Re: [Xen-devel] [PATCH] pci: clear host_maskall field on assign

[Jan Beulich <jbeulich@xxxxxxxx>]

On 05.10.2019 01:58, Chao Gao wrote:
> On Wed, Oct 02, 2019 at 12:49:35PM +0200, Roger Pau Monne wrote:
>> The current implementation of host_maskall makes it sticky across
>> assign and deassign calls, which means that once a guest forces Xen to
>> set host_maskall the maskall bit is not going to be cleared until a
>> call to PHYSDEVOP_prepare_msix is performed. Such call however
>> shouldn't be part of the normal flow when doing PCI passthrough, and
>> hence the flag needs to be cleared when assigning in order to prevent
>> host_maskall being carried over from previous assignations.
>>
>> Note that other mask fields, like guest_masked or the entry maskbit
>> are already reset when the msix capability is initialized. Also note
>> that doing the reset of host_maskall there would allow the guest to
>> reset such field by enabling and disabling MSIX, which is not
>> intended.
>>
>> Signed-off-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>
>> ---
>> Cc: Chao Gao <chao.gao@xxxxxxxxx>
>> Cc: "Spassov, Stanislav" <stanspas@xxxxxxxxx>
>> Cc: Pasi Kärkkäinen <pasik@xxxxxx>
>> ---
>> Chao, Stanislav, can you please check if this patch fixes your
>> issues?
> 
> I am glad to. For your testing, you can just kill qemu and destroy the
> guest. Then maskall bit of a pass-thru device will be set. And in this
> case, try to recreate the guest and check whether the maskall bit is
> cleared in guest.
> 
> The solution is similar to my v1 [1]. One question IMO (IIRC, it is why
> I changed to another approach) is: why not do such reset at deivce
> deassignment such that dom0 can use a clean device. Otherwise, the
> device won't work after being unbound from pciback. But I am not so
> sure, I can check it next Tuesday.

I too did think about this, but aiui pciback needs to issue
PHYSDEVOP_release_msix anyway, and Dom0 would then re-setup MSI-X
"from scratch", i.e. we'd clear the flag anyway in
msix_capability_init() due to msix->used_entries being zero at
the first (of possibly several) invocation(s).

Jan

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