Linux Audio

Check our new training course

Loading...
v6.2
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright 2007 Andi Kleen, SUSE Labs.
 
  4 *
  5 * This contains most of the x86 vDSO kernel-side code.
  6 */
  7#include <linux/mm.h>
  8#include <linux/err.h>
  9#include <linux/sched.h>
 10#include <linux/sched/task_stack.h>
 11#include <linux/slab.h>
 12#include <linux/init.h>
 13#include <linux/random.h>
 14#include <linux/elf.h>
 15#include <linux/cpu.h>
 16#include <linux/ptrace.h>
 17#include <linux/time_namespace.h>
 18
 19#include <asm/pvclock.h>
 20#include <asm/vgtod.h>
 21#include <asm/proto.h>
 22#include <asm/vdso.h>
 23#include <asm/vvar.h>
 24#include <asm/tlb.h>
 25#include <asm/page.h>
 
 26#include <asm/desc.h>
 27#include <asm/cpufeature.h>
 28#include <clocksource/hyperv_timer.h>
 29
 30#undef _ASM_X86_VVAR_H
 31#define EMIT_VVAR(name, offset)	\
 32	const size_t name ## _offset = offset;
 33#include <asm/vvar.h>
 34
 35struct vdso_data *arch_get_vdso_data(void *vvar_page)
 36{
 37	return (struct vdso_data *)(vvar_page + _vdso_data_offset);
 38}
 39#undef EMIT_VVAR
 40
 41unsigned int vclocks_used __read_mostly;
 42
 43#if defined(CONFIG_X86_64)
 44unsigned int __read_mostly vdso64_enabled = 1;
 45#endif
 46
 47void __init init_vdso_image(const struct vdso_image *image)
 48{
 49	BUG_ON(image->size % PAGE_SIZE != 0);
 50
 51	apply_alternatives((struct alt_instr *)(image->data + image->alt),
 52			   (struct alt_instr *)(image->data + image->alt +
 53						image->alt_len));
 54}
 55
 56static const struct vm_special_mapping vvar_mapping;
 57struct linux_binprm;
 58
 59static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
 60		      struct vm_area_struct *vma, struct vm_fault *vmf)
 
 
 
 
 
 
 
 
 
 
 61{
 62	const struct vdso_image *image = vma->vm_mm->context.vdso_image;
 
 
 
 
 63
 64	if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
 65		return VM_FAULT_SIGBUS;
 
 
 
 66
 67	vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
 68	get_page(vmf->page);
 69	return 0;
 70}
 
 71
 72static void vdso_fix_landing(const struct vdso_image *image,
 73		struct vm_area_struct *new_vma)
 74{
 75#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 76	if (in_ia32_syscall() && image == &vdso_image_32) {
 77		struct pt_regs *regs = current_pt_regs();
 78		unsigned long vdso_land = image->sym_int80_landing_pad;
 79		unsigned long old_land_addr = vdso_land +
 80			(unsigned long)current->mm->context.vdso;
 81
 82		/* Fixing userspace landing - look at do_fast_syscall_32 */
 83		if (regs->ip == old_land_addr)
 84			regs->ip = new_vma->vm_start + vdso_land;
 85	}
 86#endif
 87}
 88
 89static int vdso_mremap(const struct vm_special_mapping *sm,
 90		struct vm_area_struct *new_vma)
 91{
 92	const struct vdso_image *image = current->mm->context.vdso_image;
 93
 94	vdso_fix_landing(image, new_vma);
 95	current->mm->context.vdso = (void __user *)new_vma->vm_start;
 
 
 
 96
 97	return 0;
 
 98}
 99
100#ifdef CONFIG_TIME_NS
101/*
102 * The vvar page layout depends on whether a task belongs to the root or
103 * non-root time namespace. Whenever a task changes its namespace, the VVAR
104 * page tables are cleared and then they will re-faulted with a
105 * corresponding layout.
106 * See also the comment near timens_setup_vdso_data() for details.
107 */
108int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
109{
110	struct mm_struct *mm = task->mm;
111	struct vm_area_struct *vma;
112	VMA_ITERATOR(vmi, mm, 0);
113
114	mmap_read_lock(mm);
115	for_each_vma(vmi, vma) {
116		unsigned long size = vma->vm_end - vma->vm_start;
117
118		if (vma_is_special_mapping(vma, &vvar_mapping))
119			zap_page_range(vma, vma->vm_start, size);
120	}
121	mmap_read_unlock(mm);
122
 
 
123	return 0;
124}
125#endif
126
127static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 
 
 
 
