[Linuxppc-users] Live VM disk removal failure

[Tyrel Datwyler]

On 07/05/2017 11:41 PM, Michael Ellerman wrote:
> Ricardo Marin Matinata <rmm at br.ibm.com> writes:
> 
>> Also, "for the archives", the reason you won't need rpadlpar_io module 
>> (and linux-image-extra-*) on 16.10 and onwards is because ppc64-diag no 
>> longer requires it in more recent versions.
> 
> No longer requires it under some circumstances AFAICS?
> 
>   static int check_kmods(void)
>   {
>   	struct stat sbuf;
>   	int rc;
>   
>   	/* We only need to do this for PHB/SLOT/PCI operations */
>   	if (usr_drc_type == DRC_TYPE_PCI || usr_drc_type == DRC_TYPE_PHB ||
>   	    usr_drc_type == DRC_TYPE_SLOT)
>   		return 0;
>   
>   	/* We don't use rpadlar_io/rpaphp for PCI operations run with the
>   	 * -v / virtio flag, which relies on generic PCI rescan instead
>   	 */
>   	if (usr_drc_type == DRC_TYPE_PCI && pci_virtio)
>   		return 0;
> 
> 
> Or is it always run with "-v" these days?

Only under qemu/kvm. This was originally a work around as rpadlpar was PowerVM centric in
that it only ever expected one device slot under a PHB. In the qemu device model we
present mulitple slots under a PHB. So, the quick and dirty solution at the time was just
to bypass the rpadlpar module and trigger a generic scan of the PCI bus from userspace.

I've since fixed the multiple slot registration in the rpadlpar module. However, I don't
think anybody has really done any testing under qemu without the "-v" flag. I'm in the
process of moving the PCI hotplug code from drmgr in userspace down into the kernel. I
would like to remove the "-v" flag from the drmgr invocation under qemu once that code
goes upstream.

-Tyrel

> 
> cheers
> _______________________________________________
> Linuxppc-users mailing list
> Linuxppc-users at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-users
>