Merge branch 'x86/cpufeature' into irq/numa
[sfrench/cifs-2.6.git] / arch / x86 / vdso / vdso32-setup.c
1 /*
2  * (C) Copyright 2002 Linus Torvalds
3  * Portions based on the vdso-randomization code from exec-shield:
4  * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
5  *
6  * This file contains the needed initializations to support sysenter.
7  */
8
9 #include <linux/init.h>
10 #include <linux/smp.h>
11 #include <linux/thread_info.h>
12 #include <linux/sched.h>
13 #include <linux/gfp.h>
14 #include <linux/string.h>
15 #include <linux/elf.h>
16 #include <linux/mm.h>
17 #include <linux/err.h>
18 #include <linux/module.h>
19
20 #include <asm/cpufeature.h>
21 #include <asm/msr.h>
22 #include <asm/pgtable.h>
23 #include <asm/unistd.h>
24 #include <asm/elf.h>
25 #include <asm/tlbflush.h>
26 #include <asm/vdso.h>
27 #include <asm/proto.h>
28
29 enum {
30         VDSO_DISABLED = 0,
31         VDSO_ENABLED = 1,
32         VDSO_COMPAT = 2,
33 };
34
35 #ifdef CONFIG_COMPAT_VDSO
36 #define VDSO_DEFAULT    VDSO_COMPAT
37 #else
38 #define VDSO_DEFAULT    VDSO_ENABLED
39 #endif
40
41 #ifdef CONFIG_X86_64
42 #define vdso_enabled                    sysctl_vsyscall32
43 #define arch_setup_additional_pages     syscall32_setup_pages
44 #endif
45
46 /*
47  * This is the difference between the prelinked addresses in the vDSO images
48  * and the VDSO_HIGH_BASE address where CONFIG_COMPAT_VDSO places the vDSO
49  * in the user address space.
50  */
51 #define VDSO_ADDR_ADJUST        (VDSO_HIGH_BASE - (unsigned long)VDSO32_PRELINK)
52
53 /*
54  * Should the kernel map a VDSO page into processes and pass its
55  * address down to glibc upon exec()?
56  */
57 unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
58
59 static int __init vdso_setup(char *s)
60 {
61         vdso_enabled = simple_strtoul(s, NULL, 0);
62
63         return 1;
64 }
65
66 /*
67  * For consistency, the argument vdso32=[012] affects the 32-bit vDSO
68  * behavior on both 64-bit and 32-bit kernels.
69  * On 32-bit kernels, vdso=[012] means the same thing.
70  */
71 __setup("vdso32=", vdso_setup);
72
73 #ifdef CONFIG_X86_32
74 __setup_param("vdso=", vdso32_setup, vdso_setup, 0);
75
76 EXPORT_SYMBOL_GPL(vdso_enabled);
77 #endif
78
79 static __init void reloc_symtab(Elf32_Ehdr *ehdr,
80                                 unsigned offset, unsigned size)
81 {
82         Elf32_Sym *sym = (void *)ehdr + offset;
83         unsigned nsym = size / sizeof(*sym);
84         unsigned i;
85
86         for(i = 0; i < nsym; i++, sym++) {
87                 if (sym->st_shndx == SHN_UNDEF ||
88                     sym->st_shndx == SHN_ABS)
89                         continue;  /* skip */
90
91                 if (sym->st_shndx > SHN_LORESERVE) {
92                         printk(KERN_INFO "VDSO: unexpected st_shndx %x\n",
93                                sym->st_shndx);
94                         continue;
95                 }
96
97                 switch(ELF_ST_TYPE(sym->st_info)) {
98                 case STT_OBJECT:
99                 case STT_FUNC:
100                 case STT_SECTION:
101                 case STT_FILE:
102                         sym->st_value += VDSO_ADDR_ADJUST;
103                 }
104         }
105 }
106
107 static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset)
108 {
109         Elf32_Dyn *dyn = (void *)ehdr + offset;
110
111         for(; dyn->d_tag != DT_NULL; dyn++)
112                 switch(dyn->d_tag) {
113                 case DT_PLTGOT:
114                 case DT_HASH:
115                 case DT_STRTAB:
116                 case DT_SYMTAB:
117                 case DT_RELA:
118                 case DT_INIT:
119                 case DT_FINI:
120                 case DT_REL:
121                 case DT_DEBUG:
122                 case DT_JMPREL:
123                 case DT_VERSYM:
124                 case DT_VERDEF:
125                 case DT_VERNEED:
126                 case DT_ADDRRNGLO ... DT_ADDRRNGHI:
127                         /* definitely pointers needing relocation */
128                         dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
129                         break;
130
131                 case DT_ENCODING ... OLD_DT_LOOS-1:
132                 case DT_LOOS ... DT_HIOS-1:
133                         /* Tags above DT_ENCODING are pointers if
134                            they're even */
135                         if (dyn->d_tag >= DT_ENCODING &&
