xref: /linux/arch/s390/kernel/vdso.c (revision 43347d56c8d9dd732cee2f8efd384ad21dd1f6c4)
1 /*
2  * vdso setup for s390
3  *
4  *  Copyright IBM Corp. 2008
5  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License (version 2 only)
9  * as published by the Free Software Foundation.
10  */
11 
12 #include <linux/init.h>
13 #include <linux/errno.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/mm.h>
17 #include <linux/smp.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/slab.h>
21 #include <linux/user.h>
22 #include <linux/elf.h>
23 #include <linux/security.h>
24 #include <linux/bootmem.h>
25 #include <linux/compat.h>
26 #include <asm/asm-offsets.h>
27 #include <asm/pgtable.h>
28 #include <asm/processor.h>
29 #include <asm/mmu.h>
30 #include <asm/mmu_context.h>
31 #include <asm/sections.h>
32 #include <asm/vdso.h>
33 #include <asm/facility.h>
34 
35 #ifdef CONFIG_COMPAT
36 extern char vdso32_start, vdso32_end;
37 static void *vdso32_kbase = &vdso32_start;
38 static unsigned int vdso32_pages;
39 static struct page **vdso32_pagelist;
40 #endif
41 
42 extern char vdso64_start, vdso64_end;
43 static void *vdso64_kbase = &vdso64_start;
44 static unsigned int vdso64_pages;
45 static struct page **vdso64_pagelist;
46 
47 /*
48  * Should the kernel map a VDSO page into processes and pass its
49  * address down to glibc upon exec()?
50  */
51 unsigned int __read_mostly vdso_enabled = 1;
52 
53 static int vdso_fault(const struct vm_special_mapping *sm,
54 		      struct vm_area_struct *vma, struct vm_fault *vmf)
55 {
56 	struct page **vdso_pagelist;
57 	unsigned long vdso_pages;
58 
59 	vdso_pagelist = vdso64_pagelist;
60 	vdso_pages = vdso64_pages;
61 #ifdef CONFIG_COMPAT
62 	if (is_compat_task()) {
63 		vdso_pagelist = vdso32_pagelist;
64 		vdso_pages = vdso32_pages;
65 	}
66 #endif
67 
68 	if (vmf->pgoff >= vdso_pages)
69 		return VM_FAULT_SIGBUS;
70 
71 	vmf->page = vdso_pagelist[vmf->pgoff];
72 	get_page(vmf->page);
73 	return 0;
74 }
75 
76 static int vdso_mremap(const struct vm_special_mapping *sm,
77 		       struct vm_area_struct *vma)
78 {
79 	unsigned long vdso_pages;
80 
81 	vdso_pages = vdso64_pages;
82 #ifdef CONFIG_COMPAT
83 	if (is_compat_task())
84 		vdso_pages = vdso32_pages;
85 #endif
86 
87 	if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
88 		return -EINVAL;
89 
90 	if (WARN_ON_ONCE(current->mm != vma->vm_mm))
91 		return -EFAULT;
92 
93 	current->mm->context.vdso_base = vma->vm_start;
94 	return 0;
95 }
96 
97 static const struct vm_special_mapping vdso_mapping = {
98 	.name = "[vdso]",
99 	.fault = vdso_fault,
100 	.mremap = vdso_mremap,
101 };
102 
103 static int __init vdso_setup(char *s)
104 {
105 	unsigned long val;
106 	int rc;
107 
108 	rc = 0;
109 	if (strncmp(s, "on", 3) == 0)
110 		vdso_enabled = 1;
111 	else if (strncmp(s, "off", 4) == 0)
112 		vdso_enabled = 0;
113 	else {
114 		rc = kstrtoul(s, 0, &val);
115 		vdso_enabled = rc ? 0 : !!val;
116 	}
117 	return !rc;
118 }
119 __setup("vdso=", vdso_setup);
120 
121 /*
122  * The vdso data page
123  */
124 static union {
125 	struct vdso_data	data;
126 	u8			page[PAGE_SIZE];
127 } vdso_data_store __page_aligned_data;
128 struct vdso_data *vdso_data = &vdso_data_store.data;
129 
130 /*
131  * Setup vdso data page.
