arch/powerpc/oprofile/cell/vma_map.c

   1 /*
   2  * Cell Broadband Engine OProfile Support
   3  *
   4  * (C) Copyright IBM Corporation 2006
   5  *
   6  * Author: Maynard Johnson <maynardj@us.ibm.com>
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License
  10  * as published by the Free Software Foundation; either version
  11  * 2 of the License, or (at your option) any later version.
  12  */
  13
  14 /* The code in this source file is responsible for generating
  15  * vma-to-fileOffset maps for both overlay and non-overlay SPU
  16  * applications.
  17  */
  18
  19 #include <linux/mm.h>
  20 #include <linux/string.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/elf.h>
  23 #include "pr_util.h"
  24
  25
  26 void vma_map_free(struct vma_to_fileoffset_map *map)
  27 {
  28         while (map) {
  29                 struct vma_to_fileoffset_map *next = map->next;
  30                 kfree(map);
  31                 map = next;
  32         }
  33 }
  34
  35 unsigned int
  36 vma_map_lookup(struct vma_to_fileoffset_map *map, unsigned int vma,
  37                const struct spu *aSpu, int *grd_val)
  38 {
  39         /*
  40          * Default the offset to the physical address + a flag value.
  41          * Addresses of dynamically generated code can't be found in the vma
  42          * map.  For those addresses the flagged value will be sent on to
  43          * the user space tools so they can be reported rather than just
  44          * thrown away.
  45          */
  46         u32 offset = 0x10000000 + vma;
  47         u32 ovly_grd;
  48
  49         for (; map; map = map->next) {
  50                 if (vma < map->vma || vma >= map->vma + map->size)
  51                         continue;
  52
  53                 if (map->guard_ptr) {
  54                         ovly_grd = *(u32 *)(aSpu->local_store + map->guard_ptr);
  55                         if (ovly_grd != map->guard_val)
  56                                 continue;
  57                         *grd_val = ovly_grd;
  58                 }
  59                 offset = vma - map->vma + map->offset;
  60                 break;
  61         }
  62
  63         return offset;
  64 }
  65
  66 static struct vma_to_fileoffset_map *
  67 vma_map_add(struct vma_to_fileoffset_map *map, unsigned int vma,
  68             unsigned int size, unsigned int offset, unsigned int guard_ptr,
  69             unsigned int guard_val)
  70 {
  71         struct vma_to_fileoffset_map *new =
  72                 kzalloc(sizeof(struct vma_to_fileoffset_map), GFP_KERNEL);
  73         if (!new) {
  74                 printk(KERN_ERR "SPU_PROF: %s, line %d: malloc failed\n",
  75                        __FUNCTION__, __LINE__);
  76                 vma_map_free(map);
  77                 return NULL;
  78         }
  79
  80         new->next = map;
  81         new->vma = vma;
  82         new->size = size;
  83         new->offset = offset;
  84         new->guard_ptr = guard_ptr;
  85         new->guard_val = guard_val;
  86
  87         return new;
  88 }
  89
  90
  91 /* Parse SPE ELF header and generate a list of vma_maps.
