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