On some environments, such as certain SR-IOV VF configurations, RoCE
isn't supported for mlx4 Ethernet ports. Currently the driver will
not open IB device on that port.
This is problematic since we do want user-space RAW Ethernet QPs functionality
to remain in place. For that end, enhance the relevant driver flows such that we
do create a device instance in that case.
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The kernel commit cited below restructured ib device management
so that the device kobject is initialized in ib_alloc_device.
As part of the restructuring, the kobject is now initialized in
procedure ib_alloc_device, and is later added to the device hierarchy
in the ib_register_device call stack, in procedure
ib_device_register_sysfs (which calls device_add).
However, in the ib_device_register_sysfs error flow, if an error
occurs following the call to device_add, the cleanup procedure
device_unregister is called. This call results in the device object
being deleted -- which results in various use-after-free crashes.
The correct cleanup call is device_del -- which undoes device_add
without deleting the device object.
The device object will then (correctly) be deleted in the
ib_register_device caller's error cleanup flow, when the caller invokes
ib_dealloc_device.
Fixes: 55aeed0654 ("IB/core: Make ib_alloc_device init the kobject")
Cc: <stable@vger.kernel.org> # v4.2+
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Before calling ipoib_stop, rtnl_lock should be taken, then
the flow clears the IPOIB_FLAG_ADMIN_UP and IPOIB_FLAG_OPER_UP
flags, and waits for mcast completion if IPOIB_MCAST_FLAG_BUSY
is set.
On the other hand, the flow of multicast join task initializes
a mcast completion, sets the IPOIB_MCAST_FLAG_BUSY and calls
ipoib_mcast_join. If IPOIB_FLAG_OPER_UP flag is not set, this
call returns EINVAL without setting the mcast completion and
leads to a deadlock.
ipoib_stop |
| |
clear_bit(IPOIB_FLAG_ADMIN_UP) |
| |
Context Switch |
| ipoib_mcast_join_task
| |
| spin_lock_irq(lock)
| |
| init_completion(mcast)
| |
| set_bit(IPOIB_MCAST_FLAG_BUSY)
| |
| Context Switch
| |
clear_bit(IPOIB_FLAG_OPER_UP) |
| |
spin_lock_irqsave(lock) |
| |
Context Switch |
| ipoib_mcast_join
| return (-EINVAL)
| |
| spin_unlock_irq(lock)
| |
| Context Switch
| |
ipoib_mcast_dev_flush |
wait_for_completion(mcast) |
ipoib_stop will wait for mcast completion for ever, and will
not release the rtnl_lock. As a result panic occurs with the
following trace:
[13441.639268] Call Trace:
[13441.640150] [<ffffffff8168b579>] schedule+0x29/0x70
[13441.641038] [<ffffffff81688fc9>] schedule_timeout+0x239/0x2d0
[13441.641914] [<ffffffff810bc017>] ? complete+0x47/0x50
[13441.642765] [<ffffffff810a690d>] ? flush_workqueue_prep_pwqs+0x16d/0x200
[13441.643580] [<ffffffff8168b956>] wait_for_completion+0x116/0x170
[13441.644434] [<ffffffff810c4ec0>] ? wake_up_state+0x20/0x20
[13441.645293] [<ffffffffa05af170>] ipoib_mcast_dev_flush+0x150/0x190 [ib_ipoib]
[13441.646159] [<ffffffffa05ac967>] ipoib_ib_dev_down+0x37/0x60 [ib_ipoib]
[13441.647013] [<ffffffffa05a4805>] ipoib_stop+0x75/0x150 [ib_ipoib]
Fixes: 08bc327629 ("IB/ipoib: fix for rare multicast join race condition")
Signed-off-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Update the broadcast address in the priv->broadcast object when the
Pkey value changes in index 0, otherwise the multicast GID value will
keep the previous value of the PKey, and will not be updated.
