The new ptrace(2) features allow to change truss(1) to systematically
operate on the process descriptors instead of pids.
Allocate the global kqueue that tracks all noted children
by pdopenpid()-ing them and adding to the kqueue with
EVFILT_PROCDESC/NOTE_PDSIGCHLD. The activated knote triggers the
pdwait() call to return the child tracing info. This replaces the
waitid(P_ALL) call in the non-capsicumized truss(1) eventloop.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D58094
Import rdma-core upstream commit ff01da2c5ac2 ("verbs: Fix typo in copying
IBV_FLOW_SPEC_UDP/TCP 'val'").
The TCP/UDP filter value was copied using the size of the IPv4 filter,
overrunning the destination subobject.
Also accept inner MPLS flow specs, matching 6d6f29721eea ("verbs: Allow
creation of inner MPLS flow spec"). Upstream deliberately has no inner
case for GRE - it is a tunnel header, matched only in the outer stack - so
that part of the review comment is not applied.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 30/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Allow verbs applications packet steering of GRE tunneled traffic.
Adding GRE flow specification based on RFC 2890.
GRE consists of flags, protocol and key fields.
IPv4 protocol 47 (IPPROTO_GRE) can be used when GRE packets are
encapsulated in IPv4.
b) verbs: Add MPLS flow specification filter
Add MPLS flow specification based on RFC 3032.
MPLS spec defined with label field which includes the
label value and additional parameters such as: BoS, TC and TTL.
MPLS allows stacking multiple labels in sequence.
In addition, the MPLS header can be encapsulated on top of different
layers, e.g.: ETH, IP (rfc4023), UDP (rfc7510), GRE (rfc4023).
Therefore, when using the flow creation verb, the application should
organize the spec filters list in the command in an ordered manner, such
that reflects the actual protocol stack of the packet, to determine the
exact position of the MPLS headers in the protocol stack.
c) mlx5: Report MPLS tunnel offload capabilities through mlx5 direct
verbs
This patch exposes the mlx5 device's capability to offload MPLS based
tunnel protocols via DV API.
The possible protocols are:
- Control word + MPLS over GRE.
- Control word + MPLS over UDP.
Differential revision: https://reviews.freebsd.org/D32205
MFC after: 1 month
Import Linux upstream commit a93b632c4531 ("IB/mlx5: Fix GRE flow
specification").
Honour the user-supplied GRE protocol mask instead of hard-coding 0xffff,
so wildcard and partial masks work.
Also give flow_spec_data[] 8-byte alignment, so this userspace copy of
struct ib_uverbs_flow_spec_hdr matches the kernel UAPI on 32-bit builds.
The kernel header has always used __aligned_u64 here and only our copy
diverged, so there is no upstream commit for that half. The attribute is
spelled out instead of introducing an __aligned_u64 macro, which
<infiniband/types.h> does not provide and which libbnxtre already defines
for itself.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 29/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Add ib_uverbs_flow_spec_gre to define the rule to match GRE
encapsulation protocol.
The spec includes the generic specs header, type, size and reserved
fields while the filter itself is defined as ib_uverbs_flow_gre_filter
and includes:
- Checksum present bit, key present bit and version bits in a single
16bit field.
- Protocol type field - Indicates the ether protocol type of the
encapsulated payload.
- Key field - present if key bit is set and contains an application
specific key value.
b) IB/uverbs: Expose MPLS flow spec to user-kernel ABI header
Add ib_uverbs_flow_spec_mpls to define the rule to match MPLS
protocol.
The spec includes the generic specs header, type, size and reserved
fields while the filter itself is defined as ib_uverbs_flow_mpls_filter
and includes a single 32bit field named 'label' which consists of:
Bits 0:19 - The MPLS label.
Bits 20:22 - Traffic class field.
Bit 23 - Bottom of stack bit.
Bits 24:31 - Time to live (TTL) field.
c) IB/mlx5: Add support for GRE flow specification
This patch introduces support for the GRE flow spec and
allowing the creation of rules based on the protocol and
key fields that are part of GRE protocol header.
d) IB/mlx5: Expose MPLS related tunneling offloads
This patch reports the device's capbilities to offload
encapsulated MPLS tunnel protocols to user-space:
- Capability to offload MPLS over GRE.
- Capability to offload MPLS over UDP.
