On 06/29/2017 08:55 PM, Eric Blake wrote:
> On 06/28/2017 07:05 AM, Vladimir Sementsov-Ogievskiy wrote:
>> Add bdrv_dirty_bitmap_deserialize_ones() function, which is needed for
>> qcow2 bitmap loading, to handle unallocated bitmap parts, marked as
>> all-ones.
>>

>> + * hbitmap_deserialize_ones
>> + * @hb: HBitmap to operate on.
>> + * @start: First bit to restore.
>> + * @count: Number of bits to restore.
> 
> This part is accurate (the dirty-bitmap is using an underlying bitmap
> with "one bit per sector" before my series, afterwards it will be "one
> bit per byte", remembering that hbitmap really stores only one bit per
> granularity multiple of the underlying unit), if incomplete (the code
> asserts that things are aligned, but doesn't document that the caller
> must pass in aligned values); but again, that's matching the
> pre-existing deserialize_zeroes code.

Okay, I looked again; the documentation for
hbitmap_serialization_granularity() has a blanket statement that all
other hbitmap_serialization_* functions taking a start and count must be
aligned. Indirect, but at least documented, so I retract my statement
about the docs being incomplete.

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