relocs: sync with the Linux kernel
authorH. Peter Anvin <hpa@zytor.com>
Tue, 29 May 2012 08:02:54 +0000 (01:02 -0700)
committerH. Peter Anvin <hpa@zytor.com>
Tue, 29 May 2012 08:06:10 +0000 (01:06 -0700)
Sync the relocs tool with the Linux kernel.  The new version of this
tool correctly verifies that any absolute symbol is either listed as
allowed absolute or is listed as relative.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
com32/tools/Makefile
com32/tools/include/tools/le_byteshift.h [new file with mode: 0644]
com32/tools/relocs.c

index 7badabd..0161baf 100644 (file)
@@ -15,6 +15,8 @@ include $(MAKEDIR)/build.mk
 
 BINS    = relocs
 
+INCLUDES += -I./include
+
 all : $(BINS)
 
 relocs : relocs.o
diff --git a/com32/tools/include/tools/le_byteshift.h b/com32/tools/include/tools/le_byteshift.h
new file mode 100644 (file)
index 0000000..c99d45a
--- /dev/null
@@ -0,0 +1,70 @@
+#ifndef _TOOLS_LE_BYTESHIFT_H
+#define _TOOLS_LE_BYTESHIFT_H
+
+#include <linux/types.h>
+
+static inline __u16 __get_unaligned_le16(const __u8 *p)
+{
+       return p[0] | p[1] << 8;
+}
+
+static inline __u32 __get_unaligned_le32(const __u8 *p)
+{
+       return p[0] | p[1] << 8 | p[2] << 16 | p[3] << 24;
+}
+
+static inline __u64 __get_unaligned_le64(const __u8 *p)
+{
+       return (__u64)__get_unaligned_le32(p + 4) << 32 |
+              __get_unaligned_le32(p);
+}
+
+static inline void __put_unaligned_le16(__u16 val, __u8 *p)
+{
+       *p++ = val;
+       *p++ = val >> 8;
+}
+
+static inline void __put_unaligned_le32(__u32 val, __u8 *p)
+{
+       __put_unaligned_le16(val >> 16, p + 2);
+       __put_unaligned_le16(val, p);
+}
+
+static inline void __put_unaligned_le64(__u64 val, __u8 *p)
+{
+       __put_unaligned_le32(val >> 32, p + 4);
+       __put_unaligned_le32(val, p);
+}
+
+static inline __u16 get_unaligned_le16(const void *p)
+{
+       return __get_unaligned_le16((const __u8 *)p);
+}
+
+static inline __u32 get_unaligned_le32(const void *p)
+{
+       return __get_unaligned_le32((const __u8 *)p);
+}
+
+static inline __u64 get_unaligned_le64(const void *p)
+{
+       return __get_unaligned_le64((const __u8 *)p);
+}
+
+static inline void put_unaligned_le16(__u16 val, void *p)
+{
+       __put_unaligned_le16(val, p);
+}
+
+static inline void put_unaligned_le32(__u32 val, void *p)
+{
+       __put_unaligned_le32(val, p);
+}
+
+static inline void put_unaligned_le64(__u64 val, void *p)
+{
+       __put_unaligned_le64(val, p);
+}
+
+#endif /* _TOOLS_LE_BYTESHIFT_H */
index 2474206..f06af6e 100644 (file)
@@ -1,6 +1,3 @@
-/*
- * This file is taken from the Linux kernel and is distributed under GPL v2.
- */
 #include <stdio.h>
 #include <stdarg.h>
 #include <stdlib.h>
 #define USE_BSD
 #include <endian.h>
 #include <regex.h>
-#include <sys/types.h>
+#include <tools/le_byteshift.h>
+
+static void die(char *fmt, ...);
 
 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
 static Elf32_Ehdr ehdr;
 static unsigned long reloc_count, reloc_idx;
 static unsigned long *relocs;
+static unsigned long reloc16_count, reloc16_idx;
+static unsigned long *relocs16;
 
 struct section {
        Elf32_Shdr     shdr;
@@ -29,60 +30,89 @@ struct section {
 };
 static struct section *secs;
 