Differential revision: https://reviews.freebsd.org/D32204
MFC after: 1 month
Fix is based on upstream rdma-core commit ee54f9d5348f ("mlx5: Add
loopback flags to QP creation").
Validate create_flags against a supported-flags mask instead of rejecting
anything that is not MLX5DV_QP_CREATE_TUNNEL_OFFLOADS. A create_flags of
0 is now accepted, unknown bits are rejected, and the vendor flag is OR-ed
in rather than assigned. The loopback flags that commit also adds do not
exist in this tree, so only the validation is taken.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 28/31] FreeBSD OFED support for DPDK MLX5 PMD
a) In order to enable offloading such as checksum and LRO for incoming
tunneling traffic, the QP should be created with tunnel offloads flag -
MLX5DV_QP_CREATE_TUNNEL_OFFLOAD.
b) Reports capability of which tunneling type supports
the tunneling offloads.
c) verbs: Add support in RSS of the inner packet
Some user space application would like to do RSS on the inner packet
fields instead of the outer. When user will set the IBV_RX_HASH_INNER
bit with one of the other hash fields, then the RSS will be on the inner
packet.
Differential revision: https://reviews.freebsd.org/D32203
MFC after: 1 month
[PATCH 27/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Add needed definitions to allow creation of a Multi-Packet RQ
using the mlx5 direct verbs interface.
In order to create a Multi-Packet RQ, one needs to provide a
mlx5dv_wq_init_attr containing the following information in its
striding_rq_attrs struct:
- single_stride_log_num_of_bytes: log of size of each stride
- single_wqe_log_num_of_strides: log of number of strides per WQE
- two_byte_shift_en: When enabled, hardware pads 2 bytes of zeros
before writing the message to memory (e.g. for IP alignment).
b) Add a helper function to verify 64 bit comp mask
The common check for a mask is as follows:
if (comp_mask & ~COMP_MASK_SUPPORTED_VALUES)
return EINVAL;
This can cause an issue when using 64 bit mask if the supported variable
is signed 32 bit: It will be bitwise inverted and then zeroed to 64
bits. Hence, wrong bits in the mask that exceed 32 bits will not raise
an error but would be ignored.
To fix this, a helper function is added in driver.h to be used by providers
code and fix wrong mask checks where the above was found to be applicable.
Differential revision: https://reviews.freebsd.org/D32202
MFC after: 1 month
[PATCH 26/31] FreeBSD OFED support for DPDK MLX5 PMD
A Multi-Packet RQ is a receive queue where multiple packets are
written to the same WQE. Each message starts in the beginning of a
stride. The total size of the scatter elements of each WQE is
determined upon RQ creation and all the posted WQEs should meet the
determined size.
A Multi-Packet RQ reduces the number of needed post-recv operations
thus increasing performance.
It reduces memory footprint by allowing each packet to consume a
different number of strides instead of the whole WR.
Differential revision: https://reviews.freebsd.org/D32201
MFC after: 1 month
[PATCH 25/31] FreeBSD OFED support for DPDK MLX5 PMD
Struct mlx5_ib_striding_rq_caps was not aligned to 64 bit as
it should have been. Add a 32 bit reserved field.
Differential revision: https://reviews.freebsd.org/D32200
MFC after: 1 month
Document the full struct mlx5dv_context (cqe_comp_caps, sw_parsing_caps)
and the mlx5dv_context_comp_mask enum in mlx5dv_query_device.3, and note
that these caps are only valid when the matching comp_mask bit is asked
for on input and returned on output.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 24/31] FreeBSD OFED support for DPDK MLX5 PMD
Software parsing (SWP) is a feature that can be used to instruct the
device to stop using its internal parser and to parse packets on the
transmit path according to offsets set for each packet.
Through this feature, the device allows the handling of checksum and LSO
by the hardware according to the location of IP and TCP/UDP headers.
Report various SW parsing capabilities and supported QP types through
mlx5 direct verbs interface.
Differential revision: https://reviews.freebsd.org/D32199
MFC after: 1 month
Reject TM-SRQ creation when tm_cap.max_ops is 0, so the internal command
QP's send queue is not sized to zero work-queue entries.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 23/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Expose tag matching capabilities and show them in ibv_devinfo.
b) verbs: Introduce tag matching SRQ
Introducing tag matching SRQ (TM-SRQ), which retains basic semantic of
regular SRQ, reports completions to own CQ, and has additional tag
based message receiving mechanism.
