On 06.09.2023 22:13, Bjorn Helgaas wrote: > On Wed, Sep 06, 2023 at 11:00:19PM +0300, Andy Shevchenko wrote: >> On Wed, Sep 06, 2023 at 10:59:20PM +0300, Andy Shevchenko wrote: >>> On Wed, Sep 06, 2023 at 02:06:23PM -0500, Bjorn Helgaas wrote: >>>> On Wed, Aug 16, 2023 at 05:21:15PM +0000, Bartosz Pawlowski wrote: >> ... >> >>>>> +/* >>>>> + * Intel IPU E2000 revisions before C0 implement incorrect endianness >>>>> + * in ATS Invalidate Request message body. Although there is existing software >>>>> + * workaround for this issue, it is not functionally complete (no 5-lvl paging >>>>> + * support) and it requires changes in all IOMMU implementations supporting >>>>> + * ATS. Therefore, disabling ATS seems to be more reasonable. >>>> Can we reference the commit that added the existing software >>>> workaround? >>> See below. >> Oh, I meant the second question here, i.e. >> >>>> It sounds like systems that (a) don't require 5-level paging and (b) >>>> use an IOMMU implementation that include the appropriate changes might >>>> still be able to use ATS? Is there a way for them to do that? >> ^^^ this one. > Sorry, I'm missing your point here. > > The comment mentions an existing partial software workaround. > Presumably that was added by some commit, and it would be helpful to > know which one. > > The comment also suggests that if the software workaround were > completed (or if a system didn't require 5-level paging) and it had > related changes to its IOMMU driver, we could still use ATS even on > hardware with this defect. > > So I'm wondering if there's a way for an IOMMU driver that has the > required changes and can tell that we're not using 5-level paging can > override this quirk that disables ATS. > > Maybe we want to unconditionally disable ATS on these broken devices. > In that case, I think we should just completely drop the comments > about the software workaround and IOMMU driver changes because they > wouldn't be relevant. This software workaround was shared only with E2000 customers and not publicly due to it's complexity and tricky design, so I think we cannot refer to it and I'll drop this comment. Regards, Bartosz
