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/system.h>
  29 #include <asm/processor.h>
  30 #include <asm/mmu.h>
  31 #include <asm/mmu_context.h>
  32 #include <asm/sections.h>
  33 #include <asm/vdso.h>
  34
  35 #if defined(CONFIG_32BIT) || defined(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 #ifdef CONFIG_64BIT
  43 extern char vdso64_start, vdso64_end;
  44 static void *vdso64_kbase = &vdso64_start;
  45 static unsigned int vdso64_pages;
  46 static struct page **vdso64_pagelist;
  47 #endif /* CONFIG_64BIT */
  48
  49 /*
  50  * Should the kernel map a VDSO page into processes and pass its
  51  * address down to glibc upon exec()?
  52  */
  53 unsigned int __read_mostly vdso_enabled = 1;
  54
  55 static int __init vdso_setup(char *s)
  56 {
  57         unsigned long val;
  58         int rc;
  59
  60         rc = 0;
  61         if (strncmp(s, "on", 3) == 0)
  62                 vdso_enabled = 1;
  63         else if (strncmp(s, "off", 4) == 0)
  64                 vdso_enabled = 0;
  65         else {
  66                 rc = strict_strtoul(s, 0, &val);
  67                 vdso_enabled = rc ? 0 : !!val;
  68         }
  69         return !rc;
  70 }
  71 __setup("vdso=", vdso_setup);
  72
  73 /*
  74  * The vdso data page
  75  */
  76 static union {
  77         struct vdso_data        data;
  78         u8                      page[PAGE_SIZE];
  79 } vdso_data_store __page_aligned_data;
  80 struct vdso_data *vdso_data = &vdso_data_store.data;
  81
  82 /*
  83  * Setup vdso data page.
  84  */
  85 static void vdso_init_data(struct vdso_data *vd)
  86 {
  87         vd->ectg_available = user_mode != HOME_SPACE_MODE && test_facility(31);
  88 }
  89
  90 #ifdef CONFIG_64BIT
  91 /*
  92  * Setup per cpu vdso data page.
  93  */
  94 static void vdso_init_per_cpu_data(int cpu, struct vdso_per_cpu_data *vpcd)
  95 {
  96 }
  97
  98 /*
  99  * Allocate/free per cpu vdso data.
 100  */
 101 #define SEGMENT_ORDER   2
 102
 103 int vdso_alloc_per_cpu(int cpu, struct _lowcore *lowcore)
 104 {
 105         unsigned long segment_table, page_table, page_frame;
 106         u32 *psal, *aste;
 107         int i;
 108
 109         lowcore->vdso_per_cpu_data = __LC_PASTE;
 110
 111         if (user_mode == HOME_SPACE_MODE || !vdso_enabled)
 112                 return 0;
 113
 114         segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
 115         page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
 116         page_frame = get_zeroed_page(GFP_KERNEL);
 117         if (!segment_table || !page_table || !page_frame)
 118                 goto out;
 119
 120         clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
 121                     PAGE_SIZE << SEGMENT_ORDER);
 122         clear_table((unsigned long *) page_table, _PAGE_TYPE_EMPTY,
 123                     256*sizeof(unsigned long));
 124
 125         *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
 126         *(unsigned long *) page_table = _PAGE_RO + page_frame;
 127
 128         psal = (u32 *) (page_table + 256*sizeof(unsigned long));
 129         aste = psal + 32;
 130
 131         for (i = 4; i < 32; i += 4)
 132                 psal[i] = 0x80000000;
 133
 134         lowcore->paste[4] = (u32)(addr_t) psal;
 135         psal[0] = 0x20000000;
 136         psal[2] = (u32)(addr_t) aste;
 137         *(unsigned long *) (aste + 2) = segment_table +
 138                 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
 139         aste[4] = (u32)(addr_t) psal;
 140         lowcore->vdso_per_cpu_data = page_frame;
 141
 142         vdso_init_per_cpu_data(cpu, (struct vdso_per_cpu_data *) page_frame);
 143         return 0;
 144
 145 out:
 146         free_page(page_frame);
 147         free_page(page_table);
 148         free_pages(segment_table, SEGMENT_ORDER);
 149         return -ENOMEM;
 150 }
 151
 152 void vdso_free_per_cpu(int cpu, struct _lowcore *lowcore)
 153 {
 154         unsigned long segment_table, page_table, page_frame;
 155         u32 *psal, *aste;
 156
 157         if (user_mode == HOME_SPACE_MODE || !vdso_enabled)
 158                 return;
 159
 160         psal = (u32 *)(addr_t) lowcore->paste[4];
 161         aste = (u32 *)(addr_t) psal[2];
 162         segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
 163         page_table = *(unsigned long *) segment_table;
 164         page_frame = *(unsigned long *) page_table;
 165
 166         free_page(page_frame);
 167         free_page(page_table);
 168         free_pages(segment_table, SEGMENT_ORDER);
 169 }
 170
 171 static void __vdso_init_cr5(void *dummy)
 172 {
 173         unsigned long cr5;
 174
 175         cr5 = offsetof(struct _lowcore, paste);
 176         __ctl_load(cr5, 5, 5);
 177 }
 178
 179 static void vdso_init_cr5(void)
 180 {
 181         if (user_mode != HOME_SPACE_MODE && vdso_enabled)
 182                 on_each_cpu(__vdso_init_cr5, NULL, 1);
 183 }
 184 #endif /* CONFIG_64BIT */
 185
 186 /*
 187  * This is called from binfmt_elf, we create the special vma for the
 188  * vDSO and insert it into the mm struct tree
 189  */
 190 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 191 {
 192         struct mm_struct *mm = current->mm;
 193         struct page **vdso_pagelist;
 194         unsigned long vdso_pages;
 195         unsigned long vdso_base;
 196         int rc;
 197
 198         if (!vdso_enabled)
 199                 return 0;
 200         /*
 201          * Only map the vdso for dynamically linked elf binaries.