Detailed description for the TM-SRQ usage was added into
Documentation/tag_matching.md
c) mlx5: Add support to tag matching SRQ type
Create command QP for a tag-matching SRQ. Command QP used for
inserting/removing entries from the tag matching list. This command QP
is hidden from the users in the mlx5_srq structure.
New verb ibv_post_srq_ops() will be added in next patch to use it.
Differential revision: https://reviews.freebsd.org/D32198
MFC after: 1 month
Import Linux upstream commit 4e2b53a5cb5a ("IB/mlx5: Report inner RSS
capability").
Define MLX5_RX_HASH_INNER as (1UL << 31). Shifting 1 into the sign bit of
a plain int is undefined behaviour, and user space compiles this header
too. The main.c half of that commit is already carried by the preceding
fixup.
Sponsored by: NVidia networking
MFC after: 1 month
Drop the unused MLX5_IB_QP_TUNNEL_OFFLOAD flag. Tunnel offload is driven
by qp->tunnel_offload_en, so the enum bit was dead and misleading.
Also advertise MLX5_RX_HASH_INNER in query_device()'s
rss_caps.rx_hash_fields_mask, so user space can discover inner RSS.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 22/31] FreeBSD OFED support for DPDK MLX5 PMD
a) The device can support receive Stateless Offloads for the inner
packet's fields only when the packet is processed by TIR which is
enabled to support tunneling. Otherwise, the device treats the
packet as an ordinary non-tunneling packet and receive offloads
can be done only for the outer packet's field.
In order to enable receive Stateless Offloading support for incoming
tunneling traffic the TIR should be created with tunneled_offload_en.
Tunneling offloads is supported only be raw ethernet QP.
This patch includes:
- New QP creation flag for tunneling offloads.
- Reports device capabilities.
b) IB/mlx5: Add support for RSS on the inner packet
Some user space application would like to do RSS on the inner
packet fields instead on the outer.
When MLX5_RX_HASH_INNER is set with one or more of the other
hash fields, then the RSS will be done using the inner packet.
Differential revision: https://reviews.freebsd.org/D32196
MFC after: 1 month
Import Linux upstream commit ccc870879027 ("IB/mlx5: Allow creation of a
multi-packet RQ").
Only the create_rq() part was missing: the striding RQ parameters we
validate and store on the WQ never reached the firmware. Program
two_byte_shift_en, single_stride_log_num_of_bytes and
single_wqe_log_num_of_strides, with the two log values encoded relative to
their MLX5_MIN_* bases so the 3-bit firmware fields do not truncate them.
Upstream renamed these WQ context fields later on, so the names used here
are the ones this tree's mlx5_ifc.h still carries.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 21/31] FreeBSD OFED support for DPDK MLX5 PMD
a) This patch reports the device's striding RQ capabilities to
the user-space:
- min/max_single_stride_log_num_of_bytes: Log of min/max number of
bytes in a single stride.
- min/max_single_wqe_log_num_of_strides: Log of min/max number of
strides in a single WQE.
- supported_qpts: A bit mask to know which QP types support multi-
packet RQ, for now only Raw Packet QPs.
b) Allow creation of a multi-packet receive queue.
In order to create a multi-packet RQ, the following fields in
the mlx5_ib_rwq should be set:
- log_num_strides: Log of number of strides per WQE
- single_stride_log_num_of_bytes: Log of a single stride size
- two_byte_shift_en: When enabled, hardware pads 2 bytes of zeros
before writing the message to memory (e.g. for the IP alignment).
Differential revision: https://reviews.freebsd.org/D32195
MFC after: 1 month
Interpret the kernel's raw multi_pkt_send_wqe value (0, 1 or 3) instead of
masking it against a bitmask the kernel never emits, which silently
disabled MPW.
Sponsored by: NVidia networking
MFC after: 1 month
PATCH 20/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Use flag MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED to indicate hardware
supports multi packet WQE and it's enabled in SQ context.
Flag MLX5DV_CONTEXT_FLAGS_MPW is deprecated, shall not be used
in new applications.
b) Report if enhanced multi packet send WQE is supported through mlx5
direct verbs.
