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

Re: [Xen-devel] [PATCH v8 07/11] vpci/bars: add handlers to map the BARs

>>> On 27.02.18 at 12:57, <roger.pau@xxxxxxxxxx> wrote:
> On Tue, Feb 27, 2018 at 03:01:44AM -0700, Jan Beulich wrote:
>> >>> On 27.02.18 at 10:21, <roger.pau@xxxxxxxxxx> wrote:
>> > With the current approach in the unmap case there will be stale
>> > mappings left behind.
>> > 
>> > I guess it's better then to not modify the memory decoding bit at all
>> > until the operation finishes. That also rises the question of whether
>> > the memory decoding bit should be modified if p2m mapping/unmapping
>> > reports an error.
>> > 
>> > Should we also attempt to rollback failed map/unmap operations? What
>> > happens if the rollback also fails?
>> It is in particular this last question why I don't think rollback makes
>> sense. If there's any failure, I think the decode bit should be sticky
>> clear; we may want (need?) to invent some magical mechanism to
>> get a device back out of this state later on. But that way stale
>> mappings are not an immediate problem (I think).
> The only problem I see with this approach is that if an error happens
> in the unmap case we will disable memory decoding and leave some stale
> p2m mappings. Then the guest might change the position of the BARs,
> and those stale mappings would be completely forgotten.

Well, the implication was that together with the decode enable bit
not being allowed to be turned off would be that some recording
would perhaps be needed of the stale mappings. Allowing the
decode enable bit to be set again would then depend on the stale
mappings first being dealt with. As long as the decode enable bit
can't be set, the owning domain playing with the BARs is of no


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.