Re: [Qemu-devel] [PATCH] fixup! block-migration: acquire AioContext as necessary

[Fam Zheng <famz@redhat.com>]

On Tue, 02/23 10:42, Paolo Bonzini wrote:
> 
> 
> On 23/02/2016 06:39, Fam Zheng wrote:
> > BTW, could you also explain the blk_mig_lock() question (*) I had?
> > 
> > *: https://lists.gnu.org/archive/html/qemu-devel/2016-02/msg03317.html
> 
> Sorry, missed that:
> 
> > 
> >> @@ -597,21 +627,28 @@ static void block_migration_cleanup(void *opaque)
> >>  {
> >>      BlkMigDevState *bmds;
> >>      BlkMigBlock *blk;
> >> +    AioContext *ctx;
> >>  
> >>      bdrv_drain_all();
> >>  
> >>      unset_dirty_tracking();
> >>  
> >> -    blk_mig_lock();
> > 
> > Why is it okay to skip the blk_mig_lock() for block_mig_state.bmds_list?
> 
> The bmds_list is not protected by blk_mig_lock:
> 
>     /* Written during setup phase.  Can be read without a lock.  */
>     int blk_enable;
>     int shared_base;
>     QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list;
>     int64_t total_sector_sum;
>     bool zero_blocks;
> 
> block_migration_cleanup is called from qemu_savevm_state_cleanup, when
> there can be no concurrency between the migration thread and the I/O
> thread.  In fact the call of qemu_savevm_state_cleanup might as well be
> moved out of the migration thread.

Thanks!

Fam