arch/arm/kernel/vdso.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Adapted from arm64 version.
   4  *
   5  * Copyright (C) 2012 ARM Limited
   6  * Copyright (C) 2015 Mentor Graphics Corporation.
   7  */
   8
   9 #include <linux/cache.h>
  10 #include <linux/elf.h>
  11 #include <linux/err.h>
  12 #include <linux/kernel.h>
  13 #include <linux/mm.h>
  14 #include <linux/of.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>
  22 #include <asm/page.h>
  23 #include <asm/vdso.h>
  24 #include <asm/vdso_datapage.h>
  25 #include <clocksource/arm_arch_timer.h>
  26
  27 #define MAX_SYMNAME     64
  28
  29 static struct page **vdso_text_pagelist;
  30
  31 extern char vdso_start[], vdso_end[];
  32
  33 /* Total number of pages needed for the data and text portions of the VDSO. */
  34 unsigned int vdso_total_pages __ro_after_init;
  35
  36 /*
  37  * The VDSO data page.
  38  */
  39 static union vdso_data_store vdso_data_store __page_aligned_data;
  40 static struct vdso_data *vdso_data = &vdso_data_store.data;
  41
  42 static struct page *vdso_data_page __ro_after_init;
  43 static const struct vm_special_mapping vdso_data_mapping = {
  44         .name = "[vvar]",
  45         .pages = &vdso_data_page,
  46 };
  47
  48 static int vdso_mremap(const struct vm_special_mapping *sm,
  49                 struct vm_area_struct *new_vma)
  50 {
  51         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
  52         unsigned long vdso_size;
  53
  54         /* without VVAR page */
  55         vdso_size = (vdso_total_pages - 1) << PAGE_SHIFT;
  56
  57         if (vdso_size != new_size)
  58                 return -EINVAL;
  59
  60         current->mm->context.vdso = new_vma->vm_start;
  61
  62         return 0;
  63 }
  64
  65 static struct vm_special_mapping vdso_text_mapping __ro_after_init = {
  66         .name = "[vdso]",
  67         .mremap = vdso_mremap,
  68 };
  69
  70 struct elfinfo {
  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 */
  75 };
  76
  77 /* Cached result of boot-time check for whether the arch timer exists,
  78  * and if so, whether the virtual counter is useable.
  79  */
  80 static bool cntvct_ok __ro_after_init;
  81
  82 static bool __init cntvct_functional(void)
  83 {
  84         struct device_node *np;
  85         bool ret = false;
  86
  87         if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
  88                 goto out;
  89
  90         /* The arm_arch_timer core should export
  91          * arch_timer_use_virtual or similar so we don't have to do
  92          * this.
  93          */
  94         np = of_find_compatible_node(NULL, NULL, "arm,armv7-timer");
  95         if (!np)
  96                 goto out_put;
  97
  98         if (of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
  99                 goto out_put;
 100
 101         ret = true;
 102
 103 out_put:
 104         of_node_put(np);
 105 out:
 106         return ret;
 107 }
 108
 109 static void * __init find_section(Elf32_Ehdr *ehdr, const char *name,
 110                                   unsigned long *size)
 111 {
 112         Elf32_Shdr *sechdrs;
 113         unsigned int i;
 114         char *secnames;
 115
 116         /* Grab section headers and strings so we can tell who is who */
 117         sechdrs = (void *)ehdr + ehdr->e_shoff;
 118         secnames = (void *)ehdr + sechdrs[ehdr->e_shstrndx].sh_offset;
 119
 120         /* Find the section they want */
 121         for (i = 1; i < ehdr->e_shnum; i++) {
 122                 if (strcmp(secnames + sechdrs[i].sh_name, name) == 0) {
 123                         if (size)
 124                                 *size = sechdrs[i].sh_size;
 125                         return (void *)ehdr + sechdrs[i].sh_offset;
 126                 }
 127         }
 128
 129         if (size)
 130                 *size = 0;
