| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
The current log is not very useful as it does not log the device
name since it it is prior to registration -
(unnamed net_device) (uninitialized): Setting up device
Modify to log after the device registration -
rmnet1: rmnet dev created
Signed-off-by: Subash Abhinov Kasiviswanathan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This was used only in the original patch series where the IOCTLs were
present and is no longer in use.
Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Signed-off-by: Subash Abhinov Kasiviswanathan <[email protected]>
Cc: Dan Williams <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
rmnet_rtnl_validate() was checking for upto mux_id 254, however the
rmnet_devices devices could hold upto 32 entries only. Fix this by
increasing the size of the rmnet_devices.
Fixes: ceed73a2cf4a ("drivers: net: ethernet: qualcomm: rmnet: Initial implementation")
Signed-off-by: Subash Abhinov Kasiviswanathan <[email protected]>
Cc: Dan Williams <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When we moved to updating representors from a workqueue grabbing
the RTNL somehow got lost in the process. Restore it, and make
sure RCU lock is not held while we are grabbing the RTNL. RCU
protects the representor table, so since we will be under RTNL
we can drop RCU lock as soon as we find the netdev pointer.
RTNL is needed for the dev_set_mtu() call.
Fixes: 2dff19622421 ("nfp: process MTU updates from firmware flower app")
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
It's reasonable to assume that if user selects to build the NFP
driver all offload capabilities will be enabled by default.
Change the CONFIG_NFP_APP_FLOWER to default to enabled.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Use dev_consume_skb_any() in place of dev_kfree_skb_any()
when control frame has been successfully processed in flower
and on the driver's main TX completion path.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Since representors are now created with a separate callback
start/stop app callbacks can be moved again to their original
location. They are intended to app-specific init/clean up
over the control channel.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Create representors after lower vNIC is registered and destroy
them before it is destroyed. Move the code out of start/stop
callbacks directly into vnic_init/clean callbacks. Make sure
SR-IOV callbacks don't try to create representors when lower
device does not exist.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
We currently only have one app callback for vNIC creation
and destruction. This is insufficient, because some actions
have to be taken before netdev is registered, after it's
registered and after it's unregistered. Old callbacks
were really corresponding to alloc/free actions. Rename
them and add proper init/clean. Apps using representors
will be able to use new callbacks to manage lifetime of
upper devices.
Signed-off-by: Jakub Kicinski <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5-updates-2017-09-03
This series from Tariq includes micro data path optimization for mlx5e
netdevice driver.
Mainly Tariq introduces the following changes to NAPI and RX handling
path of the driver:
- RX ring structure reorganizing
- Trivial code refactoring and optimization
- NAPI busy-poll for when fast UMR is in progress
- Non-atomic state operations in NAPI context
- Remove unnecessary fields from fast path structures
- page-cache micro optimization
- Rely on NAPI to avoid missing an IRQ for RX/TX shared NAPI contexts
- Stop NAPI when irq changes affinity
- Distribute RSS table among all RX rings
====================
Signed-off-by: David S. Miller <[email protected]>
|
|
This patch introduces callbacks and tunnel type to offload GRE tunnels.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
struct mlxsw_sp_rif is a router-private structure, and therefore
everything related to it is as well: parameters, and derived RIF types
including loopbacks. IPIP module needs access to some details of
loopback interfaces, but exporting all the RIF shebang would create too
large an interface.
So instead export just the bare minimum necessary: accessors for RIF
index and underlay VRF ID.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
These traps are generated for packets that fail checks for source IP,
encapsulation type, or GRE key. Trap these packets to CPU for follow-up
handling by the kernel, which will send ICMP destination unreachable
responses.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The local route that points at IPIP's underlay device (decap route) can
be present long before the GRE device. Thus when an encap route is
added, it's necessary to look inside the underlay FIB if the decap route
is already present. If so, the current trap offload needs to be
withdrawn and replaced with a decap offload.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Unlike encapsulation, which is represented by a next hop forwarding to
an IPIP tunnel, decapsulation is a type of local route. It is created
for local routes whose prefix corresponds to the local address of one of
offloaded IPIP tunnels. When the tunnel is removed (i.e. all the encap
next hops are removed), the decap offload is migrated back to a trap for
resolution in slow path.
This patch assumes that decap route is already present when encap route
is added. A follow-up patch will fix this issue.
