[v4,1/2] Fix `add-symbol-file -o ... -s ...` address mapping bug

Message ID 20260926064742.29327-2-dragorn421@gmail.com
State New
Headers
Series Fix `add-symbol-file -o ... -s ...` address mapping bug |

Checks

Context Check Description
linaro-tcwg-bot/tcwg_gdb_build--master-aarch64 success Build passed
linaro-tcwg-bot/tcwg_gdb_build--master-arm success Build passed
linaro-tcwg-bot/tcwg_gdb_check--master-aarch64 success Test passed
linaro-tcwg-bot/tcwg_gdb_check--master-arm success Test passed

Commit Message

Dragorn421 Sept. 26, 2026, 6:47 a.m. UTC
  The 'add-symbol-file' command allows users to map sections explicitly
("-s SECTION ADDR") and apply an offset ("-o OFFSET"). The documentation
explains:

   If an optional @var{offset} is specified, it is added to the start
   address of each section, except those for which the address was
   specified explicitly.

But those two options currently may not be supplied together in the
same command invocation without encountering buggy behavior.

For example if mapping the .text section from main.o at 0x1234:

```
(gdb) add-symbol-file main.o -s .text 0x1234
add symbol table from file "main.o" at
          .text_addr = 0x1234
(y or n) y
Reading symbols from main.o...
(No debugging symbols found in main.o)
(gdb) info files
...
          0x0000000000001030 - 0x0000000000001040 is .plt.got
          0x0000000000001234 - 0x000000000000132c is .text
          0x0000000000001138 - 0x0000000000001145 is .fini
...
```

`info files` does then report the expect address for .text

However this breaks when combining with the -o "set offset for other
unspecified sections" option:

```
(gdb) add-symbol-file main.o -o 0xFF000000 -s .text 0x1234
add symbol table from file "main.o" at
          .text_addr = 0x1234
with other sections offset by 0xff000000
(y or n) y
Reading symbols from main.o...
(No debugging symbols found in main.o)
(gdb) info files
...
          0x00000000ff001030 - 0x00000000ff001040 is .plt.got
          0x0000000000001040 - 0x0000000000001138 is .text
          0x00000000ff001138 - 0x00000000ff001145 is .fini
...
```

We notice here the -o option was correctly used per the addresses of
e.g. the `.fini` section, but the .text section address is now wrong.

This behavior boils down to the `set_objfile_default_section_offset`
function assuming it can pass 0 as `objfile_relocate`'s `offsets`
entries to not modify the `-s` mappings previously set by the
`symbol_file_add` call in `add_symbol_file_command`. When in fact
`objfile_relocate` does not check for 0, it only checks for the new
offset being the same as the current one.

The fix is to change `set_objfile_default_section_offset` to pass
the current offset instead of 0 for sections that should be unaltered.

Reviewed-By: Keith Seitz <keiths@redhat.com>
---
 gdb/symfile.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
  

Patch

diff --git a/gdb/symfile.c b/gdb/symfile.c
index 017f7a49d8d..a5f4e64fa6b 100644
--- a/gdb/symfile.c
+++ b/gdb/symfile.c
@@ -2139,8 +2139,8 @@  set_objfile_default_section_offset (struct objfile *objf,
     = addrs_section_sort (objf_addrs);
 
   /* Walk the BFD section list, and if a matching section is found in
-     ADDRS_SORTED_LIST, set its offset to zero to keep its address
-     unchanged.
+     ADDRS_SORTED_LIST, set its offset to its current offset to keep
+     its address unchanged.
 
      Note that both lists may contain multiple sections with the same
      name, and then the sections from ADDRS are matched in BFD order
@@ -2163,7 +2163,8 @@  set_objfile_default_section_offset (struct objfile *objf,
 	}
 
       if (cmp == 0)
-	offsets[objf_sect->sectindex] = 0;
+	offsets[objf_sect->sectindex]
+	  = objf->section_offsets[objf_sect->sectindex];
     }
 
   /* Apply the new section offsets.  */