The original commit of this patch has a munged log message that is
missing several of the tags the original author intended to be on the
patch. This was due to patchworks misinterpreting a cut-n-paste
separator line as an end of message line and munging the mbox that was
used to import the patch:
https://patchwork.kernel.org/patch/10264089/
The original patch will be reapplied with a fixed commit message so the
proper tags are applied.
This reverts commit aa0de36a40.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Zero out the fields of the struct ib_pd for user mode pds so that
users querying pds via nldev will not get garbage. For simplicity,
use kzalloc() to allocate the mlx4_ib_pd struct.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Setting iova, length, and page_size allows this information to be
seen via NLDEV netlink queries, which can aid in user rdma debugging.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Some of the struct ib_mr fields weren't getting initialized. This was
benign, but will cause problems when dumping the mr resource via
nldev/restrack.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Implement RDMA nldev netlink interface to get detailed CM_ID information.
Because cm_id's are attached to rdma devices in various work queue
contexts, the pid and task information at restrak_add() time is sometimes
not useful. For example, an nvme/f host connection cm_id ends up being
bound to a device in a work queue context and the resulting pid at attach
time no longer exists after connection setup. So instead we mark all
cm_id's created via the rdma_ucm as "user", and all others as "kernel".
This required tweaking the restrack code a little. It also required
wrapping some rdma_cm functions to allow passing the module name string.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Move struct rdma_id_private to a new header cma_priv.h so the resource
tracking services in core/nldev.c can read useful information about cm_ids.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Create a common dumpit function that can be used by all common resource
types. This reduces code replication and simplifies the code as we add
more resource types.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This struct is involved in the user API for mlx4 and should not be hidden
inside a driver header file.
Fixes: 09d208b258 ("IB/mlx4: Add report for RSS capabilities by vendor channel")
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
mlx5-updates-2018-02-28-1 (IPSec-1)
This series consists of some fixes and refactors for the mlx5 drivers,
especially around the FPGA and flow steering. Most of them are trivial
fixes and are the foundation of allowing IPSec acceleration from user-space.
We use flow steering abstraction in order to accelerate IPSec packets.
When a user creates a steering rule, [s]he states that we'll carry an
encrypt/decrypt flow action (using a specific configuration) for every
packet which conforms to a certain match. Since currently offloading these
packets is done via FPGA, we'll add another set of flow steering ops.
These ops will execute the required FPGA commands and then call the
standard steering ops.
In order to achieve this, we need that the commands will get all the
required information. Therefore, we pass the fte object and embed the
flow_action struct inside the fte. In addition, we add the shim layer
that will later be used for alternating between the standard and the
FPGA steering commands.
Some fixes, like " net/mlx5e: Wait for FPGA command responses with a timeout"
are very relevant for user-space applications, as these applications could
be killed, but we still want to wait for the FPGA and update the kernel's
database.
Regards,
Aviad and Matan
Signed-off-by: Doug Ledford <dledford@redhat.com>
In send_atomic_ack function, it is not necessary to make a
skb_clone. To gain better performance (high throughput and
low latency), this skb_clone is removed.
The following tests are made.
server client
--------- ---------
|1.1.1.1|<----rxe-channel--->|1.1.1.2|
--------- ---------
On server: rping -s -a 1.1.1.1 -v -C 1000 -S 512
On client: rping -c -a 1.1.1.1 -v -C 1000 -S 512
The kernel config CONFIG_DEBUG_KMEMLEAK is enabled on both server
and client.
This test runs for several hours. There is no memory leak and the whole
system can work well.
Based on the above network, the following tests are made.
Server: ibv_rc_pingpong -d rxe0 -g 1
Client: ibv_rc_pingpong -d rxe0 -g 1 1.1.1.1
The test results on Server(10 tests are made).
Before:
Throughput is 137.07 Mbit/sec
Latency is 517.76 usec/iter
After:
Throughput is 148.85 Mbit/sec
Latency is 476.64 usec/iter
The throughput is enhanced and the latency is reduced.
CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add a parameter for configuring the port on which the ib_srpt driver
listens for incoming RDMA/CM connections, namely
/sys/kernel/config/target/srpt/discovery_auth/rdma_cm_port. The default
value for this parameter is 0 which means "do not listen for incoming
RDMA/CM connections". Add RDMA/CM support to all code that handles
connection state changes. Modify srpt_init_nodeacl() such that ACLs can
be configured for IPv4 and IPv6 addresses.
