lkml.org 
[lkml]   [2021]   [Apr]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    Date
    From
    SubjectRe: [PATCH rdma-next 01/10] RDMA: Add access flags to ib_alloc_mr() and ib_mr_pool_init()
    On Tue, Apr 06, 2021 at 04:15:52PM +0200, Christoph Hellwig wrote:
    > On Tue, Apr 06, 2021 at 11:04:37AM -0300, Jason Gunthorpe wrote:
    > > It might be idiodic, but I have to keep the uverbs thing working
    > > too.
    > >
    > > There is a lot of assumption baked in to all the drivers that
    > > user/kernel is the same thing, we'd have to go in and break this.
    > >
    > > Essentially #2 ends up as deleting IB_ACCESS_RELAXED_ORDERING kernel
    > > side and instead doing some IB_ACCESS_DISABLE_RO in kernel,
    > > translating uverbs IBV_ACCESS_* to this then finding and inverting all
    > > the driver logic and also finding and unblocking all the places that
    > > enforce valid access flags in the drivers. It is complicated enough
    >
    > Inverting the polarity of a flag at the uapi boundary is pretty
    > trivial and we already do it all over the kernel.

    Yes, but the complexity is how the drivers are constructed they are
    designed to reject flags they don't know about..

    Hum, it looks like someone has already been in here and we now have a
    IB_ACCESS_OPTIONAL concept.

    Something like this would be the starting point:

    diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
    index bed4cfe50554f7..fcb107df0eefc6 100644
    --- a/include/rdma/ib_verbs.h
    +++ b/include/rdma/ib_verbs.h
    @@ -1440,9 +1440,11 @@ enum ib_access_flags {
    IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
    IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
    IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
    - IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,

    IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
    + _IB_ACCESS_RESERVED1 = IB_UVERBS_ACCESS_RELAXED_ORDERING,
    + IB_ACCESS_DISABLE_RELAXED_ORDERING,
    +
    IB_ACCESS_SUPPORTED =
    ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
    };
    However I see only EFA actually uses IB_ACCESS_OPTIONAL, so the lead
    up would be to audit all the drivers to process optional access_flags
    properly. Maybe this was done, but I don't see much evidence of it..

    Sigh. It is a big mess cleaning adventure in drivers really.

    > Do we actually ever need the strict ordering semantics in the kernel?

    No, only for uverbs.

    Jason

    \
     
     \ /
      Last update: 2021-04-06 16:41    [W:3.765 / U:0.200 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site