Re: [PATCH] VT-d: fix PCI device detach from virtual machine

[Weidong Han <weidong.han@intel.com>]

David Woodhouse wrote:
> On Thu, 2009-02-26 at 17:31 +0800, Han, Weidong wrote:
>   
>> When assign a device behind conventional PCI bridge or PCIe to
>> PCI/PCI-x bridge to a domain, it must assign its bridge and may
>> also need to assign secondary interface to the same domain. 
>>
>> Dependent assignment is already there, but dependent
>> deassignment is missed when detach device from virtual machine.
>> This results in conventional PCI device assignment failure after
>> it has been assigned once. This patch addes dependent
>> deassignment, and fixes the issue. 
>>     
>
> Um, this code makes my head hurt.
>
> Why are we doing this in the first place? Because the IOMMU works on the
> source-id in PCIe transactions, the pci_find_upstream_pcie_bridge()
> function effectively tells us which PCI device our own device will be
> masquerading as, for the purposes of DMA.
>
> So why do we bother setting up a context in the IOMMU for the device
> itself, when no DMA will ever appear to come from this device? And
>   
if the device is behind PCI Express-to-PCI/PCI-X bridge, the source-id 
may be the device bdf or the
source-id provided by the bridge. so it needs to map the device itself.
> likewise why do we bother setting up a context for intermediate PCI
> bridges?
>   
I'm not sure if the intermediate PCI bridges are necessary. need to 
check PCI spec.
> Why not just jump straight to the 'DMA proxy' device, and use that
> _only_?
>   
What's the 'DMA proxy' device? is it the upstream pcie-to-pci bridge?
> We'll have to cope with multiple devices behind the same 'proxy', but it
> looks like our handling of that is totally screwed already...  what
> happens right now if you have two PCI devices behind the same PCIe-PCI
> bridge, and try to attach both of them to different domains... or both
> to the _same_ domain, in fact, and then detach just one of them. I think
> the answer to the latter question is that your newly-added
> iommu_detach_dependent_devices() routine will tear down the mapping on
> the 'proxy' device and faults will start happening for the device which
> is supposed to still be mapped?
>   
all the device behind a pcie-to-pci bridge must be co-assigned to a 
single domain. So it also require users to detach them together.

Regards,
Weidong
> Confused... and tempted to rip it all out and start over.
>
>