This leads to interface state down because the interface will keep the
old PKey value.
For example, in SR-IOV environment, if the PF changes the value of PKey
index 0 for one of the VFs, then the VF receives PKey change event that
triggers heavy flush. This flush calls update_parent_pkey that update the
broadcast object and its relevant members. If in this case the multicast
GID will not be updated, the interface state will be down.
Fixes: c290414169 ("IPoIB: Fix pkey change flow for virtualization environments")
Signed-off-by: Feras Daoud <ferasda@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In RC QP there is no need to resolve the outgoing interface
for each packet, as this does not change during QP life cycle.
Instead cache the interface on the socket and use that one.
This improves performance by 12% by sparing redundant
calls to rxe_find_route.
ib_send_bw -d rxe0 -x 1 -n 9000 -e -s $((1024 * 1024 )) -l 100
----------------------------------------------------------------------------------------
| | bytes | iterations | BW peak[MB/sec] | BW average[MB/sec] | MsgRate[Mpps] |
----------------------------------------------------------------------------------------
| before | 1048576 | 9000 | inf | 551.21 | 0.000551 |
| after | 1048576 | 9000 | inf | 615.54 | 0.000616 |
----------------------------------------------------------------------------------------
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Function rxe_rcv is used internally in RXE and don't need to be
exported. This patch removes such export declaration.
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch avoids RNR NAK timer and retransmit timer initialization and
cleanup for non RC QPs (such as UD QP, GSI QP).
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Expose new counters using the get_hw_stats callback.
We expose the following counters:
+---------------------+----------------------------------------+
| Name | Description |
|---------------------+----------------------------------------|
|sent_pkts | number of sent pkts |
|---------------------+----------------------------------------|
|rcvd_pkts | number of received packets |
|---------------------+----------------------------------------|
|out_of_sequence | number of errors due to packet |
| | transport sequence number |
|---------------------+----------------------------------------|
|duplicate_request | number of received duplicated packets. |
| | A request that previously executed is |
| | named duplicated. |
|---------------------+----------------------------------------|
|rcvd_rnr_err | number of received RNR by completer |
|---------------------+----------------------------------------|
|send_rnr_err | number of sent RNR by responder |
|---------------------+----------------------------------------|
|rcvd_seq_err | number of out of sequence packets |
| | received |
|---------------------+----------------------------------------|
|ack_deffered | number of deferred handling of ack |
| | packets. |
|---------------------+----------------------------------------|
|retry_exceeded_err | number of times retry exceeded |
|---------------------+----------------------------------------|
|completer_retry_err | number of times completer decided to |
| | retry |
|---------------------+----------------------------------------|
|send_err | number of failed send packet |
+---------------------+----------------------------------------+
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A couple spots were missed in the original patch to implement this
change. Add those spots.
Fixes: a9a42886d0 (cxgb4: Convert PDBG to pr_debug)
Signed-off-by: Doug Ledford <dledford@redhat.com>
* Multiplications for the size determination of memory allocations
indicated that array data structures should be processed.
Thus use the corresponding function "kcalloc".
This issue was detected by using the Coccinelle software.
* Replace the specification of data types by pointer dereferences
to make the corresponding size determinations a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
* A multiplication for the size determination of a memory allocation
indicated that an array data structure should be processed.
Thus use the corresponding function "kmalloc_array".
This issue was detected by using the Coccinelle software.
* Replace the specification of a data structure by a pointer dereference
to make the corresponding size determination a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Replace the specification of a data structure by a reference to
the desired member as the parameter for the operator "sizeof" to make
the corresponding size determination a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
* Pass a product for a call of the function "vmalloc_user" without storing
it in an intermediate variable.
* Delete the local variable "memsize" which became unnecessary with
this refactoring.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
* Multiplications for the size determination of memory allocations
indicated that array data structures should be processed.
Thus reuse the corresponding function "kcalloc".