-static void die(char *fmt, ...)
-{
-       va_list ap;
-       va_start(ap, fmt);
-       vfprintf(stderr, fmt, ap);
-       va_end(ap);
-       exit(1);
-}
+enum symtype {
+       S_ABS,
+       S_REL,
+       S_SEG,
+       S_LIN,
+       S_NSYMTYPES
+};
 
+static const char * const sym_regex_kernel[S_NSYMTYPES] = {
 /*
- * Following symbols have been audited.  Don't warn user about
+ * Following symbols have been audited. There values are constant and do
+ * not change if bzImage is loaded at a different physical address than
+ * the address for which it has been compiled. Don't warn user about
  * absolute relocations present w.r.t these symbols.
  */
+       [S_ABS] =
+       "^(__.*_len|__.*_dwords)$",
 
-/* True absolute relocations */
+/*
+ * These symbols are known to be relative, even if the linker marks them
+ * as absolute (typically defined outside any section in the linker script.)
+ */
+       [S_REL] =
+       "^(__.*_start|__.*_end|_end|_[se](text|data))$",
+};
 
-static const char safe_abs_regex[] =
-"^(__.*_len|__.*_dwords)$";
-static regex_t safe_abs_regex_c;
 
-static int is_safe_abs_reloc(const char *sym_name)
-{
-       return !regexec(&safe_abs_regex_c, sym_name, 0, NULL, 0);
-}
+static const char * const sym_regex_realmode[S_NSYMTYPES] = {
+/*
+ * These are 16-bit segment symbols when compiling 16-bit code.
+ */
+       [S_SEG] =
+       "^real_mode_seg$",
 
-/* These are relative even though the linker marks them absolute */
+/*
+ * These are offsets belonging to segments, as opposed to linear addresses,
+ * when compiling 16-bit code.
+ */
+       [S_LIN] =
+       "^pa_",
+};
 
-static const char safe_rel_regex[] =
-"^(__.*_start|__.*_end|_end|_[se](text|data))$";
-static regex_t safe_rel_regex_c;
+static const char * const *sym_regex;
 
-static int is_safe_rel_reloc(const char *sym_name)
+static regex_t sym_regex_c[S_NSYMTYPES];
+static int is_reloc(enum symtype type, const char *sym_name)
 {
-       return !regexec(&safe_rel_regex_c, sym_name, 0, NULL, 0);
+       return sym_regex[type] &&
+               !regexec(&sym_regex_c[type], sym_name, 0, NULL, 0);
 }
 
-static void regex_init(void)
+static void regex_init(int use_real_mode)
 {
-       char errbuf[128];
-       int err;
+        char errbuf[128];
+        int err;
+       int i;
 
-       err = regcomp(&safe_abs_regex_c, safe_abs_regex,
-                     REG_EXTENDED|REG_NOSUB);
-       if (err) {
-               regerror(err, &safe_abs_regex_c, errbuf, sizeof errbuf);
-               die("%s", errbuf);
-       }
+       if (use_real_mode)
+               sym_regex = sym_regex_realmode;
+       else
+               sym_regex = sym_regex_kernel;
 
-       err = regcomp(&safe_rel_regex_c, safe_rel_regex,
-                     REG_EXTENDED|REG_NOSUB);
-       if (err) {
-               regerror(err, &safe_rel_regex_c, errbuf, sizeof errbuf);
-               die("%s", errbuf);
-       }
+       for (i = 0; i < S_NSYMTYPES; i++) {
+               if (!sym_regex[i])
+                       continue;
+
+               err = regcomp(&sym_regex_c[i], sym_regex[i],
+                             REG_EXTENDED|REG_NOSUB);
+
+               if (err) {
+                       regerror(err, &sym_regex_c[i], errbuf, sizeof errbuf);
+                       die("%s", errbuf);
+               }
+        }
+}
+
+static void die(char *fmt, ...)
+{
+       va_list ap;
+       va_start(ap, fmt);
+       vfprintf(stderr, fmt, ap);
+       va_end(ap);
+       exit(1);
 }
 
 static const char *sym_type(unsigned type)
@@ -153,13 +183,16 @@ static const char *rel_type(unsigned type)
                REL_TYPE(R_386_RELATIVE),
                REL_TYPE(R_386_GOTOFF),
                REL_TYPE(R_386_GOTPC),
+               REL_TYPE(R_386_8),
+               REL_TYPE(R_386_PC8),
+               REL_TYPE(R_386_16),
+               REL_TYPE(R_386_PC16),
 #undef REL_TYPE
        };
-       const char *name = NULL;
-       if (type < ARRAY_SIZE(type_name))
+       const char *name = "unknown type rel type name";
+       if (type < ARRAY_SIZE(type_name) && type_name[type]) {
                name = type_name[type];
-       if (!name)
-               name = "unknown";
+       }
        return name;
 }
 
