https://github.com/stevana/hs-mmap-mutablebytearray
https://github.com/stevana/hs-mmap-mutablebytearray
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/stevana/hs-mmap-mutablebytearray
- Owner: stevana
- License: mit
- Created: 2022-01-03T09:54:15.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-01-04T11:12:41.000Z (over 3 years ago)
- Last Synced: 2025-02-08T11:34:58.032Z (4 months ago)
- Language: Haskell
- Size: 11.7 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# hs-mmap-mutablebytearray
This repo contains a minimum example of how combining `mmap` with page aligned and
pinned `MutableByteArray#`s leads to a segfault.
It's not clear, to me, if the bug is in this repo or in GHC.
Tested with latest GHC available in `ghcup` at the time, which is GHC version
9.2.1.
The following steps should let you reproduce the segfault locally:
```bash
git clone https://github.com/stevana/hs-mmap-mutablebytearray.git
cd hs-mmap-mutablebytearray
make
make ulimit # enable core dumps
make strace # run the program under strace
make gdb # debug the core dump
```
Some observations:
* `mmap` by itself doesn't seem to be the problem, see the `worksMmap` example
in the `Mmap` module, and run it with `make works`. This example uses the
`posix_memalign` syscall to allocate page size aligned memory (it needs to be
page aligened for `mmap` to work, and the `Foregin` modules don't provide a
way to do that as far as I could tell.)
* Depending on how big we make the `mmap`ed `MutableByteArray#` we get segfaults
at different places. If the bytearray is of the same size as the page size (as
is the case when run with `make`), then `gdb` and `backtrace` tell us that the
segfault is caused by:
https://gitlab.haskell.org/ghc/ghc/-/blob/master/rts/sm/BlockAlloc.c#L833 .
Whereas if the size of the byte array is smaller (as in the case when run with
`make small`), e.g. 16 bytes, then `backtrace` in `gdb` mentions
`stg_newAlignedPinnedByteArrayzh`:
https://gitlab.haskell.org/ghc/ghc/-/blob/master/rts/PrimOps.cmm#L113 .
After asking in #haskell and #ghc on IRC the following was pointed out to me:
* A similar request about being able to `mmap` byte arrays has been
discussed in the following ticket:
https://gitlab.haskell.org/ghc/ghc/-/issues/17747
* The problems can perhaps be avoided by first `mmap`ing a `Ptr`
then using `fetchAndWordAddr#` rather than working with byte arrays.