arch/arm/kernel/vdso.c

   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 /* Total number of pages needed for the data and text portions of the VDSO. */
  43 unsigned int vdso_total_pages __ro_after_init;
  44
  45 /*
  46  * The VDSO data page.
  47  */
  48 static union vdso_data_store vdso_data_store __page_aligned_data;
  49 static struct vdso_data *vdso_data = &vdso_data_store.data;
  50
  51 static struct page *vdso_data_page __ro_after_init;
  52 static const struct vm_special_mapping vdso_data_mapping = {
  53         .name = "[vvar]",
  54         .pages = &vdso_data_page,
  55 };
  56
  57 static struct vm_special_mapping vdso_text_mapping __ro_after_init = {
  58         .name = "[vdso]",
  59 };
  60
  61 struct 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  */
  71 static bool cntvct_ok __ro_after_init;
  72
  73 static 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
  94 out_put:
  95         of_node_put(np);
  96 out:
  97         return ret;
  98 }
  99
 100 static 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
 125 static 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
 145 static 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
 156 static 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                 vdso_nullpatch_one(&einfo, "__vdso_clock_getres");
 175         }
 176 }
 177
 178 static int __init vdso_init(void)
 179 {
 180         unsigned int text_pages;
 181         int i;
 182
 183         if (memcmp(&vdso_start, "\177ELF", 4)) {
 184                 pr_err("VDSO is not a valid ELF object!\n");
 185                 return -ENOEXEC;
 186         }
 187
 188         text_pages = (&vdso_end - &vdso_start) >> PAGE_SHIFT;
 189         pr_debug("vdso: %i text pages at base %p\n", text_pages, &vdso_start);
 190
 191         /* Allocate the VDSO text pagelist */
 192         vdso_text_pagelist = kcalloc(text_pages, sizeof(struct page *),
 193                                      GFP_KERNEL);
 194         if (vdso_text_pagelist == NULL)
 195                 return -ENOMEM;
 196
 197         /* Grab the VDSO data page. */
 198         vdso_data_page = virt_to_page(vdso_data);
 199
 200         /* Grab the VDSO text pages. */
 201         for (i = 0; i < text_pages; i++) {
 202                 struct page *page;
 203
 204                 page = virt_to_page(&vdso_start + i * PAGE_SIZE);
 205                 vdso_text_pagelist[i] = page;
 206         }
 207
 208         vdso_text_mapping.pages = vdso_text_pagelist;
 209
 210         vdso_total_pages = 1; /* for the data/vvar page */
 211         vdso_total_pages += text_pages;
 212
 213         cntvct_ok = cntvct_functional();
 214
 215         patch_vdso(&vdso_start);
 216
 217         return 0;
 218 }
 219 arch_initcall(vdso_init);
 220
 221 static int install_vvar(struct mm_struct *mm, unsigned long addr)
 222 {
 223         struct vm_area_struct *vma;
 224
 225         vma = _install_special_mapping(mm, addr, PAGE_SIZE,
 226                                        VM_READ | VM_MAYREAD,
 227                                        &vdso_data_mapping);
 228
 229         return PTR_ERR_OR_ZERO(vma);
 230 }
 231
 232 /* assumes mmap_sem is write-locked */
 233 void arm_install_vdso(struct mm_struct *mm, unsigned long addr)
 234 {
 235         struct vm_area_struct *vma;
 236         unsigned long len;
 237
 238         mm->context.vdso = 0;
 239
 240         if (vdso_text_pagelist == NULL)
 241                 return;
 242
 243         if (install_vvar(mm, addr))
 244                 return;
 245
 246         /* Account for vvar page. */
 247         addr += PAGE_SIZE;
 248         len = (vdso_total_pages - 1) << PAGE_SHIFT;
 249
 250         vma = _install_special_mapping(mm, addr, len,
 251                 VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
 252                 &vdso_text_mapping);
 253
 254         if (!IS_ERR(vma))
 255                 mm->context.vdso = addr;
 256 }
 257
 258 static void vdso_write_begin(struct vdso_data *vdata)
 259 {
 260         ++vdso_data->tb_seq_count;
 261         smp_wmb(); /* Pairs with smp_rmb in vdso_read_retry */
 262 }
 263
 264 static void vdso_write_end(struct vdso_data *vdata)
 265 {
 266         smp_wmb(); /* Pairs with smp_rmb in vdso_read_begin */
 267         ++vdso_data->tb_seq_count;
 268 }
 269
 270 static bool tk_is_cntvct(const struct timekeeper *tk)
 271 {
 272         if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
 273                 return false;
 274
 275         if (strcmp(tk->tkr_mono.clock->name, "arch_sys_counter") != 0)
 276                 return false;
 277
 278         return true;
 279 }
 280
 281 /**
 282  * update_vsyscall - update the vdso data page
 283  *
 284  * Increment the sequence counter, making it odd, indicating to
 285  * userspace that an update is in progress.  Update the fields used
 286  * for coarse clocks and, if the architected system timer is in use,
 287  * the fields used for high precision clocks.  Increment the sequence
 288  * counter again, making it even, indicating to userspace that the
 289  * update is finished.
