| Message ID | 20210830165302.60225-1-olga.kornievskaia@gmail.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | revisit RMDA XDR padding management | expand |
On Mon, Aug 30, 2021 at 12:53 PM Olga Kornievskaia <olga.kornievskaia@gmail.com> wrote: > > From: Olga Kornievskaia <kolga@netapp.com> > > We have previously revisited how XDR padding management was done > for the RDMA read chunks. > > This patch series attempts to do the same for the RDMA read chunks > and altogether remove the options of doing an implicit roundup. Apologies for typos but this is about RDMA write chunks. > According to the RFC 8166 client "SHOULD NOT" include additional > space for XDR roundup padding. Furthermore, server MUST NOT write > XDR padding into the a write chunk. Given those two statement and > a patch "NFS: Always provide aligned buffers to the RPC read layers", > there is no reason for the client to look at the tail and assume > there is any padding data for which it needs to allocate space for. > > The only operation that this patch series effects is an NFS read. > All non-read ops that might use a write chunk are setup to use > reply chunk if reply is larger than inline threshold, not a write > chunk. > > > > *** SUBJECT HERE *** > > *** BLURB HERE *** > > Olga Kornievskaia (2): > xprtrdma: xdr pad optimization revisted again > xprtrdma: remove re_implicit_roundup xprt_rdma_pad_optimize > > net/sunrpc/xprtrdma/rpc_rdma.c | 15 --------------- > net/sunrpc/xprtrdma/transport.c | 8 -------- > net/sunrpc/xprtrdma/verbs.c | 2 -- > net/sunrpc/xprtrdma/xprt_rdma.h | 6 ------ > 4 files changed, 31 deletions(-) > > -- > 2.27.0 >
From: Olga Kornievskaia <kolga@netapp.com> We have previously revisited how XDR padding management was done for the RDMA read chunks. This patch series attempts to do the same for the RDMA read chunks and altogether remove the options of doing an implicit roundup. According to the RFC 8166 client "SHOULD NOT" include additional space for XDR roundup padding. Furthermore, server MUST NOT write XDR padding into the a write chunk. Given those two statement and a patch "NFS: Always provide aligned buffers to the RPC read layers", there is no reason for the client to look at the tail and assume there is any padding data for which it needs to allocate space for. The only operation that this patch series effects is an NFS read. All non-read ops that might use a write chunk are setup to use reply chunk if reply is larger than inline threshold, not a write chunk. *** SUBJECT HERE *** *** BLURB HERE *** Olga Kornievskaia (2): xprtrdma: xdr pad optimization revisted again xprtrdma: remove re_implicit_roundup xprt_rdma_pad_optimize net/sunrpc/xprtrdma/rpc_rdma.c | 15 --------------- net/sunrpc/xprtrdma/transport.c | 8 -------- net/sunrpc/xprtrdma/verbs.c | 2 -- net/sunrpc/xprtrdma/xprt_rdma.h | 6 ------ 4 files changed, 31 deletions(-)