Reliable and fast DWARF-based stack unwinding

Abstract

Debug information, usually encoded in the DWARF format, is a hidden and obscure component of our computing infrastructure. Debug information is obviously used by debuggers, but it also plays a key role in program analysis tools, and, most surprisingly, it can be relied upon by the runtime of high-level programming languages. For instance the C++ runtime leverages DWARF stack unwind tables to implement exceptions! Alas, generating debug information adds significant burden to compiler implementations, and the debug information itself can be pervaded by subtle bugs, making the whole infrastructure unreliable. Additionally, interpreting the debug tables is a time-consuming task and, for some applications as sampling profilers, it turns out to be a performance bottleneck. In this paper we focus on the DWARF .eh_frame table, that enables stack unwinding in absence of framepointers. We will describe two techniques to perform validation and synthesis of the DWARF stack unwinding tables, and their implementation for the x86_64 architecture. The validation tool has proven effective for compiler and inline assembly testing, while the synthesis tool can generate DWARF unwind tables for arbitrary binaries lacking debug information. Additionally, we report on a technique to precompile unwind tables into native x86_64 code, which we have implemented and integrated into libunwind, resulting in 11x-25x DWARF-based unwind speedups. CCS Concepts: • Software and its engineering → Compilers; Software testing and debugging.

Retrofitting effect handlers onto OCaml

“…With these changes, we get the same backtrace for the meander program from §2.3, modulo runtime system functions due to effect handlers. We have verified the correctness of our DWARF directives using the verification tool from Bastian et al [2].…”

“…Our design retains OCaml's compatibility with program analysis tools that inspect the stack using DWARF unwind tables. We have validated our DWARF unwind tables with the assistance of an automated validator tool [2]. • a formal operational semantics for the effect handler implementation in OCaml.…”

“…While the complete details of DWARF stack unwinding is beyond the scope of the paper, it is beneficial to know how DWARF unwind tables are constructed in order to appreciate our solution. We refer the interested reader to Bastian et al [2] for a good overview of DWARF stack unwinding.…”

“…Despite the correct DWARF unwind information, using DWARF to record call stack information in perf only captures the call stack of the current fiber in Multicore OCaml. Since stack unwinding using DWARF is slow due to bytecode interpretation overhead, perf dumps the (user) call stack when sampled [2]. This only includes the frames from the current fiber.…”

“…This is a limitation of perf and not of our stack layout. Bastian et al [2] report on a technique to pre-compile the unwind table to assembly, which speeds up DWARF-based unwinding by up to 25×. With this technique, perf can unwind the stack at sample points rather than dumping the call stack, which would capture the complete backtrace rather than just the current fiber.…”