summaryrefslogtreecommitdiffstats
path: root/src/epa.cpp
diff options
context:
space:
mode:
authorFrédéric Brière <fbriere@fbriere.net>2019-12-27 02:26:38 -0500
committerFrédéric Brière <fbriere@fbriere.net>2019-12-27 02:28:05 -0500
commit91de36717a119f9501698af97550bfcdffd2875a (patch)
tree077f6a77d25d3c19935da32ab5dabec0a417e0c9 /src/epa.cpp
parent3d126cd9a7f9029e199e9220e2ffa08ac4e23227 (diff)
downloadtwinkle-91de36717a119f9501698af97550bfcdffd2875a.tar
twinkle-91de36717a119f9501698af97550bfcdffd2875a.tar.gz
twinkle-91de36717a119f9501698af97550bfcdffd2875a.tar.lz
twinkle-91de36717a119f9501698af97550bfcdffd2875a.tar.xz
twinkle-91de36717a119f9501698af97550bfcdffd2875a.zip
Introduce read-write-update locks and guards to prevent deadlocks
This converts t_rwmutex and t_rwmutex_guard into a read-write-update[*] lock and guard, in an attempt to circumvent the various deadlocks that were introduced with the addition of lines_mtx in 38bb6b7. [*] For more details, see https://stackoverflow.com/a/18785300 and http://lkml.iu.edu/hypermail/linux/kernel/0004.3/0117.html. Note that this is not a real fix; this would require analyzing and refactoring phone.cpp, which is well beyond my abilities. This is at best a workaround that appears to conveniently dodge all the deadlocks I've encountered so far. (It would have been more proper to introduce a separate class for this purpose, but this would have required modifying over 80 lines just to change one type for another. As phone_users_mtx is the only other instance of this class, the impact of subverting t_rwmutex directly is minimal.)
Diffstat (limited to 'src/epa.cpp')
0 files changed, 0 insertions, 0 deletions