Maximal message should be used as a limit to the max message payload allowed,
without the headers. The ConnectX-3 check is done against this value includes
the headers. When the payload is 4K this will cause the NIC to drop packets.
Increase maximal message to 8K as workaround, this shouldn't change current
behaviour because we continue to set the MTU to 4k.
To reproduce;
set MTU to 4296 on the corresponding interface, for example:
ifconfig eth0 mtu 4296 (both server and client)
On server:
ib_send_bw -c UD -d mlx4_0 -s 4096 -n 1000000 -i1 -m 4096
On client:
ib_send_bw -d mlx4_0 -c UD <server_ip> -s 4096 -n 1000000 -i 1 -m 4096
Fixes: 6e0d733d92 ("IB/mlx4: Allow 4K messages for UD QPs")
Signed-off-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Taking advantage of the optimization which was introduced in previous
commit ("IB/mlx4: Use optimal numbers of MTT entries") to optimize the
MTT usage for QP and CQ.
Signed-off-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Optimize the device performance by assigning multiple physical pages,
which are contiguous, to a single MTT. As a result, the number of MTTs
is reduced and in turn save cache misses of MTTs.
Signed-off-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When working over a RoCE network, the UDP source port should be set only
for statically connected QPs (RC, UC and XRC).
Fixes: 2811ba51b0 ("IB/mlx5: Add RoCE fields to Address Vector")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The UMR's QP is created by calling mlx5_ib_create_qp directly, and
therefore the send CQ and the recv CQ on the ibqp weren't assigned.
Assign them right after calling the mlx5_ib_create_qp to assure
that any access to those pointers will work as expected and won't
crash the system as might happen as part of reset flow.
Fixes: e126ba97db ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Smatch tool reports the following error:
drivers/infiniband/hw/cxgb4/qp.c:1886
c4iw_create_qp() error: we previously assumed 'ucontext'
could be null (see line 1804)
Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The Broadcom driver produces the following compilation warning
drivers/infiniband/hw/bnxt_re/ib_verbs.c:
In function ‘bnxt_re_create_ah’:
drivers/infiniband/hw/bnxt_re/ib_verbs.c:668:6:
warning: variable ‘vlan_tag’ set but not used [-Wunused-but-set-variable]
u16 vlan_tag;
Let's remove it till vlan_tag will be implemented properly.
Cc: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add the PCI write end padding flag to device_cap_flags enum and set it
during mlx5_ib_query_device so it will be reported to user-space.
During WQ/QP creation, set that capability for WQ/QP if user requested
it and HW supports it.
PCI write end padding modification is not supported for now. There's no
such flag for a QP but for a WQ, create and modify use the same flag.
Return an error if PCI write end padding flag is set during modify_wq.
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There are root complexes that are able to optimize their
performance when incoming data is multiple full cache lines.
PCI write end padding is the device's ability to pad the ending of
incoming packets (scatter) to full cache line such that the last
upstream write generated by an incoming packet will be a full cache
line.
Add a relevant entry to ib_device_cap_flags to report such capability
of an RDMA device.
Add the QP and WQ create flags:
* A QP/WQ created with a scatter end padding flag will cause
HW to pad the last upstream write generated by a packet to cache line.
User should consider several factors before activating this feature:
- In case of high CPU memory load (which may cause PCI back pressure in
turn), if a large percent of the writes are partial cache line, this
feature should be checked as an optional solution.
- This feature might reduce performance if most packets are between one
and two cache lines and PCIe throughput has reached its maximum
capacity. E.g. 65B packet from the network port will lead to 128B
write on PCIe, which may cause traffic on PCIe to reach high
throughput.
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Following crash happens, if crc algorithm couldn't be allocated:
[ 1087.989072] rdma_rxe: loaded
[ 1097.855397] PCLMULQDQ-NI instructions are not detected.
[ 1097.901220] rdma_rxe: failed to allocate crc algorithmi err:-2
[ 1097.901248] BUG: unable to handle kernel
[ 1097.901249] NULL pointer dereference
[ 1097.901250] at 0000000000000046
[...]
Reason is that rxe->tfm is assigned the error return, which will then
be used for crypto_free_shash() in rxe_cleanup. Fix by using a
temporary variable and assigning it rxe->tfm after allocation succeeded.
Fixes: cee2688e3c ("IB/rxe: Offload CRC calculation when possible")
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Chelsio cxgb4 HW is big-endian, hence there is need to properly
annotate r2 and stag fields as __be32 and not __u32 to fix the
following sparse warnings.
drivers/infiniband/hw/cxgb4/qp.c:614:16:
warning: incorrect type in assignment (different base types)
expected unsigned int [unsigned] [usertype] r2
got restricted __be32 [usertype] <noident>
drivers/infiniband/hw/cxgb4/qp.c:615:18:
warning: incorrect type in assignment (different base types)
expected unsigned int [unsigned] [usertype] stag
got restricted __be32 [usertype] <noident>
Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In the modify QP handler the base_qpn_udp field in the RSS QPC is
overwrite later by irrelevant value assignment. Hence, ingress packets
which gets to the RSS QP will be steered then to a garbage QPN.
