[for-next,v3,0/7] On-Demand Paging on SoftRoCE

Message

Daisuke Matsuda (Fujitsu) Dec. 23, 2022, 6:51 a.m. UTC

This patch series implements the On-Demand Paging feature on SoftRoCE(rxe)
driver, which has been available only in mlx5 driver[1] so far.

As an important change, it is necessary to convert triple tasklets
(requester, responder and completer) to workqueue because they must be
able to sleep in order to trigger page fault before accessing MRs. There
have been some discussions about this, and Bob Pearson posted a patchset[2]
for the conversion. I found it causes soft lockup and reported to him on
18th Nov. However, there has been no progress since then. The issue has
been a blocker to this ODP patchset for more than three months since the
first post of the RFC patchset.

As his patchset, consisting of 13 patches, is too large and complicated to
find the cause of soft lockup, I have prepared a patch that realizes the
conversion without the problem. I tried to minimize changes so that he can
easily make the changes he originally intended upon it.

[Overview]
When applications register a memory region(MR), RDMA drivers normally pin
pages in the MR so that physical addresses are never changed during RDMA
communication. This requires the MR to fit in physical memory and
inevitably leads to memory pressure. On the other hand, On-Demand Paging
(ODP) allows applications to register MRs without pinning pages. They are
paged-in when the driver requires and paged-out when the OS reclaims. As a
result, it is possible to register a large MR that does not fit in physical
memory without taking up so much physical memory.

[Why to add this feature?]
We, Fujitsu, have contributed to RDMA with a view to using it with
persistent memory. Persistent memory can host a filesystem that allows
applications to read/write files directly without involving page cache.
This is called FS-DAX(filesystem direct access) mode. There is a problem
that data on DAX-enabled filesystem cannot be duplicated with software RAID
or other hardware methods. Data replication with RDMA, which features
high-speed connections, is the best solution for the problem.

However, there is a known issue that hinders using RDMA with FS-DAX. When
RDMA operations to a file and update of the file metadata are processed
concurrently on the same node, illegal memory accesses can be executed,
disregarding the updated metadata. This is because RDMA operations do not
go through page cache but access data directly. There was an effort[3] to
solve this problem, but it was rejected in the end. Though there is no
general solution available, it is possible to work around the problem using
the ODP feature. It enables the kernel driver to update metadata before
processing RDMA operations.

We have enhanced the rxe to expedite the usage of persistent memory. Our
contribution to rxe includes RDMA Atomic write[4] and RDMA Flush[5]. With
them being merged along with ODP, an environment will be ready for
developers to create and test software for RDMA with FS-DAX. There is a
library(librpma)[6] being developed for this purpose. This environment can
be used by anybody without any special hardware but an ordinary computer
with a normal NIC though it is inferior to hardware implementations in
terms of performance.

[Design considerations]
ODP has been available only in mlx5, but functions and data structures
that can be used commonly are provided in ib_uverbs(infiniband/core). The
interfaces are heavily dependent on HMM infrastructure[7], and this
patchset use them as much as possible. While mlx5 has both Explicit and
Implicit ODP features along with prefetch feature, this patchset implements
the Explicit ODP feature only.

As responder and completer sleep, it is more likely that packet drop occurs
because of overflow in receiver queue. There are multiple queues involved,
but, as SoftRoCE uses UDP, the most important one would be the UDP buffers.
The size can be configured in net.core.rmem_default and net.core.rmem_max
sysconfig parameters. Users should change these values in case of packet
drop, but page fault would be typically not so long as to cause the
problem.

[How does ODP work?]
"struct ib_umem_odp" is used to manage pages. It is created for each
ODP-enabled MR on its registration. This struct holds a pair of arrays
(dma_list/pfn_list) that serve as a driver page table. DMA addresses and
PFNs are stored in the driver page table. They are updated on page-in and
page-out, both of which use the common interfaces in the ib_uverbs layer.

Page-in can occur when requester, responder or completer access an MR in
order to process RDMA operations. If they find that the pages being
accessed are not present on physical memory or requisite permissions are
not set on the pages, they provoke page fault to make pages present with
proper permissions and at the same time update the driver page table. After
confirming the presence of the pages, they execute memory access such as
read, write or atomic operations.

Page-out is triggered by page reclaim or filesystem events (e.g. metadata
update of a file that is being used as an MR). When creating an ODP-enabled
MR, the driver registers an MMU notifier callback. When the kernel issues a
page invalidation notification, the callback is provoked to unmap DMA
addresses and update the driver page table. After that, the kernel releases
the pages.

[Supported operations]
All traditional operations are supported on RC connection. The new Atomic
write[4] and RDMA Flush[5] operations are not included in this patchset. I
shall post them later after this patchset is merged. On UD connection,
Send, Recv, and SRQ-Recv are supported.