Note: incoming connection requests are only accepted for ports that
have been enabled. See also the "if (!sport->enabled)" code in the
connection request handler. See also the following configfs attribute:
/sys/kernel/config/target/srpt/$port/$port/enable.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Saeed Mahameed says:
====================
mlx5-updates-2018-02-28-1 (IPSec-1)
This series consists of some fixes and refactors for the mlx5 drivers,
especially around the FPGA and flow steering. Most of them are trivial
fixes and are the foundation of allowing IPSec acceleration from user-space.
We use flow steering abstraction in order to accelerate IPSec packets.
When a user creates a steering rule, [s]he states that we'll carry an
encrypt/decrypt flow action (using a specific configuration) for every
packet which conforms to a certain match. Since currently offloading these
packets is done via FPGA, we'll add another set of flow steering ops.
These ops will execute the required FPGA commands and then call the
standard steering ops.
In order to achieve this, we need that the commands will get all the
required information. Therefore, we pass the fte object and embed the
flow_action struct inside the fte. In addition, we add the shim layer
that will later be used for alternating between the standard and the
FPGA steering commands.
Some fixes, like " net/mlx5e: Wait for FPGA command responses with a timeout"
are very relevant for user-space applications, as these applications could
be killed, but we still want to wait for the FPGA and update the kernel's
database.
Regards,
Aviad and Matan
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The user can provide very large cqe_size which will cause to integer
overflow as it can be seen in the following UBSAN warning:
Signed-off-by: Doug Ledford <dledford@redhat.com>
Users of ucma are supposed to provide size of option level,
in most paths it is supposed to be equal to u8 or u16, but
it is not the case for the IB path record, where it can be
multiple of struct ib_path_rec_data.
This patch takes simplest possible approach and prevents providing
values more than possible to allocate.
Reported-by: syzbot+a38b0e9f694c379ca7ce@syzkaller.appspotmail.com
Fixes: 7ce86409ad ("RDMA/ucma: Allow user space to set service type")
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
resolved_dev returned might be NULL as ifindex is transient number.
Ignoring NULL check of resolved_dev might crash the kernel.
Therefore perform NULL check before accessing resolved_dev.
Additionally rdma_resolve_ip_route() invokes addr_resolve() which
performs check and address translation for loopback ifindex.
Therefore, checking it again in rdma_resolve_ip_route() is not helpful.
Therefore, the code is simplified to avoid IFF_LOOPBACK check.
Fixes: 200298326b ("IB/core: Validate route when we init ah")
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The has_tag member will indicate whether a tag action was specified
in flow specification.
A flow tag 0 = MLX5_FS_DEFAULT_FLOW_TAG is assumed a valid flow tag
that is currently used by mlx5 RDMA driver, whereas in HW flow_tag = 0
means that the user doesn't care about flow_tag. HW always provide
a flow_tag = 0 if all flow tags requested on a specific flow are 0.
So we need a way (in the driver) to differentiate between a user really
requesting flow_tag = 0 and a user who does not care, in order to be
able to report conflicting flow tags on a specific flow.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Aviad Yehezkel <aviadye@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Group and pass all function arguments of parse_flow_attr call in one
common struct mlx5_flow_act.
This patch passes all the action arguments of parse_flow_attr in one common
struct mlx5_flow_act. It allows us to scale the number of actions without adding
new arguments to the function.
Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Jason Gunthorpe <jgg@mellanox.com>
Hitting the following hardlockup due to a race condition in
error CQE processing.
