search

facebookarchive / BOLT

Binary Optimization and Layout Tool - A linux command-line utility used for optimizing performance of binaries
2.51k stars 176 forks source link

updateDWARFObjectAddressRanges: nullify low pc #268

Closed yota9 closed 2 years ago

yota9 commented 2 years ago

In case the case the DW_AT_ranges tag already exists for the object the low pc values won't be updated and will be incorrect in after-bolt binaries.

Vladislav Khmelevsky, Advanced Software Technology Lab, Huawei

yota9 commented 2 years ago

This is needed for the case if we have the input binary like this:

< 0><0x0000000b>  DW_TAG_compile_unit
                    DW_AT_name                  sync
                    DW_AT_language              DW_LANG_Go
                    DW_AT_stmt_list             0x00000000
                    DW_AT_low_pc                0x000a3c40
                    DW_AT_ranges                0x00000000
                ranges: 2 at .debug_ranges offset 0 (0x00000000) (32 bytes)
                        [ 0] range entry    0x00000000 0x0000129a
                        [ 1] range end      0x00000000 0x00000000
                    DW_AT_comp_dir              .
                    DW_AT_producer              Go cmd/compile go1.17.3; -shared regabi
                    <Unknown AT value 0x2905>   sync

The DW_AT_low_pc won't be updated and the address will be incorrect for the output. As I understand the gdb will use it's value as the range base, so the debugging won't work as expected (e.g. show sources & etc). After the patching the range entries will be written in absolute values and the low_pc must be zeroed. After BOLT with this patch:

                    DW_AT_low_pc                0x00000000
                    DW_AT_ranges                0x00000010
                ranges: 21 at .debug_ranges offset 16 (0x00000010) (336 bytes)
                        [ 0] range entry    0x00457056 0x004572e3
                        [ 1] range entry    0x00457300 0x00457344
                        [ 2] range entry    0x00457344 0x00457419
                        [ 3] range entry    0x0045741a 0x0045754c
                    .....
ayermolo commented 2 years ago

Hi.

Thanks for the update. There is an internal patch that I am working on that changes how we handle .debug_info from update to "re-write". This also affects that code. Would you mind waiting until that's in and rebasing on it?

yota9 commented 2 years ago

Hello @ayermolo ! Sure, also could you please tell me what do you think about #267 patch? Or you mean to rebase both the patches? I hope there won't be a lot of changes in these places :)

ayermolo commented 2 years ago

It looks fine. That whole mess with indirect will be simplified. We will be directly creating patches with forms we want.

ayermolo commented 2 years ago

Would it be possible to add a test for this?

yota9 commented 2 years ago

@aaupov I've found one more problem, after the "Rewrite of .debug_info section" patch I've got error in dwarfdump, didn't investigate it yet: dwarfdump ERROR: dwarf_child: DW_DLE_ABBREV_MISSING(394) Unable to find abbrev for DIE

ayermolo commented 2 years ago

@yota9 can you file a bug with repro, and assign it to me?

yota9 commented 2 years ago

@yota9 can you file a bug with repro, and assign it to me?

Sure, https://github.com/facebookincubator/BOLT/issues/274 but I don't have rights to assign issues :) Thanks!

yota9 commented 2 years ago

Re-opened in https://reviews.llvm.org/D117216