arch/s390/kernel/vdso.c

   1 /*
   2  * vdso setup for s390
   3  *
   4  *  Copyright IBM Corp. 2008
   5  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
   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 only)
   9  * as published by the Free Software Foundation.
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/errno.h>
  14 #include <linux/sched.h>
  15 #include <linux/kernel.h>
  16 #include <linux/mm.h>
  17 #include <linux/smp.h>
  18 #include <linux/stddef.h>
  19 #include <linux/unistd.h>
  20 #include <linux/slab.h>
  21 #include <linux/user.h>
  22 #include <linux/elf.h>
  23 #include <linux/security.h>
  24 #include <linux/bootmem.h>
  25 #include <linux/compat.h>
  26 #include <asm/asm-offsets.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/processor.h>
  29 #include <asm/mmu.h>
  30 #include <asm/mmu_context.h>
  31 #include <asm/sections.h>
  32 #include <asm/vdso.h>
  33 #include <asm/facility.h>
  34
  35 #ifdef CONFIG_COMPAT
  36 extern char vdso32_start, vdso32_end;
  37 static void *vdso32_kbase = &vdso32_start;
  38 static unsigned int vdso32_pages;
  39 static struct page **vdso32_pagelist;
  40 #endif
  41
  42 extern char vdso64_start, vdso64_end;
  43 static void *vdso64_kbase = &vdso64_start;
  44 static unsigned int vdso64_pages;
  45 static struct page **vdso64_pagelist;
  46
  47 /*
  48  * Should the kernel map a VDSO page into processes and pass its
  49  * address down to glibc upon exec()?
  50  */
  51 unsigned int __read_mostly vdso_enabled = 1;
  52
  53 static int __init vdso_setup(char *s)
  54 {
  55         unsigned long val;
  56         int rc;
  57
  58         rc = 0;
  59         if (strncmp(s, "on", 3) == 0)
  60                 vdso_enabled = 1;
  61         else if (strncmp(s, "off", 4) == 0)
  62                 vdso_enabled = 0;
  63         else {
  64                 rc = kstrtoul(s, 0, &val);
  65                 vdso_enabled = rc ? 0 : !!val;
  66         }
  67         return !rc;
  68 }
  69 __setup("vdso=", vdso_setup);
  70
  71 /*
  72  * The vdso data page
  73  */
  74 static union {
  75         struct vdso_data        data;
  76         u8                      page[PAGE_SIZE];
  77 } vdso_data_store __page_aligned_data;
  78 struct vdso_data *vdso_data = &vdso_data_store.data;
  79
  80 /*
  81  * Setup vdso data page.
  82  */
  83 static void __init vdso_init_data(struct vdso_data *vd)
  84 {
  85         vd->ectg_available = test_facility(31);
  86 }
  87
  88 /*
  89  * Allocate/free per cpu vdso data.
  90  */
  91 #define SEGMENT_ORDER   2
  92
  93 int vdso_alloc_per_cpu(struct lowcore *lowcore)
  94 {
  95         unsigned long segment_table, page_table, page_frame;
  96         struct vdso_per_cpu_data *vd;
  97         u32 *psal, *aste;
  98         int i;
  99
 100         lowcore->vdso_per_cpu_data = __LC_PASTE;
 101
 102         if (!vdso_enabled)
 103                 return 0;
 104
 105         segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
 106         page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
 107         page_frame = get_zeroed_page(GFP_KERNEL);
 108         if (!segment_table || !page_table || !page_frame)
 109                 goto out;
 110
 111         /* Initialize per-cpu vdso data page */
 112         vd = (struct vdso_per_cpu_data *) page_frame;
 113         vd->cpu_nr = lowcore->cpu_nr;
 114         vd->node_id = cpu_to_node(vd->cpu_nr);
 115
 116         /* Set up access register mode page table */
 117         clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
 118                     PAGE_SIZE << SEGMENT_ORDER);
 119         clear_table((unsigned long *) page_table, _PAGE_INVALID,
 120                     256*sizeof(unsigned long));
 121
 122         *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
 123         *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
 124
 125         psal = (u32 *) (page_table + 256*sizeof(unsigned long));
 126         aste = psal + 32;
 127
 128         for (i = 4; i < 32; i += 4)
 129                 psal[i] = 0x80000000;
 130
 131         lowcore->paste[4] = (u32)(addr_t) psal;
 132         psal[0] = 0x02000000;
 133         psal[2] = (u32)(addr_t) aste;
 134         *(unsigned long *) (aste + 2) = segment_table +
 135                 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
 136         aste[4] = (u32)(addr_t) psal;
 137         lowcore->vdso_per_cpu_data = page_frame;
 138
 139         return 0;
 140
 141 out:
 142         free_page(page_frame);
 143         free_page(page_table);
 144         free_pages(segment_table, SEGMENT_ORDER);
 145         return -ENOMEM;
 146 }
 147
 148 void vdso_free_per_cpu(struct lowcore *lowcore)
 149 {
 150         unsigned long segment_table, page_table, page_frame;
 151         u32 *psal, *aste;
 152
 153         if (!vdso_enabled)
 154                 return;
 155
 156         psal = (u32 *)(addr_t) lowcore->paste[4];
 157         aste = (u32 *)(addr_t) psal[2];
 158         segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
 159         page_table = *(unsigned long *) segment_table;
 160         page_frame = *(unsigned long *) page_table;
 161
 162         free_page(page_frame);
 163         free_page(page_table);
 164         free_pages(segment_table, SEGMENT_ORDER);
 165 }
 166
 167 static void vdso_init_cr5(void)
 168 {
 169         unsigned long cr5;
 170
 171         if (!vdso_enabled)
 172                 return;
 173         cr5 = offsetof(struct lowcore, paste);
 174         __ctl_load(cr5, 5, 5);
 175 }
 176
 177 /*
 178  * This is called from binfmt_elf, we create the special vma for the
 179  * vDSO and insert it into the mm struct tree
 180  */
 181 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 182 {
 183         struct mm_struct *mm = current->mm;
 184         struct page **vdso_pagelist;
 185         unsigned long vdso_pages;
 186         unsigned long vdso_base;
 187         int rc;
 188
 189         if (!vdso_enabled)
 190                 return 0;
 191         /*
 192          * Only map the vdso for dynamically linked elf binaries.
