Re: [Xen-devel] [PATCH v2 00/45] New VGIC(-v2) implementation

[Julien Grall <julien.grall@xxxxxxx>]

Hi,

On 03/15/2018 08:30 PM, Andre Przywara wrote:
> tl;dr: Coarse changelog below, individual patches have changelogs as
> well. git branch:
> http://www.linux-arm.org/git?p=xen-ap.git;a=shortlog;h=refs/heads/vgic-new/v2
> git://linux-arm.org/xen-ap.git branch vgic-new/v2
>
> Another update, addressing the review comments. Nothing too outstanding this
> time, the most interesting changes include:
> - removing the split-out preparatory patches, which are already merged
> - changing the setting and clearing of _IRQ_INPROGRESS
> - including Julien's LR access rework series
> - restricting new level IRQ handling to the new VGIC
> - fix multiple SGI handling (mimicing the recent Linux/KVM patch)
> - ASSERTing that h/w IRQs stay connected to their virtual IRQs
> - directly update h/w affinity, without taking the desc lock
> - restrict 8K struct vcpu to new VGIC and ARM64
> - use separate Makefile for vgic/ directory
> - many minor changes to address whitespace issues and usage of unsigned,
>    also extending comments
>
> A summarising changelog can be found below, each individual patch has
> its own changelog as well.
>
> There are some things that have (still) not been covered yet:
> - struct VCPU still allocates two pages on ARM64 when using the new VGIC now.
> We could try to look if we can allocate some parts of struct vcpu instead of
> embedding sub-structures into it.
> - vGICv3 support is not implemented, but should be fairly straight-forward to
> add, as the design incorporated this already. Will look at this next.
> - There is a possible DOS vector on the VCPU ap_list, which holds pending
> vIRQs. A guest can make this list rather long, which forces the hypervisor
> to hold the list lock when iterating the list. This should be bounded by
> the number of emulated vIRQs though, and there are ideas how to mitigate
> this issue. Those fixes would be posted on top as fixes later.
> - There is no ITS support, though the VGIC code itself is more ready for that
> than the old VGIC ever was. However due to differences between the Xen
> and KVM architecture the ITS bits are not easy to port over to Xen.
> - Do we need to call vgic_evtchn_irq_pending() in
> local_event_needs_delivery_nomask()? The event channel IRQ should be covered
> by the VGIC already.
Thank you for summarising the open questions! I will log them on jira 
once the series is merged.
Meanwhile, I have committed patches #1 - #8.

Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel