1 // SPDX-License-Identifier: GPL-2.0-only
3 * Adapted from arm64 version.
5 * Copyright (C) 2012 ARM Limited
6 * Copyright (C) 2015 Mentor Graphics Corporation.
9 #include <linux/cache.h>
10 #include <linux/elf.h>
11 #include <linux/err.h>
12 #include <linux/kernel.h>
15 #include <linux/printk.h>
16 #include <linux/slab.h>
17 #include <linux/timekeeper_internal.h>
18 #include <linux/vmalloc.h>
19 #include <asm/arch_timer.h>
20 #include <asm/barrier.h>
21 #include <asm/cacheflush.h>
24 #include <asm/vdso_datapage.h>
25 #include <clocksource/arm_arch_timer.h>
27 #define MAX_SYMNAME 64
29 static struct page **vdso_text_pagelist;
31 extern char vdso_start[], vdso_end[];
33 /* Total number of pages needed for the data and text portions of the VDSO. */
34 unsigned int vdso_total_pages __ro_after_init;
39 static union vdso_data_store vdso_data_store __page_aligned_data;
40 static struct vdso_data *vdso_data = &vdso_data_store.data;
42 static struct page *vdso_data_page __ro_after_init;
43 static const struct vm_special_mapping vdso_data_mapping = {
45 .pages = &vdso_data_page,
48 static int vdso_mremap(const struct vm_special_mapping *sm,
49 struct vm_area_struct *new_vma)
51 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
52 unsigned long vdso_size;
54 /* without VVAR page */
55 vdso_size = (vdso_total_pages - 1) << PAGE_SHIFT;
57 if (vdso_size != new_size)
60 current->mm->context.vdso = new_vma->vm_start;
65 static struct vm_special_mapping vdso_text_mapping __ro_after_init = {
67 .mremap = vdso_mremap,
71 Elf32_Ehdr *hdr; /* ptr to ELF */
72 Elf32_Sym *dynsym; /* ptr to .dynsym section */
73 unsigned long dynsymsize; /* size of .dynsym section */
74 char *dynstr; /* ptr to .dynstr section */
77 /* Cached result of boot-time check for whether the arch timer exists,
78 * and if so, whether the virtual counter is useable.
80 static bool cntvct_ok __ro_after_init;
82 static bool __init cntvct_functional(void)
84 struct device_node *np;
87 if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
90 /* The arm_arch_timer core should export
91 * arch_timer_use_virtual or similar so we don't have to do
94 np = of_find_compatible_node(NULL, NULL, "arm,armv7-timer");
96 np = of_find_compatible_node(NULL, NULL, "arm,armv8-timer");
100 if (of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
111 static void * __init find_section(Elf32_Ehdr *ehdr, const char *name,
118 /* Grab section headers and strings so we can tell who is who */
119 sechdrs = (void *)ehdr + ehdr->e_shoff;
120 secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
122 /* Find the section they want */
123 for (i = 1; i < ehdr->e_shnum; i++) {
124 if (strcmp(secnames + sechdrs[i].sh_name, name) == 0) {
126 *size = sechdrs[i].sh_size;
127 return (void *)ehdr + sechdrs[i].sh_offset;
136 static Elf32_Sym * __init find_symbol(struct elfinfo *lib, const char *symname)
140 for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
141 char name[MAX_SYMNAME], *c;
143 if (lib->dynsym[i].st_name == 0)
145 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
147 c = strchr(name, '@');
150 if (strcmp(symname, name) == 0)
151 return &lib->dynsym[i];
156 static void __init vdso_nullpatch_one(struct elfinfo *lib, const char *symname)
160 sym = find_symbol(lib, symname);
167 static void __init patch_vdso(void *ehdr)
169 struct elfinfo einfo;
171 einfo = (struct elfinfo) {
175 einfo.dynsym = find_section(einfo.hdr, ".dynsym", &einfo.dynsymsize);
176 einfo.dynstr = find_section(einfo.hdr, ".dynstr", NULL);
178 /* If the virtual counter is absent or non-functional we don't
179 * want programs to incur the slight additional overhead of
180 * dispatching through the VDSO only to fall back to syscalls.
183 vdso_nullpatch_one(&einfo, "__vdso_gettimeofday");
184 vdso_nullpatch_one(&einfo, "__vdso_clock_gettime");
188 static int __init vdso_init(void)
190 unsigned int text_pages;
193 if (memcmp(vdso_start, "\177ELF", 4)) {
194 pr_err("VDSO is not a valid ELF object!\n");
198 text_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
200 /* Allocate the VDSO text pagelist */
201 vdso_text_pagelist = kcalloc(text_pages, sizeof(struct page *),
203 if (vdso_text_pagelist == NULL)
206 /* Grab the VDSO data page. */
207 vdso_data_page = virt_to_page(vdso_data);
209 /* Grab the VDSO text pages. */
210 for (i = 0; i < text_pages; i++) {
213 page = virt_to_page(vdso_start + i * PAGE_SIZE);
214 vdso_text_pagelist[i] = page;
217 vdso_text_mapping.pages = vdso_text_pagelist;
219 vdso_total_pages = 1; /* for the data/vvar page */
220 vdso_total_pages += text_pages;
222 cntvct_ok = cntvct_functional();
224 patch_vdso(vdso_start);
228 arch_initcall(vdso_init);
230 static int install_vvar(struct mm_struct *mm, unsigned long addr)
232 struct vm_area_struct *vma;
234 vma = _install_special_mapping(mm, addr, PAGE_SIZE,
235 VM_READ | VM_MAYREAD,
238 return PTR_ERR_OR_ZERO(vma);
241 /* assumes mmap_sem is write-locked */
242 void arm_install_vdso(struct mm_struct *mm, unsigned long addr)
244 struct vm_area_struct *vma;
247 mm->context.vdso = 0;
249 if (vdso_text_pagelist == NULL)
252 if (install_vvar(mm, addr))
255 /* Account for vvar page. */
257 len = (vdso_total_pages - 1) << PAGE_SHIFT;
259 vma = _install_special_mapping(mm, addr, len,
260 VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
264 mm->context.vdso = addr;
267 static void vdso_write_begin(struct vdso_data *vdata)
269 ++vdso_data->seq_count;
270 smp_wmb(); /* Pairs with smp_rmb in vdso_read_retry */
273 static void vdso_write_end(struct vdso_data *vdata)
275 smp_wmb(); /* Pairs with smp_rmb in vdso_read_begin */
276 ++vdso_data->seq_count;
279 static bool tk_is_cntvct(const struct timekeeper *tk)
281 if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
284 if (tk->tkr_mono.clock->archdata.clock_mode != VDSO_CLOCKMODE_ARCHTIMER)
291 * update_vsyscall - update the vdso data page
293 * Increment the sequence counter, making it odd, indicating to
294 * userspace that an update is in progress. Update the fields used
295 * for coarse clocks and, if the architected system timer is in use,
296 * the fields used for high precision clocks. Increment the sequence
297 * counter again, making it even, indicating to userspace that the
298 * update is finished.
300 * Userspace is expected to sample seq_count before reading any other
301 * fields from the data page. If seq_count is odd, userspace is
302 * expected to wait until it becomes even. After copying data from
303 * the page, userspace must sample seq_count again; if it has changed
304 * from its previous value, userspace must retry the whole sequence.
306 * Calls to update_vsyscall are serialized by the timekeeping core.
308 void update_vsyscall(struct timekeeper *tk)
310 struct timespec64 *wtm = &tk->wall_to_monotonic;
313 /* The entry points have been zeroed, so there is no
314 * point in updating the data page.
319 vdso_write_begin(vdso_data);
321 vdso_data->tk_is_cntvct = tk_is_cntvct(tk);
322 vdso_data->xtime_coarse_sec = tk->xtime_sec;
323 vdso_data->xtime_coarse_nsec = (u32)(tk->tkr_mono.xtime_nsec >>
325 vdso_data->wtm_clock_sec = wtm->tv_sec;
326 vdso_data->wtm_clock_nsec = wtm->tv_nsec;
328 if (vdso_data->tk_is_cntvct) {
329 vdso_data->cs_cycle_last = tk->tkr_mono.cycle_last;
330 vdso_data->xtime_clock_sec = tk->xtime_sec;
331 vdso_data->xtime_clock_snsec = tk->tkr_mono.xtime_nsec;
332 vdso_data->cs_mult = tk->tkr_mono.mult;
333 vdso_data->cs_shift = tk->tkr_mono.shift;
334 vdso_data->cs_mask = tk->tkr_mono.mask;
337 vdso_write_end(vdso_data);
339 flush_dcache_page(virt_to_page(vdso_data));
342 void update_vsyscall_tz(void)
344 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
345 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
346 flush_dcache_page(virt_to_page(vdso_data));