132  */
133 static void __init vdso_init_data(struct vdso_data *vd)
134 {
135 	vd->ectg_available = test_facility(31);
136 }
137 
138 /*
139  * Allocate/free per cpu vdso data.
140  */
141 #define SEGMENT_ORDER	2
142 
143 /*
144  * The initial vdso_data structure for the boot CPU. Eventually
145  * it is replaced with a properly allocated structure in vdso_init.
146  * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
147  * pointer is required to be able to return from an interrupt or
148  * program check. See the exit paths in entry.S.
149  */
150 struct vdso_data boot_vdso_data __initdata;
151 
152 void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
153 {
154 	lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
155 }
156 
157 int vdso_alloc_per_cpu(struct lowcore *lowcore)
158 {
159 	unsigned long segment_table, page_table, page_frame;
160 	struct vdso_per_cpu_data *vd;
161 	u32 *psal, *aste;
162 	int i;
163 
164 	lowcore->vdso_per_cpu_data = __LC_PASTE;
165 
166 	if (!vdso_enabled)
167 		return 0;
168 
169 	segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
170 	page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
171 	page_frame = get_zeroed_page(GFP_KERNEL);
172 	if (!segment_table || !page_table || !page_frame)
173 		goto out;
174 	arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
175 	arch_set_page_dat(virt_to_page(page_table), 0);
176 
177 	/* Initialize per-cpu vdso data page */
178 	vd = (struct vdso_per_cpu_data *) page_frame;
179 	vd->cpu_nr = lowcore->cpu_nr;
180 	vd->node_id = cpu_to_node(vd->cpu_nr);
181 
182 	/* Set up access register mode page table */
183 	memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
184 	memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
185 
186 	*(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
187 	*(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
188 
189 	psal = (u32 *) (page_table + 256*sizeof(unsigned long));
190 	aste = psal + 32;
191 
192 	for (i = 4; i < 32; i += 4)
193 		psal[i] = 0x80000000;
194 
195 	lowcore->paste[4] = (u32)(addr_t) psal;
196 	psal[0] = 0x02000000;
197 	psal[2] = (u32)(addr_t) aste;
198 	*(unsigned long *) (aste + 2) = segment_table +
199 		_ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
200 	aste[4] = (u32)(addr_t) psal;
201 	lowcore->vdso_per_cpu_data = page_frame;
202 
203 	return 0;
204 
205 out:
206 	free_page(page_frame);
207 	free_page(page_table);
208 	free_pages(segment_table, SEGMENT_ORDER);
209 	return -ENOMEM;
210 }
211 
212 void vdso_free_per_cpu(struct lowcore *lowcore)
213 {
214 	unsigned long segment_table, page_table, page_frame;
215 	u32 *psal, *aste;
216 
217 	if (!vdso_enabled)
218 		return;
219 
220 	psal = (u32 *)(addr_t) lowcore->paste[4];
221 	aste = (u32 *)(addr_t) psal[2];
222 	segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
223 	page_table = *(unsigned long *) segment_table;
224 	page_frame = *(unsigned long *) page_table;
225 
226 	free_page(page_frame);
227 	free_page(page_table);
228 	free_pages(segment_table, SEGMENT_ORDER);
229 }
230 
231 static void vdso_init_cr5(void)
232 {
233 	unsigned long cr5;
234 
235 	if (!vdso_enabled)
236 		return;
237 	cr5 = offsetof(struct lowcore, paste);
238 	__ctl_load(cr5, 5, 5);
239 }
240 
241 /*
242  * This is called from binfmt_elf, we create the special vma for the
243  * vDSO and insert it into the mm struct tree
244  */
245 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
246 {
247 	struct mm_struct *mm = current->mm;
248 	struct vm_area_struct *vma;
249 	unsigned long vdso_pages;
250 	unsigned long vdso_base;
251 	int rc;
252 
253 	if (!vdso_enabled)
254 		return 0;
255 	/*
256 	 * Only map the vdso for dynamically linked elf binaries.