Note that this patch only supports IPv4 underlay. Support for IPv6
underlay will be subject to follow-up work apart from this patchset.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Add the missing bits to recognize IPv6 next hops as IPIP ones to enable
offloading of IPv6 overlay encapsulation.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This introduces some common code for tracking of offloaded IP-in-IP
tunnels, and support for offloading IPv4 overlay encapsulating routes in
particular. A follow-up patch will introduce IPv6 overlay as well.
Offloaded tunnels are kept in a linked list of mlxsw_sp_ipip_entry
objects hooked up in mlxsw_sp_router. A network device that represents
the tunnel is used as a key to look up the corresponding IPIP entry.
Note that in the future, more general keying mechanism will be needed,
because parts of the tunnel information can be provided by the route.
IPIP entries are reference counted, because several next hops may end up
using the same tunnel, and we only want to offload it once.
Encapsulation path hooks into next hop handling. Routes that forward to
a tunnel are now considered gateway routes, thus giving them the same
treatment that other remote routes get. An IPIP next hop type is
introduced.
Details of individual tunnel types are kept in an array of
mlxsw_sp_ipip_ops objects. If a tunnel type doesn't match any of the
known tunnel types, the next-hop is not considered an IPIP next hop.
The list of IPIP tunnel types is currently empty, follow-up patches will
add support for GRE. Traffic to IPIP tunnel types that are not
explicitly recognized by the driver traps and is handled in slow path.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In the router, some next hops may reference an encapsulating netdevice,
such as GRE or IPIP. To properly offload these next hops, mlxsw needs to
keep track of whether a given next hop is a regular Ethernet entry, or
an IP-in-IP tunneling entry.
To facilitate this book-keeping, add a type field to struct
mlxsw_sp_nexthop. There is, as of this patch, only one next hop type:
MLXSW_SP_NEXTHOP_TYPE_ETH. Follow-up patches will introduce the IP-in-IP
variant.
There are several places where next hops are initialized in the IPv4
path. Instead of replicating the logic at every one of them, factor it
out to a function mlxsw_sp_nexthop4_type_init(). The corresponding fini
is actually protocol-neutral, so put it to mlxsw_sp_nexthop_type_fini(),
but create a corresponding protocoled _fini function that dispatches to
the protocol-neutral one.
The IPv6 path is simpler, but for symmetry with IPv4, create the same
suite of functions with corresponding logic.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
IPv6 counterpart of the previous patch: introduce a function to
determine whether a given route is a gateway route.
The new function takes a mlxsw_sp argument which follow-up patches will
use. Thus mlxsw_sp_fib6_entry_type_set() got that argument as well.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
For IPv4 IP-in-IP offload, routes that direct traffic to IP-in-IP
devices need to be considered gateway routes as well. That involves a
bit more logic, so extract the current test to a separate function,
where the logic can be later added.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When offloading L3 tunnels, an adjacency entry is created that loops the
packet back into the underlay router. Loopback interfaces then hold the
corresponding information and are created for IP-in-IP netdevices.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Loopback RIFs, which will be introduced in a follow-up patch, differ
from other RIFs in that they do not have a FID associated with them.
To support this, demote FID allocation from mlxsw_sp_rif_create to
configure op of the existing RIF types, and likewise the FID release
from mlxsw_sp_rif_destroy to deconfigure op.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Details of individual tunnel types are kept in an array of
mlxsw_sp_ipip_ops objects. Follow-up patches will use the list to
determine whether a constructed RIF should be a loopback, and to decide
whether a next hop references a tunnel.
The list is currently empty, follow-up patches will add support for GRE.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The spectrum_ipip module that will be introduced in the follow-up
patches needs to know the data type.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To support IPIP, the driver needs to be able to construct an IPIP
adjacency. Change mlxsw_reg_ratr_pack to take an adjacency type as an
argument. Adjust the one existing caller.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Unlike other interface types, loopback RIFs do not have MAC address. So
drop the corresponding argument from mlxsw_reg_ritr_pack() and move it
to a new function. Call that from callers of mlxsw_reg_ritr_pack.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The RTDP register is used for configuring the tunnel decap properties of
NVE and IPinIP.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
To implement IP-in-IP decapsulation, Spectrum uses LPM entries of type
IP2ME with tunnel validity bit and tunnel pointer set. The necessary
register fields are already available, so add a function to pack the
RALUE as appropriate.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
This enum is used with reg_ratr_trap_id, so move it next to the register
definition.