Differential revision: https://reviews.freebsd.org/D32194
MFC after: 1 month
Match a doorbell to its page with a range check instead of a page-boundary
mask, so mlx5_free_db() also works for external allocations that are not
page aligned and no longer leaks the doorbell page.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 19/31] FreeBSD OFED support for DPDK MLX5 PMD
Add a new DV API mlx5dv_set_context_attr() to enable setting an external
memory allocator. This API will allow the application to use specific
decisions about the memory allocation of HW resources (e.g. DV objects).
Some examples are managing numa pinning per object, managing a
hugepages resource pool, shared memory regions.
Also extend mlx5dv_get_qp() to return UAR mapping info. This can allow a
process to share its doorbell access with secondary process by re-mmap
the UAR address on the device and make it accessible as a user space
address.
Differential revision: https://reviews.freebsd.org/D32193
MFC after: 1 month
Actually provide the two ABI versions this commit describes: export
mlx5dv_init_obj@@MLX5_1.2 with the current behaviour and a compat
mlx5dv_init_obj@MLX5_1.0 that restores the historical 'void **' value, and
list the symbol under both versions in libmlx5.map. Binaries linked
against the old version keep resolving.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 18/31] FreeBSD OFED support for DPDK MLX5 PMD
Provide two implementations of mlx5dv_init_obj, one that has the
historical behaviour that has existed until now of returning the
void **uar and a new version that returns the 'void *' version
renamed to cq_uar.
Differential revision: https://reviews.freebsd.org/D32192
MFC after: 1 month
Report sw_parsing_caps using the same condition that enables software
parsing on the SQ (both eth_net_offloads and swp), so what we advertise
matches what the driver actually does.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 17/31] FreeBSD OFED support for DPDK MLX5 PMD
Software parsing (SWP) is a feature that can be used to instruct the
device to stop using its internal parser and to parse packets on the
transmit path according to offsets set for each packets.
Through this feature, the device allows the handling of checksum and
LSO by the hardware according to the location of IP and TCP/UDP
headers.
Enable SW parsing on Raw Ethernet send queue by default if firmware
supports it and report these capabilities to user space.
Differential revision: https://reviews.freebsd.org/D32191
MFC after: 1 month
Fix is based on upstream rdma-core commit 14a0fc824f16 ("rdma-core/irdma:
Implement device supported verb APIs").
Use ib_wr->invalidate_rkey instead of ib_wr->imm_data for
IBV_WR_SEND_WITH_INV and IBV_WR_LOCAL_INV. The two share a union, but
imm_data is __be32 while the rkey to invalidate is host order, so with the
new endian annotations this path reads the wrong member. Upstream irdma
has always used invalidate_rkey here.
Sponsored by: NVidia networking
MFC after: 1 month
PATCH 16/31] FreeBSD OFED support for DPDK MLX5 PMD
a) verbs: Annotate ibv_wc helpers with endian
This follows the scheme as used in the wc by introducing a
ibv_wc_read_invalidated_rkey to access the host endian
invalidated_rkey value with proper annotations.
This is just an inline wrapper to allow sparse to work sensibly,
not really a good reason to add another driver entry point.
b) verbs: Annotate ibv_send_wr with endian
This follows the scheme as used in the wc by introducing a
transparent union in ibv_send_wr to indicate the invalidate_rkey
is in host endian.
c) mlx5: Add sparse annotations.
Differential revision: https://reviews.freebsd.org/D32190
MFC after: 1 month
[PATCH 15/31] FreeBSD OFED support for DPDK MLX5 PMD
Enable mlx5 based hardware to report packet pacing capabilities
from kernel to user space. Packet pacing allows to limit the rate to any
number between the maximum and minimum, based on user settings.
The capabilities are exposed to user space through query_device by uhw.
The following capabilities are reported:
- The maximum and minimum rate limit in kbps supported by packet pacing.
- Bitmap showing which QP types are supported by packet pacing operation.
Differential revision: https://reviews.freebsd.org/D32189
MFC after: 1 month
Import Linux upstream commits 37da2a03c036 ("RDMA/mlx5: Use proper spec
flow label type") and 0f750966dca8 ("IB/mlx5: Add inner spec and IPv6
validation in user's flow attribute list").
The first widens set_flow_label()'s mask and value to u32, so the 20-bit
IPv6 flow label is no longer truncated to 8 bits. The second validates
the ethertype of inner flow specs as well as outer ones, and adds the IPv6
ethertype checking this tree was missing.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 14/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Add support to receive specific Vxlan packet in ConnectX-4.
b) Add support to match packet fields which are tunneled,
i.e. support matching the header of the inner packet which is the result of
or bit operation of the original header and the IB_FLOW_SPEC_INNER type.
