xref: /openbmc/linux/arch/s390/kernel/vdso.c (revision 37744fee)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * vdso setup for s390
4  *
5  *  Copyright IBM Corp. 2008
6  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
7  */
8 
9 #include <linux/init.h>
10 #include <linux/errno.h>
11 #include <linux/sched.h>
12 #include <linux/kernel.h>
13 #include <linux/mm.h>
14 #include <linux/smp.h>
15 #include <linux/stddef.h>
16 #include <linux/unistd.h>
17 #include <linux/slab.h>
18 #include <linux/user.h>
19 #include <linux/elf.h>
20 #include <linux/security.h>
21 #include <linux/memblock.h>
22 #include <linux/compat.h>
23 #include <asm/asm-offsets.h>
24 #include <asm/pgtable.h>
25 #include <asm/processor.h>
26 #include <asm/mmu.h>
27 #include <asm/mmu_context.h>
28 #include <asm/sections.h>
29 #include <asm/vdso.h>
30 #include <asm/facility.h>
31 
32 extern char vdso64_start, vdso64_end;
33 static void *vdso64_kbase = &vdso64_start;
34 static unsigned int vdso64_pages;
35 static struct page **vdso64_pagelist;
36 
37 /*
38  * Should the kernel map a VDSO page into processes and pass its
39  * address down to glibc upon exec()?
40  */
41 unsigned int __read_mostly vdso_enabled = 1;
42 
43 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
44 		      struct vm_area_struct *vma, struct vm_fault *vmf)
45 {
46 	struct page **vdso_pagelist;
47 	unsigned long vdso_pages;
48 
49 	vdso_pagelist = vdso64_pagelist;
50 	vdso_pages = vdso64_pages;
51 
52 	if (vmf->pgoff >= vdso_pages)
53 		return VM_FAULT_SIGBUS;
54 
55 	vmf->page = vdso_pagelist[vmf->pgoff];
56 	get_page(vmf->page);
57 	return 0;
58 }
59 
60 static int vdso_mremap(const struct vm_special_mapping *sm,
61 		       struct vm_area_struct *vma)
62 {
63 	unsigned long vdso_pages;
64 
65 	vdso_pages = vdso64_pages;
66 
67 	if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
68 		return -EINVAL;
69 
70 	if (WARN_ON_ONCE(current->mm != vma->vm_mm))
71 		return -EFAULT;
72 
73 	current->mm->context.vdso_base = vma->vm_start;
74 	return 0;
75 }
76 
77 static const struct vm_special_mapping vdso_mapping = {
78 	.name = "[vdso]",
79 	.fault = vdso_fault,
80 	.mremap = vdso_mremap,
81 };
82 
83 static int __init vdso_setup(char *str)
84 {
85 	bool enabled;
86 
87 	if (!kstrtobool(str, &enabled))
88 		vdso_enabled = enabled;
89 	return 1;
90 }
91 __setup("vdso=", vdso_setup);
92 
93 /*
94  * The vdso data page
95  */
96 static union {
97 	struct vdso_data	data;
98 	u8			page[PAGE_SIZE];
99 } vdso_data_store __page_aligned_data;
100 struct vdso_data *vdso_data = &vdso_data_store.data;
101 
102 /*
103  * Setup vdso data page.
104  */
105 static void __init vdso_init_data(struct vdso_data *vd)
106 {
107 	vd->ectg_available = test_facility(31);
108 }
109 
110 /*
111  * Allocate/free per cpu vdso data.
112  */
113 #define SEGMENT_ORDER	2
114 
115 /*
116  * The initial vdso_data structure for the boot CPU. Eventually
117  * it is replaced with a properly allocated structure in vdso_init.
118  * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
119  * pointer is required to be able to return from an interrupt or
120  * program check. See the exit paths in entry.S.
