On 5/2/24 09:33, Peter Krempa wrote:
On Thu, May 02, 2024 at 08:52:17 +0200, Michal Privoznik wrote:
> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com>
> ---
>
> Rebased version of the patch sent earlier, because the file was changed
> meanwhile.
Sorry for that. I've changed what Jirka requested and forgot to push my
patches on Tuesday :/
>
> NEWS.rst | 22 ++++++++++++++++++++++
> 1 file changed, 22 insertions(+)
>
> diff --git a/NEWS.rst b/NEWS.rst
> index 7ced82e3da..554ca559a2 100644
> --- a/NEWS.rst
> +++ b/NEWS.rst
> @@ -22,6 +22,13 @@ v10.3.0 (unreleased)
> USB address is now automatically assigned to USB network devices thus they
> can be used without manual configuration.
>
> + * conf: Introduce memReserve to <controller/>
> +
> + Some PCI devices have large non-prefetchable memory. This is not a problem
> + for coldplug because firmware sets up such devices properly. But it may be
> + a problem for hotplug devices. To resolve this, new ``memReserve``
> + attribute is introduced which allows overriding value computed by firmware.
The last sentence at first glance seems to contradict the previous
statement that the firmware does stuff correctly. I don't know the
details but is really firmware the problem when hotplugging? Isn't this
more of a hint to the guest OS?
I'm no expert in that area either, but the way I understand it: during
cold boot, FW calculates some values which may be then too small during
hotplug. If the device was there from the beginning (i.e. during cold
boot) FW would handle the device correctly (i.e. compute correct
values). But since FW can't foresee the future it stays conservative.
But users can! Hence the knob.
I'll try to rephrase the last sentence (or whole paragraph).
As I don't have a better suggestion:
Reviewed-by: Peter Krempa <pkrempa(a)redhat.com>
Thanks.
Michal