The patch fixes this by skipping the above assignment when a RSS QP is
modified, also, the RSS context's attributes assignments are relocated
just before the context is posted to avoid future issues like this.
Additionally, this patch takes the opportunity to change the code to be
disciplined to the device's manual and assigns the RSS QP context just at
RESET to INIT transition.
Fixes:3078f5f1bd8b ("IB/mlx4: Add support for RSS QP")
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The mlx4's RSS patches submission missed a report of RSS capabilities
which should be reported by the vendor channel in query_device.
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The RDMA/umem uses generic RB-trees macros to generate various ib_umem
access functions. The generation is performed with INTERVAL_TREE_DEFINE
macro, which allows one of two modes: declare all functions as static or
declare none of the function to be static.
The second mode of operation produces the following sparse errors:
drivers/infiniband/core/umem_rbtree.c:69:1:
warning: symbol 'rbt_ib_umem_iter_first' was not declared.
Should it be static?
drivers/infiniband/core/umem_rbtree.c:69:1:
warning: symbol 'rbt_ib_umem_iter_next' was not declared.
Should it be static?
Code relocation together with declaration of such functions to be
"static" solves the issue.
Because there is no need to have separate file for two functions,
let's consolidate umem_rtree.c and umem_odp.c into one file.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The cmd_sn field of CQ doorbell inits for 0. It should be
increment on each first db rung after a completion Event.
if the cmd_sn of notify doorbell Adjacent two times is the
same, the hardware will distinguish it for the same notify
request and update its type according to the priority level
of next event and solicited event.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The calculation of hem index are different between hns_roce_table_get
and hns_roce_table_find. When the table chunk size of TRRL is not
divisible by object size, it will faile to find the trrl table.
This patch is to update the calculation of the hem index in the
hns_roce_table_find to the same as which in the hns_roce_table_get.
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
the owner need to be set when posting sqwqe in hip08 RoCE.
The owner be used according to the below algorithm:
The value of owner should be 1 in the first lap, it
should be 0 in the second lap and in turn.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In hip08 RoCE, it need to add the sq_invld_flg field
in QP context for RoCE hardware.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The ack timeout's value in qp context shall be a 5-bit value
and be assgined by users. When at of qpc is set zero, the
timer is disabled.
When attr_mask set for IB_QP_TIMEOUT, The ack timeout field
is effective.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
If the extend sges exist, the sq_cur_sge_blk_addr field in QPC
(qp context) should be configured.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When sig_type of qpc is non-selectable, all sq's wqes will produce
cqe and not depend on the cqe attribute of wqe. When sig_type of
qpc is selectable, The cqe attribute of wqe will decide whether to
produce the cqe.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When send flags is IB_SEND_SOLICITED, the se(solicated event)
field of sqwqe will be set.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When post wr for mixed rdma operation, we need to use fence
mechanism to keep the correct execute order.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The TRRL(Target RDMA Read/aTOMIC List) record the information
of receiving RDMA READ or ATOMIC operation in hip08. It will
be used the hardware. The driver need to assign a continuous
physical address for trrl_ba field of qp context.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The irrl(initiator RDMA Read/Atomic list) base address of qp
context is assigned for addr[63:6]. This patch mainly fixed
it.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The hardware vendors need to generate RoCEv1 or RoCEv2
packet according to the sgid type configured.
Besides, update the gid table size for hip08 RoCE
device.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
HNS_ROCE_CAP_FALG_ROCE_V1_V2 is added for selecting capability of
RoCE in hns driver. When HNS_ROCE_CAP_FALG_ROCE_V1_V2 is set,
driver will inform ib core that the related hns device can support
RoCEv2, and ib core can generate the gid of the related type.
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds rereg mr support for hip08.
Signed-off-by: Shaobo Xu <xushaobo2@huawei.com>
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Signed-off-by: Lijun Ou <oulijun@huawei.com>
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The code that allocates the tx ring in the sdma code fails to take
advantage of kvzalloc variations.
Fix by converting to use kvzalloc_node.
Reported-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The driver parameter num_user_contexts controls global behavior and
should not be modified by the driver.
This patch eliminates modification of num_user_contexts by using a
local variable to keep track of the value.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Kamenee Arumugam <kamenee.arumugam@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The only use for the urg interrupt is for priority PSM packets.
There is no reason for this interrupt to be enabled for kernel
contexts.
