Fix PR server/20414 - x32 gdbserver always crashes inferior
authorPedro Alves <palves@redhat.com>
Tue, 26 Jul 2016 21:21:27 +0000 (22:21 +0100)
committerPedro Alves <palves@redhat.com>
Tue, 26 Jul 2016 21:21:27 +0000 (22:21 +0100)
Debugging an x32 process with an x32 gdbserver always results in:

 (gdb) c
 Continuing.

 Program received signal SIGSEGV, Segmentation fault.
 0xf7de9600 in _dl_debug_state () from target:/libx32/ld-linux-x32.so.2
 (gdb)

Looking at the remote debug logs reveals the problem, here:

  Packet received: T05swbreak:;06:a0d4ffff00000000;07:b8d3ffff00000000;10:0096def701000000;thread:p7d7a.7d7a;core:1;
                                                                          ^^^^^^^^^^^^^^^^

The underlined value is the expedited value of RIP (in little endian).
But notice that 01 in 0x01f7de9600, while gdb says the program stopped
at 0xf7de9600.  0x01ffffffff is over 32 bits, which indicates that
something wen't wrong somewhere in gdbserver.

The problem turns out to be in gdbserver's x86_get_pc / x86_set_pc
routines, where "unsigned long" is used assuming that it can fit a
64-bit value, while unsigned long is actually 32-bit on x32.  The
result is that collect_register_by_name / supply_register_by_name end
up reading/writing random bytes off the stack.

Fix this by using explicit uint64_t instead of unsigned long.
For consistency, use uint32_t instead of unsigned int in the 32-bit
paths.

gdb/gdbserver/ChangeLog:
2016-07-26  Pedro Alves  <palves@redhat.com>

PR server/20414
* linux-x86-low.c (x86_get_pc, x86_set_pc): Use uint64_t instead
of unsigned long for 64-bit registers and use uint32_t instead of
unsigned int for 32-bit registers.

gdb/gdbserver/ChangeLog
gdb/gdbserver/linux-x86-low.c

index 5e6fb7e105bee7ea838c1f0aaa7ad4e077a72617..dfdbc219b2bafbe23d696b887eb4ae159d85bdd8 100644 (file)
@@ -1,3 +1,10 @@
+2016-07-26  Pedro Alves  <palves@redhat.com>
+
+       PR server/20414
+       * linux-x86-low.c (x86_get_pc, x86_set_pc): Use uint64_t instead
+       of unsigned long for 64-bit registers and use uint32_t instead of
+       unsigned int for 32-bit registers.
+
 2016-07-26  Pedro Alves  <palves@redhat.com>
 
        * linux-x86-low.c (x86_siginfo_fixup): Rename 'native' parameter
index 4e6dbce6a4959f70258a216648cb8b20d4fde0c5..d6b67c1b885537732cb6778cb1bca5e014613680 100644 (file)
@@ -427,13 +427,15 @@ x86_get_pc (struct regcache *regcache)
 
   if (use_64bit)
     {
-      unsigned long pc;
+      uint64_t pc;
+
       collect_register_by_name (regcache, "rip", &pc);
       return (CORE_ADDR) pc;
     }
   else
     {
-      unsigned int pc;
+      uint32_t pc;
+
       collect_register_by_name (regcache, "eip", &pc);
       return (CORE_ADDR) pc;
     }
@@ -446,12 +448,14 @@ x86_set_pc (struct regcache *regcache, CORE_ADDR pc)
 
   if (use_64bit)
     {
-      unsigned long newpc = pc;
+      uint64_t newpc = pc;
+
       supply_register_by_name (regcache, "rip", &newpc);
     }
   else
     {
-      unsigned int newpc = pc;
+      uint32_t newpc = pc;
+
       supply_register_by_name (regcache, "eip", &newpc);
     }
 }
This page took 0.044953 seconds and 4 git commands to generate.