@@ -189,7 +222,7 @@ static const char *sym_name(const char *sym_strtab, Elf32_Sym *sym)
                name = sym_strtab + sym->st_name;
        }
        else {
-               name = sec_name(secs[sym->st_shndx].shdr.sh_name);
+               name = sec_name(sym->st_shndx);
        }
        return name;
 }
@@ -428,7 +461,7 @@ static void print_absolute_symbols(void)
        printf("\n");
 }
 
-static int print_absolute_relocs(FILE *f)
+static void print_absolute_relocs(void)
 {
        int i, printed = 0;
 
@@ -472,17 +505,18 @@ static int print_absolute_relocs(FILE *f)
                         * Before warning check if this absolute symbol
                         * relocation is harmless.
                         */
-                       if (is_safe_abs_reloc(name) ||
-                           is_safe_rel_reloc(name))
+                       if (is_reloc(S_ABS, name) || is_reloc(S_REL, name))
                                continue;
 
                        if (!printed) {
-                               fprintf(f, "Unknown absolute relocations present\n");
-                               fprintf(f, "Offset     Info     Type     Sym.Value Sym.Name\n");
+                               printf("WARNING: Absolute relocations"
+                                       " present\n");
+                               printf("Offset     Info     Type     Sym.Value "
+                                       "Sym.Name\n");
                                printed = 1;
                        }
 
-                       fprintf(f, "%08x %08x %10s %08x  %s\n",
+                       printf("%08x %08x %10s %08x  %s\n",
                                rel->r_offset,
                                rel->r_info,
                                rel_type(ELF32_R_TYPE(rel->r_info)),
@@ -492,12 +526,11 @@ static int print_absolute_relocs(FILE *f)
        }
 
        if (printed)
-               fputc('\n', f);
-
-       return printed;
+               printf("\n");
 }
 
-static void walk_relocs(void (*visit)(Elf32_Rel *rel, Elf32_Sym *sym))
+static void walk_relocs(void (*visit)(Elf32_Rel *rel, Elf32_Sym *sym),
+                       int use_real_mode)
 {
        int i;
        /* Walk through the relocations */
@@ -522,31 +555,71 @@ static void walk_relocs(void (*visit)(Elf32_Rel *rel, Elf32_Sym *sym))
                        Elf32_Rel *rel;
                        Elf32_Sym *sym;
                        unsigned r_type;
+                       const char *symname;
+                       int shn_abs;
+
                        rel = &sec->reltab[j];
                        sym = &sh_symtab[ELF32_R_SYM(rel->r_info)];
                        r_type = ELF32_R_TYPE(rel->r_info);
-                       /* Don't visit relocations to absolute symbols */
-                       if (sym->st_shndx == SHN_ABS &&
-                           !is_safe_rel_reloc(sym_name(sym_strtab, sym)))
-                               continue;
+
+                       shn_abs = sym->st_shndx == SHN_ABS;
 