This issue was detected by using the Coccinelle software.
* Replace the specification of a data type by a pointer dereference
to make the corresponding size determination a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
* A multiplication for the size determination of a memory allocation
indicated that an array data structure should be processed.
Thus reuse the corresponding function "kcalloc".
This issue was detected by using the Coccinelle software.
* Replace the specification of a data type by a pointer dereference
to make the corresponding size determination a bit safer according to
the Linux coding style convention.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Use a more typical logging style.
Miscellanea:
o Obsolete the c4iw_debug module parameter
o Coalesce formats
o Realign arguments
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Convert printks to pr_<level>
Miscellanea:
o Coalesce formats
o Realign arguments
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Using the normal mechanism, not an indirected one, is clearer.
Miscellanea:
o Coalesce formats
o Realign arguments
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Convert printks to pr_<level>
Miscellanea:
o Coalesce formats
o Realign arguments
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
IPoIB driver now uses the new set of callback functions.
If the hardware provider supports the new ipoib_options implementation,
the driver uses the callbacks in its data path flows, otherwise it uses the
driver default implementation for all data flows in its code.
The default implementation wasn't change and it is exactly as it was before
introduction of acceleration support.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Make ipoib_priv point to netdev_priv where the code calls netdev_priv.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Change of function parameter name from qpn to be dqpn.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch is preparing the netdev part at the IPoIB driver to be able
to use the ipoib_options.
It deals with the two flows from the .ndo: ipoib_open and ipoib_stop.
The code is rearranged as follows:
* All operations which deal with the hardware resources, (for example
change QP state, post-receive etc.) are performed in one place.
* All operations that are control oriented (like restart multicast task,
start the reap_ah etc.) are performed in separate place.
The functions that deal with the hardware resources now located at
__ipoib_ib_dev_open for the ipoib_open flow and __ipoib_ib_dev_stop
for ipoib_stop.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch prepares init and teardown flows so we can call them
through ipoib_options function pointers.
It arranges that area of code as the following:
* All operations which deal with the resource allocation/deletion
are performed in one place.
* All operations that are control oriented, meaning that they are not
connected to a specific hardware, are performed in a separate place.
The operations for allocation of hardware resources are now in the
function ipoib_dev_init_default, and the deletion of all the resources
are in ipoib_dev_uninit_default
The only exception is the creation of the PD object,
which is used both for resource allocation (create QP etc.)
and for control flows like creating AH.
It also does:
* Move creation of rx_ring and tx_ring to be in the resources
allocation area.
* Move the function ipoib_ib_dev_open that does the open device
to the control area instead of the dev_init which creates resources.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add RDMA netdev interface to ib device structure allowing RDMA
netdev devices to be allocated by ib clients.
The idea is to allow to providers to optimize IPoIB data path.
New struct that includes functions and data member is exposed.
It exposes set of callback functions for handling data path flows
in IPoIB driver.
Each provider can support these set of functions in order
to optimize its specific data path, and let IPoIB to leverage
its data path.
There is an assumption, that providers should give the full set
of functions and not only part of them, in order to work properly.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
HFI1 VNIC SDMA support enables transmission of VNIC packets over SDMA.
Map VNIC queues to SDMA engines and support halting and wakeup of the
VNIC queues.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
HFI1 HW specific support for VNIC functionality.
Dynamically allocate a set of contexts for VNIC when the first vnic
port is instantiated. Allocate VNIC contexts from user contexts pool
and return them back to the same pool while freeing up. Set aside
enough MSI-X interrupts for VNIC contexts and assign them when the
contexts are allocated. On the receive side, use an RSM rule to
spread TCP/UDP streams among VNIC contexts.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add support to create and free OPA_VNIC rdma netdev devices.
Implement netstack interface functionality including xmit_skb,
receive side NAPI etc. Also implement rdma netdev control functions.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VEMA function interfaces with the Infiniband MAD stack to exchange the
management information packets with the Ethernet Manager (EM).
