s390/vdso: don't clear PG_reserved
[sfrench/cifs-2.6.git] / arch / s390 / kernel / vdso.c
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 #ifdef CONFIG_COMPAT
33 extern char vdso32_start, vdso32_end;
34 static void *vdso32_kbase = &vdso32_start;
35 static unsigned int vdso32_pages;
36 static struct page **vdso32_pagelist;
37 #endif
38
39 extern char vdso64_start, vdso64_end;
40 static void *vdso64_kbase = &vdso64_start;
41 static unsigned int vdso64_pages;
42 static struct page **vdso64_pagelist;
43
44 /*
45  * Should the kernel map a VDSO page into processes and pass its
46  * address down to glibc upon exec()?
47  */
48 unsigned int __read_mostly vdso_enabled = 1;
49
50 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
51                       struct vm_area_struct *vma, struct vm_fault *vmf)
52 {
53         struct page **vdso_pagelist;
54         unsigned long vdso_pages;
55
56         vdso_pagelist = vdso64_pagelist;
57         vdso_pages = vdso64_pages;
58 #ifdef CONFIG_COMPAT
59         if (vma->vm_mm->context.compat_mm) {
60                 vdso_pagelist = vdso32_pagelist;
61                 vdso_pages = vdso32_pages;
62         }
63 #endif
64
65         if (vmf->pgoff >= vdso_pages)
66                 return VM_FAULT_SIGBUS;
67
68         vmf->page = vdso_pagelist[vmf->pgoff];
69         get_page(vmf->page);
70         return 0;
71 }
72
73 static int vdso_mremap(const struct vm_special_mapping *sm,
74                        struct vm_area_struct *vma)
75 {
76         unsigned long vdso_pages;
77
78         vdso_pages = vdso64_pages;
79 #ifdef CONFIG_COMPAT
80         if (vma->vm_mm->context.compat_mm)
81                 vdso_pages = vdso32_pages;
82 #endif
83
84         if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
85                 return -EINVAL;
86
87         if (WARN_ON_ONCE(current->mm != vma->vm_mm))
88                 return -EFAULT;
89
90         current->mm->context.vdso_base = vma->vm_start;
91         return 0;
92 }
93
94 static const struct vm_special_mapping vdso_mapping = {
95         .name = "[vdso]",
96         .fault = vdso_fault,
97         .mremap = vdso_mremap,
98 };
99
100 static int __init vdso_setup(char *s)
101 {
102         unsigned long val;
103         int rc;
104
105         rc = 0;
106         if (strncmp(s, "on", 3) == 0)
107                 vdso_enabled = 1;
108         else if (strncmp(s, "off", 4) == 0)
109                 vdso_enabled = 0;
110         else {
111                 rc = kstrtoul(s, 0, &val);
112                 vdso_enabled = rc ? 0 : !!val;
113         }
114         return !rc;
115 }
116 __setup("vdso=", vdso_setup);
117
118 /*
119  * The vdso data page
120  */
121 static union {
122         struct vdso_data        data;
123         u8                      page[PAGE_SIZE];
124 } vdso_data_store __page_aligned_data;
125 struct vdso_data *vdso_data = &vdso_data_store.data;
126
127 /*
128  * Setup vdso data page.
129  */
130 static void __init vdso_init_data(struct vdso_data *vd)
131 {
132         vd->ectg_available = test_facility(31);
133 }
134
135 /*
136  * Allocate/free per cpu vdso data.
137  */
138 #define SEGMENT_ORDER   2
139
140 /*
141  * The initial vdso_data structure for the boot CPU. Eventually
142  * it is replaced with a properly allocated structure in vdso_init.
143  * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
144  * pointer is required to be able to return from an interrupt or
145  * program check. See the exit paths in entry.S.
