2 * Copyright 2007 Andi Kleen, SUSE Labs.
3 * Subject to the GPL, v.2
5 * This contains most of the x86 vDSO kernel-side code.
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/init.h>
12 #include <linux/random.h>
13 #include <linux/elf.h>
14 #include <linux/cpu.h>
15 #include <linux/ptrace.h>
16 #include <asm/pvclock.h>
17 #include <asm/vgtod.h>
18 #include <asm/proto.h>
23 #include <asm/cpufeature.h>
25 #if defined(CONFIG_X86_64)
26 unsigned int __read_mostly vdso64_enabled = 1;
29 void __init init_vdso_image(const struct vdso_image *image)
31 BUG_ON(image->size % PAGE_SIZE != 0);
33 apply_alternatives((struct alt_instr *)(image->data + image->alt),
34 (struct alt_instr *)(image->data + image->alt +
41 * Put the vdso above the (randomized) stack with another randomized
42 * offset. This way there is no hole in the middle of address space.
43 * To save memory make sure it is still in the same PTE as the stack
44 * top. This doesn't give that many random bits.
46 * Note that this algorithm is imperfect: the distribution of the vdso
47 * start address within a PMD is biased toward the end.
49 * Only used for the 64-bit and x32 vdsos.
51 static unsigned long vdso_addr(unsigned long start, unsigned len)
56 unsigned long addr, end;
60 * Round up the start address. It can start out unaligned as a result
61 * of stack start randomization.
63 start = PAGE_ALIGN(start);
65 /* Round the lowest possible end address up to a PMD boundary. */
66 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
67 if (end >= TASK_SIZE_MAX)
72 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
73 addr = start + (offset << PAGE_SHIFT);
79 * Forcibly align the final address in case we have a hardware
80 * issue that requires alignment for performance reasons.
82 addr = align_vdso_addr(addr);
88 static int vdso_fault(const struct vm_special_mapping *sm,
89 struct vm_area_struct *vma, struct vm_fault *vmf)
91 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
93 if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
94 return VM_FAULT_SIGBUS;
96 vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
101 static void vdso_fix_landing(const struct vdso_image *image,
102 struct vm_area_struct *new_vma)
104 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
105 if (in_ia32_syscall() && image == &vdso_image_32) {
106 struct pt_regs *regs = current_pt_regs();
107 unsigned long vdso_land = image->sym_int80_landing_pad;
108 unsigned long old_land_addr = vdso_land +
109 (unsigned long)current->mm->context.vdso;
111 /* Fixing userspace landing - look at do_fast_syscall_32 */
112 if (regs->ip == old_land_addr)
113 regs->ip = new_vma->vm_start + vdso_land;
118 static int vdso_mremap(const struct vm_special_mapping *sm,
119 struct vm_area_struct *new_vma)
121 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
122 const struct vdso_image *image = current->mm->context.vdso_image;
124 if (image->size != new_size)
127 if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
130 vdso_fix_landing(image, new_vma);
131 current->mm->context.vdso = (void __user *)new_vma->vm_start;
136 static int vvar_fault(const struct vm_special_mapping *sm,
137 struct vm_area_struct *vma, struct vm_fault *vmf)
139 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
144 return VM_FAULT_SIGBUS;
146 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
147 image->sym_vvar_start;
150 * Sanity check: a symbol offset of zero means that the page
151 * does not exist for this vdso image, not that the page is at
152 * offset zero relative to the text mapping. This should be
153 * impossible here, because sym_offset should only be zero for
154 * the page past the end of the vvar mapping.
157 return VM_FAULT_SIGBUS;
159 if (sym_offset == image->sym_vvar_page) {
160 ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
161 __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
162 } else if (sym_offset == image->sym_pvclock_page) {
163 struct pvclock_vsyscall_time_info *pvti =
164 pvclock_pvti_cpu0_va();
165 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
168 (unsigned long)vmf->virtual_address,
169 __pa(pvti) >> PAGE_SHIFT);
173 if (ret == 0 || ret == -EBUSY)
174 return VM_FAULT_NOPAGE;
176 return VM_FAULT_SIGBUS;
180 * Add vdso and vvar mappings to current process.
181 * @image - blob to map
182 * @addr - request a specific address (zero to map at free addr)
184 static int map_vdso(const struct vdso_image *image, unsigned long addr)
186 struct mm_struct *mm = current->mm;
187 struct vm_area_struct *vma;
188 unsigned long text_start;
191 static const struct vm_special_mapping vdso_mapping = {
194 .mremap = vdso_mremap,
196 static const struct vm_special_mapping vvar_mapping = {
201 if (down_write_killable(&mm->mmap_sem))
204 addr = get_unmapped_area(NULL, addr,
205 image->size - image->sym_vvar_start, 0, 0);
206 if (IS_ERR_VALUE(addr)) {
211 text_start = addr - image->sym_vvar_start;
212 current->mm->context.vdso = (void __user *)text_start;
213 current->mm->context.vdso_image = image;
216 * MAYWRITE to allow gdb to COW and set breakpoints
218 vma = _install_special_mapping(mm,
222 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
230 vma = _install_special_mapping(mm,
232 -image->sym_vvar_start,
233 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
239 do_munmap(mm, text_start, image->size);
244 current->mm->context.vdso = NULL;
245 current->mm->context.vdso_image = NULL;
248 up_write(&mm->mmap_sem);
252 static int map_vdso_randomized(const struct vdso_image *image)
254 unsigned long addr = vdso_addr(current->mm->start_stack,
255 image->size - image->sym_vvar_start);
256 return map_vdso(image, addr);
259 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
260 static int load_vdso32(void)
262 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
265 return map_vdso(&vdso_image_32, 0);
270 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
275 return map_vdso_randomized(&vdso_image_64);
279 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
282 #ifdef CONFIG_X86_X32_ABI
283 if (test_thread_flag(TIF_X32)) {
286 return map_vdso_randomized(&vdso_image_x32);
289 #ifdef CONFIG_IA32_EMULATION
290 return load_vdso32();
297 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
299 return load_vdso32();
304 static __init int vdso_setup(char *s)
306 vdso64_enabled = simple_strtoul(s, NULL, 0);
309 __setup("vdso=", vdso_setup);
313 static void vgetcpu_cpu_init(void *arg)
315 int cpu = smp_processor_id();
316 struct desc_struct d = { };
317 unsigned long node = 0;
319 node = cpu_to_node(cpu);
321 if (static_cpu_has(X86_FEATURE_RDTSCP))
322 write_rdtscp_aux((node << 12) | cpu);
325 * Store cpu number in limit so that it can be loaded
326 * quickly in user space in vgetcpu. (12 bits for the CPU
327 * and 8 bits for the node)
329 d.limit0 = cpu | ((node & 0xf) << 12);
331 d.type = 5; /* RO data, expand down, accessed */
332 d.dpl = 3; /* Visible to user code */
333 d.s = 1; /* Not a system segment */
334 d.p = 1; /* Present */
335 d.d = 1; /* 32-bit */
337 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
340 static int vgetcpu_online(unsigned int cpu)
342 return smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
345 static int __init init_vdso(void)
347 init_vdso_image(&vdso_image_64);
349 #ifdef CONFIG_X86_X32_ABI
350 init_vdso_image(&vdso_image_x32);
353 /* notifier priority > KVM */
354 return cpuhp_setup_state(CPUHP_AP_X86_VDSO_VMA_ONLINE,
355 "AP_X86_VDSO_VMA_ONLINE", vgetcpu_online, NULL);
357 subsys_initcall(init_vdso);
358 #endif /* CONFIG_X86_64 */