The combination of IB_FLOW_SPEC_INNER | IB_FLOW_SPEC_VXLAN_TUNNEL is not
needed to be checked, because the IB core has this check already.
Dfiferential revision: https://reviews.freebsd.org/D32188
MFC after: 1 month
PATCH 13/31] FreeBSD OFED support for DPDK MLX5 PMD
The capabilities whether hardware support multi packet WQE or not is
exposed to user space through query_device by uhw.
Differential revision: https://reviews.freebsd.org/D32187
MFC after: 1 month
[PATCH 12/31] FreeBSD OFED support for DPDK MLX5 PMD
a) When enabling many VFs, the total amount of DMA mappings increase
significantly. This causes DMA allocations to take a lot of time
since they are serialized in the kernel.
As a result the driver enters into fatal condition due to
timeout and the system hangs. To recover from this we disable
MR cache for VFs. This also resolves Rx-Error MLX5 PMD issue.
PFs will still have a full cache and VFs cache can be manipulated
as usual after driver load.
b) The fast_registration length is used to convey length for memory
registrations through UMR which can be of any size up to 2^64.
Change the length type to be u64.
Differential revision: https://reviews.freebsd.org/D32186
MFC after: 1 month
Use designated initializers for .class and .class_mask in PCI_VDEVICE(),
and set .driver_data explicitly on the VF entries, rather than relying on
positional initializers.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 11/31] FreeBSD OFED support for DPDK MLX5 PMD
The MLX5 core driver recognizes a VF device as a PF device
in FreeBSD VM, because mlx5_core_pci_table was not having VF
flag MLX5_PCI_DEV_IS_VF. Added MLX5_PCI_DEV_IS_VF flag in
mlx5_core_pci_table.
Updated PCI_VDEVICE macro in LinuxKPI for class and class_mask,
same as linux.
Differential revision: https://reviews.freebsd.org/D32185
MFC after: 1 month
[PATCH 10/31] FreeBSD OFED support for DPDK MLX5 PMD
Adding UIO driver that exposes the synthetic network device
resources to userspace for application such as DPDK can drive it.
Added rescind CB capability to the VMBUS driver. Whenever the
hypervisor wants to close the device channel, VMBUS on receiving
this msg, calls the CB of the device driver (hv_uio in our case).
Differential revision: https://reviews.freebsd.org/D32184
MFC after: 1 month
Mark the new "ifname" sysctl MPSAFE and use strlcpy() so the interface
name it returns is always NUL-terminated.
Sponsored by: NVidia networking
MFC after: 1 month
PATCH 09/31] FreeBSD OFED support for DPDK MLX5 PMD
HyperV is not enabling ROCE by default causing the OFED to
not create few SYSCTL variables that publish kernel interface
name. Due to this MLX5 PMD is not able to read kernel
interface name to get MAC address.
Added new sysctl variable in MLX5_EN driver to map the device
name to the interface name.
Differential revision: https://reviews.freebsd.org/D32183
MFC after: 1 month
Hold a reference on the netdev recorded in dev->roce.netdev and drop it
when the entry is replaced or torn down, and register the netdev notifier
before the initial interface scan. Otherwise the stored pointer can
dangle, and an interface appearing during the scan can be missed.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 08/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Refactor the netdev notifier registration into a small helper
function.
This is a pre-step towards having mlx5 IB device over an Ethernet port
which doesn't support RoCE. Also, renamed the de-registration helper
and the new helper as netdev notifier and not roce, to make it clear
this is not only used with roce.
b) Rename RoCE related helpers to reflect being Eth ones
This is a pre-step towards having mlx5 IB device also over Eth ports
where
RoCE is not supported. We change the roce enable/disable and roce_lag
init/fini function names to have _eth instead of _roce.
c) Support RAW Ethernet when RoCE is disabled
On some environments, such as certain SRIOV VF configurations, RoCE is
not supported for mlx5 Ethernet ports. Currently, the driver will not
open IB device on that port.
This is problematic, since we do want user-space RAW Ethernet
(RAW_PACKET 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.
Differential revision: https://reviews.freebsd.org/D32182
MFC after: 1 month
Fix is based on upstream Linux commit a22ed86cff36 ("IB/mlx5: Add drop
flow steering rule support").