 131         return NULL;
 132 }
 133
 134 static Elf32_Sym * __init find_symbol(struct elfinfo *lib, const char *symname)
 135 {
 136         unsigned int i;
 137
 138         for (i = 0; i < (lib->dynsymsize / sizeof(Elf32_Sym)); i++) {
 139                 char name[MAX_SYMNAME], *c;
 140
 141                 if (lib->dynsym[i].st_name == 0)
 142                         continue;
 143                 strlcpy(name, lib->dynstr + lib->dynsym[i].st_name,
 144                         MAX_SYMNAME);
 145                 c = strchr(name, '@');
 146                 if (c)
 147                         *c = 0;
 148                 if (strcmp(symname, name) == 0)
 149                         return &lib->dynsym[i];
 150         }
 151         return NULL;
 152 }
 153
 154 static void __init vdso_nullpatch_one(struct elfinfo *lib, const char *symname)
 155 {
 156         Elf32_Sym *sym;
 157
 158         sym = find_symbol(lib, symname);
 159         if (!sym)
 160                 return;
 161
 162         sym->st_name = 0;
 163 }
 164
 165 static void __init patch_vdso(void *ehdr)
 166 {
 167         struct elfinfo einfo;
 168
 169         einfo = (struct elfinfo) {
 170                 .hdr = ehdr,
 171         };
 172
 173         einfo.dynsym = find_section(einfo.hdr, ".dynsym", &einfo.dynsymsize);
 174         einfo.dynstr = find_section(einfo.hdr, ".dynstr", NULL);
 175
 176         /* If the virtual counter is absent or non-functional we don't
 177          * want programs to incur the slight additional overhead of
 178          * dispatching through the VDSO only to fall back to syscalls.
 179          */
 180         if (!cntvct_ok) {
 181                 vdso_nullpatch_one(&einfo, "__vdso_gettimeofday");
 182                 vdso_nullpatch_one(&einfo, "__vdso_clock_gettime");
 183         }
 184 }
 185
 186 static int __init vdso_init(void)
 187 {
 188         unsigned int text_pages;
 189         int i;
 190
 191         if (memcmp(vdso_start, "\177ELF", 4)) {
 192                 pr_err("VDSO is not a valid ELF object!\n");
 193                 return -ENOEXEC;
 194         }
 195
 196         text_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
 197         pr_debug("vdso: %i text pages at base %p\n", text_pages, vdso_start);
 198
 199         /* Allocate the VDSO text pagelist */
 200         vdso_text_pagelist = kcalloc(text_pages, sizeof(struct page *),
 201                                      GFP_KERNEL);
 202         if (vdso_text_pagelist == NULL)
 203                 return -ENOMEM;
 204
 205         /* Grab the VDSO data page. */
 206         vdso_data_page = virt_to_page(vdso_data);
 207
 208         /* Grab the VDSO text pages. */
 209         for (i = 0; i < text_pages; i++) {
 210                 struct page *page;
 211
 212                 page = virt_to_page(vdso_start + i * PAGE_SIZE);
 213                 vdso_text_pagelist[i] = page;
 214         }
 215
 216         vdso_text_mapping.pages = vdso_text_pagelist;
 217
 218         vdso_total_pages = 1; /* for the data/vvar page */
 219         vdso_total_pages += text_pages;
 220
 221         cntvct_ok = cntvct_functional();
 222
 223         patch_vdso(vdso_start);
 224
 225         return 0;
 226 }
 227 arch_initcall(vdso_init);
 228
 229 static int install_vvar(struct mm_struct *mm, unsigned long addr)
 230 {
 231         struct vm_area_struct *vma;
 232
 233         vma = _install_special_mapping(mm, addr, PAGE_SIZE,
 234                                        VM_READ | VM_MAYREAD,
 235                                        &vdso_data_mapping);
 236
 237         return PTR_ERR_OR_ZERO(vma);
 238 }
 239
 240 /* assumes mmap_sem is write-locked */
 241 void arm_install_vdso(struct mm_struct *mm, unsigned long addr)
 242 {
 243         struct vm_area_struct *vma;
 244         unsigned long len;
 245
 246         mm->context.vdso = 0;
