1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
|
# In attr_file_free, the locks are acquired in the opposite order in which they
# are normally acquired. This is probably something worth fixing to have a
# consistent lock hierarchy that is easy to understand.
deadlock:attr_cache_lock
# git_mwindow_file_register has the possibility of evicting some files from the
# global cache. In order to avoid races and closing files that are currently
# being accessed, before evicting any file it will attempt to acquire that
# file's lock. Finally, git_mwindow_file_register is typically called with a
# file lock held, because the caller will use the fd in the mwf immediately
# after registering it. This causes ThreadSanitizer to observe different orders
# of acquisition of the mutex (which implies a possibility of a deadlock),
# _but_ since the files are added to the cache after other files have been
# evicted, there cannot be a case where mwf A is trying to be registered while
# evicting mwf B concurrently and viceversa: at most one of them can be present
# in the cache.
deadlock:git_mwindow_file_register
# When invoking the time/timezone functions from git_signature_now(), they
# access libc methods that need to be instrumented to correctly analyze the
# data races.
called_from_lib:libc.so.6
# TODO(#5592): Investigate and fix this. It can be triggered by the `thread`
# test suite.
race:git_filter_list__load_ext
|