136                             (dyn->d_tag & 1) == 0)
137                                 dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
138                         break;
139
140                 case DT_VERDEFNUM:
141                 case DT_VERNEEDNUM:
142                 case DT_FLAGS_1:
143                 case DT_RELACOUNT:
144                 case DT_RELCOUNT:
145                 case DT_VALRNGLO ... DT_VALRNGHI:
146                         /* definitely not pointers */
147                         break;
148
149                 case OLD_DT_LOOS ... DT_LOOS-1:
150                 case DT_HIOS ... DT_VALRNGLO-1:
151                 default:
152                         if (dyn->d_tag > DT_ENCODING)
153                                 printk(KERN_INFO "VDSO: unexpected DT_tag %x\n",
154                                        dyn->d_tag);
155                         break;
156                 }
157 }
158
159 static __init void relocate_vdso(Elf32_Ehdr *ehdr)
160 {
161         Elf32_Phdr *phdr;
162         Elf32_Shdr *shdr;
163         int i;
164
165         BUG_ON(memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0 ||
166                !elf_check_arch_ia32(ehdr) ||
167                ehdr->e_type != ET_DYN);
168
169         ehdr->e_entry += VDSO_ADDR_ADJUST;
170
171         /* rebase phdrs */
172         phdr = (void *)ehdr + ehdr->e_phoff;
173         for (i = 0; i < ehdr->e_phnum; i++) {
174                 phdr[i].p_vaddr += VDSO_ADDR_ADJUST;
175
176                 /* relocate dynamic stuff */
177                 if (phdr[i].p_type == PT_DYNAMIC)
178                         reloc_dyn(ehdr, phdr[i].p_offset);
179         }
180
181         /* rebase sections */
182         shdr = (void *)ehdr + ehdr->e_shoff;
183         for(i = 0; i < ehdr->e_shnum; i++) {
184                 if (!(shdr[i].sh_flags & SHF_ALLOC))
185                         continue;
186
187                 shdr[i].sh_addr += VDSO_ADDR_ADJUST;
188
189                 if (shdr[i].sh_type == SHT_SYMTAB ||
190                     shdr[i].sh_type == SHT_DYNSYM)
191                         reloc_symtab(ehdr, shdr[i].sh_offset,
192                                      shdr[i].sh_size);
193         }
194 }
195
196 static struct page *vdso32_pages[1];
197
198 #ifdef CONFIG_X86_64
199
200 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SYSENTER32))
201 #define vdso32_syscall()        (boot_cpu_has(X86_FEATURE_SYSCALL32))
202
203 /* May not be __init: called during resume */
204 void syscall32_cpu_init(void)
205 {
206         /* Load these always in case some future AMD CPU supports
207            SYSENTER from compat mode too. */
208         checking_wrmsrl(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
209         checking_wrmsrl(MSR_IA32_SYSENTER_ESP, 0ULL);
210         checking_wrmsrl(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
211
212         wrmsrl(MSR_CSTAR, ia32_cstar_target);
213 }
214
215 #define compat_uses_vma         1
216
217 static inline void map_compat_vdso(int map)
218 {
219 }
220
221 #else  /* CONFIG_X86_32 */
222
223 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SEP))
224 #define vdso32_syscall()        (0)
225
226 void enable_sep_cpu(void)
227 {
228         int cpu = get_cpu();
229         struct tss_struct *tss = &per_cpu(init_tss, cpu);
230
231         if (!boot_cpu_has(X86_FEATURE_SEP)) {
232                 put_cpu();
233                 return;
234         }
235
236         tss->x86_tss.ss1 = __KERNEL_CS;
237         tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss;
238         wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
239         wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0);
240         wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
241         put_cpu();      
242 }
243
244 static struct vm_area_struct gate_vma;
245
246 static int __init gate_vma_init(void)
247 {
248         gate_vma.vm_mm = NULL;
249         gate_vma.vm_start = FIXADDR_USER_START;
250         gate_vma.vm_end = FIXADDR_USER_END;
251         gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
252         gate_vma.vm_page_prot = __P101;
253         /*
254          * Make sure the vDSO gets into every core dump.
255          * Dumping its contents makes post-mortem fully interpretable later
256          * without matching up the same kernel and hardware config to see
257          * what PC values meant.
