On 08/04/2017 09:32, John Snow wrote:
>
>
> On 03/23/2017 01:39 PM, Paolo Bonzini wrote:
>> All block jobs are using block_job_defer_to_main_loop as the final
>> step just before the coroutine terminates. At this point,
>> block_job_enter should do nothing, but currently it restarts
>> the freed coroutine.
>>
>> Now, the job->co states should probably be changed to an enum
>> (e.g. BEFORE_START, STARTED, YIELDED, COMPLETED) subsuming
>
> Yes, I'd love to formalize the FSM for jobs.
>
>> block_job_started, job->deferred_to_main_loop and job->busy.
>> For now, this patch eliminates the problematic reenter by
>> removing the reset of job->deferred_to_main_loop (which served
>> no purpose, as far as I could see) and checking the flag in
>> block_job_enter.
>
> Not sure -- the original commit 794f01414 makes it seem like it should
> stay so that the correct AIO context can be acquired. Probably a race as
> jobs don't often stay long once they've deferred to the main loop, but I
> think the reset is harmless as you say.
You're right. The difference is that now we pretty much expect the
deferred part to be at the end of the job:
commit bae8196d9f97916de6323e70e3e374362ee16ec4
Author: Paolo Bonzini <pbonz...@redhat.com>
Date: Thu Oct 27 12:48:50 2016 +0200
blockjob: introduce .drain callback for jobs
This is required to decouple block jobs from running in an
AioContext. With multiqueue block devices, a BlockDriverState
does not really belong to a single AioContext.
The solution is to first wait until all I/O operations are
complete; then loop in the main thread for the block job to
complete entirely.
so I'll improve the commit message to point to both 794f01414 and this one.
Paolo
>>
>> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
>> ---
>> blockjob.c | 10 ++++++++--
>> include/block/blockjob_int.h | 3 ++-
>> 2 files changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/blockjob.c b/blockjob.c
>> index 3fa2885..2d80dae 100644
>> --- a/blockjob.c
>> +++ b/blockjob.c
>> @@ -750,7 +750,14 @@ void block_job_resume_all(void)
>>
>> void block_job_enter(BlockJob *job)
>> {
>> - if (job->co && !job->busy) {
>> + if (!block_job_started(job)) {
>> + return;
>> + }
>> + if (job->deferred_to_main_loop) {
>> + return;
>> + }
>> +
>> + if (!job->busy) {
>> qemu_coroutine_enter(job->co);
>> }
>> }
>> @@ -874,7 +881,6 @@ static void block_job_defer_to_main_loop_bh(void *opaque)
>> aio_context_acquire(aio_context);
>> }
>>
>> - data->job->deferred_to_main_loop = false;
>> data->fn(data->job, data->opaque);
>>
>> if (aio_context != data->aio_context) {
>> diff --git a/include/block/blockjob_int.h b/include/block/blockjob_int.h
>> index 97ffc43..4d287ba 100644
>> --- a/include/block/blockjob_int.h
>> +++ b/include/block/blockjob_int.h
>> @@ -227,7 +227,8 @@ typedef void BlockJobDeferToMainLoopFn(BlockJob *job,
>> void *opaque);
>> * @fn: The function to run in the main loop
>> * @opaque: The opaque value that is passed to @fn
>> *
>> - * Execute a given function in the main loop with the BlockDriverState
>> + * This function must be called by the main job coroutine just before it
>> + * returns. @fn is executed in the main loop with the BlockDriverState
>> * AioContext acquired. Block jobs must call bdrv_unref(), bdrv_close(),
>> and
>> * anything that uses bdrv_drain_all() in the main loop.
>> *
>>
>
> Reviewed-by: John Snow <js...@redhat.com>
>