                        switch (r_type) {
                        case R_386_NONE:
                        case R_386_PC32:
+                       case R_386_PC16:
+                       case R_386_PC8:
                        case R_386_GOTPC:
                        case R_386_GOTOFF:
                        case R_386_GOT32:
                        case R_386_PLT32:
-                               /* Relative relocations don't need to
-                                  be adjusted */
+                               /*
+                                * NONE can be ignored and and PC relative
+                                * relocations don't need to be adjusted.
+                                */
+                               break;
+
+                       case R_386_16:
+                               symname = sym_name(sym_strtab, sym);
+                               if (!use_real_mode)
+                                       goto bad;
+                               if (shn_abs) {
+                                       if (is_reloc(S_ABS, symname))
+                                               break;
+                                       else if (!is_reloc(S_SEG, symname))
+                                               goto bad;
+                               } else {
+                                       if (is_reloc(S_LIN, symname))
+                                               goto bad;
+                                       else
+                                               break;
+                               }
+                               visit(rel, sym);
                                break;
+
                        case R_386_32:
-                               /* Visit relocations that need adjustment */
+                               symname = sym_name(sym_strtab, sym);
+                               if (shn_abs) {
+                                       if (is_reloc(S_ABS, symname))
+                                               break;
+                                       else if (!is_reloc(S_REL, symname))
+                                               goto bad;
+                               } else {
+                                       if (use_real_mode &&
+                                           !is_reloc(S_LIN, symname))
+                                               break;
+                               }
                                visit(rel, sym);
                                break;
                        default:
                                die("Unsupported relocation type: %s (%d)\n",
                                    rel_type(r_type), r_type);
+                               break;
+                       bad:
+                               symname = sym_name(sym_strtab, sym);
+                               die("Invalid %s %s relocation: %s\n",
+                                   shn_abs ? "absolute" : "relative",
+                                   rel_type(r_type), symname);
                        }
                }
        }
@@ -554,8 +627,12 @@ static void walk_relocs(void (*visit)(Elf32_Rel *rel, Elf32_Sym *sym))
 
 static void count_reloc(Elf32_Rel *rel, Elf32_Sym *sym)
 {
-       (void)rel; (void)sym;
-       reloc_count += 1;
+       (void)sym;
+
+       if (ELF32_R_TYPE(rel->r_info) == R_386_16)
+               reloc16_count++;
+       else
+               reloc_count++;
 }
 
 static void collect_reloc(Elf32_Rel *rel, Elf32_Sym *sym)
@@ -563,7 +640,10 @@ static void collect_reloc(Elf32_Rel *rel, Elf32_Sym *sym)
        (void)sym;
 
        /* Remember the address that needs to be adjusted. */
-       relocs[reloc_idx++] = rel->r_offset;
+       if (ELF32_R_TYPE(rel->r_info) == R_386_16)
+               relocs16[reloc16_idx++] = rel->r_offset;
+       else
+               relocs[reloc_idx++] = rel->r_offset;
 }
 
 static int cmp_relocs(const void *va, const void *vb)
@@ -573,23 +653,41 @@ static int cmp_relocs(const void *va, const void *vb)
        return (*a == *b)? 0 : (*a > *b)? 1 : -1;
 }
 
-static void emit_relocs(int as_text)
+static int write32(unsigned int v, FILE *f)
+{
+       unsigned char buf[4];
+
+       put_unaligned_le32(v, buf);
+       return fwrite(buf, 1, 4, f) == 4 ? 0 : -1;
+}
+
+static void emit_relocs(int as_text, int use_real_mode)
 {
        int i;
        /* Count how many relocations I have and allocate space for them. */
        reloc_count = 0;
-       walk_relocs(count_reloc);
+       walk_relocs(count_reloc, use_real_mode);
        relocs = malloc(reloc_count * sizeof(relocs[0]));
        if (!relocs) {
                die("malloc of %d entries for relocs failed\n",
                        reloc_count);
        }
+
+       relocs16 = malloc(reloc16_count * sizeof(relocs[0]));
+       if (!relocs16) {
+               die("malloc of %d entries for relocs16 failed\n",
+                       reloc16_count);
+       }
        /* Collect up the relocations */
        reloc_idx = 0;
-       walk_relocs(collect_reloc);
+       walk_relocs(collect_reloc, use_real_mode);
+
+       if (reloc16_count && !use_real_mode)
+               die("Segment relocations found but --realmode not specified\n");
 
        /* Order the relocations for more efficient processing */
        qsort(relocs, reloc_count, sizeof(relocs[0]), cmp_relocs);
+       qsort(relocs16, reloc16_count, sizeof(relocs16[0]), cmp_relocs);
 
