32_elfinfo v32;
617 struct lib64_elfinfo v64;
618
619 v32.hdr = vdso32_kbase;
620#ifdef CONFIG_PPC64
621 v64.hdr = vdso64_kbase;
622#endif
623 if (vdso_do_find_sections(&v32, &v64))
624 return -1;
625
626 if (vdso_fixup_datapage(&v32, &v64))
627 return -1;
628
629 if (vdso_fixup_features(&v32, &v64))
630 return -1;
631
632 if (vdso_fixup_alt_funcs(&v32, &v64))
633 return -1;
634
635 vdso_setup_trampolines(&v32, &v64);
636
637 return 0;
638}
639
640/*
641 * Called from setup_arch to initialize the bitmap of available
642 * syscalls in the systemcfg page
643 */
644static void __init vdso_setup_syscall_map(void)
645{
646 unsigned int i;
647 extern unsigned long *sys_call_table;
648#ifdef CONFIG_PPC64
649 extern unsigned long *compat_sys_call_table;
650#endif
651 extern unsigned long sys_ni_syscall;
652
653
654 for (i = 0; i < NR_syscalls; i++) {
655#ifdef CONFIG_PPC64
656 if (sys_call_table[i] != sys_ni_syscall)
657 vdso_data->syscall_map_64[i >> 5] |=
658 0x80000000UL >> (i & 0x1f);
659 if (compat_sys_call_table[i] != sys_ni_syscall)
660 vdso_data->syscall_map_32[i >> 5] |=
661 0x80000000UL >> (i & 0x1f);
662#else /* CONFIG_PPC64 */
663 if (sys_call_table[i] != sys_ni_syscall)
664 vdso_data->syscall_map_32[i >> 5] |=
665 0x80000000UL >> (i & 0x1f);
666#endif /* CONFIG_PPC64 */
667 }
668}
669
670#ifdef CONFIG_PPC64
671int vdso_getcpu_init(void)
672{
673 unsigned long cpu, node, val;
674
675 /*
676 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node
677 * in the next 16 bits. The VDSO uses this to implement getcpu().
678 */
679 cpu = get_cpu();
680 WARN_ON_ONCE(cpu > 0xffff);
681
682 node = cpu_to_node(cpu);
683 WARN_ON_ONCE(node > 0xffff);
684
685 val = (cpu & 0xfff) | ((node & 0xffff) << 16);
686 mtspr(SPRN_SPRG_VDSO_WRITE, val);
687 get_paca()->sprg_vdso = val;
688
689 put_cpu();
690
691 return 0;
692}
693/* We need to call this before SMP init */
694early_initcall(vdso_getcpu_init);
695#endif
696
697static int __init vdso_init(void)
698{
699 int i;
700
701#ifdef CONFIG_PPC64
702 /*
703 * Fill up the "systemcfg" stuff for backward compatibility
704 */
705 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64");
706 vdso_data->version.major = SYSTEMCFG_MAJOR;
707 vdso_data->version.minor = SYSTEMCFG_MINOR;
708 vdso_data->processor = mfspr(SPRN_PVR);
709 /*
710 * Fake the old platform number for pSeries and add
711 * in LPAR bit if necessary
712 */
713 vdso_data->platform = 0x100;
714 if (firmware_has_feature(FW_FEATURE_LPAR))
715 vdso_data->platform |= 1;
716 vdso_data->physicalMemorySize = memblock_phys_mem_size();
717 vdso_data->dcache_size = ppc64_caches.l1d.size;
718 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size;
719 vdso_data->icache_size = ppc64_caches.l1i.size;
720 vdso_data->icache_line_size = ppc64_caches.l1i.line_size;
721 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size;
722 vdso_data->icache_block_size = ppc64_caches.l1i.block_size;
723 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size;
724 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size;
725
726 /*
727 * Calculate the size of the 64 bits vDSO
728 */
729 vdso64_pages = (&vdso64_end - &vdso64_start) >> PAGE_SHIFT;
730 DBG("vdso64_kbase: %p, 0x%x pages\n", vdso64_kbase, vdso64_pages);
731#else
732 vdso_data->dcache_block_size = L1_CACHE_BYTES;
733 vdso_data->dcache_log_block_size = L1_CACHE_SHIFT;
734 vdso_data->icache_block_size = L1_CACHE_BYTES;
735 vdso_data->icache_log_block_size = L1_CACHE_SHIFT;
736#endif /* CONFIG_PPC64 */
737
738
739#ifdef CONFIG_VDSO32
740 vdso32_kbase = &vdso32_start;
741
742 /*
743 * Calculate the size of the 32 bits vDSO
744 */
745 vdso32_pages = (&vdso32_end - &vdso32_start) >> PAGE_SHIFT;
746 DBG("vdso32_kbase: %p, 0x%x pages\n", vdso32_kbase, vdso32_pages);
747#endif
748
749
750 /*
751 * Setup the syscall map in the vDOS
752 */
753 vdso_setup_syscall_map();
754
755 /*
756 * Initialize the vDSO images in memory, that is do necessary
757 * fixups of vDSO symbols, locate trampolines, etc...
758 */
759 if (vdso_setup()) {
760 printk(KERN_ERR "vDSO setup failure, not enabled !\n");
761 vdso32_pages = 0;
762#ifdef CONFIG_PPC64
763 vdso64_pages = 0;
764#endif
765 return 0;
766 }
767
768#ifdef CONFIG_VDSO32
769 /* Make sure pages are in the correct state */
770 vdso32_pagelist = kcalloc(vdso32_pages + 2, sizeof(struct page *),
771 GFP_KERNEL);
772 BUG_ON(vdso32_pagelist == NULL);
773 for (i = 0; i < vdso32_pages; i++) {
774 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
775 get_page(pg);
776 vdso32_pagelist[i] = pg;
777 }
778 vdso32_pagelist[i++] = virt_to_page(vdso_data);
779 vdso32_pagelist[i] = NULL;
780#endif
781
782#ifdef CONFIG_PPC64
783 vdso64_pagelist = kcalloc(vdso64_pages + 2, sizeof(struct page *),
784 GFP_KERNEL);
785 BUG_ON(vdso64_pagelist == NULL);
786 for (i = 0; i < vdso64_pages; i++) {
787 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
788 get_page(pg);
789 vdso64_pagelist[i] = pg;
790 }
791 vdso64_pagelist[i++] = virt_to_page(vdso_data);
792 vdso64_pagelist[i] = NULL;
793#endif /* CONFIG_PPC64 */
794
795 get_page(virt_to_page(vdso_data));
796
797 smp_wmb();
798 vdso_ready = 1;
799
800 return 0;
801}
802arch_initcall(vdso_init);