arch/arm64/kernel/vdso.c

   1 /*
   2  * VDSO implementation for AArch64 and vector page setup for AArch32.
   3  *
   4  * Copyright (C) 2012 ARM Limited
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17  *
  18  * Author: Will Deacon <will.deacon@arm.com>
  19  */
  20
  21 #include <linux/cache.h>
  22 #include <linux/clocksource.h>
  23 #include <linux/elf.h>
  24 #include <linux/err.h>
  25 #include <linux/errno.h>
  26 #include <linux/gfp.h>
  27 #include <linux/kernel.h>
  28 #include <linux/mm.h>
  29 #include <linux/sched.h>
  30 #include <linux/signal.h>
  31 #include <linux/slab.h>
  32 #include <linux/timekeeper_internal.h>
  33 #include <linux/vmalloc.h>
  34
  35 #include <asm/cacheflush.h>
  36 #include <asm/signal32.h>
  37 #include <asm/vdso.h>
  38 #include <asm/vdso_datapage.h>
  39
  40 extern char vdso_start[], vdso_end[];
  41 static unsigned long vdso_pages __ro_after_init;
  42
  43 /*
  44  * The vDSO data page.
  45  */
  46 static union {
  47         struct vdso_data        data;
  48         u8                      page[PAGE_SIZE];
  49 } vdso_data_store __page_aligned_data;
  50 struct vdso_data *vdso_data = &vdso_data_store.data;
  51
  52 #ifdef CONFIG_COMPAT
  53 /*
  54  * Create and map the vectors page for AArch32 tasks.
  55  */
  56 static struct page *vectors_page[1] __ro_after_init;
  57
  58 static int __init alloc_vectors_page(void)
  59 {
  60         extern char __kuser_helper_start[], __kuser_helper_end[];
  61         extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
  62
  63         int kuser_sz = __kuser_helper_end - __kuser_helper_start;
  64         int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
  65         unsigned long vpage;
  66
  67         vpage = get_zeroed_page(GFP_ATOMIC);
  68
  69         if (!vpage)
  70                 return -ENOMEM;
  71
  72         /* kuser helpers */
  73         memcpy((void *)vpage + 0x1000 - kuser_sz, __kuser_helper_start,
  74                 kuser_sz);
  75
  76         /* sigreturn code */
  77         memcpy((void *)vpage + AARCH32_KERN_SIGRET_CODE_OFFSET,
  78                __aarch32_sigret_code_start, sigret_sz);
  79
  80         flush_icache_range(vpage, vpage + PAGE_SIZE);
  81         vectors_page[0] = virt_to_page(vpage);
  82
  83         return 0;
  84 }
  85 arch_initcall(alloc_vectors_page);
  86
  87 int aarch32_setup_vectors_page(struct linux_binprm *bprm, int uses_interp)
  88 {
  89         struct mm_struct *mm = current->mm;
  90         unsigned long addr = AARCH32_VECTORS_BASE;
  91         static const struct vm_special_mapping spec = {
  92                 .name   = "[vectors]",
  93                 .pages  = vectors_page,
  94
  95         };
  96         void *ret;
  97
  98         if (down_write_killable(&mm->mmap_sem))
  99                 return -EINTR;
 100         current->mm->context.vdso = (void *)addr;
 101
 102         /* Map vectors page at the high address. */
 103         ret = _install_special_mapping(mm, addr, PAGE_SIZE,
 104                                        VM_READ|VM_EXEC|VM_MAYREAD|VM_MAYEXEC,
 105                                        &spec);
 106
 107         up_write(&mm->mmap_sem);
 108
 109         return PTR_ERR_OR_ZERO(ret);
 110 }
 111 #endif /* CONFIG_COMPAT */
 112
 113 static int vdso_mremap(const struct vm_special_mapping *sm,
 114                 struct vm_area_struct *new_vma)
 115 {
 116         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
 117         unsigned long vdso_size = vdso_end - vdso_start;
 118
 119         if (vdso_size != new_size)
 120                 return -EINVAL;
 121
 122         current->mm->context.vdso = (void *)new_vma->vm_start;
 123
 124         return 0;
 125 }
 126
 127 static struct vm_special_mapping vdso_spec[2] __ro_after_init = {
 128         {
 129                 .name   = "[vvar]",
 130         },
 131         {
 132                 .name   = "[vdso]",
 133                 .mremap = vdso_mremap,
 134         },
 135 };
 136
 137 static int __init vdso_init(void)
 138 {
 139         int i;
 140         struct page **vdso_pagelist;
 141         unsigned long pfn;
 142
 143         if (memcmp(vdso_start, "\177ELF", 4)) {
 144                 pr_err("vDSO is not a valid ELF object!\n");
 145                 return -EINVAL;
 146         }
 147
 148         vdso_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
