Re: [Qemu-devel] [PATCH 14/19] block: insert event-tap to bdrv_aio_writev() and bdrv_aio_flush().

[Yoshiaki Tamura]

2011/1/19 Kevin Wolf <address@hidden>:
> Am 19.01.2011 14:16, schrieb Yoshiaki Tamura:
>> 2011/1/19 Kevin Wolf <address@hidden>:
>>> Am 19.01.2011 06:44, schrieb Yoshiaki Tamura:
>>>> event-tap function is called only when it is on, and requests sent
>>>> from device emulators.
>>>>
>>>> Signed-off-by: Yoshiaki Tamura <address@hidden>
>>>> ---
>>>>  block.c |   11 +++++++++++
>>>>  1 files changed, 11 insertions(+), 0 deletions(-)
>>>>
>>>> diff --git a/block.c b/block.c
>>>> index ff2795b..85bd8b8 100644
>>>> --- a/block.c
>>>> +++ b/block.c
>>>> @@ -28,6 +28,7 @@
>>>>  #include "block_int.h"
>>>>  #include "module.h"
>>>>  #include "qemu-objects.h"
>>>> +#include "event-tap.h"
>>>>
>>>>  #ifdef CONFIG_BSD
>>>>  #include <sys/types.h>
>>>> @@ -2111,6 +2112,11 @@ BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState 
>>>> *bs, int64_t sector_num,
>>>>      if (bdrv_check_request(bs, sector_num, nb_sectors))
>>>>          return NULL;
>>>>
>>>> +    if (bs->device_name && event_tap_is_on()) {
>>>> +        return event_tap_bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
>>>> +                                         cb, opaque);
>>>> +    }
>>>> +
>>>>      if (bs->dirty_bitmap) {
>>>>          blk_cb_data = blk_dirty_cb_alloc(bs, sector_num, nb_sectors, cb,
>>>>                                           opaque);
>>>
>>> Just noticed the context here... Does this patch break block migration
>>> when event-tap is on?
>>
>> I don't think so.  event-tap will call bdrv_aio_writev() upon
>> flushing requests and it shouldn't affect block-migration.  The
>> block written after the synchronization should be marked as dirty
>> and should be sent in the next round.  Am I missing the point?
>
> No, that makes sense. I don't have a complete understanding of the whole
> series yet, so there may be well more misunderstandings on my side.

It's OK.  I'm glad that you're reviewing.

>>> Another question that came to my mind is if we really hook everything we
>>> need. I think we'll need to have a hook in bdrv_flush as well. I don't
>>> know if you do hook qemu_aio_flush and friends -  does a call cause
>>> event-tap to flush its queue? If not, a call to qemu_aio_flush might
>>> hang qemu because it's waiting for requests to complete which are
>>> actually stuck in the event-tap queue.
>>
>> No it doesn't queue at event-tap.  Marcelo pointed that we should
>> hook bdrv_aio_flush to avoid requests inversion, that made sense
>> to me.  Do we need to hook bdrv_flush for that same reason?  If
>
> bdrv_flush() is the synchronous version of bdrv_aio_flush(), so in
> general it seems likely that we need to do the same.

Hmm.  Because it's synchronous, we need to start synchronization
right away, and once done, flush requests queued in event-tap
then return.

>> we hook bdrv_flush and qemu_aio_flush, we're going loop forever
>> because the synchronization code is calling vm_stop that call
>> bdrv_flush_all and qemu_aio_flush.
>
> qemu_aio_flush doesn't invoke any bdrv_* functions, so I don't see why
> we would loop forever. It just waits for AIO requests to complete.
>
> I just looked up the code and I think the situation is a bit different
> than I thought originally: qemu_aio_flush waits only for completion of
> requests which belong to a driver that has registered using
> qemu_aio_set_fd_handler. So this means that AIO requests queued in
> event-tap are not considered in-flight requests and we won't get stuck
> in a loop. Maybe we have no problem in fact. :-)

I had the same thoughts.  We don't have to hook qemu_aio_flush.

> On the other hand, e.g. migration relies on the fact that after a
> qemu_aio_flush, all AIO requests that the device model has submitted are
> completed. I think event-tap must take care that the requests which are
> queued are not forgotten to be migrated. (Maybe the code already
> considers this, I'm just writing down what comes to my mind...)

That's where event-tap is calling qemu_aio_flush.  It should be
almost same as for live migration.  Requests queued in event-tap
are replayed on the secondary side, that is the core design of
Kemari.

Thanks,

Yoshi

>
> Kevin
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to address@hidden
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>