 193          */
 194         if (!uses_interp)
 195                 return 0;
 196
 197         vdso_pagelist = vdso64_pagelist;
 198         vdso_pages = vdso64_pages;
 199 #ifdef CONFIG_COMPAT
 200         if (is_compat_task()) {
 201                 vdso_pagelist = vdso32_pagelist;
 202                 vdso_pages = vdso32_pages;
 203         }
 204 #endif
 205         /*
 206          * vDSO has a problem and was disabled, just don't "enable" it for
 207          * the process
 208          */
 209         if (vdso_pages == 0)
 210                 return 0;
 211
 212         current->mm->context.vdso_base = 0;
 213
 214         /*
 215          * pick a base address for the vDSO in process space. We try to put
 216          * it at vdso_base which is the "natural" base for it, but we might
 217          * fail and end up putting it elsewhere.
 218          */
 219         if (down_write_killable(&mm->mmap_sem))
 220                 return -EINTR;
 221         vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
 222         if (IS_ERR_VALUE(vdso_base)) {
 223                 rc = vdso_base;
 224                 goto out_up;
 225         }
 226
 227         /*
 228          * Put vDSO base into mm struct. We need to do this before calling
 229          * install_special_mapping or the perf counter mmap tracking code
 230          * will fail to recognise it as a vDSO (since arch_vma_name fails).
 231          */
 232         current->mm->context.vdso_base = vdso_base;
 233
 234         /*
 235          * our vma flags don't have VM_WRITE so by default, the process
 236          * isn't allowed to write those pages.
 237          * gdb can break that with ptrace interface, and thus trigger COW
 238          * on those pages but it's then your responsibility to never do that
 239          * on the "data" page of the vDSO or you'll stop getting kernel
 240          * updates and your nice userland gettimeofday will be totally dead.
 241          * It's fine to use that for setting breakpoints in the vDSO code
 242          * pages though.
 243          */
 244         rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
 245                                      VM_READ|VM_EXEC|
 246                                      VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 247                                      vdso_pagelist);
 248         if (rc)
 249                 current->mm->context.vdso_base = 0;
 250 out_up:
 251         up_write(&mm->mmap_sem);
 252         return rc;
 253 }
 254
 255 const char *arch_vma_name(struct vm_area_struct *vma)
 256 {
 257         if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
 258                 return "[vdso]";
 259         return NULL;
 260 }
 261
 262 static int __init vdso_init(void)
 263 {
 264         int i;
 265
 266         if (!vdso_enabled)
 267                 return 0;
 268         vdso_init_data(vdso_data);
 269 #ifdef CONFIG_COMPAT
 270         /* Calculate the size of the 32 bit vDSO */
 271         vdso32_pages = ((&vdso32_end - &vdso32_start
 272                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 273
 274         /* Make sure pages are in the correct state */
 275         vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
 276                                   GFP_KERNEL);
 277         BUG_ON(vdso32_pagelist == NULL);
 278         for (i = 0; i < vdso32_pages - 1; i++) {
 279                 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
 280                 ClearPageReserved(pg);
 281                 get_page(pg);
 282                 vdso32_pagelist[i] = pg;
 283         }
 284         vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
 285         vdso32_pagelist[vdso32_pages] = NULL;
 286 #endif
 287
 288         /* Calculate the size of the 64 bit vDSO */
 289         vdso64_pages = ((&vdso64_end - &vdso64_start
 290                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 291
 292         /* Make sure pages are in the correct state */
 293         vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
 294                                   GFP_KERNEL);
 295         BUG_ON(vdso64_pagelist == NULL);
 296         for (i = 0; i < vdso64_pages - 1; i++) {
 297                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
 298                 ClearPageReserved(pg);
 299                 get_page(pg);
 300                 vdso64_pagelist[i] = pg;
 301         }
 302         vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
 303         vdso64_pagelist[vdso64_pages] = NULL;
 304         if (vdso_alloc_per_cpu(&S390_lowcore))
 305                 BUG();
 306         vdso_init_cr5();
 307
 308         get_page(virt_to_page(vdso_data));
 309
 310         return 0;
 311 }
 312 early_initcall(vdso_init);