 149
 150         /* Allocate the vDSO pagelist, plus a page for the data. */
 151         vdso_pagelist = kcalloc(vdso_pages + 1, sizeof(struct page *),
 152                                 GFP_KERNEL);
 153         if (vdso_pagelist == NULL)
 154                 return -ENOMEM;
 155
 156         /* Grab the vDSO data page. */
 157         vdso_pagelist[0] = phys_to_page(__pa_symbol(vdso_data));
 158
 159
 160         /* Grab the vDSO code pages. */
 161         pfn = sym_to_pfn(vdso_start);
 162
 163         for (i = 0; i < vdso_pages; i++)
 164                 vdso_pagelist[i + 1] = pfn_to_page(pfn + i);
 165
 166         vdso_spec[0].pages = &vdso_pagelist[0];
 167         vdso_spec[1].pages = &vdso_pagelist[1];
 168
 169         return 0;
 170 }
 171 arch_initcall(vdso_init);
 172
 173 int arch_setup_additional_pages(struct linux_binprm *bprm,
 174                                 int uses_interp)
 175 {
 176         struct mm_struct *mm = current->mm;
 177         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 178         void *ret;
 179
 180         vdso_text_len = vdso_pages << PAGE_SHIFT;
 181         /* Be sure to map the data page */
 182         vdso_mapping_len = vdso_text_len + PAGE_SIZE;
 183
 184         if (down_write_killable(&mm->mmap_sem))
 185                 return -EINTR;
 186         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 187         if (IS_ERR_VALUE(vdso_base)) {
 188                 ret = ERR_PTR(vdso_base);
 189                 goto up_fail;
 190         }
 191         ret = _install_special_mapping(mm, vdso_base, PAGE_SIZE,
 192                                        VM_READ|VM_MAYREAD,
 193                                        &vdso_spec[0]);
 194         if (IS_ERR(ret))
 195                 goto up_fail;
 196
 197         vdso_base += PAGE_SIZE;
 198         mm->context.vdso = (void *)vdso_base;
 199         ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
 200                                        VM_READ|VM_EXEC|
 201                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 202                                        &vdso_spec[1]);
 203         if (IS_ERR(ret))
 204                 goto up_fail;
 205
 206
 207         up_write(&mm->mmap_sem);
 208         return 0;
 209
 210 up_fail:
 211         mm->context.vdso = NULL;
 212         up_write(&mm->mmap_sem);
 213         return PTR_ERR(ret);
 214 }
 215
 216 /*
 217  * Update the vDSO data page to keep in sync with kernel timekeeping.
 218  */
 219 void update_vsyscall(struct timekeeper *tk)
 220 {
 221         u32 use_syscall = !tk->tkr_mono.clock->archdata.vdso_direct;
 222
 223         ++vdso_data->tb_seq_count;
 224         smp_wmb();
 225
 226         vdso_data->use_syscall                  = use_syscall;
 227         vdso_data->xtime_coarse_sec             = tk->xtime_sec;
 228         vdso_data->xtime_coarse_nsec            = tk->tkr_mono.xtime_nsec >>
 229                                                         tk->tkr_mono.shift;
 230         vdso_data->wtm_clock_sec                = tk->wall_to_monotonic.tv_sec;
 231         vdso_data->wtm_clock_nsec               = tk->wall_to_monotonic.tv_nsec;
 232
 233         /* Read without the seqlock held by clock_getres() */
 234         WRITE_ONCE(vdso_data->hrtimer_res, hrtimer_resolution);
 235
 236         if (!use_syscall) {
 237                 /* tkr_mono.cycle_last == tkr_raw.cycle_last */
 238                 vdso_data->cs_cycle_last        = tk->tkr_mono.cycle_last;
 239                 vdso_data->raw_time_sec         = tk->raw_sec;
 240                 vdso_data->raw_time_nsec        = tk->tkr_raw.xtime_nsec;
 241                 vdso_data->xtime_clock_sec      = tk->xtime_sec;
 242                 vdso_data->xtime_clock_nsec     = tk->tkr_mono.xtime_nsec;
 243                 vdso_data->cs_mono_mult         = tk->tkr_mono.mult;
 244                 vdso_data->cs_raw_mult          = tk->tkr_raw.mult;
 245                 /* tkr_mono.shift == tkr_raw.shift */
 246                 vdso_data->cs_shift             = tk->tkr_mono.shift;
 247         }
 248
 249         smp_wmb();
 250         ++vdso_data->tb_seq_count;
 251 }
 252
 253 void update_vsyscall_tz(void)
 254 {
 255         vdso_data->tz_minuteswest       = sys_tz.tz_minuteswest;
 256         vdso_data->tz_dsttime           = sys_tz.tz_dsttime;
 257 }