diff options
Diffstat (limited to 'results/classifier/gemma3:27b/runtime/1813307')
| -rw-r--r-- | results/classifier/gemma3:27b/runtime/1813307 | 24 |
1 files changed, 24 insertions, 0 deletions
diff --git a/results/classifier/gemma3:27b/runtime/1813307 b/results/classifier/gemma3:27b/runtime/1813307 new file mode 100644 index 000000000..f2902e8bd --- /dev/null +++ b/results/classifier/gemma3:27b/runtime/1813307 @@ -0,0 +1,24 @@ + + + +util/path.c/follow_path() does not handle "/" well + +Hello, + +I noticed that qemu does not handle "/" very well in follow_path(). + +Specifically, I was trying to run gdbserver under qemu, and it failed inside its implementation of __getcwd. + +Indeed it does something like + if (__lstat ("/", &st) < 0) +..... +and then loops from current dir toward the top using lstat("..") + +On qemu side, lstat forwards the request to follow_path() in util/path.c, and when passed "/", it returns the path in QEMU_LD_PREFIX (which was the top of my sysroot). +OTHT, the series of lstat("..") finally reaches the real device root because it's not recognized as "/" in follow_path(), so this is inconsistent and __getcwd fails. + +I suppose there's a good reason for returning QEMU_LD_PREFIX when asking for "/", but why is it so? + +If there's no good reason, maybe the behaviour could be changed to map "/" to "/" ? + +Thanks \ No newline at end of file |