 
128		      struct vm_area_struct *vma, struct vm_fault *vmf)
129{
130	const struct vdso_image *image = vma->vm_mm->context.vdso_image;
131	unsigned long pfn;
132	long sym_offset;
 
133
134	if (!image)
135		return VM_FAULT_SIGBUS;
136
137	sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
138		image->sym_vvar_start;
139
140	/*
141	 * Sanity check: a symbol offset of zero means that the page
142	 * does not exist for this vdso image, not that the page is at
143	 * offset zero relative to the text mapping.  This should be
144	 * impossible here, because sym_offset should only be zero for
145	 * the page past the end of the vvar mapping.
146	 */
147	if (sym_offset == 0)
148		return VM_FAULT_SIGBUS;
149
150	if (sym_offset == image->sym_vvar_page) {
151		struct page *timens_page = find_timens_vvar_page(vma);
152
153		pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
154
155		/*
156		 * If a task belongs to a time namespace then a namespace
157		 * specific VVAR is mapped with the sym_vvar_page offset and
158		 * the real VVAR page is mapped with the sym_timens_page
159		 * offset.
160		 * See also the comment near timens_setup_vdso_data().
161		 */
162		if (timens_page) {
163			unsigned long addr;
164			vm_fault_t err;
165
166			/*
167			 * Optimization: inside time namespace pre-fault
168			 * VVAR page too. As on timens page there are only
169			 * offsets for clocks on VVAR, it'll be faulted
170			 * shortly by VDSO code.
171			 */
172			addr = vmf->address + (image->sym_timens_page - sym_offset);
173			err = vmf_insert_pfn(vma, addr, pfn);
174			if (unlikely(err & VM_FAULT_ERROR))
175				return err;
176
177			pfn = page_to_pfn(timens_page);
178		}
179
180		return vmf_insert_pfn(vma, vmf->address, pfn);
181	} else if (sym_offset == image->sym_pvclock_page) {
182		struct pvclock_vsyscall_time_info *pvti =
183			pvclock_get_pvti_cpu0_va();
184		if (pvti && vclock_was_used(VDSO_CLOCKMODE_PVCLOCK)) {
185			return vmf_insert_pfn_prot(vma, vmf->address,
186					__pa(pvti) >> PAGE_SHIFT,
187					pgprot_decrypted(vma->vm_page_prot));
 
188		}
189	} else if (sym_offset == image->sym_hvclock_page) {
190		pfn = hv_get_tsc_pfn();
191
192		if (pfn && vclock_was_used(VDSO_CLOCKMODE_HVCLOCK))
193			return vmf_insert_pfn(vma, vmf->address, pfn);
194	} else if (sym_offset == image->sym_timens_page) {
195		struct page *timens_page = find_timens_vvar_page(vma);
196
197		if (!timens_page)
198			return VM_FAULT_SIGBUS;
199
200		pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
201		return vmf_insert_pfn(vma, vmf->address, pfn);
202	}
203
 
 
 
204	return VM_FAULT_SIGBUS;
205}
206
207static const struct vm_special_mapping vdso_mapping = {
208	.name = "[vdso]",
209	.fault = vdso_fault,
210	.mremap = vdso_mremap,
211};
212static const struct vm_special_mapping vvar_mapping = {
213	.name = "[vvar]",
214	.fault = vvar_fault,
215};
216
217/*
218 * Add vdso and vvar mappings to current process.
219 * @image          - blob to map
220 * @addr           - request a specific address (zero to map at free addr)
221 */
222static int map_vdso(const struct vdso_image *image, unsigned long addr)
223{
224	struct mm_struct *mm = current->mm;
225	struct vm_area_struct *vma;
226	unsigned long text_start;
227	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
228
229	if (mmap_write_lock_killable(mm))
230		return -EINTR;
231
232	addr = get_unmapped_area(NULL, addr,
233				 image->size - image->sym_vvar_start, 0, 0);
234	if (IS_ERR_VALUE(addr)) {
235		ret = addr;
236		goto up_fail;
237	}
238
239	text_start = addr - image->sym_vvar_start;
 
