i386: move xen
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / i386 / kernel / sysenter_32.c
1 /*
2 * linux/arch/i386/kernel/sysenter.c
3 *
4 * (C) Copyright 2002 Linus Torvalds
5 * Portions based on the vdso-randomization code from exec-shield:
6 * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
7 *
8 * This file contains the needed initializations to support sysenter.
9 */
10
11 #include <linux/init.h>
12 #include <linux/smp.h>
13 #include <linux/thread_info.h>
14 #include <linux/sched.h>
15 #include <linux/gfp.h>
16 #include <linux/string.h>
17 #include <linux/elf.h>
18 #include <linux/mm.h>
19 #include <linux/err.h>
20 #include <linux/module.h>
21
22 #include <asm/cpufeature.h>
23 #include <asm/msr.h>
24 #include <asm/pgtable.h>
25 #include <asm/unistd.h>
26 #include <asm/elf.h>
27 #include <asm/tlbflush.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 /*
42 * Should the kernel map a VDSO page into processes and pass its
43 * address down to glibc upon exec()?
44 */
45 unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
46
47 EXPORT_SYMBOL_GPL(vdso_enabled);
48
49 static int __init vdso_setup(char *s)
50 {
51 vdso_enabled = simple_strtoul(s, NULL, 0);
52
53 return 1;
54 }
55
56 __setup("vdso=", vdso_setup);
57
58 extern asmlinkage void sysenter_entry(void);
59
60 static __init void reloc_symtab(Elf32_Ehdr *ehdr,
61 unsigned offset, unsigned size)
62 {
63 Elf32_Sym *sym = (void *)ehdr + offset;
64 unsigned nsym = size / sizeof(*sym);
65 unsigned i;
66
67 for(i = 0; i < nsym; i++, sym++) {
68 if (sym->st_shndx == SHN_UNDEF ||
69 sym->st_shndx == SHN_ABS)
70 continue; /* skip */
71
72 if (sym->st_shndx > SHN_LORESERVE) {
73 printk(KERN_INFO "VDSO: unexpected st_shndx %x\n",
74 sym->st_shndx);
75 continue;
76 }
77
78 switch(ELF_ST_TYPE(sym->st_info)) {
79 case STT_OBJECT:
80 case STT_FUNC:
81 case STT_SECTION:
82 case STT_FILE:
83 sym->st_value += VDSO_HIGH_BASE;
84 }
85 }
86 }
87
88 static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset)
89 {
90 Elf32_Dyn *dyn = (void *)ehdr + offset;
91
92 for(; dyn->d_tag != DT_NULL; dyn++)
93 switch(dyn->d_tag) {
94 case DT_PLTGOT:
95 case DT_HASH:
96 case DT_STRTAB:
97 case DT_SYMTAB:
98 case DT_RELA:
99 case DT_INIT:
100 case DT_FINI:
101 case DT_REL:
102 case DT_DEBUG:
103 case DT_JMPREL:
104 case DT_VERSYM:
105 case DT_VERDEF:
106 case DT_VERNEED:
107 case DT_ADDRRNGLO ... DT_ADDRRNGHI:
108 /* definitely pointers needing relocation */
109 dyn->d_un.d_ptr += VDSO_HIGH_BASE;
110 break;
111
112 case DT_ENCODING ... OLD_DT_LOOS-1:
113 case DT_LOOS ... DT_HIOS-1:
114 /* Tags above DT_ENCODING are pointers if
115 they're even */
116 if (dyn->d_tag >= DT_ENCODING &&
117 (dyn->d_tag & 1) == 0)
118 dyn->d_un.d_ptr += VDSO_HIGH_BASE;
119 break;
120
121 case DT_VERDEFNUM:
122 case DT_VERNEEDNUM:
123 case DT_FLAGS_1:
124 case DT_RELACOUNT:
125 case DT_RELCOUNT:
126 case DT_VALRNGLO ... DT_VALRNGHI:
127 /* definitely not pointers */
128 break;
129
130 case OLD_DT_LOOS ... DT_LOOS-1:
131 case DT_HIOS ... DT_VALRNGLO-1:
132 default:
133 if (dyn->d_tag > DT_ENCODING)
134 printk(KERN_INFO "VDSO: unexpected DT_tag %x\n",
135 dyn->d_tag);
136 break;
137 }
138 }
139
140 static __init void relocate_vdso(Elf32_Ehdr *ehdr)
141 {
142 Elf32_Phdr *phdr;
143 Elf32_Shdr *shdr;
144 int i;
145
146 BUG_ON(memcmp(ehdr->e_ident, ELFMAG, 4) != 0 ||
147 !elf_check_arch(ehdr) ||
148 ehdr->e_type != ET_DYN);
149
150 ehdr->e_entry += VDSO_HIGH_BASE;
151
152 /* rebase phdrs */
153 phdr = (void *)ehdr + ehdr->e_phoff;
154 for (i = 0; i < ehdr->e_phnum; i++) {
155 phdr[i].p_vaddr += VDSO_HIGH_BASE;
156
157 /* relocate dynamic stuff */
158 if (phdr[i].p_type == PT_DYNAMIC)
159 reloc_dyn(ehdr, phdr[i].p_offset);
160 }
161
162 /* rebase sections */
163 shdr = (void *)ehdr + ehdr->e_shoff;
164 for(i = 0; i < ehdr->e_shnum; i++) {
165 if (!(shdr[i].sh_flags & SHF_ALLOC))
166 continue;
167
168 shdr[i].sh_addr += VDSO_HIGH_BASE;
169
170 if (shdr[i].sh_type == SHT_SYMTAB ||
171 shdr[i].sh_type == SHT_DYNSYM)
172 reloc_symtab(ehdr, shdr[i].sh_offset,
173 shdr[i].sh_size);
174 }
175 }
176
177 void enable_sep_cpu(void)
178 {
179 int cpu = get_cpu();
180 struct tss_struct *tss = &per_cpu(init_tss, cpu);
181
182 if (!boot_cpu_has(X86_FEATURE_SEP)) {
183 put_cpu();
184 return;
185 }
186
187 tss->x86_tss.ss1 = __KERNEL_CS;
188 tss->x86_tss.esp1 = sizeof(struct tss_struct) + (unsigned long) tss;
189 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
190 wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.esp1, 0);
191 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) sysenter_entry, 0);
192 put_cpu();
193 }
194
195 static struct vm_area_struct gate_vma;
196
197 static int __init gate_vma_init(void)
198 {
199 gate_vma.vm_mm = NULL;
200 gate_vma.vm_start = FIXADDR_USER_START;
201 gate_vma.vm_end = FIXADDR_USER_END;
202 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
203 gate_vma.vm_page_prot = __P101;
204 /*
205 * Make sure the vDSO gets into every core dump.