While at it, drop the enumerator initializers.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
So far, adjacencies have always been of type Ethernet (with value of 0),
and thus there was no need to explicitly support RATR type. However to
support IP-in-IP adjacencies, this type and a suite of IP-in-IP-specific
attributes need to be added.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Update the register so that loopback RIFs can be created and loopback
properties specified.
Signed-off-by: Petr Machata <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
When using a mac address described in the device tree, a check is made
to see if it is valid. When it's not, no fallback is defined. This
patches tries to get the mac address from h/w (or use a random one if
the h/w one isn't valid) when the dt mac address isn't valid.
Signed-off-by: Antoine Tenart <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The MAC retrieval logic is using a variable to store an h/w stored mac
address and checks this mac against invalid ones before using it. But
the mac address is only read from h/w when using PPv2.1. So when using
PPv2.2 it defaults to its init state.
This patches fixes the logic to only check if the h/w mac is valid when
actually retrieving a mac from h/w.
Signed-off-by: Antoine Tenart <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The MAC retrieval has a quite complicated logic (which is broken). Moves
it to its own function to prepare for patches fixing its logic, so that
reviews are easier.
Signed-off-by: Antoine Tenart <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull rdma updates from Doug Ledford:
"This is a big pull request.
Of note is that I'm sending you the new ioctl API for the rdma
subsystem. We put it up on linux-api@, but didn't get much response.
The API is complex, but it solves two different problems in one go:
1) The bi-directional nature of the RDMA file write calls, which
created the security hole we had to handle (and for which the fix
is now causing problems for systems in production, we were a bit
over zealous in the fix and the ability to open a device, then
fork, then create new queue pairs on the device and use them is
broken).
2) The bloat caused by different vendors implementing extensions to
the base verbs API. Each vendor's hardware is slightly different,
and the hardware might be suitable for one extension but not
another.
By the time we add generic extensions for all the different ways
that the different hardware can offload things, the API becomes
bloated. Things like our completion structs have started to exceed
a cache line in size because of all the elements needed to support
this. That in turn shows up heavily in the performance graphs with
a noticable drop in performance on 100Gigabit links as our
completion structs go from occupying one cache line to 1+.
This API makes things like the completion structs modular in a
very similar way to netlink so that your structs can only include
the items needed for the offloads/features you are actually using
on a given queue pair. In that way we support everything, but only
use what we need, and our structs stay smaller.
The ioctl API is better explained by the posting on linux-api@ than I
can explain it here, so I'll just leave it at that.
The rest of the pull request is typical stuff.
Updates for 4.14 kernel merge window
- Lots of hfi1 driver updates (mixed with a few qib and core updates
as well)
- rxe updates
- various mlx updates
- Set default roce type to RoCEv2
- Several larger fixes for bnxt_re that were too big for -rc
- Several larger fixes for qedr that, likewise, were too big for -rc
- Misc core changes
- Make the hns_roce driver compilable on arches other than aarch64 so
we can more easily debug build issues related to it
- Add rdma-netlink infrastructure updates
- Add automatic IRQ affinity infrastructure
- Add 32bit lid support
- Lots of misc fixes across the subsystem from random people
- Autoloading of RDMA netlink modules
- PCI pool cleanups from Romain Perier
- mlx5 driver feature additions and fixes
- Hardware tag matchine feature
- Fix sleeping in atomic when resolving roce ah
- Add experimental ioctl interface as posted to linux-api@"
* tag 'for-linus-ioctl' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (328 commits)
IB/core: Expose ioctl interface through experimental Kconfig
IB/core: Assign root to all drivers
IB/core: Add completion queue (cq) object actions
IB/core: Add legacy driver's user-data
IB/core: Export ioctl enum types to user-space
IB/core: Explicitly destroy an object while keeping uobject
IB/core: Add macros for declaring methods and attributes
IB/core: Add uverbs merge trees functionality
IB/core: Add DEVICE object and root tree structure
IB/core: Declare an object instead of declaring only type attributes
IB/core: Add new ioctl interface
RDMA/vmw_pvrdma: Fix a signedness
RDMA/vmw_pvrdma: Report network header type in WC
IB/core: Add might_sleep() annotation to ib_init_ah_from_wc()
IB/cm: Fix sleeping in atomic when RoCE is used
IB/core: Add support to finalize objects in one transaction
IB/core: Add a generic way to execute an operation on a uobject
Documentation: Hardware tag matching
IB/mlx5: Support IB_SRQT_TM
net/mlx5: Add XRQ support
...
|
|
The current allocation for dev->caps.spec_qps is for the size of the
pointer and not the size of the actual mlx4_spec_qps structure. Fix
this by using the correct size. Also splint allocation over a few
lines to make it cppcheck clean on overly wide lines.