  92  * A pointer to the first vma_map in the generated list
  93  * of vma_maps is returned.  */
  94 struct vma_to_fileoffset_map *create_vma_map(const struct spu *aSpu,
  95                                              unsigned long __spu_elf_start)
  96 {
  97         static const unsigned char expected[EI_PAD] = {
  98                 [EI_MAG0] = ELFMAG0,
  99                 [EI_MAG1] = ELFMAG1,
 100                 [EI_MAG2] = ELFMAG2,
 101                 [EI_MAG3] = ELFMAG3,
 102                 [EI_CLASS] = ELFCLASS32,
 103                 [EI_DATA] = ELFDATA2MSB,
 104                 [EI_VERSION] = EV_CURRENT,
 105                 [EI_OSABI] = ELFOSABI_NONE
 106         };
 107
 108         int grd_val;
 109         struct vma_to_fileoffset_map *map = NULL;
 110         void __user *spu_elf_start = (void __user *)__spu_elf_start;
 111         struct spu_overlay_info ovly;
 112         unsigned int overlay_tbl_offset = -1;
 113         Elf32_Phdr __user *phdr_start;
 114         Elf32_Shdr __user *shdr_start;
 115         Elf32_Ehdr ehdr;
 116         Elf32_Phdr phdr;
 117         Elf32_Shdr shdr, shdr_str;
 118         Elf32_Sym sym;
 119         int i, j;
 120         char name[32];
 121
 122         unsigned int ovly_table_sym = 0;
 123         unsigned int ovly_buf_table_sym = 0;
 124         unsigned int ovly_table_end_sym = 0;
 125         unsigned int ovly_buf_table_end_sym = 0;
 126         struct spu_overlay_info __user *ovly_table;
 127         unsigned int n_ovlys;
 128
 129         /* Get and validate ELF header.  */
 130
 131         if (copy_from_user(&ehdr, spu_elf_start, sizeof (ehdr)))
 132                 goto fail;
 133
 134         if (memcmp(ehdr.e_ident, expected, EI_PAD) != 0) {
 135                 printk(KERN_ERR "SPU_PROF: "
 136                        "%s, line %d: Unexpected e_ident parsing SPU ELF\n",
 137                        __FUNCTION__, __LINE__);
 138                 goto fail;
 139         }
 140         if (ehdr.e_machine != EM_SPU) {
 141                 printk(KERN_ERR "SPU_PROF: "
 142                        "%s, line %d: Unexpected e_machine parsing SPU ELF\n",
 143                        __FUNCTION__,  __LINE__);
 144                 goto fail;
 145         }
 146         if (ehdr.e_type != ET_EXEC) {
 147                 printk(KERN_ERR "SPU_PROF: "
 148                        "%s, line %d: Unexpected e_type parsing SPU ELF\n",
 149                        __FUNCTION__, __LINE__);
 150                 goto fail;
 151         }
 152         phdr_start = spu_elf_start + ehdr.e_phoff;
 153         shdr_start = spu_elf_start + ehdr.e_shoff;
 154
 155         /* Traverse program headers.  */
 156         for (i = 0; i < ehdr.e_phnum; i++) {
 157                 if (copy_from_user(&phdr, phdr_start + i, sizeof(phdr)))
 158                         goto fail;
 159
 160                 if (phdr.p_type != PT_LOAD)
 161                         continue;
 162                 if (phdr.p_flags & (1 << 27))
 163                         continue;
 164
 165                 map = vma_map_add(map, phdr.p_vaddr, phdr.p_memsz,
 166                                   phdr.p_offset, 0, 0);
 167                 if (!map)
 168                         goto fail;
 169         }
 170
 171         pr_debug("SPU_PROF: Created non-overlay maps\n");
 172         /* Traverse section table and search for overlay-related symbols.  */
 173         for (i = 0; i < ehdr.e_shnum; i++) {
 174                 if (copy_from_user(&shdr, shdr_start + i, sizeof(shdr)))
 175                         goto fail;
 176
 177                 if (shdr.sh_type != SHT_SYMTAB)
 178                         continue;
 179                 if (shdr.sh_entsize != sizeof (sym))
 180                         continue;
 181
 182                 if (copy_from_user(&shdr_str,
 183                                    shdr_start + shdr.sh_link,
 184                                    sizeof(shdr)))
 185                         goto fail;
 186
 187                 if (shdr_str.sh_type != SHT_STRTAB)
 188                         goto fail;;
 189
 190                 for (j = 0; j < shdr.sh_size / sizeof (sym); j++) {
 191                         if (copy_from_user(&sym, spu_elf_start +
 192                                                  shdr.sh_offset +
 193                                                  j * sizeof (sym),
 194                                            sizeof (sym)))
 195                                 goto fail;
 196