121  */
122 struct vdso_data boot_vdso_data __initdata;
123 
124 void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
125 {
126 	lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
127 }
128 
129 int vdso_alloc_per_cpu(struct lowcore *lowcore)
130 {
131 	unsigned long segment_table, page_table, page_frame;
132 	struct vdso_per_cpu_data *vd;
133 
134 	segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
135 	page_table = get_zeroed_page(GFP_KERNEL);
136 	page_frame = get_zeroed_page(GFP_KERNEL);
137 	if (!segment_table || !page_table || !page_frame)
138 		goto out;
139 	arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
140 	arch_set_page_dat(virt_to_page(page_table), 0);
141 
142 	/* Initialize per-cpu vdso data page */
143 	vd = (struct vdso_per_cpu_data *) page_frame;
144 	vd->cpu_nr = lowcore->cpu_nr;
145 	vd->node_id = cpu_to_node(vd->cpu_nr);
146 
147 	/* Set up page table for the vdso address space */
148 	memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
149 	memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
150 
151 	*(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
152 	*(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
153 
154 	lowcore->vdso_asce = segment_table +
155 		_ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
156 	lowcore->vdso_per_cpu_data = page_frame;
157 
158 	return 0;
159 
160 out:
161 	free_page(page_frame);
162 	free_page(page_table);
163 	free_pages(segment_table, SEGMENT_ORDER);
164 	return -ENOMEM;
165 }
166 
167 void vdso_free_per_cpu(struct lowcore *lowcore)
168 {
169 	unsigned long segment_table, page_table, page_frame;
170 
171 	segment_table = lowcore->vdso_asce & PAGE_MASK;
172 	page_table = *(unsigned long *) segment_table;
173 	page_frame = *(unsigned long *) page_table;
174 
175 	free_page(page_frame);
176 	free_page(page_table);
177 	free_pages(segment_table, SEGMENT_ORDER);
178 }
179 
180 /*
181  * This is called from binfmt_elf, we create the special vma for the
182  * vDSO and insert it into the mm struct tree
183  */
184 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
185 {
186 	struct mm_struct *mm = current->mm;
187 	struct vm_area_struct *vma;
188 	unsigned long vdso_pages;
189 	unsigned long vdso_base;
190 	int rc;
191 
192 	if (!vdso_enabled)
193 		return 0;
194 
195 	if (is_compat_task())
196 		return 0;
197 
198 	vdso_pages = vdso64_pages;
199 	/*
200 	 * vDSO has a problem and was disabled, just don't "enable" it for
201 	 * the process
202 	 */
203 	if (vdso_pages == 0)
204 		return 0;
205 
206 	/*
207 	 * pick a base address for the vDSO in process space. We try to put
208 	 * it at vdso_base which is the "natural" base for it, but we might
209 	 * fail and end up putting it elsewhere.
210 	 */
211 	if (down_write_killable(&mm->mmap_sem))
212 		return -EINTR;
213 	vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
214 	if (IS_ERR_VALUE(vdso_base)) {
215 		rc = vdso_base;
216 		goto out_up;
217 	}
218 
219 	/*
220 	 * our vma flags don't have VM_WRITE so by default, the process
221 	 * isn't allowed to write those pages.
222 	 * gdb can break that with ptrace interface, and thus trigger COW
223 	 * on those pages but it's then your responsibility to never do that
224 	 * on the "data" page of the vDSO or you'll stop getting kernel
225 	 * updates and your nice userland gettimeofday will be totally dead.
226 	 * It's fine to use that for setting breakpoints in the vDSO code
227 	 * pages though.
228 	 */
229 	vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
230 				       VM_READ|VM_EXEC|
231 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
232 				       &vdso_mapping);
233 	if (IS_ERR(vma)) {
234 		rc = PTR_ERR(vma);
235 		goto out_up;
236 	}
237 
238 	current->mm->context.vdso_base = vdso_base;
239 	rc = 0;
240 
241 out_up:
242 	up_write(&mm->mmap_sem);
243 	return rc;
244 }
245 
246 static int __init vdso_init(void)
247 {
248 	int i;
249 
250 	vdso_init_data(vdso_data);
251 
252 	/* Calculate the size of the 64 bit vDSO */
253 	vdso64_pages = ((&vdso64_end - &vdso64_start
254 			 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
255 
256 	/* Make sure pages are in the correct state */
257 	vdso64_pagelist = kcalloc(vdso64_pages + 1, sizeof(struct page *),
258 				  GFP_KERNEL);
259 	BUG_ON(vdso64_pagelist == NULL);
260 	for (i = 0; i < vdso64_pages - 1; i++) {
261 		struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
262 		get_page(pg);
263 		vdso64_pagelist[i] = pg;
264 	}
265 	vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
266 	vdso64_pagelist[vdso64_pages] = NULL;
267 	if (vdso_alloc_per_cpu(&S390_lowcore))
268 		BUG();
269 
270 	get_page(virt_to_page(vdso_data));
271 
272 	return 0;
273 }
274 early_initcall(vdso_init);
275