Re: [PATCH v3 2/2] vfio: platform: Add generic DT reset support

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Philipp,

On 13/04/18 11:22, Philipp Zabel wrote:
[..]
> That also means it is impossible to use just one of the devices that
> share a reset line for vfio individually, while the other ones are still
> in use by the host. Currently the reset line is a shared resource
> similar to the iommu for devices in the same iommu_group.
> 
> Is there any mechanism in vfio that would allow modeling other shared
> resources apart from iommu?

No we only check the VFIO group viability at IOMMU level.
> 
> [...]
>>> For some of those it may be possible, but that is basically just a work-
>>> around for reality not matching expectations. There may be other cases
>>> where devices sharing a reset line are not even in the same parent node
>>> because they are controlled via a different bus. In general, I don't
>>> think it is feasible or desirable to force grouping of devices that
>>> share the same reset line into a common parent node.
>>
>> At least for Renesas R-Car SoCs, I think this is feasible, as all affected
>> devices are currently grouped under the same /soc node.
>> I added subnodes for all devices sharing resets (one for pwm, 4 for USB2,
>> and one for USB3; display doesn't have resets yet), and it still boots ;-)
> 
> Is this grouping enough to make sure all of the pwm/usb2/usb3 devices
> are only ever configured for vfio use together?
> 
> Assuming I have pwm[1-4] all sharing the same reset line, and I want
> pwm2 to be used by a vfio guest, I first have to make sure that all of
> pwm[1-4] are unbound, releasing their shared resets, before vfio-
> platform can request the same reset line as exclusive.
> 
> Thinking about it, if the pwm drivers keep their requested reset control
> around for the duration the device is bound, the reset controller
> framework should already kind of handle this - while any of the shared
> reset control handles is kept around, any exclusive request for the same
> reset control will fail with -EBUSY (and the other way around).
> But that requires all drivers to request the reset control during probe
> and release it during remove.
> 
>> However, ehci_platform_probe() cannot get its (optional) resets anymore.
>> Probably the reset controller framework needs to be taught to look for
>> shared resets in the parent node, too?
> 
> Hm, a generic framework shouldn't do such a thing, the parent node could
> be covered by a completely different binding.
> 
>>> My suggestion would be to relax the language in the reset.txt DT
>>> bindings doc.
>>
>> Which is fine to keep the status quo with the hardware designers, but makes
>> it less likely for non-whitelisted generic reset controller support to
>> become acceptable for the vfio people...
> 
> I still may be missing context, but I fail to see how
> 
> 	pwm@0 {
> 		resets = <&shared_reset_control>;
> 	};
> 
> 	pwm@1 {
> 		resets = <&shared_reset_control>;
> 	};
> 
> ->
> 
> 	pwms {
> 		resets = <&shared_reset_control>;
> 
> 		pwm@0 {
> 		};
> 
> 		pwm@1 {
> 		};
> 	};
> 
> makes any difference here, unless pwms gets bound to an actual driver
> that is used for vfio?

I don't think we are ready to assign pwms with VFIO as Alex emphasized
VFIO was meant to be used with IOMMU and I guess those devices do not
belong to any iommu group.

Thanks

Eric
> 
> regards
> Philipp
> 



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux