1 /*
2 * Adapted from arm64 version.
3 *
4 * Copyright (C) 2012 ARM Limited
5 * Copyright (C) 2015 Mentor Graphics Corporation.
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 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 */
19
20 #include <linux/cache.h>
21 #include <linux/elf.h>
22 #include <linux/err.h>
23 #include <linux/kernel.h>
24 #include <linux/mm.h>
25 #include <linux/of.h>
26 #include <linux/printk.h>
27 #include <linux/slab.h>
28 #include <linux/timekeeper_internal.h>
29 #include <linux/vmalloc.h>
30 #include <asm/arch_timer.h>
31 #include <asm/barrier.h>
32 #include <asm/cacheflush.h>
33 #include <asm/page.h>
34 #include <asm/vdso.h>
35 #include <asm/vdso_datapage.h>
36 #include <clocksource/arm_arch_timer.h>
37
38 #define MAX_SYMNAME 64
39
40 static struct page **vdso_text_pagelist;
41
42 extern char vdso_start[], vdso_end[];
43
44 /* Total number of pages needed for the data and text portions of the VDSO. */
45 unsigned int vdso_total_pages __ro_after_init;
46
47 /*
48 * The VDSO data page.
49 */
50 static union vdso_data_store vdso_data_store __page_aligned_data;
51 static struct vdso_data *vdso_data = &vdso_data_store.data;
52
53 static struct page *vdso_data_page __ro_after_init;
54 static const struct vm_special_mapping vdso_data_mapping = {
55 .name = "[vvar]",
56 .pages = &vdso_data_page,
57 };
58
59 static struct vm_special_mapping vdso_text_mapping __ro_after_init = {
60 .name = "[vdso]",
61 };
62
63 struct elfinfo {
64 Elf32_Ehdr *hdr; /* ptr to ELF */
65 Elf32_Sym *dynsym; /* ptr to .dynsym section */
66 unsigned long dynsymsize; /* size of .dynsym section */
67 char *dynstr; /* ptr to .dynstr section */
68 };
69
70 /* Cached result of boot-time check for whether the arch timer exists,
71 * and if so, whether the virtual counter is useable.
72 */
73 static bool cntvct_ok __ro_after_init;
74
cntvct_functional(void)75 static bool __init cntvct_functional(void)
76 {
77 struct device_node *np;
78 bool ret = false;
79
80 if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
81 goto out;
82
83 /* The arm_arch_timer core should export
84 * arch_timer_use_virtual or similar so we don't have to do
85 * this.
86 */
87 np = of_find_compatible_node(NULL, NULL, "arm,armv7-timer");
88 if (!np)
89 np = of_find_compatible_node(NULL, NULL, "arm,armv8-timer");
90 if (!np)
91 goto out_put;
92
93 if (of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
94 goto out_put;
95
96 ret = true;
97
98 out_put:
99 of_node_put(np);
100 out:
101 return ret;
102 }
103
find_section(Elf32_Ehdr * ehdr,const char * name,unsigned long * size)104 static void * __init find_section(Elf32_Ehdr *ehdr, const char *name,
105 unsigned long *size)
106 {
107 Elf32_Shdr *sechdrs;
108 unsigned int i;
109 char *secnames;
110
111 /* Grab section headers and strings so we can tell who is who */
112 sechdrs = (void *)ehdr + ehdr->e_shoff;
113 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
114
115 /* Find the section they want */
116 for (i = 1; i < ehdr->e_shnum; i++) {
117 if (strcmp(secnames + sechdrs[i].sh_name, name) == 0) {
118 if (size)
119 *size = sechdrs[i].sh_size;
120 return (void *)ehdr + sechdrs[i].sh_offset;
121 }
122 }
123
124 if (size)
125 *size = 0;
126 return NULL;
127 }
128
find_symbol(struct elfinfo * lib,const char * symname)129 static Elf32_Sym * __init find_symbol(struct elfinfo *lib, const char *symname)
130 {
131 unsigned int i;
132
133 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
134 char name[MAX_SYMNAME], *c;
135
136 if (lib->dynsym[i].st_name == 0)
137 continue;
138 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
139 MAX_SYMNAME);
140 c = strchr(name, '@');
141 if (c)
142 *c = 0;
143 if (strcmp(symname, name) == 0)
144 return &lib->dynsym[i];
145 }
146 return NULL;
147 }
148
vdso_nullpatch_one(struct elfinfo * lib,const char * symname)149 static void __init vdso_nullpatch_one(struct elfinfo *lib, const char *symname)
150 {
151 Elf32_Sym *sym;
152
153 sym = find_symbol(lib, symname);
154 if (!sym)
155 return;
156
157 sym->st_name = 0;
158 }
159
patch_vdso(void * ehdr)160 static void __init patch_vdso(void *ehdr)
161 {
162 struct elfinfo einfo;
163
164 einfo = (struct elfinfo) {
165 .hdr = ehdr,
166 };
167
168 einfo.dynsym = find_section(einfo.hdr, ".dynsym", &einfo.dynsymsize);
169 einfo.dynstr = find_section(einfo.hdr, ".dynstr", NULL);
170
171 /* If the virtual counter is absent or non-functional we don't
172 * want programs to incur the slight additional overhead of
173 * dispatching through the VDSO only to fall back to syscalls.
174 */
175 if (!cntvct_ok) {
176 vdso_nullpatch_one(&einfo, "__vdso_gettimeofday");
177 vdso_nullpatch_one(&einfo, "__vdso_clock_gettime");
178 }
179 }
180
vdso_init(void)181 static int __init vdso_init(void)
182 {
183 unsigned int text_pages;
184 int i;
185
186 if (memcmp(vdso_start, "\177ELF", 4)) {
187 pr_err("VDSO is not a valid ELF object!\n");
188 return -ENOEXEC;
189 }
190
191 text_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
192 pr_debug("vdso: %i text pages at base %p\n", text_pages, vdso_start);
193
194 /* Allocate the VDSO text pagelist */
195 vdso_text_pagelist = kcalloc(text_pages, sizeof(struct page *),
196 GFP_KERNEL);
197 if (vdso_text_pagelist == NULL)
198 return -ENOMEM;
199
200 /* Grab the VDSO data page. */
201 vdso_data_page = virt_to_page(vdso_data);
202
203 /* Grab the VDSO text pages. */
204 for (i = 0; i < text_pages; i++) {
205 struct page *page;
206
207 page = virt_to_page(vdso_start + i * PAGE_SIZE);
208 vdso_text_pagelist[i] = page;
209 }
210
211 vdso_text_mapping.pages = vdso_text_pagelist;
212
213 vdso_total_pages = 1; /* for the data/vvar page */
214 vdso_total_pages += text_pages;
215
216 cntvct_ok = cntvct_functional();
217
218 patch_vdso(vdso_start);
219
220 return 0;
221 }
222 arch_initcall(vdso_init);
223
install_vvar(struct mm_struct * mm,unsigned long addr)224 static int install_vvar(struct mm_struct *mm, unsigned long addr)
225 {
226 struct vm_area_struct *vma;
227
228 vma = _install_special_mapping(mm, addr, PAGE_SIZE,
229 VM_READ | VM_MAYREAD,
230 &vdso_data_mapping);
231
232 return PTR_ERR_OR_ZERO(vma);
233 }
234
235 /* assumes mmap_sem is write-locked */
arm_install_vdso(struct mm_struct * mm,unsigned long addr)236 void arm_install_vdso(struct mm_struct *mm, unsigned long addr)
237 {
238 struct vm_area_struct *vma;
239 unsigned long len;
240
241 mm->context.vdso = 0;
242
243 if (vdso_text_pagelist == NULL)
244 return;
245
246 if (install_vvar(mm, addr))
247 return;
248
249 /* Account for vvar page. */
250 addr += PAGE_SIZE;
251 len = (vdso_total_pages - 1) << PAGE_SHIFT;
252
253 vma = _install_special_mapping(mm, addr, len,
254 VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
255 &vdso_text_mapping);
256
257 if (!IS_ERR(vma))
258 mm->context.vdso = addr;
259 }
260
vdso_write_begin(struct vdso_data * vdata)261 static void vdso_write_begin(struct vdso_data *vdata)
262 {
263 ++vdso_data->seq_count;
264 smp_wmb(); /* Pairs with smp_rmb in vdso_read_retry */
265 }
266
vdso_write_end(struct vdso_data * vdata)267 static void vdso_write_end(struct vdso_data *vdata)
268 {
269 smp_wmb(); /* Pairs with smp_rmb in vdso_read_begin */
270 ++vdso_data->seq_count;
271 }
272
tk_is_cntvct(const struct timekeeper * tk)273 static bool tk_is_cntvct(const struct timekeeper *tk)
274 {
275 if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
276 return false;
277
278 if (strcmp(tk->tkr_mono.clock->name, "arch_sys_counter") != 0)
279 return false;
280
281 return true;
282 }
283
284 /**
285 * update_vsyscall - update the vdso data page
286 *
287 * Increment the sequence counter, making it odd, indicating to
288 * userspace that an update is in progress. Update the fields used
289 * for coarse clocks and, if the architected system timer is in use,
290 * the fields used for high precision clocks. Increment the sequence
291 * counter again, making it even, indicating to userspace that the
292 * update is finished.
293 *
294 * Userspace is expected to sample seq_count before reading any other
295 * fields from the data page. If seq_count is odd, userspace is
296 * expected to wait until it becomes even. After copying data from
297 * the page, userspace must sample seq_count again; if it has changed
298 * from its previous value, userspace must retry the whole sequence.
299 *
300 * Calls to update_vsyscall are serialized by the timekeeping core.
301 */
update_vsyscall(struct timekeeper * tk)302 void update_vsyscall(struct timekeeper *tk)
303 {
304 struct timespec64 *wtm = &tk->wall_to_monotonic;
305
306 if (!cntvct_ok) {
307 /* The entry points have been zeroed, so there is no
308 * point in updating the data page.
309 */
310 return;
311 }
312
313 vdso_write_begin(vdso_data);
314
315 vdso_data->tk_is_cntvct = tk_is_cntvct(tk);
316 vdso_data->xtime_coarse_sec = tk->xtime_sec;
317 vdso_data->xtime_coarse_nsec = (u32)(tk->tkr_mono.xtime_nsec >>
318 tk->tkr_mono.shift);
319 vdso_data->wtm_clock_sec = wtm->tv_sec;
320 vdso_data->wtm_clock_nsec = wtm->tv_nsec;
321
322 if (vdso_data->tk_is_cntvct) {
323 vdso_data->cs_cycle_last = tk->tkr_mono.cycle_last;
324 vdso_data->xtime_clock_sec = tk->xtime_sec;
325 vdso_data->xtime_clock_snsec = tk->tkr_mono.xtime_nsec;
326 vdso_data->cs_mult = tk->tkr_mono.mult;
327 vdso_data->cs_shift = tk->tkr_mono.shift;
328 vdso_data->cs_mask = tk->tkr_mono.mask;
329 }
330
331 vdso_write_end(vdso_data);
332
333 flush_dcache_page(virt_to_page(vdso_data));
334 }
335
update_vsyscall_tz(void)336 void update_vsyscall_tz(void)
337 {
338 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
339 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
340 flush_dcache_page(virt_to_page(vdso_data));
341 }
342