The IB verbs objects are counted by the kernel and ib_core ensures that
deallocate PD will success so it will be called once all other objects
that depends on PD will be released. This is achieved by managing various
reference counters on such objects.
The mlx5 driver didn't follow this standard flow when allowed DEVX objects
that are not managed by ib_core to be interleaved with the ones under
ib_core responsibility.
In such interleaved scenarios deallocate command can fail and ib_core will
leave uobject in internal DB and attempt to clean it later to free
resources anyway.
This change partially restores returned value from dealloc_pd() for all
drivers, but keeping in mind that non-DEVX devices and kernel verbs paths
shouldn't fail.
Fixes: 21a428a019 ("RDMA: Handle PD allocations by IB/core")
Link: https://lore.kernel.org/r/20200907120921.476363-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Currently it triggers a WARN_ON and then goes ahead and destroys the
uobject anyhow, leaking any driver memory.
The only place that leaks driver memory should be during FD close() in
uverbs_destroy_ufile_hw().
Drivers are only allowed to fail destroy uobjects if they guarantee
destroy will eventually succeed. uverbs_destroy_ufile_hw() provides the
loop to give the driver that chance.
Link: https://lore.kernel.org/r/20200902081708.746631-1-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The original function returns unsigned long and 0 on failure.
Fixes: 4a35339958 ("RDMA/umem: Add API to find best driver supported page size in an MR")
Link: https://lore.kernel.org/r/0-v1-982a13cc5c6d+501ae-fix_best_pgsz_stub_jgg@nvidia.com
Reviewed-by: Gal Pressman <galpress@amazon.com>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The UDP source port number in RoCE v2 is used to create entropy for
network routers (ECMP), load balancers and 802.3ad link aggregation
switching that are not aware of RoCE IB headers. Considering that the IB
core has achieved a new interface to get a hashed value of it, the fixed
value of it in QPC and UD WQE in hns driver could be fixed and the port
number is to be set dynamically now.
For QPC of RC, the value could be hashed from flow_lable if the user pass
it in or from remote qpn and local qpn. For WQE of UD, it is set according
to fl or as a random value.
Link: https://lore.kernel.org/r/1598002289-8611-1-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In almost all cases rdma_accept() is called under the handler_mutex by
ULPs from their handler callbacks. The one exception was ucma which did
not get the handler_mutex.
To improve the understand-ability of the locking scheme obtain the mutex
for ucma as well.
This improves how ucma works by allowing it to directly use handler_mutex
for some of its internal locking against the handler callbacks intead of
the global file->mut lock.
There does not seem to be a serious bug here, other than a DISCONNECT event
can be delivered concurrently with accept succeeding.
Link: https://lore.kernel.org/r/20200818120526.702120-7-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The "domain" argument is constant and modern device (mlx5) doesn't support
anything except IB_FLOW_DOMAIN_USER, so delete this extra parameter and
simplify code.
Link: https://lore.kernel.org/r/20200730081235.1581127-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Smaller set of RDMA updates. A smaller number of 'big topics' with the
majority of changes being driver updates.
- Driver updates for hfi1, rxe, mlx5, hns, qedr, usnic, bnxt_re
- Removal of dead or redundant code across the drivers
- RAW resource tracker dumps to include a device specific data blob for
device objects to aide device debugging
- Further advance the IOCTL interface, remove the ability to turn it off.
Add QUERY_CONTEXT, QUERY_MR, and QUERY_PD commands
- Remove stubs related to devices with no pkey table
- A shared CQ scheme to allow multiple ULPs to share the CQ rings of a
device to give higher performance
- Several more static checker, syzkaller and rare crashers fixed
-----BEGIN PGP SIGNATURE-----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=0z2L
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"A quiet cycle after the larger 5.8 effort. Substantially cleanup and
driver work with a few smaller features this time.