It interfaces with the OPA VNIC netdev function to SET/GET the management
information. The information exchanged with the EM includes class port
details, encapsulation configuration, various counters, unicast and
multicast MAC list and the MAC table. It also supports sending traps
to the EM.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Sudeep Dutt <sudeep.dutt@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VNIC EMA interface functions are the management interfaces to the OPA
VNIC netdev. Add support to add and remove VNIC ports. Implement the
required GET/SET management interface functions and processing of new
management information. Add support to send trap notifications upon various
events like interface status change, unicast/multicast mac list update and
mac address change.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VNIC MAC table contains the MAC address to DLID mappings provided by
the Ethernet manager. During transmission, the MAC table provides the MAC
address to DLID translation. Implement MAC table using simple hash list.
Also provide support to update/query the MAC table by Ethernet manager.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VNIC driver statistics support maintains various counters including
standard netdev counters and the Ethernet manager defined counters.
Add the Ethtool hook to read the counters.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Define VNIC EM MAD structures and the associated macros. These structures
are used for information exchange between VNIC EM agent (EMA) on the host
and the Ethernet manager. These include the virtual ethernet switch (vesw)
port information, vesw port mac table, summay and error counters,
vesw port interface mac lists and the EMA trap.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
OPA VNIC netdev function supports Ethernet functionality over Omni-Path
fabric by encapsulating Ethernet packets inside Omni-Path packet header.
It allocates a rdma netdev device and interfaces with the network stack to
provide standard Ethernet network interfaces. It overrides HFI1 device's
netdev operations where it is required.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Sudeep Dutt <sudeep.dutt@intel.com>
Signed-off-by: Andrzej Kacprowski <andrzej.kacprowski@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Previously, ib_uverbs_event_file was suffixed by _file as it contained
the actual file information. Since it's now only used as base struct
for ib_uverbs_async_event_file and ib_uverbs_completion_event_file,
we change its name to ib_uverbs_event_queue. This represents its
logical role better.
Fixes: 1e7710f3f6 ('IB/core: Change completion channel to use the reworked objects schema')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Previously, we inferred the events size in ib_uverbs_event_read by
using the is_async flag. Instead of that, we pass the event size
directly.
Fixes: 1e7710f3f6 ('IB/core: Change completion channel to use the reworked objects schema')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Instead of having uverbs_uobject_put both in the error flow and the
good flow, we unite them.
Fixes: fd3c7904db ('IB/core: Change idr objects to use the new schema')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Currently, we initialize all fields of ib_uobject straight after
allocation. Therefore, a kmalloc was sufficient. Since ib_uobject
could be embedded in a type specific structure, we nullify it to
spare programmer errors.
Fixes: 3832125624 ('IB/core: Add support for idr types')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The only scenario where this function was called while the lock is
already taken is in the context cleanup scenario. Thus, in order not
to pass the lock state to this function, we just call the remove logic
straight from the cleanup context function.
Fixes: 3832125624 ('IB/core: Add support for idr types')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
We rename the "write" flags to "exclusive", as it's used for both
WRITE and DESTROY actions.
Fixes: 3832125624 ('IB/core: Add support for idr types')
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Saeed Mahameed says:
====================
Mellanox, mlx5 RDMA net device support
This series provides the lower level mlx5 support of RDMA netdevice
creation API [1] suggested and introduced by Intel's HFI OPA VNIC
netdevice driver [2], to enable IPoIB mlx5 RDMA netdevice creation.
mlx5 IPoIB RDMA netdev will serve as an acceleration netdevice for the current
IPoIB ULP generic netdevice, providing:
- mlx5 RSS support.
- mlx5 HW RX,TX offloads (checksum, TSO, LRO, etc ..).
- Full mlx5 HW features transparent to the ULP itself.
