mold: A Modern Linker

This is a repository of a linker I'm currently developing as an independent project for my Masters degree.

My goal is to make a linker that is almost as fast as concatenating object files with cat command. Concretely speaking, I want to use the linker to link a Chromium executable (about 1.8 GiB in size) just in 2 seconds. LLVM's lld, the fastest open-source linker which I originally created a few years ago, takes about 12 seconds to link Chromium on my machine. So the goal is 6x performance bump over lld. I don't know if I can ever achieve that, but it's worth a try. I need to create something anyway to earn units to graduate, and I want to (at least try to) create something useful.

I have quite a few new ideas as to how to achieve that speedup, though they are still just random unproved thoughts which need to be implemented and tested with benchmarks. Here is a brain dump:

Background

Even though lld has significantly improved the situation, linking is still one of the slowest steps in a build. It is in particular annoying when I changed one line of code and had to wait for a few seconds or even more for a linker to complete. It should be instantaneous. There's a need for a faster linker.
The number of cores on a PC has increased a lot lately, and this trend is expected to continue. However, the existing linkers can't take an advantage of that because they don't scale well for more cores. I have a 64-core/128-thread machine, so my goal is to create a linker that uses the CPU nicely. mold should be much faster than other linkers on 4 or 8-core machines too, though.
It looks to me that the designs of the existing linkers are somewhat similar, and I believe there are a lot of drastically different designs that haven't been explored yet. Develoeprs generally don't care about linkers as long as they work correctly, and they don't even think about creating a new one. So there may be lots of low hanging fruits there in this area.

Basic design

In order to achieve a cat-like performance, the most important thing is to fix the layout of an output file as quickly as possible, so that we can start copying actual data from input object files to an output executable/shared library file.
Copying data from input files to an output file is I/O-bounded, so there should be room for doing computationally-intensive tasks while copying data from one file to another.
After the first invocation of the linker, the linker should not exit but instead become a daemon to keep parsed input files in memory. The daemonized linker keeps an eye on the build directories using inotify(2), and as soon as a new file is created or an exiting file is updated, it reloads a file to memory.
Daemonizing alone wouldn't make the linker magically faster. We need to split the linker into two in such a way that the latter half of the process finishes as quickly as possible by speculatively parsing and preprocessing input files in the first half of the process. The key factor of success would be to design nice data structures that allows us to offload as much processing as possible from the second to the first half.
One of the most time-consuming stage among linker stages is symbol resolution. To resolve symbols, we basically have to throw all symbol strings into a hash table to match undefined symbols with defined symbols. But this can be done in the daemon rather than after the actual command line is specified using string interning.
Object files may contain a special section called a mergeable string section. The section contains lots of null-terminated strings, and the linker is expected to gather all mergeable string sections and merge their contents. So, if two object files contain the same string literal, for example, the resulting output will contain a single merged string. This step is time-consuming, but string merging can be done in the daemon using string interning.
Static archives (.a files) contain object files, but the static archive's string table contains only defined symbols of member object files and lacks other types of symbols. That makes static archives unsuitable for speculative parsing. The daemon should ignore the string table of static archive and directly read all member object files of all archives to get the whole picture of all possible input files.
If there's a relocation that uses a GOT of a symbol, then we have to create a GOT entry for that symbol. Otherwise, we shouldn't. That means we need to scan all relocation tables to fix the length and the contents of a .got section. This is perhaps time-consuming, but we can do that while copying data from input files to an output file. After the data copy is done, we can attach a .got section at the end of the output file.
Many linkers support incremental linking, but I think that's a hack to work around the slowness of regular linking. I want to focus on making the regular linking extremely fast.

Flow of Control

Step 1

Read all object files into memory and intern all symbol strings, comdat signature strings, mergeable strings and exception frame signature strings.

Step 2

Resolve all symbols to match undefined symbols with defined symbols.

Step 3

Eliminate unused archive member object files.

Step 4

Eliminate duplicate comdat groups.

Step 5 (do them in parallell)

Create output sections and add input sections to them.
Scan all relocations to fix the sizes of .plt, .got, .got.plt, .dynstr, .rela.dyn and .rela.plt sections.
Scan all mergeable strings to fix the sizes of mergeable sections. Also assign offsets within a section to mergeable strings.
Scan all .eh_frame's to fix the size of .eh_frame section. Also compute the size of .eh_frame_hdrs section.

Step 6

Assign file offsets in an output file and addresses in the virtual address space to all input and output sections.

Step 7 (do them in parallel)

Copy input sections to output sections.
Copy mergeable strings to merged output sections.
Scan all symbols to assign them regular, PLT and GOT addresses, and then fill .plt, .got, .plt, dynstr, rela.dyn and rela.plt sections.
Fill .eh_frame and .eh_frame_hdr sections.

Step 8

Apply relocations to copied input sections in an output file.

Compatibility

GNU ld, GNU gold and LLVM lld support essentially the same set of command line options and features. mold doesn't have to be completely compatible with them. As long as it can be used for linking large user-land programs, I'm fine with that. It is OK to leave some command line options unimplemented; if mold is blazingly fast, other projects would still be happy to adopt it by modifying their projects' build files.
I don't want to support the linker script language in mold because it's so complicated and inevitably slows down the linker. User-land programs rarely use linker scripts, so it shouldn't be a roadblock for most projects.
mold emits Linux executables and runs only on Linux. I won't avoid Unix-ism when writing code (e.g. I'll probably use fork(2)). I don't want to think about portability until mold becomes a thing that's worth to be ported.

Details

If we aim to the 2 seconds goal for Chromium, every millisecond counts. We can't ignore the latency of process exit. If we mmap a lot of files, _exit(2) is not instantaneous but takes a few hundred milliseconds because the kernel has to clean up a lot of resources. As a workaround, we should organize the linker command as two processes; the first process forks the second process, and the second process does the actual work. As soon as the second process writes a result file to a filesystem, it notifies the first process, and the first process exits. The second process can take time to exit, because it is not an interactive process.
Intel Threading Building Blocks (TBB) is a good library for parallel execution and has several concurrent containers. We are particularly interested in using parallel_for_each and concurrent_hash_map.
The output from the linker should be deterministic for the sake of build reproducibility and ease of debugging. This might add a little bit of overhead to the linker, but that shouldn't be too much.

Size of the problem

When linking Chrome, a linker reads 3,430,966,844 bytes of data in total. The data contains the following items:

Data item	Number
Relocations	62,024,719
Relocations against SHF_ALLOC sections	39,496,375
Sections	27,543,225
Symbols	23,953,607
Public defined symbols	10,512,135
Regular sections (*1)	10,345,314
Comdat groups	9,914,510
Mergeable strings	1,579,996
Public undefined symbols	1,428,149
Object files	30,723

(*1) Sections that have to be copied from input object files to an output file. Sections that contain relocations or symbols are for example excluded.

9.2 KiB Raw Blame History