 
240
241	/*
242	 * MAYWRITE to allow gdb to COW and set breakpoints
243	 */
244	vma = _install_special_mapping(mm,
245				       text_start,
246				       image->size,
247				       VM_READ|VM_EXEC|
248				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
249				       &vdso_mapping);
250
251	if (IS_ERR(vma)) {
252		ret = PTR_ERR(vma);
253		goto up_fail;
254	}
255
256	vma = _install_special_mapping(mm,
257				       addr,
258				       -image->sym_vvar_start,
259				       VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
260				       VM_PFNMAP,
261				       &vvar_mapping);
262
263	if (IS_ERR(vma)) {
264		ret = PTR_ERR(vma);
265		do_munmap(mm, text_start, image->size, NULL);
266	} else {
267		current->mm->context.vdso = (void __user *)text_start;
268		current->mm->context.vdso_image = image;
269	}
270
271up_fail:
272	mmap_write_unlock(mm);
273	return ret;
274}
275
276#ifdef CONFIG_X86_64
277/*
278 * Put the vdso above the (randomized) stack with another randomized
279 * offset.  This way there is no hole in the middle of address space.
280 * To save memory make sure it is still in the same PTE as the stack
281 * top.  This doesn't give that many random bits.
282 *
283 * Note that this algorithm is imperfect: the distribution of the vdso
284 * start address within a PMD is biased toward the end.
285 *
286 * Only used for the 64-bit and x32 vdsos.
287 */
288static unsigned long vdso_addr(unsigned long start, unsigned len)
289{
290	unsigned long addr, end;
291	unsigned offset;
292
293	/*
294	 * Round up the start address.  It can start out unaligned as a result
295	 * of stack start randomization.
296	 */
297	start = PAGE_ALIGN(start);
298
299	/* Round the lowest possible end address up to a PMD boundary. */
300	end = (start + len + PMD_SIZE - 1) & PMD_MASK;
301	if (end >= TASK_SIZE_MAX)
302		end = TASK_SIZE_MAX;
303	end -= len;
304
305	if (end > start) {
306		offset = get_random_u32_below(((end - start) >> PAGE_SHIFT) + 1);
307		addr = start + (offset << PAGE_SHIFT);
308	} else {
309		addr = start;
310	}
311
312	/*
313	 * Forcibly align the final address in case we have a hardware
314	 * issue that requires alignment for performance reasons.
315	 */
316	addr = align_vdso_addr(addr);
317
318	return addr;
319}
320
321static int map_vdso_randomized(const struct vdso_image *image)
322{
323	unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
324
325	return map_vdso(image, addr);
326}
327#endif
328
329int map_vdso_once(const struct vdso_image *image, unsigned long addr)
330{
331	struct mm_struct *mm = current->mm;
332	struct vm_area_struct *vma;
333	VMA_ITERATOR(vmi, mm, 0);
334
335	mmap_write_lock(mm);
336	/*
337	 * Check if we have already mapped vdso blob - fail to prevent
338	 * abusing from userspace install_special_mapping, which may
339	 * not do accounting and rlimit right.
340	 * We could search vma near context.vdso, but it's a slowpath,
341	 * so let's explicitly check all VMAs to be completely sure.
342	 */
343	for_each_vma(vmi, vma) {
344		if (vma_is_special_mapping(vma, &vdso_mapping) ||
345				vma_is_special_mapping(vma, &vvar_mapping)) {
346			mmap_write_unlock(mm);
347			return -EEXIST;
348		}
349	}
350	mmap_write_unlock(mm);
351
352	return map_vdso(image, addr);
353}
354
355#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
356static int load_vdso32(void)
357{
358	if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
359		return 0;
360
361	return map_vdso(&vdso_image_32, 0);
362}
363#endif
364
365#ifdef CONFIG_X86_64
366int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
367{
368	if (!vdso64_enabled)
369		return 0;
370
371	return map_vdso_randomized(&vdso_image_64);
372}
373
374#ifdef CONFIG_COMPAT
375int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
376				       int uses_interp, bool x32)
377{
378#ifdef CONFIG_X86_X32_ABI
379	if (x32) {
380		if (!vdso64_enabled)
381			return 0;
382		return map_vdso_randomized(&vdso_image_x32);
 
383	}
384#endif
385#ifdef CONFIG_IA32_EMULATION
386	return load_vdso32();
387#else
388	return 0;
389#endif
390}
391#endif
392#else
393int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
394{
395	return load_vdso32();
396}
397#endif
398
399bool arch_syscall_is_vdso_sigreturn(struct pt_regs *regs)
 
400{
401#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
402	const struct vdso_image *image = current->mm->context.vdso_image;
403	unsigned long vdso = (unsigned long) current->mm->context.vdso;
 
 
404
405	if (in_ia32_syscall() && image == &vdso_image_32) {
406		if (regs->ip == vdso + image->sym_vdso32_sigreturn_landing_pad ||
407		    regs->ip == vdso + image->sym_vdso32_rt_sigreturn_landing_pad)
408			return true;
409	}
 
 
 
410#endif
411	return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
412}
413
414#ifdef CONFIG_X86_64
415static __init int vdso_setup(char *s)
416{
417	vdso64_enabled = simple_strtoul(s, NULL, 0);
418	return 1;
 
