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