I'm running out of time and will be offline for the next two weeks, so I'm just sending out what I have right now. This is probably not ready to be merged yet, but if need be, someone else can pick it up. Otherwise I'll do that myself when I return.
This is related to the following bug reports: https://bugzilla.redhat.com/show_bug.cgi?id=1614623 https://bugzilla.redhat.com/show_bug.cgi?id=1601212 Essentially it tries to fix a few deadlocks into which we run when running two commit jobs (probably just one example for problematic scenarios) on nodes that are in a non-mainloop AioContext (i.e. with dataplane). Obviously, besides verifying that the code changes are actually correct as they are, we also want to add some test cases for this. I suppose test-bdrv-drain should test for the individual bugs, and a qemu-iotests case could try the higher level scenario of multiple commit jobs with dataplane. Kevin Wolf (5): blockjob: Wake up BDS when job becomes idle tests: Acquire AioContext around job_finish_sync() job: Drop AioContext lock around aio_poll() block: Drop AioContext lock in bdrv_drain_poll_top_level() [WIP] Lock AioContext in bdrv_co_drain_bh_cb() include/block/blockjob.h | 13 +++++++++++++ include/qemu/job.h | 9 +++++++++ block/io.c | 31 ++++++++++++++++++++++++++++++- blockjob.c | 18 ++++++++++++++++++ job.c | 10 ++++++++++ tests/test-bdrv-drain.c | 6 ++++++ tests/test-blockjob.c | 6 ++++++ 7 files changed, 92 insertions(+), 1 deletion(-) -- 2.13.6