        /* Print the relocations */
        if (as_text) {
@@ -598,61 +696,83 @@ static void emit_relocs(int as_text)
                 */
                printf(".section \".data.reloc\",\"a\"\n");
                printf(".balign 4\n");
-               for (i = 0; i < reloc_count; i++) {
-                       printf("\t .long 0x%08lx\n", relocs[i]);
+               if (use_real_mode) {
+                       printf("\t.long %lu\n", reloc16_count);
+                       for (i = 0; i < reloc16_count; i++)
+                               printf("\t.long 0x%08lx\n", relocs16[i]);
+                       printf("\t.long %lu\n", reloc_count);
+                       for (i = 0; i < reloc_count; i++) {
+                               printf("\t.long 0x%08lx\n", relocs[i]);
+                       }
+               } else {
+                       /* Print a stop */
+                       printf("\t.long 0x%08lx\n", (unsigned long)0);
+                       for (i = 0; i < reloc_count; i++) {
+                               printf("\t.long 0x%08lx\n", relocs[i]);
+                       }
                }
+
                printf("\n");
        }
        else {
-               unsigned char buf[4];
-               /* Now print each relocation */
-               for (i = 0; i < reloc_count; i++) {
-                       buf[0] = (relocs[i] >>  0) & 0xff;
-                       buf[1] = (relocs[i] >>  8) & 0xff;
-                       buf[2] = (relocs[i] >> 16) & 0xff;
-                       buf[3] = (relocs[i] >> 24) & 0xff;
-                       fwrite(buf, 4, 1, stdout);
+               if (use_real_mode) {
+                       write32(reloc16_count, stdout);
+                       for (i = 0; i < reloc16_count; i++)
+                               write32(relocs16[i], stdout);
+                       write32(reloc_count, stdout);
+
+                       /* Now print each relocation */
+                       for (i = 0; i < reloc_count; i++)
+                               write32(relocs[i], stdout);
+               } else {
+                       /* Print a stop */
+                       write32(0, stdout);
+
+                       /* Now print each relocation */
+                       for (i = 0; i < reloc_count; i++) {
+                               write32(relocs[i], stdout);
+                       }
                }
-               /* Print a stop */
-               memset(buf, 0, sizeof buf);
-               fwrite(buf, 4, 1, stdout);
        }
 }
 
 static void usage(void)
 {
-       die("relocs [--abs-syms |--abs-relocs | --text] vmlinux\n");
+       die("relocs [--abs-syms|--abs-relocs|--text|--realmode] vmlinux\n");
 }
 
 int main(int argc, char **argv)
 {
        int show_absolute_syms, show_absolute_relocs;
-       int as_text;
+       int as_text, use_real_mode;
        const char *fname;
        FILE *fp;
        int i;
-       int err = 0;
 
        show_absolute_syms = 0;
        show_absolute_relocs = 0;
        as_text = 0;
+       use_real_mode = 0;
        fname = NULL;
        for (i = 1; i < argc; i++) {
                char *arg = argv[i];
                if (*arg == '-') {
-                       if (strcmp(argv[1], "--abs-syms") == 0) {
+                       if (strcmp(arg, "--abs-syms") == 0) {
                                show_absolute_syms = 1;
                                continue;
                        }
-
-                       if (strcmp(argv[1], "--abs-relocs") == 0) {
+                       if (strcmp(arg, "--abs-relocs") == 0) {
                                show_absolute_relocs = 1;
                                continue;
                        }
-                       else if (strcmp(argv[1], "--text") == 0) {
+                       if (strcmp(arg, "--text") == 0) {
                                as_text = 1;
                                continue;
                        }
+                       if (strcmp(arg, "--realmode") == 0) {
+                               use_real_mode = 1;
+                               continue;
+                       }
                }
                else if (!fname) {
                        fname = arg;
@@ -663,10 +783,7 @@ int main(int argc, char **argv)
        if (!fname) {
                usage();
        }
-
-
-       regex_init();
-
+       regex_init(use_real_mode);
        fp = fopen(fname, "r");
        if (!fp) {
                die("Cannot open %s: %s\n",
@@ -682,10 +799,9 @@ int main(int argc, char **argv)
                return 0;
        }
        if (show_absolute_relocs) {
-               print_absolute_relocs(stdout);
+               print_absolute_relocs();
                return 0;
        }
-       err = print_absolute_relocs(stderr);
-       emit_relocs(as_text);
-       return err;
+       emit_relocs(as_text, use_real_mode);
+       return 0;
 }