[26146.879798] bnxt_en 0000:04:00.0: QPLIB: FP: CQ Processed Req
[26146.886346] bnxt_en 0000:04:00.0: QPLIB: wr_id[1251] = 0x0 with status 0xa
[26156.350935] NMI watchdog: Watchdog detected hard LOCKUP on cpu 4
[26156.357470] Modules linked in: nfsd auth_rpcgss nfs_acl lockd grace
[26156.447957] CPU: 4 PID: 3413 Comm: kworker/4:1H Kdump: loaded
[26156.457994] Hardware name: Dell Inc. PowerEdge R430/0CN7X8,
[26156.466390] Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
[26156.472639] Call Trace:
[26156.475379] <NMI> [<ffffffff98d0d722>] dump_stack+0x19/0x1b
[26156.481833] [<ffffffff9873f775>] watchdog_overflow_callback+0x135/0x140
[26156.489341] [<ffffffff9877f237>] __perf_event_overflow+0x57/0x100
[26156.496256] [<ffffffff98787c24>] perf_event_overflow+0x14/0x20
[26156.502887] [<ffffffff9860a580>] intel_pmu_handle_irq+0x220/0x510
[26156.509813] [<ffffffff98d16031>] perf_event_nmi_handler+0x31/0x50
[26156.516738] [<ffffffff98d1790c>] nmi_handle.isra.0+0x8c/0x150
[26156.523273] [<ffffffff98d17be8>] do_nmi+0x218/0x460
[26156.528834] [<ffffffff98d16d79>] end_repeat_nmi+0x1e/0x7e
[26156.534980] [<ffffffff987089c0>] ? native_queued_spin_lock_slowpath+0x1d0/0x200
[26156.543268] [<ffffffff987089c0>] ? native_queued_spin_lock_slowpath+0x1d0/0x200
[26156.551556] [<ffffffff987089c0>] ? native_queued_spin_lock_slowpath+0x1d0/0x200
[26156.559842] <EOE> [<ffffffff98d083e4>] queued_spin_lock_slowpath+0xb/0xf
[26156.567555] [<ffffffff98d15690>] _raw_spin_lock+0x20/0x30
[26156.573696] [<ffffffffc08381a1>] bnxt_qplib_lock_buddy_cq+0x31/0x40 [bnxt_re]
[26156.581789] [<ffffffffc083bbaa>] bnxt_qplib_poll_cq+0x43a/0xf10 [bnxt_re]
[26156.589493] [<ffffffffc083239b>] bnxt_re_poll_cq+0x9b/0x760 [bnxt_re]
The issue happens if RQ poll_cq or SQ poll_cq or Async error event tries to
put the error QP in flush list. Since SQ and RQ of each error qp are added
to two different flush list, we need to protect it using locks of
corresponding CQs. Difference in order of acquiring the lock in
SQ poll_cq and RQ poll_cq can cause a hard lockup.
Revisits the locking strategy and removes the usage of qplib_cq.hwq.lock.
Instead of this lock, introduces qplib_cq.flush_lock to handle
addition/deletion of QPs in flush list. Also, always invoke the flush_lock
in order (SQ CQ lock first and then RQ CQ lock) to avoid any potential
deadlock.
Other than the poll_cq context, the movement of QP to/from flush list can
be done in modify_qp context or from an async error event from HW.
Synchronize these operations using the bnxt_re verbs layer CQ locks.
To achieve this, adds a call back to the HW abstraction layer(qplib) to
bnxt_re ib_verbs layer in case of async error event. Also, removes the
buddy cq functions as it is no longer required.
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fix warning limit for kernel stack consumption:
drivers/infiniband/core/cq.c: In function 'ib_process_cq_direct':
drivers/infiniband/core/cq.c:78:1: error: the frame size of 1032 bytes
is larger than 1024 bytes [-Werror=frame-larger-than=]
Using smaller ib_wc array on the stack brings us comfortably below that
limit again.
Fixes: 246d8b184c ("IB/cq: Don't force IB_POLL_DIRECT poll context for ib_process_cq_direct")
Reported-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Sergey Gorenko <sergeygo@mellanox.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
"err" is either zero or possibly uninitialized here. It should be
-EINVAL.
Fixes: 427c1e7bcd ("{IB, net}/mlx5: Move the modify QP operation table to mlx5_ib")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The series that introduced dual port RoCE mode assumed that we don't have
a dual port HCA that use the mlx5 driver, this is not the case for
Connect-IB HCAs. This reasoning led to assigning 1 as the native port
index which causes issue when the second port is used.
For example query_pkey() when called on the second port will return values
of the first port. Make sure that we assign the right port index as the
native port index.
Fixes: 32f69e4be2 ("{net, IB}/mlx5: Manage port association for multiport RoCE")
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The commit cited below added a gid_type field (RoCEv1 or RoCEv2)
to GID properties.
When adding GIDs, this gid_type field was copied over to the
hardware gid table. However, when deleting GIDs, the gid_type field
was not copied over to the hardware gid table.
