Allow display of negative offsets in print_address_symbolic()
authorKevin Buettner <kevinb@redhat.com>
Mon, 3 Jun 2019 01:12:28 +0000 (18:12 -0700)
committerKevin Buettner <kevinb@redhat.com>
Tue, 30 Jul 2019 16:19:13 +0000 (09:19 -0700)
commita1530dc7319595b5980e8501092782724c946351
tree3f80dd7d3fb7e006cdf1989570fb035ee6c612b2
parent2906593ffecef89f8d64e0f1ca21494be71d0ebd
Allow display of negative offsets in print_address_symbolic()

When examining addresses associated with blocks with non-contiguous
address ranges, it's not uncommon to see large positive offsets which,
for some address width, actually represent a smaller negative offset.
Here's an example taken from the test case (using the
dw2-ranges-func-lo-cold executable):

    (gdb) x/5i foo_cold
       0x40110d <foo+4294967277>: push   %rbp
       0x40110e <foo+4294967278>: mov    %rsp,%rbp
       0x401111 <foo+4294967281>: callq  0x401106 <baz>
       0x401116 <foo+4294967286>: nop
       0x401117 <foo+4294967287>: pop    %rbp

This commit, in conjuction with an earlier patch from this series, causes
cases like the above to be displayed like this (below) instead:

(gdb) x/5i foo_cold
   0x40110d <foo_cold>: push   %rbp
   0x40110e <foo-18>: mov    %rsp,%rbp
   0x401111 <foo-15>: callq  0x401106 <baz>
   0x401116 <foo-10>: nop
   0x401117 <foo-9>: pop    %rbp

Note that the address of foo_cold is now (due to another patch) being
displayed as <foo_cold> instead of <foo+BigOffset>.  The subsequent
lines are shown as negative offsets from foo.

Disassembly using the "disassemble" command is somewhat affected by
these changes:

Before:

(gdb) disassemble foo_cold
Dump of assembler code for function foo:
Address range 0x401120 to 0x40113b:
   0x0000000000401120 <+0>: push   %rbp
   0x0000000000401121 <+1>: mov    %rsp,%rbp
   0x0000000000401124 <+4>: callq  0x401119 <bar>
   0x0000000000401129 <+9>: mov    0x2ef1(%rip),%eax        # 0x404020 <e>
   0x000000000040112f <+15>: test   %eax,%eax
   0x0000000000401131 <+17>: je     0x401138 <foo+24>
   0x0000000000401133 <+19>: callq  0x40110d <foo+4294967277>
   0x0000000000401138 <+24>: nop
   0x0000000000401139 <+25>: pop    %rbp
   0x000000000040113a <+26>: retq
Address range 0x40110d to 0x401119:
   0x000000000040110d <+-19>: push   %rbp
   0x000000000040110e <+-18>: mov    %rsp,%rbp
   0x0000000000401111 <+-15>: callq  0x401106 <baz>
   0x0000000000401116 <+-10>: nop
   0x0000000000401117 <+-9>: pop    %rbp
   0x0000000000401118 <+-8>: retq
End of assembler dump.

After:

(gdb) disassemble foo_cold
Dump of assembler code for function foo:
Address range 0x401120 to 0x40113b:
   0x0000000000401120 <+0>: push   %rbp
   0x0000000000401121 <+1>: mov    %rsp,%rbp
   0x0000000000401124 <+4>: callq  0x401119 <bar>
   0x0000000000401129 <+9>: mov    0x2ef1(%rip),%eax        # 0x404020 <e>
   0x000000000040112f <+15>: test   %eax,%eax
   0x0000000000401131 <+17>: je     0x401138 <foo+24>
   0x0000000000401133 <+19>: callq  0x40110d <foo_cold>
   0x0000000000401138 <+24>: nop
   0x0000000000401139 <+25>: pop    %rbp
   0x000000000040113a <+26>: retq
Address range 0x40110d to 0x401119:
   0x000000000040110d <-19>: push   %rbp
   0x000000000040110e <-18>: mov    %rsp,%rbp
   0x0000000000401111 <-15>: callq  0x401106 <baz>
   0x0000000000401116 <-10>: nop
   0x0000000000401117 <-9>: pop    %rbp
   0x0000000000401118 <-8>: retq
End of assembler dump.

Note that negative offsets are now displayed without the leading "+".
Also, the callq to foo_cold is now displayed as such instead of a callq
to foo with a large positive offset.

gdb/ChangeLog:

* printcmd.c (print_address_symbolic): Print negative offsets.
(build_address_symbolic): Force signed arithmetic when computing
offset.
gdb/ChangeLog
gdb/printcmd.c