Re: [PATCH RFC 1/2] RDMA/rxe: don't allow registering !PAGE_SIZE mr

[Zhu Yanjun]

在 2023/10/30 16:13, Zhijian Li (Fujitsu) 写道:
> On 27/10/2023 16:17, Zhu Yanjun wrote:
> > 在 2023/10/27 13:41, Li Zhijian 写道:
> > > mr->page_list only encodes *page without page offset, when
> > > page_size != PAGE_SIZE, we cannot restore the address with a wrong
> > > page_offset.
> > >
> > > Note that this patch will break some ULPs that try to register 4K
> > > MR when PAGE_SIZE is not 4K.
> > > SRP and nvme over RXE is known to be impacted.
> >
> > When ULP uses folio or compound page, ULP can not work well with RXE after this commit is applied.
> >
> > Perhaps removing page_size set in RXE is a good solution because page_size is set twice, firstly page_size is set in infiniband/core, secondly it is set in RXE.
>
> Does The RXE one mean rxe_mr_init(), I think rxe_reg_user_mr() requires this.

Please read the discussions carefully. This problem has been discussed.

Best Regards,
Zhu Yanjun

>    48 static void rxe_mr_init(int access, struct rxe_mr *mr)
>    49 {
>    50         u32 key = mr->elem.index << 8 | rxe_get_next_key(-1);
>    51
>    52         /* set ibmr->l/rkey and also copy into private l/rkey
>    53          * for user MRs these will always be the same
>    54          * for cases where caller 'owns' the key portion
>    55          * they may be different until REG_MR WQE is executed.
>    56          */
>    57         mr->lkey = mr->ibmr.lkey = key;
>    58         mr->rkey = mr->ibmr.rkey = key;
>    59
>    60         mr->access = access;
>    61         mr->ibmr.page_size = PAGE_SIZE;
>    62         mr->page_mask = PAGE_MASK;
>    63         mr->page_shift = PAGE_SHIFT;
>    64         mr->state = RXE_MR_STATE_INVALID;
>    65 }
>
>
> Thanks
> Zhijian
>
> > When folio or compound page is used in ULP, it is very possible that page_size in infiniband/core is different from the page_size in RXE
> >
> > Not sure what problem this difference will cause.
> >
> > Zhu Yanjun
> >
> > > Signed-off-by: Li Zhijian <lizhijian@xxxxxxxxxxx>
> > > ---
> > >    drivers/infiniband/sw/rxe/rxe_mr.c | 6 ++++++
> > >    1 file changed, 6 insertions(+)
> > >
> > > diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c
> > > index f54042e9aeb2..61a136ea1d91 100644
> > > --- a/drivers/infiniband/sw/rxe/rxe_mr.c
> > > +++ b/drivers/infiniband/sw/rxe/rxe_mr.c
> > > @@ -234,6 +234,12 @@ int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sgl,
> > >        struct rxe_mr *mr = to_rmr(ibmr);
> > >        unsigned int page_size = mr_page_size(mr);
> > > +    if (page_size != PAGE_SIZE) {
> > > +        rxe_info_mr(mr, "Unsupported MR with page_size %u, expect %lu\n",
> > > +               page_size, PAGE_SIZE);
> > > +        return -EOPNOTSUPP;
> > > +    }
> > > +
> > >        mr->nbuf = 0;
> > >        mr->page_shift = ilog2(page_size);
> > >        mr->page_mask = ~((u64)page_size - 1);