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 <asm/pvclock.h>
16 #include <asm/vgtod.h>
17 #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 const struct vm_special_mapping text_mapping = {
106 static int vvar_fault(const struct vm_special_mapping *sm,
107 struct vm_area_struct *vma, struct vm_fault *vmf)
109 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
114 return VM_FAULT_SIGBUS;
116 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
117 image->sym_vvar_start;
120 * Sanity check: a symbol offset of zero means that the page
121 * does not exist for this vdso image, not that the page is at
122 * offset zero relative to the text mapping. This should be
123 * impossible here, because sym_offset should only be zero for
124 * the page past the end of the vvar mapping.
127 return VM_FAULT_SIGBUS;
129 if (sym_offset == image->sym_vvar_page) {
130 ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
131 __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
132 } else if (sym_offset == image->sym_hpet_page) {
133 #ifdef CONFIG_HPET_TIMER
134 if (hpet_address && vclock_was_used(VCLOCK_HPET)) {
135 ret = vm_insert_pfn_prot(
137 (unsigned long)vmf->virtual_address,
138 hpet_address >> PAGE_SHIFT,
139 pgprot_noncached(PAGE_READONLY));
142 } else if (sym_offset == image->sym_pvclock_page) {
143 struct pvclock_vsyscall_time_info *pvti =
144 pvclock_pvti_cpu0_va();
145 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
148 (unsigned long)vmf->virtual_address,
149 __pa(pvti) >> PAGE_SHIFT);
153 if (ret == 0 || ret == -EBUSY)
154 return VM_FAULT_NOPAGE;
156 return VM_FAULT_SIGBUS;
159 static int map_vdso(const struct vdso_image *image, bool calculate_addr)
161 struct mm_struct *mm = current->mm;
162 struct vm_area_struct *vma;
163 unsigned long addr, text_start;
165 static const struct vm_special_mapping vvar_mapping = {
170 if (calculate_addr) {
171 addr = vdso_addr(current->mm->start_stack,
172 image->size - image->sym_vvar_start);
177 down_write(&mm->mmap_sem);
179 addr = get_unmapped_area(NULL, addr,
180 image->size - image->sym_vvar_start, 0, 0);
181 if (IS_ERR_VALUE(addr)) {
186 text_start = addr - image->sym_vvar_start;
187 current->mm->context.vdso = (void __user *)text_start;
188 current->mm->context.vdso_image = image;
191 * MAYWRITE to allow gdb to COW and set breakpoints
193 vma = _install_special_mapping(mm,
197 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
205 vma = _install_special_mapping(mm,
207 -image->sym_vvar_start,
208 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
219 current->mm->context.vdso = NULL;
221 up_write(&mm->mmap_sem);
225 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
226 static int load_vdso32(void)
228 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
231 return map_vdso(&vdso_image_32, false);
236 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
241 return map_vdso(&vdso_image_64, true);
245 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
248 #ifdef CONFIG_X86_X32_ABI
249 if (test_thread_flag(TIF_X32)) {
253 return map_vdso(&vdso_image_x32, true);
256 #ifdef CONFIG_IA32_EMULATION
257 return load_vdso32();
264 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
266 return load_vdso32();
271 static __init int vdso_setup(char *s)
273 vdso64_enabled = simple_strtoul(s, NULL, 0);
276 __setup("vdso=", vdso_setup);
280 static void vgetcpu_cpu_init(void *arg)
282 int cpu = smp_processor_id();
283 struct desc_struct d = { };
284 unsigned long node = 0;
286 node = cpu_to_node(cpu);
288 if (static_cpu_has(X86_FEATURE_RDTSCP))
289 write_rdtscp_aux((node << 12) | cpu);
292 * Store cpu number in limit so that it can be loaded
293 * quickly in user space in vgetcpu. (12 bits for the CPU
294 * and 8 bits for the node)
296 d.limit0 = cpu | ((node & 0xf) << 12);
298 d.type = 5; /* RO data, expand down, accessed */
299 d.dpl = 3; /* Visible to user code */
300 d.s = 1; /* Not a system segment */
301 d.p = 1; /* Present */
302 d.d = 1; /* 32-bit */
304 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
308 vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
310 long cpu = (long)arg;
312 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
313 smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
318 static int __init init_vdso(void)
320 init_vdso_image(&vdso_image_64);
322 #ifdef CONFIG_X86_X32_ABI
323 init_vdso_image(&vdso_image_x32);
326 cpu_notifier_register_begin();
328 on_each_cpu(vgetcpu_cpu_init, NULL, 1);
329 /* notifier priority > KVM */
330 __hotcpu_notifier(vgetcpu_cpu_notifier, 30);
332 cpu_notifier_register_done();
336 subsys_initcall(init_vdso);
337 #endif /* CONFIG_X86_64 */