146  */
147 struct vdso_data boot_vdso_data __initdata;
148
149 void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
150 {
151         lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
152 }
153
154 int vdso_alloc_per_cpu(struct lowcore *lowcore)
155 {
156         unsigned long segment_table, page_table, page_frame;
157         struct vdso_per_cpu_data *vd;
158
159         segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
160         page_table = get_zeroed_page(GFP_KERNEL);
161         page_frame = get_zeroed_page(GFP_KERNEL);
162         if (!segment_table || !page_table || !page_frame)
163                 goto out;
164         arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
165         arch_set_page_dat(virt_to_page(page_table), 0);
166
167         /* Initialize per-cpu vdso data page */
168         vd = (struct vdso_per_cpu_data *) page_frame;
169         vd->cpu_nr = lowcore->cpu_nr;
170         vd->node_id = cpu_to_node(vd->cpu_nr);
171
172         /* Set up page table for the vdso address space */
173         memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
174         memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
175
176         *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
177         *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
178
179         lowcore->vdso_asce = segment_table +
180                 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
181         lowcore->vdso_per_cpu_data = page_frame;
182
183         return 0;
184
185 out:
186         free_page(page_frame);
187         free_page(page_table);
188         free_pages(segment_table, SEGMENT_ORDER);
189         return -ENOMEM;
190 }
191
192 void vdso_free_per_cpu(struct lowcore *lowcore)
193 {
194         unsigned long segment_table, page_table, page_frame;
195
196         segment_table = lowcore->vdso_asce & PAGE_MASK;
197         page_table = *(unsigned long *) segment_table;
198         page_frame = *(unsigned long *) page_table;
199
200         free_page(page_frame);
201         free_page(page_table);
202         free_pages(segment_table, SEGMENT_ORDER);
203 }
204
205 /*
206  * This is called from binfmt_elf, we create the special vma for the
207  * vDSO and insert it into the mm struct tree
208  */
209 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
210 {
211         struct mm_struct *mm = current->mm;
212         struct vm_area_struct *vma;
213         unsigned long vdso_pages;
214         unsigned long vdso_base;
215         int rc;
216
217         if (!vdso_enabled)
218                 return 0;
219         /*
220          * Only map the vdso for dynamically linked elf binaries.
221          */
222         if (!uses_interp)
223                 return 0;
224
225         vdso_pages = vdso64_pages;
226 #ifdef CONFIG_COMPAT
227         mm->context.compat_mm = is_compat_task();
228         if (mm->context.compat_mm)
229                 vdso_pages = vdso32_pages;
230 #endif
231         /*
232          * vDSO has a problem and was disabled, just don't "enable" it for
233          * the process
234          */
235         if (vdso_pages == 0)
236                 return 0;
237
238         /*
239          * pick a base address for the vDSO in process space. We try to put
240          * it at vdso_base which is the "natural" base for it, but we might
241          * fail and end up putting it elsewhere.
242          */
243         if (down_write_killable(&mm->mmap_sem))
244                 return -EINTR;
245         vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
246         if (IS_ERR_VALUE(vdso_base)) {
247                 rc = vdso_base;
248                 goto out_up;
249         }
250
251         /*
252          * our vma flags don't have VM_WRITE so by default, the process
253          * isn't allowed to write those pages.
254          * gdb can break that with ptrace interface, and thus trigger COW
255          * on those pages but it's then your responsibility to never do that
256          * on the "data" page of the vDSO or you'll stop getting kernel
257          * updates and your nice userland gettimeofday will be totally dead.
258          * It's fine to use that for setting breakpoints in the vDSO code
259          * pages though.
260          */
261         vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
262                                        VM_READ|VM_EXEC|
263                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
264                                        &vdso_mapping);
265         if (IS_ERR(vma)) {
266                 rc = PTR_ERR(vma);
267                 goto out_up;
268         }
269
270         current->mm->context.vdso_base = vdso_base;
271         rc = 0;
272
273 out_up:
274         up_write(&mm->mmap_sem);
275         return rc;
276 }
277
278 static int __init vdso_init(void)
279 {
280         int i;
281
282         vdso_init_data(vdso_data);
283 #ifdef CONFIG_COMPAT
284         /* Calculate the size of the 32 bit vDSO */
285         vdso32_pages = ((&vdso32_end - &vdso32_start
286                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
287
288         /* Make sure pages are in the correct state */
289         vdso32_pagelist = kcalloc(vdso32_pages + 1, sizeof(struct page *),
290                                   GFP_KERNEL);
291         BUG_ON(vdso32_pagelist == NULL);
292         for (i = 0; i < vdso32_pages - 1; i++) {
293                 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
294                 get_page(pg);
295                 vdso32_pagelist[i] = pg;
296         }
297         vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
298         vdso32_pagelist[vdso32_pages] = NULL;
299 #endif
300
301         /* Calculate the size of the 64 bit vDSO */
302         vdso64_pages = ((&vdso64_end - &vdso64_start
303                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
304
305         /* Make sure pages are in the correct state */
306         vdso64_pagelist = kcalloc(vdso64_pages + 1, sizeof(struct page *),
307                                   GFP_KERNEL);
308         BUG_ON(vdso64_pagelist == NULL);
309         for (i = 0; i < vdso64_pages - 1; i++) {
310                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
311                 get_page(pg);
312                 vdso64_pagelist[i] = pg;
313         }
314         vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
315         vdso64_pagelist[vdso64_pages] = NULL;
316         if (vdso_alloc_per_cpu(&S390_lowcore))
317                 BUG();
318
319         get_page(virt_to_page(vdso_data));
320
321         return 0;
322 }
323 early_initcall(vdso_init);