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>
24 #if defined(CONFIG_X86_64)
25 unsigned int __read_mostly vdso64_enabled = 1;
28 void __init init_vdso_image(const struct vdso_image *image)
31 int npages = (image->size) / PAGE_SIZE;
33 BUG_ON(image->size % PAGE_SIZE != 0);
34 for (i = 0; i < npages; i++)
35 image->text_mapping.pages[i] =
36 virt_to_page(image->data + i*PAGE_SIZE);
38 apply_alternatives((struct alt_instr *)(image->data + image->alt),
39 (struct alt_instr *)(image->data + image->alt +
46 * Put the vdso above the (randomized) stack with another randomized
47 * offset. This way there is no hole in the middle of address space.
48 * To save memory make sure it is still in the same PTE as the stack
49 * top. This doesn't give that many random bits.
51 * Note that this algorithm is imperfect: the distribution of the vdso
52 * start address within a PMD is biased toward the end.
54 * Only used for the 64-bit and x32 vdsos.
56 static unsigned long vdso_addr(unsigned long start, unsigned len)
61 unsigned long addr, end;
65 * Round up the start address. It can start out unaligned as a result
66 * of stack start randomization.
68 start = PAGE_ALIGN(start);
70 /* Round the lowest possible end address up to a PMD boundary. */
71 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
72 if (end >= TASK_SIZE_MAX)
77 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
78 addr = start + (offset << PAGE_SHIFT);
84 * Forcibly align the final address in case we have a hardware
85 * issue that requires alignment for performance reasons.
87 addr = align_vdso_addr(addr);
93 static int map_vdso(const struct vdso_image *image, bool calculate_addr)
95 struct mm_struct *mm = current->mm;
96 struct vm_area_struct *vma;
97 unsigned long addr, text_start;
99 static struct page *no_pages[] = {NULL};
100 static struct vm_special_mapping vvar_mapping = {
104 struct pvclock_vsyscall_time_info *pvti;
106 if (calculate_addr) {
107 addr = vdso_addr(current->mm->start_stack,
108 image->size - image->sym_vvar_start);
113 down_write(&mm->mmap_sem);
115 addr = get_unmapped_area(NULL, addr,
116 image->size - image->sym_vvar_start, 0, 0);
117 if (IS_ERR_VALUE(addr)) {
122 text_start = addr - image->sym_vvar_start;
123 current->mm->context.vdso = (void __user *)text_start;
126 * MAYWRITE to allow gdb to COW and set breakpoints
128 vma = _install_special_mapping(mm,
132 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
133 &image->text_mapping);
140 vma = _install_special_mapping(mm,
142 -image->sym_vvar_start,
151 if (image->sym_vvar_page)
152 ret = remap_pfn_range(vma,
153 text_start + image->sym_vvar_page,
154 __pa_symbol(&__vvar_page) >> PAGE_SHIFT,
161 #ifdef CONFIG_HPET_TIMER
162 if (hpet_address && image->sym_hpet_page) {
163 ret = io_remap_pfn_range(vma,
164 text_start + image->sym_hpet_page,
165 hpet_address >> PAGE_SHIFT,
167 pgprot_noncached(PAGE_READONLY));
174 pvti = pvclock_pvti_cpu0_va();
175 if (pvti && image->sym_pvclock_page) {
176 ret = remap_pfn_range(vma,
177 text_start + image->sym_pvclock_page,
178 __pa(pvti) >> PAGE_SHIFT,
188 current->mm->context.vdso = NULL;
190 up_write(&mm->mmap_sem);
194 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
195 static int load_vdso32(void)
197 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
200 return map_vdso(&vdso_image_32, false);
205 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
210 return map_vdso(&vdso_image_64, true);
214 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
217 #ifdef CONFIG_X86_X32_ABI
218 if (test_thread_flag(TIF_X32)) {
222 return map_vdso(&vdso_image_x32, true);
225 #ifdef CONFIG_IA32_EMULATION
226 return load_vdso32();
233 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
235 return load_vdso32();
240 static __init int vdso_setup(char *s)
242 vdso64_enabled = simple_strtoul(s, NULL, 0);
245 __setup("vdso=", vdso_setup);
249 static void vgetcpu_cpu_init(void *arg)
251 int cpu = smp_processor_id();
252 struct desc_struct d = { };
253 unsigned long node = 0;
255 node = cpu_to_node(cpu);
257 if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
258 write_rdtscp_aux((node << 12) | cpu);
261 * Store cpu number in limit so that it can be loaded
262 * quickly in user space in vgetcpu. (12 bits for the CPU
263 * and 8 bits for the node)
265 d.limit0 = cpu | ((node & 0xf) << 12);
267 d.type = 5; /* RO data, expand down, accessed */
268 d.dpl = 3; /* Visible to user code */
269 d.s = 1; /* Not a system segment */
270 d.p = 1; /* Present */
271 d.d = 1; /* 32-bit */
273 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
277 vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
279 long cpu = (long)arg;
281 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
282 smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
287 static int __init init_vdso(void)
289 init_vdso_image(&vdso_image_64);
291 #ifdef CONFIG_X86_X32_ABI
292 init_vdso_image(&vdso_image_x32);
295 cpu_notifier_register_begin();
297 on_each_cpu(vgetcpu_cpu_init, NULL, 1);
298 /* notifier priority > KVM */
299 __hotcpu_notifier(vgetcpu_cpu_notifier, 30);
301 cpu_notifier_register_done();
305 subsys_initcall(init_vdso);
306 #endif /* CONFIG_X86_64 */