 
 
 
419}
420__setup("vdso=", vdso_setup);
421
422static int __init init_vdso(void)
423{
424	BUILD_BUG_ON(VDSO_CLOCKMODE_MAX >= 32);
425
426	init_vdso_image(&vdso_image_64);
427
428#ifdef CONFIG_X86_X32_ABI
429	init_vdso_image(&vdso_image_x32);
430#endif
 
 
 
 
 
 
 
 
431
432	return 0;
433}
434subsys_initcall(init_vdso);
435#endif /* CONFIG_X86_64 */
v4.6
 
  1/*
  2 * Copyright 2007 Andi Kleen, SUSE Labs.
  3 * Subject to the GPL, v.2
  4 *
  5 * This contains most of the x86 vDSO kernel-side code.
  6 */
  7#include <linux/mm.h>
  8#include <linux/err.h>
  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>
 18#include <asm/vdso.h>
 19#include <asm/vvar.h>
 
 20#include <asm/page.h>
 21#include <asm/hpet.h>
 22#include <asm/desc.h>
 23#include <asm/cpufeature.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 24
 25#if defined(CONFIG_X86_64)
 26unsigned int __read_mostly vdso64_enabled = 1;
 27#endif
 28
 29void __init init_vdso_image(const struct vdso_image *image)
 30{
 31	BUG_ON(image->size % PAGE_SIZE != 0);
 32
 33	apply_alternatives((struct alt_instr *)(image->data + image->alt),
 34			   (struct alt_instr *)(image->data + image->alt +
 35						image->alt_len));
 36}
 37
 
 38struct linux_binprm;
 39
 40/*
 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.
 45 *
 46 * Note that this algorithm is imperfect: the distribution of the vdso
 47 * start address within a PMD is biased toward the end.
 48 *
 49 * Only used for the 64-bit and x32 vdsos.
 50 */
 51static unsigned long vdso_addr(unsigned long start, unsigned len)
 52{
 53#ifdef CONFIG_X86_32
 54	return 0;
 55#else
 56	unsigned long addr, end;
 57	unsigned offset;
 58
 59	/*
 60	 * Round up the start address.  It can start out unaligned as a result
 61	 * of stack start randomization.
 62	 */
 63	start = PAGE_ALIGN(start);
 64
 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)
 68		end = TASK_SIZE_MAX;
 69	end -= len;
 70
 71	if (end > start) {
 72		offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
 73		addr = start + (offset << PAGE_SHIFT);
 74	} else {
 75		addr = start;
 
 
 
 
 
 
 
 
 76	}
 
 
 
 
 
 
 
 77
 78	/*
 79	 * Forcibly align the final address in case we have a hardware
 80	 * issue that requires alignment for performance reasons.
 81	 */
 82	addr = align_vdso_addr(addr);
 83
 84	return addr;
 85#endif
 86}
 87
 88static int vdso_fault(const struct vm_special_mapping *sm,
 89		      struct vm_area_struct *vma, struct vm_fault *vmf)
 
 
 
 
 
 
 
 90{
 91	const struct vdso_image *image = vma->vm_mm->context.vdso_image;
 
 
 
 
 
 
 92
 93	if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
 94		return VM_FAULT_SIGBUS;
 
 
 95
 96	vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
 97	get_page(vmf->page);
 98	return 0;
 99}
 
100
101static const struct vm_special_mapping text_mapping = {
102	.name = "[vdso]",
103	.fault = vdso_fault,
104};
105
106static int vvar_fault(const struct vm_special_mapping *sm,
107		      struct vm_area_struct *vma, struct vm_fault *vmf)
108{
109	const struct vdso_image *image = vma->vm_mm->context.vdso_image;
 
110	long sym_offset;
111	int ret = -EFAULT;
112
113	if (!image)
114		return VM_FAULT_SIGBUS;
115
116	sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
117		image->sym_vvar_start;
118
119	/*
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.
125	 */
126	if (sym_offset == 0)
127		return VM_FAULT_SIGBUS;
128
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(
136				vma,
137				(unsigned long)vmf->virtual_address,
138				hpet_address >> PAGE_SHIFT,
139				pgprot_noncached(PAGE_READONLY));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
140		}
141#endif
 
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)) {
146			ret = vm_insert_pfn(
147				vma,
148				(unsigned long)vmf->virtual_address,
149				__pa(pvti) >> PAGE_SHIFT);
150		}
 
 
 
 
 
 
 
 
 
 
 
 
 