 247
 248         if (vdso_text_pagelist == NULL)
 249                 return;
 250
 251         if (install_vvar(mm, addr))
 252                 return;
 253
 254         /* Account for vvar page. */
 255         addr += PAGE_SIZE;
 256         len = (vdso_total_pages - 1) << PAGE_SHIFT;
 257
 258         vma = _install_special_mapping(mm, addr, len,
 259                 VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
 260                 &vdso_text_mapping);
 261
 262         if (!IS_ERR(vma))
 263                 mm->context.vdso = addr;
 264 }
 265
 266 static void vdso_write_begin(struct vdso_data *vdata)
 267 {
 268         ++vdso_data->seq_count;
 269         smp_wmb(); /* Pairs with smp_rmb in vdso_read_retry */
 270 }
 271
 272 static void vdso_write_end(struct vdso_data *vdata)
 273 {
 274         smp_wmb(); /* Pairs with smp_rmb in vdso_read_begin */
 275         ++vdso_data->seq_count;
 276 }
 277
 278 static bool tk_is_cntvct(const struct timekeeper *tk)
 279 {
 280         if (!IS_ENABLED(CONFIG_ARM_ARCH_TIMER))
 281                 return false;
 282
 283         if (!tk->tkr_mono.clock->archdata.vdso_direct)
 284                 return false;
 285
 286         return true;
 287 }
 288
 289 /**
 290  * update_vsyscall - update the vdso data page
 291  *
 292  * Increment the sequence counter, making it odd, indicating to
 293  * userspace that an update is in progress.  Update the fields used
 294  * for coarse clocks and, if the architected system timer is in use,
 295  * the fields used for high precision clocks.  Increment the sequence
 296  * counter again, making it even, indicating to userspace that the
 297  * update is finished.
 298  *
 299  * Userspace is expected to sample seq_count before reading any other
 300  * fields from the data page.  If seq_count is odd, userspace is
 301  * expected to wait until it becomes even.  After copying data from
 302  * the page, userspace must sample seq_count again; if it has changed
 303  * from its previous value, userspace must retry the whole sequence.
 304  *
 305  * Calls to update_vsyscall are serialized by the timekeeping core.
 306  */
 307 void update_vsyscall(struct timekeeper *tk)
 308 {
 309         struct timespec64 *wtm = &tk->wall_to_monotonic;
 310
 311         if (!cntvct_ok) {
 312                 /* The entry points have been zeroed, so there is no
 313                  * point in updating the data page.
 314                  */
 315                 return;
 316         }
 317
 318         vdso_write_begin(vdso_data);
 319
 320         vdso_data->tk_is_cntvct                 = tk_is_cntvct(tk);
 321         vdso_data->xtime_coarse_sec             = tk->xtime_sec;
 322         vdso_data->xtime_coarse_nsec            = (u32)(tk->tkr_mono.xtime_nsec >>
 323                                                         tk->tkr_mono.shift);
 324         vdso_data->wtm_clock_sec                = wtm->tv_sec;
 325         vdso_data->wtm_clock_nsec               = wtm->tv_nsec;
 326
 327         if (vdso_data->tk_is_cntvct) {
 328                 vdso_data->cs_cycle_last        = tk->tkr_mono.cycle_last;
 329                 vdso_data->xtime_clock_sec      = tk->xtime_sec;
 330                 vdso_data->xtime_clock_snsec    = tk->tkr_mono.xtime_nsec;
 331                 vdso_data->cs_mult              = tk->tkr_mono.mult;
 332                 vdso_data->cs_shift             = tk->tkr_mono.shift;
 333                 vdso_data->cs_mask              = tk->tkr_mono.mask;
 334         }
 335
 336         vdso_write_end(vdso_data);
 337
 338         flush_dcache_page(virt_to_page(vdso_data));
 339 }
 340
 341 void update_vsyscall_tz(void)
 342 {
 343         vdso_data->tz_minuteswest       = sys_tz.tz_minuteswest;
 344         vdso_data->tz_dsttime           = sys_tz.tz_dsttime;
 345         flush_dcache_page(virt_to_page(vdso_data));
 346 }