I have created my own custom section to be added at the end into a
binary upon compilation which contains address of all basic blocks. As
the final address of the basic block is not known until link time, I
collect the MCSymbol* Symbol Values per BB in a temp array and at the
in the custom section and emit it (emitSymbolValue) into my section
I have found people looking at slightly different uses of the basic
block address but running into the same problem as me when compiling.
$clang calc_pi.c -o calc_pi
fatal error: error in backend: Undefined temporary symbol
However when only compiling (incompletely via -S),
e.g. $clang calc_pi.c -o calc_pi -S ,
it does compile with the expected outputs and the basic block labels
properly generated in my section. I get a file with proper labels
TLDR: compiles correctly, will not link successfully to make a binary
because the symbol of the basic block in the .text section doesn't
exist. How do I ensure the Basic Block Symbol does not get destroyed
until my section is read in as well during linking?
Followup Question: How do I ensure that after my section has been
fleshed out with addresses, that I may then destroy those symbol
values as intended.
I feel it has something to do with the liveness of the basic blocks
symbols however I do not know where they are "destroyed" to prevent
this from happening to allow my custom section containing the basic
block labels to also be "converted to addresses" as in its original
I don't think file is actually "compiling fine", the error message indicates that it is failing while trying to write the object file.
Looking at the assembly file I think you are emitting the section name but I don't see any labels inside your section (unless you mean __custome_section is a label). If your Basic block have no labels I think whatever object writer you use will not create symbols for it. Perhaps you should look around that area.
My assembly file shows that my new section (__custom_section) is being
created and populated with the basic block labels (.quad .LBB0_0,
.LBB0_1, ... etc) and other statistical data I am inputting (.long
19). If I compile without adding these basic block labels (.LBB0_0,
etc) my application compiles, links, and runs successfully as the
vanilla application would. Are you saying .LBB0_0, etc. is not an
actual basic block label used by LLVM? Inspecting assembly of other
parts of the file, I see branches reference these "things" (e.g. jge
Thanks for the response.
Problem Solved! I'll reply in case some looks for something similar in
the future. Artem pointed out:
AFAIK the symbols starting with ".L" are not exported and thus inaccessible during linking.
There a special hack for that somewhere.
Indeed this is true, after discussing with teammates, these Basic
block symbols are local labels (e.g. .LBB0_0, .LBB0_1 ...) are not
exported and not defined any more once computing the symbol table
"Undefined temporary symbol" in function ELFObjectWriter::computeSymbolTable
The work around I came up with the help of my teammate was to create a
new symbol, borrowing the Basic Blocks name with a custom prefix and
emiting a label for that newly created symbol. This was implemented in
EmitFunctionBody(). Then, adding this label to my custom section.
Compilation and linking worked giving preserving program behavior and
my custom section contained the correct address of each functions
basic blocks when inspecting the disassembly.
Thank you to Artem and Sameer for your help super much!