Re: [Qemu-devel] [PATCH v10 09/17] qcow2: Optimize write zero of unaligned tail cluster

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1247 bytes --]

On 04/28/2017 04:24 PM, Eric Blake wrote:

>>> +echo
>>> +echo == unaligned image tail cluster, no allocation needed ==
>>> +
>>> +CLUSTER_SIZE=1024 TEST_IMG="$TEST_IMG.base" _make_test_img $((size + 1024))
>>
>> Any reason for the CLUSTER_SIZE? It passes with 64 kB as well, and I
>> actually think that would be a better test because as it is, the image's
>> "unaligned" tail is exactly one cluster (so it isn't really unaligned).
> 
> Uhhh - copy-and-paste?  You're right, that 1024 is too small for what
> I'm actually doing with it.  :(

Actually, the whole test defaults to 4k clusters except when overridden,
but I did learn while hammering at things that we don't have a nice way
to tell that a backing file slightly shorter than the active file
behaves as though we read all zeros for the difference. I'm thinking of
proposing an RFC patch introducing BDRV_BLOCK_EOF, which gets set any
time bdrv_get_block_status() clamps the returns *pnum because it hit
EOF, to see what optimizations fall out elsewhere in the tree as a
result.  (It may not be worth it in the end, but we'll see).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]