Skip to content
  1. Jul 09, 2019
    • Yamin Friedman's avatar
      linux/dim: Implement RDMA adaptive moderation (DIM) · f4915455
      Yamin Friedman authored 
      

RDMA DIM implements a different algorithm from net DIM and is based on
completions which is how we can implement interrupt moderation in RDMA.

The algorithm optimizes for number of completions and ratio between
completions and events. In order to avoid long latencies, the
implementation performs fast reduction of moderation level when the
traffic changes.

Signed-off-by: default avatarYamin Friedman <yaminf@mellanox.com>
Reviewed-by: default avatarMax Gurtovoy <maxg@mellanox.com>
Reviewed-by: default avatarSagi Grimberg <sagi@grimberg.me>
Signed-off-by: default avatarSaeed Mahameed <saeedm@mellanox.com>
Signed-off-by: default avatarLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      f4915455
    • Jason Gunthorpe's avatar
      Merge tag 'blk-dim-v2' into rdma.git for-next · 2ef38e38
      Jason Gunthorpe authored 
      

Generic DIM

From: Tal Gilboa and Yamin Fridman

Implement net DIM over a generic DIM library, add RDMA DIM

dim.h lib exposes an implementation of the DIM algorithm for
dynamically-tuned interrupt moderation for networking interfaces.

We want a similar functionality for other protocols, which might need to
optimize interrupts differently. Main motivation here is DIM for NVMf
storage protocol.

Current DIM implementation prioritizes reducing interrupt overhead over
latency. Also, in order to reduce DIM's own overhead, the algorithm might
take some time to identify it needs to change profiles. While this is
acceptable for networking, it might not work well on other scenarios.

Here we propose a new structure to DIM. The idea is to allow a slightly
modified functionality without the risk of breaking Net DIM behavior for
netdev. We verified there are no degradations in current DIM behavior with
the modified solution.

Suggested solution:
- Common logic is implemented in lib/dim/dim.c
- Net DIM (existing) logic is implemented in lib/dim/net_dim.c, which uses
  the common logic in dim.c
- Any new DIM logic will be implemented in "lib/dim/new_dim.c".
  This new implementation will expose modified versions of profiles,
  dim_step() and dim_decision().
- DIM API is declared in include/linux/dim.h for all implementations.

Pros for this solution are:
- Zero impact on existing net_dim implementation and usage
- Relatively more code reuse (compared to two separate solutions)
- Increased extensibility

Required for dependencies in the next series.

Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      2ef38e38
    • Danit Goldberg's avatar
      IB/mlx5: Report correctly tag matching rendezvous capability · 89705e92
      Danit Goldberg authored 
      Userspace expects the IB_TM_CAP_RC bit to indicate that the device
supports RC transport tag matching with rendezvous offload. However the
firmware splits this into two capabilities for eager and rendezvous tag
matching.

Only if the FW supports both modes should userspace be told the tag
matching capability is available.

Cc: <stable@vger.kernel.org> # 4.13
Fixes: eb761894

 ("IB/mlx5: Fill XRQ capabilities")
Signed-off-by: default avatarDanit Goldberg <danitg@mellanox.com>
Reviewed-by: default avatarYishai Hadas <yishaih@mellanox.com>
Reviewed-by: default avatarArtemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: default avatarLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      89705e92
    • Mauro Carvalho Chehab's avatar
      docs: infiniband: add it to the driver-api bookset · a3a400da
      Mauro Carvalho Chehab authored 
      

While this contains some uAPI stuff, it was intended to be read by a
kernel doc. So, let's not move it to a different dir, but, instead, just
add it to the driver-api bookset.

Signed-off-by: default avatarMauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      a3a400da
    • Jason Gunthorpe's avatar
      Merge branch 'vhca-tunnel' into rdma.git for-next · 20893d9d
      Jason Gunthorpe authored 
      Max Gurtovoy says:

====================
Those two patches introduce VHCA tunnel mechanism to DEVX interface
needed for Bluefield SOC. See extensive commit messages for more
information.
====================

Based on the mlx5-next branch from
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux for
dependencies

* branch 'vcha-tunnel':
  IB/mlx5: Implement VHCA tunnel mechanism in DEVX
  net/mlx5: Introduce VHCA tunnel device capability
      20893d9d
    • Max Gurtovoy's avatar
      IB/mlx5: Implement VHCA tunnel mechanism in DEVX · b6142608
      Max Gurtovoy authored 
      

This mechanism will allow function-A to perform operations "on behalf" of
function-B via tunnel object. Function-A will have privileges for creating
and using this tunnel object.

For example, in the device emulation feature presented in Bluefield-1 SoC,
using device emulation capability, one can present NVMe function to the
host OS.

Since the NVMe function doesn't have a normal command interface to the HCA
HW, here is a need to create a channel that will be able to issue commands
"on behalf" of this function.

This channel is the VHCA_TUNNEL general object. The emulation software
will create this tunnel for every managed function and issue commands via
devx general cmd interface using the appropriate tunnel ID. When devX
context will receive a command with non-zero vhca_tunnel_id, it will pass
the command as-is down to the HCA.

All the validation, security and resource tracking of the commands and the
created tunneled objects is in the responsibility of the HCA FW. When a
VHCA_TUNNEL object destroyed, the device will issue an internal
FLR (function level reset) to the emulated function associated with this
tunnel. This will destroy all the created resources using the tunnel
mechanism.

Signed-off-by: default avatarMax Gurtovoy <maxg@mellanox.com>
Reviewed-by: default avatarYishai Hadas <yishaih@mellanox.com>
Signed-off-by: default avatarLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      b6142608
    • Jason Gunthorpe's avatar
      RDMA/rvt: Do not use a kernel header in the ABI · f10ff380
      Jason Gunthorpe authored 
      rvt was using ib_sge as part of it's ABI, which is not allowed. Introduce
a new struct with the same layout and use it instead.

Fixes: dabac6e4

 ("IB/hfi1: Move receive work queue struct into uapi directory")
Reported-by: default avatarStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      f10ff380
  3. Jul 08, 2019
    • Jason Gunthorpe's avatar
      RDMA/siw: Fix DEFINE_PER_CPU compilation when ARCH_NEEDS_WEAK_PER_CPU · 4c7d6dcd
      Jason Gunthorpe authored 
      The initializer for the variable cannot be inside the macro (and zero
initialization isn't needed anyhow).

include/linux/percpu-defs.h:92:33: warning: '__pcpu_unique_use_cnt' initialized and declared 'extern'
  extern __PCPU_DUMMY_ATTRS char __pcpu_unique_##name;  \
                                 ^~~~~~~~~~~~~~
include/linux/percpu-defs.h:115:2: note: in expansion of macro 'DEFINE_PER_CPU_SECTION'
  DEFINE_PER_CPU_SECTION(type, name, "")
  ^~~~~~~~~~~~~~~~~~~~~~
drivers/infiniband/sw/siw/siw_main.c:129:8: note: in expansion of macro 'DEFINE_PER_CPU'
 static DEFINE_PER_CPU(atomic_t, use_cnt = ATOMIC_INIT(0));
        ^~~~~~~~~~~~~~

Also the rules for PER_CPU require the variable names to be globally
unique, so prefix them with siw_

Fixes: b9be6f18 ("rdma/siw: transmit path")
Fixes: bdcf26bf

 ("rdma/siw: network and RDMA core interface")
Reported-by: default avatarStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
      4c7d6dcd
  5. Jul 07, 2019
  7. Jul 05, 2019