258          */
259         gate_vma.vm_flags |= VM_ALWAYSDUMP;
260         return 0;
261 }
262
263 #define compat_uses_vma         0
264
265 static void map_compat_vdso(int map)
266 {
267         static int vdso_mapped;
268
269         if (map == vdso_mapped)
270                 return;
271
272         vdso_mapped = map;
273
274         __set_fixmap(FIX_VDSO, page_to_pfn(vdso32_pages[0]) << PAGE_SHIFT,
275                      map ? PAGE_READONLY_EXEC : PAGE_NONE);
276
277         /* flush stray tlbs */
278         flush_tlb_all();
279 }
280
281 #endif  /* CONFIG_X86_64 */
282
283 int __init sysenter_setup(void)
284 {
285         void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
286         const void *vsyscall;
287         size_t vsyscall_len;
288
289         vdso32_pages[0] = virt_to_page(syscall_page);
290
291 #ifdef CONFIG_X86_32
292         gate_vma_init();
293 #endif
294
295         if (vdso32_syscall()) {
296                 vsyscall = &vdso32_syscall_start;
297                 vsyscall_len = &vdso32_syscall_end - &vdso32_syscall_start;
298         } else if (vdso32_sysenter()){
299                 vsyscall = &vdso32_sysenter_start;
300                 vsyscall_len = &vdso32_sysenter_end - &vdso32_sysenter_start;
301         } else {
302                 vsyscall = &vdso32_int80_start;
303                 vsyscall_len = &vdso32_int80_end - &vdso32_int80_start;
304         }
305
306         memcpy(syscall_page, vsyscall, vsyscall_len);
307         relocate_vdso(syscall_page);
308
309         return 0;
310 }
311
312 /* Setup a VMA at program startup for the vsyscall page */
313 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
314 {
315         struct mm_struct *mm = current->mm;
316         unsigned long addr;
317         int ret = 0;
318         bool compat;
319
320         if (vdso_enabled == VDSO_DISABLED)
321                 return 0;
322
323         down_write(&mm->mmap_sem);
324
325         /* Test compat mode once here, in case someone
326            changes it via sysctl */
327         compat = (vdso_enabled == VDSO_COMPAT);
328
329         map_compat_vdso(compat);
330
331         if (compat)
332                 addr = VDSO_HIGH_BASE;
333         else {
334                 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
335                 if (IS_ERR_VALUE(addr)) {
336                         ret = addr;
337                         goto up_fail;
338                 }
339         }
340
341         if (compat_uses_vma || !compat) {
342                 /*
343                  * MAYWRITE to allow gdb to COW and set breakpoints
344                  *
345                  * Make sure the vDSO gets into every core dump.
346                  * Dumping its contents makes post-mortem fully
347                  * interpretable later without matching up the same
348                  * kernel and hardware config to see what PC values
349                  * meant.
350                  */
351                 ret = install_special_mapping(mm, addr, PAGE_SIZE,
352                                               VM_READ|VM_EXEC|
353                                               VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
354                                               VM_ALWAYSDUMP,
355                                               vdso32_pages);
356
357                 if (ret)
358                         goto up_fail;
359         }
360
361         current->mm->context.vdso = (void *)addr;
362         current_thread_info()->sysenter_return =
363                 VDSO32_SYMBOL(addr, SYSENTER_RETURN);
364
365   up_fail:
366         up_write(&mm->mmap_sem);
367
368         return ret;
369 }
370
371 #ifdef CONFIG_X86_64
372
373 __initcall(sysenter_setup);
374
375 #ifdef CONFIG_SYSCTL
376 /* Register vsyscall32 into the ABI table */
377 #include <linux/sysctl.h>
378
379 static ctl_table abi_table2[] = {
380         {
381                 .procname       = "vsyscall32",
382                 .data           = &sysctl_vsyscall32,
383                 .maxlen         = sizeof(int),
384                 .mode           = 0644,
385                 .proc_handler   = proc_dointvec
386         },
387         {}
388 };
389
390 static ctl_table abi_root_table2[] = {
391         {
392                 .ctl_name = CTL_ABI,
393                 .procname = "abi",
394                 .mode = 0555,
395                 .child = abi_table2
396         },
397         {}
398 };
399
400 static __init int ia32_binfmt_init(void)
401 {
402         register_sysctl_table(abi_root_table2);
403         return 0;
404 }
405 __initcall(ia32_binfmt_init);
406 #endif
407
408 #else  /* CONFIG_X86_32 */
409
410 const char *arch_vma_name(struct vm_area_struct *vma)
411 {
412         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
413                 return "[vdso]";
414         return NULL;
415 }
416
417 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
418 {
419         struct mm_struct *mm = tsk->mm;
420
421         /* Check to see if this task was created in compat vdso mode */
422         if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
423                 return &gate_vma;
424         return NULL;
425 }
426
427 int in_gate_area(struct task_struct *task, unsigned long addr)
428 {
429         const struct vm_area_struct *vma = get_gate_vma(task);
430
431         return vma && addr >= vma->vm_start && addr < vma->vm_end;
432 }
433
434 int in_gate_area_no_task(unsigned long addr)
435 {
436         return 0;
437 }
438
439 #endif  /* CONFIG_X86_64 */