Initialise is_drop to false so a non-DROP flow cannot inherit stack
garbage, and pass no destination to mlx5_add_flow_rules() for drop rules
instead of one with num_dest = 1, which the flow steering core
dereferences. Both were lost in the original backport.
Derive the destination count from dst as well, so the dont-trap path,
which calls create_flow_rule() with a NULL destination, no longer asks the
core to walk a destination array that is not there.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 07/31] FreeBSD OFED support for DPDK MLX5 PMD
A drop rule is described by an action drop and no destination.
If a user specified IB_FLOW_SPEC_ACTION_DROP then set the action
to MLX5_FLOW_CONTEXT_ACTION_DROP and clear the destination.
Differential revision: https://reviews.freebsd.org/D32181
MFC after: 1 month
Import Linux upstream commit 572f46bf947c ("IB/mlx5: Refactor CQE
compression response").
Report the CQE compression capabilities only when the device supports
compression, instead of always advertising supported_format = HASH|CSUM
next to a max_num of 0.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 06/31] FreeBSD OFED support for DPDK MLX5 PMD
CQE compressing reduces PCI overhead by coalescing and compressing
multiple CQEs into a single merged CQE. Successful compressing
improves message rate especially for small packet traffic.
CQE compressing is supported for all 64B CQE formats (with certain
limitations) generated by RQ/Responder or by SQ/Requestor.
IB/mlx5: Report mlx5 CQE compression capabilities when querying device
The capabilities include:
- Max number of compressed and aggregated CQEs in a single session,
while zero means unsupported.
- For Responder, there are two formats of mini CQE: mini CQE with Rx
hash and mini CQE with checksum. They're mutual exclusive.
Differential revision: https://reviews.freebsd.org/D32180
MFC after: 1 month
[PATCH 05/31] FreeBSD OFED support for DPDK MLX5 PMD
FreeBSD doesn’t allow userspace to send ioctl call for
configuring allmulti mode. Added ALLMULTI flag in mlx5en.
In MLX5 PMD, allmulti mode will be enabled/disabled in DPDK
only if allmulti is supported by the mlx5en. it won't be changed
on network interface side.
Differential revision: https://reviews.freebsd.org/D32179
MFC after: 1 month
Fix is based on upstream Linux commit 2978975ce7f1 ("RDMA/mlx5: Process
create QP flags in one place").
Add IB_QP_CREATE_SCATTER_FCS and IB_QP_CREATE_CVLAN_STRIPPING to the
create_kernel_qp() allowlist, so a kernel RAW_PACKET QP using these flags
is no longer rejected with -EINVAL even though create_qp_common() supports
them. The flags arrived in 358e42ea66e2 and e4cc4fa7cca9, neither of
which updated this mask; upstream only fixed that later, as part of a much
larger refactor that is not worth importing, so the minimal equivalent is
done here.
Sponsored by: NVidia networking
MFC after: 1 month
[PATCH 04/31] FreeBSD OFED support for DPDK MLX5 PMD
a) Expose raw packet capabilities in the core layer to enable a device
to report it.
Two existing capabilities, scatter FCS and IP CSUM were added to this
field for a better user experience by exposing the raw packet caps
from one location.
This field will serve also for future capabilities for raw packet QP.
A new capability was introduced - cvlan stripping, which is the
device's ability to remove cvlan tag from an incoming packet and
report it in the matching work completion.
b) Check device's capabilities and report which raw packet capabilities
are supported.
c) Allow creating a WQ with cvlan stripping considering device's
capabilities. The default value was fixed to disable vlan stripping
till was asked explicitly.
In addition, allow modification of a WQ to turn on/off this property.
d) Enable creating a RAW Ethernet QP with cvlan stripping offload when
it's supported by the hardware.
e) Add support for creation of a WQ with scatter FCS capability, if
this capability is supported by the hardware.
Differential revision: https://reviews.freebsd.org/D32178
MFC after: 1 month
Import rdma-core upstream commit d2389b34ccc5 ("mlx5: Add missing include
file in mlx5dv.h").
Including <sys/types.h> keeps mlx5dv.h self-contained: off_t, used by the
new uar_mmap_offset field, is now always defined. The man page gains the
uar_mmap_offset line upstream documents in 5dd3ead7b5f5.
Sponsored by: NVidia networking
MFC after: 1 month
Konstantin Belousov