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds tx_opcode_stats to parallel the
(rx)opcode_stats in the debugfs.
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
These are the use-cases where the pkey needs to be tested to see
if a packet needs to be dropped.
a) Check if pkey is not FULL_MGMT_P_KEY or LIM_MGMT_P_KEY,
drop the packet as it's not part of the management partition.
Self-originated packets are an exception.
b) If pkey index points to FULL_MGMT_P_KEY and LIM_MGMT_P_KEY is
in the table, the packet is coming from a management node,
and the receiving node is also a management node, so it is safe
for the packet to go through.
c) If pkey index points to FULL_MGMT_P_KEY and LIM_MGMT_P_KEY is
NOT in the table, drop the packet as LIM_MGMT_P_KEY should
always be in the pkey table. It could be a misconfiguration.
d) If pkey index points to LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is
NOT in the table, it is safe for the packet to go through
since a non-management node is talking to another non-managment
node.
e) If pkey index points to LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is in
the table, drop the packet because a non-management node is
talking to a management node, and it could be an attack.
For the implementation, these rules can be simplied to only checking
for (a) and (e). There's no need to check for rule (b) as
the packet doesn't need to be dropped. Rule (c) is not possible in
the driver as LIM_MGMT_P_KEY is always in the pkey table.
Reviewed-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
__subn_get_opa_portinfo stores value returned by hfi1_get_ib_cfg() as
operational vls. hfi1_get_ib_cfg() returns vls_operational field in
hfi1_pportdata. The problem with this is that the value is always equal
to vls_supported field in hfi1_pportdata.
The logic to calculate operational_vls is to set value passed by FM
(in __subn_set_opa_portinfo routine). If no value is passed then
default value is stored in operational_vls.
Field actual_vls_operational is calculated on the basis of buffer
control table. Hence, modifying hfi1_get_ib_cfg() to return
actual_operational_vls when used with HFI1_IB_CFG_OP_VLS parameter
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Patel Jay P <jay.p.patel@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The handler for link init state (HLS_UP_INIT) notifies userspace
(update_statusp()) before enabling the device
(RCV_CTRL_RCV_PORT_ENABLE_SMASK) or setting the device state
(ppd->host_link_state). This causes a race condition where the
userspace thinks the interface is in the INIT state before the driver
has set that state.
Rework the code path to eliminate the race.
Delay setting the init state until after a HW settling period.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When host is shutting down, it invokes the shutdown hook of the
L2 driver where it would attempt to free the MSI-X vectors, but would fail
because some vectors are held by the RoCE driver.
Implement the new hook in the L2 -> RoCE interface which will be invoked so that
the RoCE driver can unregister the device and free up the MSI-X vectors it had
claimed so that L2 can proceed with it's shutdown without failure.
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The scqe.stag is actually __b32, fix it.
drivers/infiniband/hw/cxgb4/cq.c:754:52: warning: cast to restricted __be32
Cc: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.
Cc: Moni Shoua <monis@mellanox.com>
Cc: Doug Ledford <dledford@redhat.com>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In recent code, two path record entries are alwasy cleared while
allocated could be either one or two path record entries.
This leads to zero out of unallocated memory.
This fix initializes alternative path record only when alternative path
is set.
While we are at it, path record allocation doesn't check for OPA
alternative path, but rest of the code checks for OPA alternative path.
Path record allocation code doesn't check for OPA alternative LID.
This can further lead to memory corruption when only one path record is
allocated, but there is actually alternative OPA path record present in CM
request.
Cc: <stable@vger.kernel.org> # v4.12+
Fixes: 9fdca4da4d ("IB/SA: Split struct sa_path_rec based on IB and ROCE specific fields")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
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.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
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.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch updates the mlx5_ifc with the following:
- Fix tunnel_stateless_gre typo.
- max_geneve_opt_len - Maximum geneve options length.
- tunnel_stateless_geneve_rx - If set, receive Stateless
Offloads for Geneve tunneled (inner) packets are supported.
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In some benchmarks and some CPU architectures, writing the CQE on a full
cache line size improves performance by saving memory access operations
(read-modify-write) relative to partial cache line change. This patch
lets the user to configure the device to pad the CQE up to 128B in case
its content is less than 128B. Currently the driver supports only padding
for a CQE size of 128B.
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In commit 1cbe6fc86c ("IB/mlx5: Add support for CQE compressing") the
concept of CQE compression was introduced and added a support for 64B
CQE size. This change update the code to support 128B CQE size as well.
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Adding new bits in mlx5_ifc_cmd_hca_cap to get the hardware capabilities
for:
- compression_128: Support 128B CQE compression
- cqe_128_always: Support 128B CQE padding
Signed-off-by: Guy Levi <guyle@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
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).
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
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.
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Reviewed-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