257 	 */
258 	if (!uses_interp)
259 		return 0;
260 
261 	vdso_pages = vdso64_pages;
262 #ifdef CONFIG_COMPAT
263 	if (is_compat_task())
264 		vdso_pages = vdso32_pages;
265 #endif
266 	/*
267 	 * vDSO has a problem and was disabled, just don't "enable" it for
268 	 * the process
269 	 */
270 	if (vdso_pages == 0)
271 		return 0;
272 
273 	/*
274 	 * pick a base address for the vDSO in process space. We try to put
275 	 * it at vdso_base which is the "natural" base for it, but we might
276 	 * fail and end up putting it elsewhere.
277 	 */
278 	if (down_write_killable(&mm->mmap_sem))
279 		return -EINTR;
280 	vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
281 	if (IS_ERR_VALUE(vdso_base)) {
282 		rc = vdso_base;
283 		goto out_up;
284 	}
285 
286 	/*
287 	 * our vma flags don't have VM_WRITE so by default, the process
288 	 * isn't allowed to write those pages.
289 	 * gdb can break that with ptrace interface, and thus trigger COW
290 	 * on those pages but it's then your responsibility to never do that
291 	 * on the "data" page of the vDSO or you'll stop getting kernel
292 	 * updates and your nice userland gettimeofday will be totally dead.
293 	 * It's fine to use that for setting breakpoints in the vDSO code
294 	 * pages though.
295 	 */
296 	vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
297 				       VM_READ|VM_EXEC|
298 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
299 				       &vdso_mapping);
300 	if (IS_ERR(vma)) {
301 		rc = PTR_ERR(vma);
302 		goto out_up;
303 	}
304 
305 	current->mm->context.vdso_base = vdso_base;
306 	rc = 0;
307 
308 out_up:
309 	up_write(&mm->mmap_sem);
310 	return rc;
311 }
312 
313 static int __init vdso_init(void)
314 {
315 	int i;
316 
317 	if (!vdso_enabled)
318 		return 0;
319 	vdso_init_data(vdso_data);
320 #ifdef CONFIG_COMPAT
321 	/* Calculate the size of the 32 bit vDSO */
322 	vdso32_pages = ((&vdso32_end - &vdso32_start
323 			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
324 
325 	/* Make sure pages are in the correct state */
326 	vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
327 				  GFP_KERNEL);
328 	BUG_ON(vdso32_pagelist == NULL);
329 	for (i = 0; i < vdso32_pages - 1; i++) {
330 		struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
331 		ClearPageReserved(pg);
332 		get_page(pg);
333 		vdso32_pagelist[i] = pg;
334 	}
335 	vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
336 	vdso32_pagelist[vdso32_pages] = NULL;
337 #endif
338 
339 	/* Calculate the size of the 64 bit vDSO */
340 	vdso64_pages = ((&vdso64_end - &vdso64_start
341 			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
342 
343 	/* Make sure pages are in the correct state */
344 	vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
345 				  GFP_KERNEL);
346 	BUG_ON(vdso64_pagelist == NULL);
347 	for (i = 0; i < vdso64_pages - 1; i++) {
348 		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
349 		ClearPageReserved(pg);
350 		get_page(pg);
351 		vdso64_pagelist[i] = pg;
352 	}
353 	vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
354 	vdso64_pagelist[vdso64_pages] = NULL;
355 	if (vdso_alloc_per_cpu(&S390_lowcore))
356 		BUG();
357 	vdso_init_cr5();
358 
359 	get_page(virt_to_page(vdso_data));
360 
361 	return 0;
362 }
363 early_initcall(vdso_init);
364