Skip to content

kavon/llvm-statepoint-utils

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

72 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LLVM Statepoint Utilities

Build Status

LLVM's statepoint infrastructure generates stack frame layout information to enable precise garbage collection. However, the information is not in a suitable form for efficient lookups by an actual garbage collector scanning the stack. This design is intentional because "the runtime is expected to parse the [stack map] immediately after compiling a module and encode the information in its own format."

The utilities herein are designed to do just that: it can generate an efficient hash table at runtime that can be used by a garbage collector to walk the stack and find all pointers. Generating the table at runtime works around issues with position independent code, since the table is keyed on absolute return addresses. The code is pure, unadulterated C99* with no dependencies and a permissive license.

Note that this library was designed to work for programs whose stack map information was generated solely by gc.statepoint intrinsics, as these intrinsics generate specially formatted stack map records. If you're mixing patchpoint or regular stackmap intrinsics in the same code, you might need to modify the library in addition to marking call sites to differentiate them from statepoints.

The currently supported Stackmap Format is version 3, which is found in LLVM 5+.

how to build and use

  1. Run OPT_FLAG="-O3 -DNDEBUG" make
  2. Look inside dist and you should see a library file and a header file
  3. Enjoy

* almost... we rely on the packed attribute supported by popular C compilers (i.e., clang and gcc).

including these utils in your project

You can generate a single .c and corresponding .h file for inclusion in your own build system. To do this, run make unified, and the output code will be placed under build/.

a fancier implementation

To avoid having to generate the hash table each time the program starts up, you could extend this utility to instead generate a position-independent, static, callsite-offset table. For example, to lookup information about a callsite, we would:

  1. Take the return address, and subtract from it the starting address of the .text section, to obtain the callsite offset. The starting address would change on each launch because of ASLR, but it can be determined once during program startup and saved to a global variable.

  2. Use the call-site offset as the key into the statically allocated table. There are various ways of doing this, such as using a perfect hash function + a pointer array, or generating a standard hash table that is laid out statically in the data section.