 290  *
 291  * Userspace is expected to sample tb_seq_count before reading any other
 292  * fields from the data page.  If tb_seq_count is odd, userspace is
 293  * expected to wait until it becomes even.  After copying data from
 294  * the page, userspace must sample tb_seq_count again; if it has changed
 295  * from its previous value, userspace must retry the whole sequence.
 296  *
 297  * Calls to update_vsyscall are serialized by the timekeeping core.
 298  */
 299 void update_vsyscall(struct timekeeper *tk)
 300 {
 301         struct timespec64 *wtm = &tk->wall_to_monotonic;
 302
 303         if (!cntvct_ok) {
 304                 /* The entry points have been zeroed, so there is no
 305                  * point in updating the data page.
 306                  */
 307                 return;
 308         }
 309
 310         vdso_write_begin(vdso_data);
 311
 312         vdso_data->use_syscall                  = !tk_is_cntvct(tk);
 313         vdso_data->xtime_coarse_sec             = tk->xtime_sec;
 314         vdso_data->xtime_coarse_nsec            = (u32)(tk->tkr_mono.xtime_nsec >>
 315                                                         tk->tkr_mono.shift);
 316         vdso_data->wtm_clock_sec                = wtm->tv_sec;
 317         vdso_data->wtm_clock_nsec               = wtm->tv_nsec;
 318
 319         if (!vdso_data->use_syscall) {
 320                 vdso_data->cs_cycle_last        = tk->tkr_mono.cycle_last;
 321                 vdso_data->raw_time_sec         = tk->raw_sec;
 322                 vdso_data->raw_time_nsec        = tk->tkr_raw.xtime_nsec;
 323                 vdso_data->xtime_clock_sec      = tk->xtime_sec;
 324                 vdso_data->xtime_clock_snsec    = tk->tkr_mono.xtime_nsec;
 325                 vdso_data->cs_mono_mult         = tk->tkr_mono.mult;
 326                 vdso_data->cs_raw_mult          = tk->tkr_raw.mult;
 327                 /* tkr_mono.shift == tkr_raw.shift */
 328                 vdso_data->cs_shift             = tk->tkr_mono.shift;
 329                 vdso_data->cs_mask              = tk->tkr_mono.mask;
 330                 vdso_data->btm_nsec             = ktime_to_ns(tk->offs_boot);
 331         }
 332
 333         vdso_write_end(vdso_data);
 334
 335         flush_dcache_page(virt_to_page(vdso_data));
 336 }
 337
 338 void update_vsyscall_tz(void)
 339 {
 340         vdso_data->tz_minuteswest       = sys_tz.tz_minuteswest;
 341         vdso_data->tz_dsttime           = sys_tz.tz_dsttime;
 342         flush_dcache_page(virt_to_page(vdso_data));
 343 }