On Fri, Jan 09, 2026 at 01:08:33PM +0100, Fiona Ebner wrote:
> This fixes io-test 179 with qcow2. Otherwise, there would be more 'data'
> sectors.
> 
> Signed-off-by: Fiona Ebner <[email protected]>
> ---
> 
> Not sure if we should even attempt this without considering
> detect-zeroes. While it does keep 179 working, it is a change in
> behavior, since it could turn previously data sectors (with zeroes)
> into zero sectors when they are part of the head or tail of a zero
> write.
> 
> Would it even be tolerable from a performance perspective?

179 compares qemu-img map output and this makes changes to allocation status
fragile. In a case like this I would either adjust the test case and
regenerate the output, or if the output variance is now unavoidable then
a more sophisticated filter function is needed.

I'd avoid buffer_is_zero() in bdrv_co_do_zero_pwritev() since it eats
CPU cycles.

Stefan

Attachment: signature.asc
Description: PGP signature

Reply via email to