The idea here is to reuse and benefit from the already implemented mlx5e netdevice
management and channels API for both etherent and RDMA netdevices, since both IPoIB
and Ethernet netdevices share same common mlx5 HW resources (with some small
exceptions) and share most of the control/data path logic, it is more natural to
have them share the same code.
The differences between IPoIB and Ethernet netdevices can be summarized to:
Steering:
In mlx5, IPoIB traffic is sent and received from an underlay special QP, and in Ethernet
the traffic is handled by vports and vport steering is managed by e-switch or FW.
For IPoIB traffic to get steered correctly the only thing we need to do is to create RSS
HW contexts for RX and TX HW contexts for TX (similar to mlx5e) with the underlay QP attached to
them (underlay QP will be 0 in case of Ethernet).
RX,TX:
Since IPoIB traffic is different, slightly modified RX and TX handlers are required,
still we do some code reuse in data path via common helper functions.
All of the other generic netdevice and mlx5 aspects will be shared between mlx5 Ethernet
and IPoIB netdevices, e.g.
- Channels creation and handling (RQs,SQs,CQs, NAPI, interrupt moderation, etc..)
- Offloads, checksum, GRO, LRO, TSO, and more.
- netdevice logic and non Ethernet specific ndos (open/close, etc..)
In order to achieve what we want:
In patchet 1 to 3, Erez added the supported for underlay QP in mlx5_ifc and refactored
the mlx5 steering code to accept the underlay QP as a parameter for creating steering
objects and enabled flow steering for IB link.
Then we are going to use the mlx5e netdevice profile, which is already used to separate between
NIC and VF representors netdevices, to create new type of IPoIB netdevice profile.
For that, one small refactoring is required to make mlx5e netdevice profile management
more genetic and agnostic to link type which is done in patch #4.
In patch #5, we introduce ipoib.c to host all of mlx5 IPoIB (mlx5i) specific logic and a
skeleton for the IPoIB mlx5 netdevice profile, and we will start filling it in next patches,
using mlx5e already existing APIs.
Patch #6 and #7, Implement init/cleanup RX mlx5i netdev profile handlers to create mlx5 RSS
resources, same as mlx5e but without vlan and L2 steering tables.
Patch #8, Implement init/cleanup TX mlx5i netdev profile handlers, to create TX resources
same as mlx5e but with one TC (tc = 0) support.
Patch #9, Implement mlx5i open/close ndos, where we reuese the mlx5e channels API, to start/stop TX/RX channels.
Patch #10, Create the underlay QP and attach it to mlx5i RSS and TX HW contexts.
Patch #11 and #12, Break down the mlx5e xmit flow into smaller helper function and implement the
mlx5i IPoIB xmit routine.
Patch #13 and #14, Have an RX handler per netdevice profile. We already do this before this series
in a non clean way to separate between NIC netdev and VF representor RX handlers, in patch 13 we make
the RX handler generic and bound to a profile and in patch 14 we implement the IPoIB RX handlers.
Patch #15, Small cleanup to avoid e-switch with IPoIB netdev.
In order to enable mlx5 IPoIB, a merge between the IPoIB RDMA netdev offolad support [3]
- which was alread submitted to the rdma mailing list - and this series is required
plus an extra small patch [4] which will connect between both sides and actually enables the offload.
Once both patch-sets are merged into linux we will have to submit the extra small patch [4], to enable
the feature.
Thanks,
Saeed.
[1] https://patchwork.kernel.org/patch/9676637/
[2] https://lwn.net/Articles/715453/https://patchwork.kernel.org/patch/9587815/
[3] https://patchwork.kernel.org/patch/9672069/
[4] https://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux.git/commit/?id=0141db6a686e32294dee015b7d07706162ba48d8
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add check for bit IB_QP_CREATE_NETIF_QP while creating QP.
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the driver support only ethernet eswitch, and we want to
protect downstream IPoIB netdev from trying to access it in IB link.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