 202          */
 203         if (!uses_interp)
 204                 return 0;
 205
 206 #ifdef CONFIG_64BIT
 207         vdso_pagelist = vdso64_pagelist;
 208         vdso_pages = vdso64_pages;
 209 #ifdef CONFIG_COMPAT
 210         if (is_compat_task()) {
 211                 vdso_pagelist = vdso32_pagelist;
 212                 vdso_pages = vdso32_pages;
 213         }
 214 #endif
 215 #else
 216         vdso_pagelist = vdso32_pagelist;
 217         vdso_pages = vdso32_pages;
 218 #endif
 219
 220         /*
 221          * vDSO has a problem and was disabled, just don't "enable" it for
 222          * the process
 223          */
 224         if (vdso_pages == 0)
 225                 return 0;
 226
 227         current->mm->context.vdso_base = 0;
 228
 229         /*
 230          * pick a base address for the vDSO in process space. We try to put
 231          * it at vdso_base which is the "natural" base for it, but we might
 232          * fail and end up putting it elsewhere.
 233          */
 234         down_write(&mm->mmap_sem);
 235         vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
 236         if (IS_ERR_VALUE(vdso_base)) {
 237                 rc = vdso_base;
 238                 goto out_up;
 239         }
 240
 241         /*
 242          * Put vDSO base into mm struct. We need to do this before calling
 243          * install_special_mapping or the perf counter mmap tracking code
 244          * will fail to recognise it as a vDSO (since arch_vma_name fails).
 245          */
 246         current->mm->context.vdso_base = vdso_base;
 247
 248         /*
 249          * our vma flags don't have VM_WRITE so by default, the process
 250          * isn't allowed to write those pages.
 251          * gdb can break that with ptrace interface, and thus trigger COW
 252          * on those pages but it's then your responsibility to never do that
 253          * on the "data" page of the vDSO or you'll stop getting kernel
 254          * updates and your nice userland gettimeofday will be totally dead.
 255          * It's fine to use that for setting breakpoints in the vDSO code
 256          * pages though
 257          *
 258          * Make sure the vDSO gets into every core dump.
 259          * Dumping its contents makes post-mortem fully interpretable later
 260          * without matching up the same kernel and hardware config to see
 261          * what PC values meant.
 262          */
 263         rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
 264                                      VM_READ|VM_EXEC|
 265                                      VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
 266                                      VM_ALWAYSDUMP,
 267                                      vdso_pagelist);
 268         if (rc)
 269                 current->mm->context.vdso_base = 0;
 270 out_up:
 271         up_write(&mm->mmap_sem);
 272         return rc;
 273 }
 274
 275 const char *arch_vma_name(struct vm_area_struct *vma)
 276 {
 277         if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
 278                 return "[vdso]";
 279         return NULL;
 280 }
 281
 282 static int __init vdso_init(void)
 283 {
 284         int i;
 285
 286         if (!vdso_enabled)
 287                 return 0;
 288         vdso_init_data(vdso_data);
 289 #if defined(CONFIG_32BIT) || defined(CONFIG_COMPAT)
 290         /* Calculate the size of the 32 bit vDSO */
 291         vdso32_pages = ((&vdso32_end - &vdso32_start
 292                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 293
 294         /* Make sure pages are in the correct state */
 295         vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
 296                                   GFP_KERNEL);
 297         BUG_ON(vdso32_pagelist == NULL);
 298         for (i = 0; i < vdso32_pages - 1; i++) {
 299                 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
 300                 ClearPageReserved(pg);
 301                 get_page(pg);
 302                 vdso32_pagelist[i] = pg;
 303         }
 304         vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
 305         vdso32_pagelist[vdso32_pages] = NULL;
 306 #endif
 307
 308 #ifdef CONFIG_64BIT
 309         /* Calculate the size of the 64 bit vDSO */
 310         vdso64_pages = ((&vdso64_end - &vdso64_start
 311                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 312
 313         /* Make sure pages are in the correct state */
 314         vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
 315                                   GFP_KERNEL);
 316         BUG_ON(vdso64_pagelist == NULL);
 317         for (i = 0; i < vdso64_pages - 1; i++) {
 318                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
 319                 ClearPageReserved(pg);
 320                 get_page(pg);
 321                 vdso64_pagelist[i] = pg;
 322         }
 323         vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
 324         vdso64_pagelist[vdso64_pages] = NULL;
 325 #ifndef CONFIG_SMP
 326         if (vdso_alloc_per_cpu(0, &S390_lowcore))
 327                 BUG();
 328 #endif
 329         vdso_init_cr5();
 330 #endif /* CONFIG_64BIT */
 331
 332         get_page(virt_to_page(vdso_data));
 333
 334         smp_wmb();
 335
 336         return 0;
 337 }
 338 arch_initcall(vdso_init);
 339
 340 int in_gate_area_no_mm(unsigned long addr)
 341 {
 342         return 0;
 343 }
 344
 345 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 346 {
 347         return 0;
 348 }
 349
 350 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 351 {
 352         return NULL;
 353 }