 197                         if (copy_from_user(name,
 198                                            spu_elf_start + shdr_str.sh_offset +
 199                                            sym.st_name,
 200                                            20))
 201                                 goto fail;
 202
 203                         if (memcmp(name, "_ovly_table", 12) == 0)
 204                                 ovly_table_sym = sym.st_value;
 205                         if (memcmp(name, "_ovly_buf_table", 16) == 0)
 206                                 ovly_buf_table_sym = sym.st_value;
 207                         if (memcmp(name, "_ovly_table_end", 16) == 0)
 208                                 ovly_table_end_sym = sym.st_value;
 209                         if (memcmp(name, "_ovly_buf_table_end", 20) == 0)
 210                                 ovly_buf_table_end_sym = sym.st_value;
 211                 }
 212         }
 213
 214         /* If we don't have overlays, we're done.  */
 215         if (ovly_table_sym == 0 || ovly_buf_table_sym == 0
 216             || ovly_table_end_sym == 0 || ovly_buf_table_end_sym == 0) {
 217                 pr_debug("SPU_PROF: No overlay table found\n");
 218                 goto out;
 219         } else {
 220                 pr_debug("SPU_PROF: Overlay table found\n");
 221         }
 222
 223         /* The _ovly_table symbol represents a table with one entry
 224          * per overlay section.  The _ovly_buf_table symbol represents
 225          * a table with one entry per overlay region.
 226          * The struct spu_overlay_info gives the structure of the _ovly_table
 227          * entries.  The structure of _ovly_table_buf is simply one
 228          * u32 word per entry.
 229          */
 230         overlay_tbl_offset = vma_map_lookup(map, ovly_table_sym,
 231                                             aSpu, &grd_val);
 232         if (overlay_tbl_offset < 0) {
 233                 printk(KERN_ERR "SPU_PROF: "
 234                        "%s, line %d: Error finding SPU overlay table\n",
 235                        __FUNCTION__, __LINE__);
 236                 goto fail;
 237         }
 238         ovly_table = spu_elf_start + overlay_tbl_offset;
 239
 240         n_ovlys = (ovly_table_end_sym -
 241                    ovly_table_sym) / sizeof (ovly);
 242
 243         /* Traverse overlay table.  */
 244         for (i = 0; i < n_ovlys; i++) {
 245                 if (copy_from_user(&ovly, ovly_table + i, sizeof (ovly)))
 246                         goto fail;
 247
 248                 /* The ovly.vma/size/offset arguments are analogous to the same
 249                  * arguments used above for non-overlay maps.  The final two
 250                  * args are referred to as the guard pointer and the guard
 251                  * value.
 252                  * The guard pointer is an entry in the _ovly_buf_table,
 253                  * computed using ovly.buf as the index into the table.  Since
 254                  * ovly.buf values begin at '1' to reference the first (or 0th)
 255                  * entry in the _ovly_buf_table, the computation subtracts 1
 256                  * from ovly.buf.
 257                  * The guard value is stored in the _ovly_buf_table entry and
 258                  * is an index (starting at 1) back to the _ovly_table entry
 259                  * that is pointing at this _ovly_buf_table entry.  So, for
 260                  * example, for an overlay scenario with one overlay segment
 261                  * and two overlay sections:
 262                  *      - Section 1 points to the first entry of the
 263                  *        _ovly_buf_table, which contains a guard value
 264                  *        of '1', referencing the first (index=0) entry of
 265                  *        _ovly_table.
 266                  *      - Section 2 points to the second entry of the
 267                  *        _ovly_buf_table, which contains a guard value
 268                  *        of '2', referencing the second (index=1) entry of
 269                  *        _ovly_table.
 270                  */
 271                 map = vma_map_add(map, ovly.vma, ovly.size, ovly.offset,
 272                                   ovly_buf_table_sym + (ovly.buf-1) * 4, i+1);
 273                 if (!map)
 274                         goto fail;
 275         }
 276         goto out;
 277
 278  fail:
 279         map = NULL;
 280  out:
 281         return map;
 282 }