Scribd
Upload
30 views12 pages

Optimizing Linux Kernel With BOLT

BOLT (Binary Optimization and Layout Tool) is designed to optimize compiled binaries, achieving significant speed improvements through code layout optimizations. It addresses challenges specific to the Linux kernel, such as self-modifying code and custom exception handling, while providing a structured pipeline for optimization. The document outlines the usage of BOLT, including profile collection and the optimization process, as well as the differences between BOLT and traditional compiler optimizations.

Uploaded by

τῇ καλλίστῃ
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
30 views12 pages

Optimizing Linux Kernel With BOLT

BOLT (Binary Optimization and Layout Tool) is designed to optimize compiled binaries, achieving significant speed improvements through code layout optimizations. It addresses challenges specific to the Linux kernel, such as self-modifying code and custom exception handling, while providing a structured pipeline for optimization. The document outlines the usage of BOLT, including profile collection and the optimization process, as well as the differences between BOLT and traditional compiler optimizations.

Uploaded by

τῇ καλλίστῃ
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

Optimizing Linux Kernel with BOLT

Maksim Panchenko
Facebook
BOLT

Overview

• What is BOLT
• How it works
• Linux Kernel Challenges
BOLT

What is BOLT?

Binary Optimization and Layout Tool


• Binary
• Compiled and linked executable or dynamic library
• Suffers from CPU front-end stalls
• Optimization
• Double digit speedup on top of PGO+LTO
• Layout
• Code Layout is the main optimization
BOLT

Code Layout Optimizations

Un-optimized Better Best


BOLT

Clang and GCC Speedup

60.00%
49.42%
50.00%

40.00% 36.22%
29.93%
30.00%
24.35%
21.26%
20.00% 15.73%

10.00%

0.00%
Clang GCC
BOLT PGO PGO+BOLT
BOLT

How to use BOLT?


• Collect profile
• Sample run
• Production environment
• Methods
• Sampling (Linux perf tool)
• BOLT instrumentation
• Link with –emit-relocs
• Apply optimizations by running BOLT
• $ llvm-bolt a.out –data perf.data –o a.out.fast <…>
BOLT

VS Compiler

Compiler
• Uses profile for optimizations – PGO/FDO/AutoFDO
• Layout is the last step in optimization pipeline
• Profile no longer accurate
Compiler Solution
• Use two profiles
• Compile twice
• (*) Context-sensitive profiling
BOLT

BOLT Pipeline

Discover • Find functions and data in code

Disassemble • Identify instructions inside functions

Build CFG • Analyze instructions and build CFG

Read Profile • Read profile and attribute to CFG edges

Optimize • Execute local and global optimization passes

Emit • Generate code and process relocations

Rewrite • Write code to file and update ELF


BOLT

Linux Kernel Challenges for BOLT

• Self-modifying and self-patching code


• Custom unwinding and exception handling
• Re-writing the binary
BOLT

Linux Kernel Challenges for BOLT

• Alternative instruction sequences


• .altinstructions, etc.
• Exceptions
• __ex_table, .pci_fixup
• __bug_table
• ORC
• Possible approach: selective optimizations

Facebook company
BOLT

Linux Kernel Challenges for BOLT

• Rewriting the binary


• Regular ELF
• Allocate new segments(s) for code and data
• Kernel
• Pre-allocate at link time
• Expand existing segments

Facebook company
Thank you

http://github.com/facebookincubator/BOLT

You might also like