151	}
152
153	if (ret == 0 || ret == -EBUSY)
154		return VM_FAULT_NOPAGE;
155
156	return VM_FAULT_SIGBUS;
157}
158
159static int map_vdso(const struct vdso_image *image, bool calculate_addr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
160{
161	struct mm_struct *mm = current->mm;
162	struct vm_area_struct *vma;
163	unsigned long addr, text_start;
164	int ret = 0;
165	static const struct vm_special_mapping vvar_mapping = {
166		.name = "[vvar]",
167		.fault = vvar_fault,
168	};
169
170	if (calculate_addr) {
171		addr = vdso_addr(current->mm->start_stack,
172				 image->size - image->sym_vvar_start);
173	} else {
174		addr = 0;
175	}
176
177	down_write(&mm->mmap_sem);
 
178
179	addr = get_unmapped_area(NULL, addr,
180				 image->size - image->sym_vvar_start, 0, 0);
181	if (IS_ERR_VALUE(addr)) {
182		ret = addr;
183		goto up_fail;
184	}
185
186	text_start = addr - image->sym_vvar_start;
187	current->mm->context.vdso = (void __user *)text_start;
188	current->mm->context.vdso_image = image;
189
190	/*
191	 * MAYWRITE to allow gdb to COW and set breakpoints
192	 */
193	vma = _install_special_mapping(mm,
194				       text_start,
195				       image->size,
196				       VM_READ|VM_EXEC|
197				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
198				       &text_mapping);
199
200	if (IS_ERR(vma)) {
201		ret = PTR_ERR(vma);
202		goto up_fail;
203	}
204
205	vma = _install_special_mapping(mm,
206				       addr,
207				       -image->sym_vvar_start,
208				       VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
209				       VM_PFNMAP,
210				       &vvar_mapping);
211
212	if (IS_ERR(vma)) {
213		ret = PTR_ERR(vma);
214		goto up_fail;
 
 
 
215	}
216
217up_fail:
218	if (ret)
219		current->mm->context.vdso = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
220
221	up_write(&mm->mmap_sem);
222	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
223}
224
225#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
226static int load_vdso32(void)
227{
228	if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
229		return 0;
230
231	return map_vdso(&vdso_image_32, false);
232}
233#endif
234
235#ifdef CONFIG_X86_64
236int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
237{
238	if (!vdso64_enabled)
239		return 0;
240
241	return map_vdso(&vdso_image_64, true);
242}
243
244#ifdef CONFIG_COMPAT
245int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
246				       int uses_interp)
247{
248#ifdef CONFIG_X86_X32_ABI
249	if (test_thread_flag(TIF_X32)) {
250		if (!vdso64_enabled)
251			return 0;
252
253		return map_vdso(&vdso_image_x32, true);
254	}
255#endif
256#ifdef CONFIG_IA32_EMULATION
257	return load_vdso32();
258#else
259	return 0;
260#endif
261}
262#endif
263#else
264int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
265{
266	return load_vdso32();
267}
268#endif
269
270#ifdef CONFIG_X86_64
271static __init int vdso_setup(char *s)
272{
273	vdso64_enabled = simple_strtoul(s, NULL, 0);
274	return 0;
275}
276__setup("vdso=", vdso_setup);
277#endif
278
279#ifdef CONFIG_X86_64
280static void vgetcpu_cpu_init(void *arg)
281{
282	int cpu = smp_processor_id();
283	struct desc_struct d = { };
284	unsigned long node = 0;
285#ifdef CONFIG_NUMA
286	node = cpu_to_node(cpu);
287#endif
288	if (static_cpu_has(X86_FEATURE_RDTSCP))
289		write_rdtscp_aux((node << 12) | cpu);
290
291	/*
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)
295	 */
296	d.limit0 = cpu | ((node & 0xf) << 12);
297	d.limit = node >> 4;
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 */
303
304	write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
305}
306
307static int
308vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
309{
310	long cpu = (long)arg;
311
312	if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
313		smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
314
315	return NOTIFY_DONE;
316}
 
317
318static int __init init_vdso(void)
319{
 
 
320	init_vdso_image(&vdso_image_64);
321
322#ifdef CONFIG_X86_X32_ABI
323	init_vdso_image(&vdso_image_x32);
324#endif
325
326	cpu_notifier_register_begin();
327
328	on_each_cpu(vgetcpu_cpu_init, NULL, 1);
329	/* notifier priority > KVM */
330	__hotcpu_notifier(vgetcpu_cpu_notifier, 30);
331
332	cpu_notifier_register_done();
333
334	return 0;
335}
336subsys_initcall(init_vdso);
337#endif /* CONFIG_X86_64 */