As a result, when running RoCEv2, all RoCEv2 gids in the
hardware gid table were set to type RoCEv1 when any gid was deleted.
This problem would persist until the next gid was added (which would again
restore the gid_type field for all the gids in the hardware gid table).
Fix this by copying over the gid_type field to the hardware gid table
when deleting gids, so that the gid_type of all remaining gids is
preserved when a gid is deleted.
Fixes: b699a859d1 ("IB/mlx4: Add gid_type to GID properties")
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When using IPv4 addresses in RoCEv2, the GID format for the mapped
IPv4 address should be: ::ffff:<4-byte IPv4 address>.
In the cited commit, IPv4 mapped IPV6 addresses had the 3 upper dwords
zeroed out by memset, which resulted in deleting the ffff field.
However, since procedure ipv6_addr_v4mapped() already verifies that the
gid has format ::ffff:<ipv4 address>, no change is needed for the gid,
and the memset can simply be removed.
Fixes: 7e57b85c44 ("IB/mlx4: Add support for setting RoCEv2 gids in hardware")
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
iWARP does not support RDMA WRITE or SEND with immediate data.
Driver should check this before submitting to FW and return an
immediate error
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Race in qedr_poll_cq, lastest_cqe wasn't protected by lock,
leading to a case where two context's accessing poll_cq at
the same time lead to one of them having a pointer to an old
latest_cqe and reading an invalid cqe element
Signed-off-by: Amit Radzi <Amit.Radzi@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fix iWARP connect and listen to use the mapped port for
ipv4 and ipv6. Without this fixed, running on a server
that has iwpmd enabled will not use the correct port
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Assign true or false to boolean variables instead of an integer value.
This issue was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Acked-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
If a HCA supports the SG_GAPS_REG feature then fewer memory regions
are required per command. This patch reduces the number of memory
regions that is allocated per SRP session.
Signed-off-by: Sergey Gorenko <sergeygo@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Tested-by: Laurence Oberman <loberman@redhat.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Acked-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch avoids that sparse reports the following:
drivers/infiniband/hw/hfi1/driver.c:251:13: warning: context imbalance in 'rcv_hdrerr' - different lock contexts for basic block
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Continue at the bottom of a loop are removed.
Issue found using drop_continue.cocci Coccinelle script.
Signed-off-by: Arushi Singhal <arushisinghal19971997@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Storage ULPs (like NVMEoF) benefit from exposing affinity mapping
per completion vector to find the optimal multi-queue affinity
assignments. The ULPs call the verbs API ib_get_vector_affinity
introduced in commit c66cd353bb ("RDMA/core: expose affinity mappings per
completion vector") to get the underlying devices affinity mappings.
Add support in driver to expose the affinity masks per MSI-X
completion vector.
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently all CM nodes involved in a connection are
maintained in a connected_node list per dev. During
connection setup, we need to search this every time
we receive a packet on the iWARP LAN Queue (ILQ) and
this can be pretty inefficient for large number of
connections.
Fix this by organizing the CM nodes in two lists -
accelerated list and non-accelerated list. The search
on ILQ receive would be limited to only non accelerated
nodes. When a node moves to RTS, it is added to the
accelerated list.
Benchmarking ucmatose 16k connections shows a 20%
improvement in test completion time.
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently the TX pending lists for IEQ and ILQ are
handled separately. The handling of both can be
consolidated in i40iw_poll_completion.
Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The pointer rdi is being initialized with a value that is never read
and re-assigned immediately after, hence the initialization is redundant
and can be removed.
Cleans up clang warning:
drivers/infiniband/sw/rdmavt/vt.c:94:23: warning: Value stored to 'rdi'
during its initialization is never read
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
IPv6 does path selection for multipath routes deep in the lookup
functions. The next patch adds L4 hash option and needs the skb
for the forward path. To get the skb to the relevant FIB lookup
functions it needs to go through the fib rules layer, so add a
lookup_data argument to the fib_lookup_arg struct.
Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Note: This is compile only tested as I have no access to the hw. This
variable was not used anywhere in the code. Removing it saves 24 bytes.
add/remove: 0/1 grow/shrink: 0/0 up/down: 0/-24 (-24)
Function old new delta
rxe_qp_state_name 24 - -24
Total: Before=3348732, After=3348708, chg -0.00%
Signed-off-by: Hernán Gonzalez <hernan@vanguardiasur.com.ar>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