[How to test ODP?]
There are only a few resources available for testing. pyverbs testcases in
rdma-core and perftest[8] are recommendable ones. Other than them, the
ibv_rc_pingpong command can also used for testing. Note that you may have
to build perftest from upstream because older versions do not handle ODP
capabilities correctly.

The tree is available from the URL below:
https://github.com/daimatsuda/linux/tree/odp_v3

[Future work]
My next work is to enable the new Atomic write[4] and RDMA Flush[5]
operations with ODP. After that, I shall implement the prefetch feature. It
allows applications to trigger page fault using ibv_advise_mr(3) to
optimize performance. Some existing software like librpma[6] use this
feature. Additionally, I think we can also add the implicit ODP feature in
the future.


[1] [RFC 00/20] On demand paging
https://www.spinics.net/lists/linux-rdma/msg18906.html

[2] [PATCH for-next v3 00/13] Implement work queues for rdma_rxe
https://lore.kernel.org/linux-rdma/[e-mail]/

[3] [RFC PATCH v2 00/19] RDMA/FS DAX truncate proposal V1,000,002 ;-)
https://lore.kernel.org/nvdimm/[e-mail]/

[4] [PATCH v7 0/8] RDMA/rxe: Add atomic write operation
https://lore.kernel.org/linux-rdma/[e-mail]/

[5] [for-next PATCH 00/10] RDMA/rxe: Add RDMA FLUSH operation
https://lore.kernel.org/lkml/[e-mail]/

[6] librpma: Remote Persistent Memory Access Library
https://github.com/pmem/rpma

[7] Heterogeneous Memory Management (HMM)
https://www.kernel.org/doc/html/latest/mm/hmm.html

[8] linux-rdma/perftest: Infiniband Verbs Performance Tests
https://github.com/linux-rdma/perftest

[9] tests: ODP testcases for RDMA Write/Read and Atomic operations #1229
https://github.com/linux-rdma/rdma-core/pull/1229

v2->v3:
 1) Removed a patch that changes the common ib_uverbs layer.
 2) Re-implemented patches for conversion to workqueue.
 3) Fixed compile errors (happened when CONFIG_INFINIBAND_ON_DEMAND_PAGING=n).
 4) Fixed some functions that returned incorrect errors.
 5) Temporarily disabled ODP for RDMA Flush and Atomic Write.

v1->v2:
 1) Fixed a crash issue reported by Haris Iqbal.
 2) Tried to make lock patters clearer as pointed out by Romanovsky.
 3) Minor clean ups and fixes.

Daisuke Matsuda (7):
  RDMA/rxe: Convert triple tasklets to use workqueue
  RDMA/rxe: Always schedule works before accessing user MRs
  RDMA/rxe: Cleanup code for responder Atomic operations
  RDMA/rxe: Add page invalidation support
  RDMA/rxe: Allow registering MRs for On-Demand Paging
  RDMA/rxe: Add support for Send/Recv/Write/Read operations with ODP
  RDMA/rxe: Add support for the traditional Atomic operations with ODP

 drivers/infiniband/sw/rxe/Makefile    |   2 +
 drivers/infiniband/sw/rxe/rxe.c       |  27 +-
 drivers/infiniband/sw/rxe/rxe_comp.c  |  22 +-
 drivers/infiniband/sw/rxe/rxe_loc.h   |  36 ++-
 drivers/infiniband/sw/rxe/rxe_mr.c    |   7 +-
 drivers/infiniband/sw/rxe/rxe_odp.c   | 339 ++++++++++++++++++++++++++
 drivers/infiniband/sw/rxe/rxe_param.h |   2 +-
 drivers/infiniband/sw/rxe/rxe_qp.c    |   2 +-
 drivers/infiniband/sw/rxe/rxe_recv.c  |   4 +-
 drivers/infiniband/sw/rxe/rxe_req.c   |   2 +-
 drivers/infiniband/sw/rxe/rxe_resp.c  | 187 +++++++-------
 drivers/infiniband/sw/rxe/rxe_resp.h  |  46 ++++
 drivers/infiniband/sw/rxe/rxe_task.c  |  52 +++-
 drivers/infiniband/sw/rxe/rxe_task.h  |  15 +-
 drivers/infiniband/sw/rxe/rxe_verbs.c |   7 +-
 drivers/infiniband/sw/rxe/rxe_verbs.h |   2 +
 16 files changed, 638 insertions(+), 114 deletions(-)
 create mode 100644 drivers/infiniband/sw/rxe/rxe_odp.c
 create mode 100644 drivers/infiniband/sw/rxe/rxe_resp.h