206 * Dumping its contents makes post-mortem fully interpretable later
207 * without matching up the same kernel and hardware config to see
208 * what PC values meant.
209 */
210 gate_vma.vm_flags |= VM_ALWAYSDUMP;
211 return 0;
212 }
213
214 /*
215 * These symbols are defined by vsyscall.o to mark the bounds
216 * of the ELF DSO images included therein.
217 */
218 extern const char vsyscall_int80_start, vsyscall_int80_end;
219 extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
220 static struct page *syscall_pages[1];
221
222 static void map_compat_vdso(int map)
223 {
224 static int vdso_mapped;
225
226 if (map == vdso_mapped)
227 return;
228
229 vdso_mapped = map;
230
231 __set_fixmap(FIX_VDSO, page_to_pfn(syscall_pages[0]) << PAGE_SHIFT,
232 map ? PAGE_READONLY_EXEC : PAGE_NONE);
233
234 /* flush stray tlbs */
235 flush_tlb_all();
236 }
237
238 int __init sysenter_setup(void)
239 {
240 void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
241 const void *vsyscall;
242 size_t vsyscall_len;
243
244 syscall_pages[0] = virt_to_page(syscall_page);
245
246 gate_vma_init();
247
248 printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO));
249
250 if (!boot_cpu_has(X86_FEATURE_SEP)) {
251 vsyscall = &vsyscall_int80_start;
252 vsyscall_len = &vsyscall_int80_end - &vsyscall_int80_start;
253 } else {
254 vsyscall = &vsyscall_sysenter_start;
255 vsyscall_len = &vsyscall_sysenter_end - &vsyscall_sysenter_start;
256 }
257
258 memcpy(syscall_page, vsyscall, vsyscall_len);
259 relocate_vdso(syscall_page);
260
261 return 0;
262 }
263
264 /* Defined in vsyscall-sysenter.S */
265 extern void SYSENTER_RETURN;
266
267 /* Setup a VMA at program startup for the vsyscall page */
268 int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
269 {
270 struct mm_struct *mm = current->mm;
271 unsigned long addr;
272 int ret = 0;
273 bool compat;
274
275 down_write(&mm->mmap_sem);
276
277 /* Test compat mode once here, in case someone
278 changes it via sysctl */
279 compat = (vdso_enabled == VDSO_COMPAT);
280
281 map_compat_vdso(compat);
282
283 if (compat)
284 addr = VDSO_HIGH_BASE;
285 else {
286 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
287 if (IS_ERR_VALUE(addr)) {
288 ret = addr;
289 goto up_fail;
290 }
291
292 /*
293 * MAYWRITE to allow gdb to COW and set breakpoints
294 *
295 * Make sure the vDSO gets into every core dump.
296 * Dumping its contents makes post-mortem fully
297 * interpretable later without matching up the same
298 * kernel and hardware config to see what PC values
299 * meant.
300 */
301 ret = install_special_mapping(mm, addr, PAGE_SIZE,
302 VM_READ|VM_EXEC|
303 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
304 VM_ALWAYSDUMP,
305 syscall_pages);
306
307 if (ret)
308 goto up_fail;
309 }
310
311 current->mm->context.vdso = (void *)addr;
312 current_thread_info()->sysenter_return =
313 (void *)VDSO_SYM(&SYSENTER_RETURN);
314
315 up_fail:
316 up_write(&mm->mmap_sem);
317
318 return ret;
319 }
320
321 const char *arch_vma_name(struct vm_area_struct *vma)
322 {
323 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
324 return "[vdso]";
325 return NULL;
326 }
327
328 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
329 {
330 struct mm_struct *mm = tsk->mm;
331
332 /* Check to see if this task was created in compat vdso mode */
333 if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
334 return &gate_vma;
335 return NULL;
336 }
337
338 int in_gate_area(struct task_struct *task, unsigned long addr)
339 {
340 const struct vm_area_struct *vma = get_gate_vma(task);
341
342 return vma && addr >= vma->vm_start && addr < vma->vm_end;
343 }
344
345 int in_gate_area_no_task(unsigned long addr)
346 {
347 return 0;
348 }