Detected by CoverityScan, CID#1455222 ("Wrong sizeof argument")
Fixes: c73c8b1e47ca ("net/mlx4_core: Dynamically allocate structs at mlx4_slave_cap")
Signed-off-by: Colin Ian King <[email protected]>
Acked-by: Tariq Toukan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
The structures hca_param and func_cap are not being kfree'd on an error
exit path causing two memory leaks. Fix this by jumping to the existing
free memory error exit path.
Detected by CoverityScan, CID#1455219, CID#1455224 ("Resource Leak")
Fixes: c73c8b1e47ca ("net/mlx4_core: Dynamically allocate structs at mlx4_slave_cap")
Signed-off-by: Colin Ian King <[email protected]>
Acked-by: Tariq Toukan <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
In default, uniformly distribute the RSS indirection table entries
among all RX rings, rather than restricting this only to the rings
on the close NUMA node. irqbalancer would anyway dynamically override
the default affinities set to the RX rings.
This gives better multi-stream performance and CPU util.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
NAPI context keeps rescheduling on same CPU as long as it's busy.
This doesn't give the oppurtunity for changes in irq affinities
to take effect.
Fix that by calling napi_complete_done() upon a change in affinity.
This would stop the NAPI and reschedule it on the new CPU.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
We used a channel state bit MLX5E_CHANNEL_NAPI_SCHED to make
sure no NAPI is missed when a channel's napi_schedule() is called
for completion events of the different channel's resources/rings
while NAPI is currently running.
Now, as similar mechanism is implemented in kernel,
("39e6c8208d7b net: solve a NAPI race"),
we obsolete our own implementation and rely on the return value
of napi_complete_done().
This patch removes a redundant overhead of atomic bit operations.
Signed-off-by: Tariq Toukan <[email protected]>
Cc: Eric Dumazet <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
In XDP_TX flow, we now get back quicker to each page in page-cache,
and on some occasions refcount does not get back to 1 on time, causing
some costly page allocations.
Slightly increase the size of RX page-cache to significantly decrease
the chances for this to happen.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Avoid recycling an RX page if it moved to another NUMA node.
Add an ethtool counter to count such events.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
As of current design, in each NAPI, only a single UMR WQE
completion could be available in the completion queue of the
the internal control operations (ICO) send queue, in addition
to nop operations that require no actions upon completion.
This renders the consume index obsolete, as the wqe_counter
field in CQE is sufficient.
This helps removing a memory barrier, and obsoletes the need
for tracking the num_wqebbs to update the consumer counter.
In addition, remove other unused fields in icosq struct:
pdev, dma_fifo_pc, and prev_cc.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Separate the RX post WQEs function of the different RQ types.
This enables RQ type-specific optimizations in data-path.
Poll the ICOSQ completion queue only for Striding RQ,
and only when a UMR post completion could be possibly available.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
The indication for a UMR WQE in progress is needed only within
the NAPI context, and hence no races possible and no need for
the use of atomic operations.
The only place the flag is read outside of NAPI context is
in closure flow, after RQ is disabled flag is no more accessed
in NAPI.
Use a boolean instead of a bit in ring state, so that its
non-atomic set operations do not race with the atomic sets of
the other bits.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Rings enabled state change occurs in control path only, and is always
followed by a napi_sychronize(), so that following NAPIs read the
new value. This read does not need to be atomic.
The RQ auto-moderation bit is not set/cleared in data-path.
No need for atomic read, a regular read operation is sufficient.
In RQ creation time as well, there's no multiple threads trying
to access it yet, hence a regular read can be used.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
Refactor function mlx5e_lro_update_hdr() to reduce number of
branches.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
NAPI context handles different kinds of completion queues
(RX, TX, and others). Hence, upon a poll trial, some of them
might be empty.
Here we early-return upon empty completion queues, as well as
full rx buffer, and save unnecessary logic and memory barriers.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
If a UMR post is in progress, it means that there's a missing
WQE in RQ, and that a completion will be shortly available in
ICO SQ completion queue. Prefer busy-poll to handle it as soon
as possible.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|
|
The dma offset of a MPWQE (Multi-Packet WQE) in memory region
is fixed for all rounds. Calculate it once on creation time,
instead of in runtime. This also obsoletes the wqe argument in
the function.
In addition, optimize dma_info iterator calculation.
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>
|