- Driver updates for hfi1, rxe, mlx5, hns, qedr, usnic, bnxt_re
- Removal of dead or redundant code across the drivers
- RAW resource tracker dumps to include a device specific data blob
for device objects to aide device debugging
- Further advance the IOCTL interface, remove the ability to turn it
off. Add QUERY_CONTEXT, QUERY_MR, and QUERY_PD commands
- Remove stubs related to devices with no pkey table
- A shared CQ scheme to allow multiple ULPs to share the CQ rings of
a device to give higher performance
- Several more static checker, syzkaller and rare crashers fixed"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (121 commits)
RDMA/mlx5: Fix flow destination setting for RDMA TX flow table
RDMA/rxe: Remove pkey table
RDMA/umem: Add a schedule point in ib_umem_get()
RDMA/hns: Fix the unneeded process when getting a general type of CQE error
RDMA/hns: Fix error during modify qp RTS2RTS
RDMA/hns: Delete unnecessary memset when allocating VF resource
RDMA/hns: Remove redundant parameters in set_rc_wqe()
RDMA/hns: Remove support for HIP08_A
RDMA/hns: Refactor hns_roce_v2_set_hem()
RDMA/hns: Remove redundant hardware opcode definitions
RDMA/netlink: Remove CAP_NET_RAW check when dump a raw QP
RDMA/include: Replace license text with SPDX tags
RDMA/rtrs: remove WQ_MEM_RECLAIM for rtrs_wq
RDMA/rtrs-clt: add an additional random 8 seconds before reconnecting
RDMA/cma: Execute rdma_cm destruction from a handler properly
RDMA/cma: Remove unneeded locking for req paths
RDMA/cma: Using the standard locking pattern when delivering the removal event
RDMA/cma: Simplify DEVICE_REMOVAL for internal_id
RDMA/efa: Add EFA 0xefa1 PCI ID
RDMA/efa: User/kernel compatibility handshake mechanism
...
The lookaside count is improperly initialized to the size of the
Receive Queue with the additional +1. In the traces below, the
RQ size is 384, so the count was set to 385.
The lookaside count is then rarely refreshed. Note the high and
incorrect count in the trace below:
rvt_get_rwqe: [hfi1_0] wqe ffffc900078e9008 wr_id 55c7206d75a0 qpn c
qpt 2 pid 3018 num_sge 1 head 1 tail 0, count 385
rvt_get_rwqe: (hfi1_rc_rcv+0x4eb/0x1480 [hfi1] <- rvt_get_rwqe) ret=0x1
The head,tail indicate there is only one RWQE posted although the count
says 385 and we correctly return the element 0.
The next call to rvt_get_rwqe with the decremented count:
rvt_get_rwqe: [hfi1_0] wqe ffffc900078e9058 wr_id 0 qpn c
qpt 2 pid 3018 num_sge 0 head 1 tail 1, count 384
rvt_get_rwqe: (hfi1_rc_rcv+0x4eb/0x1480 [hfi1] <- rvt_get_rwqe) ret=0x1
Note that the RQ is empty (head == tail) yet we return the RWQE at tail 1,
which is not valid because of the bogus high count.
Best case, the RWQE has never been posted and the rc logic sees an RWQE
that is too small (all zeros) and puts the QP into an error state.
In the worst case, a server slow at posting receive buffers might fool
rvt_get_rwqe() into fetching an old RWQE and corrupt memory.
Fix by deleting the faulty initialization code and creating an
inline to fetch the posted count and convert all callers to use
new inline.
Fixes: f592ae3c99 ("IB/rdmavt: Fracture single lock used for posting and processing RWQEs")
Link: https://lore.kernel.org/r/20200728183848.22226.29132.stgit@awfm-01.aw.intel.com
Reported-by: Zhaojuan Guo <zguo@redhat.com>
Cc: <stable@vger.kernel.org> # 5.4.x
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Tested-by: Honggang Li <honli@redhat.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The header files in RDMA subsystem are dual licensed and can be
described by simple SPDX tag, so replace all of them at once
together with making them use the same coding style for header
guard defines.
Link: https://lore.kernel.org/r/20200719072521.135260-1-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Create the same logic flow for the write() interface as we have for the
ioctl() path by making sure that the object is committed or aborted
automatically after HW object creation.
Link: https://lore.kernel.org/r/20200719052223.75245-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Update the code to allocate and free ib_xrcd structure in the
ib_core instead of inside drivers.
Link: https://lore.kernel.org/r/20200630101855.368895-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Expose UAPI to query ucontext, this will let user space application that
didn't allocate the ucontext but has access to by owning the matching
command FD to retrieve the ucontext information.
Link: https://lore.kernel.org/r/20200630093916.332097-4-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Replace the mutex with read write semaphore and use xarray instead of
linked list for XRC target QPs. This will give faster XRC target
lookup. In addition, when QP is closed, don't insert it back to the xarray
if the destroy command failed.
Link: https://lore.kernel.org/r/20200706122716.647338-4-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
ib_alloc_xrcd() already does the required initialization, so move the
uverbs to call it and save code duplication, while cleaning the function
argument lists of that function.
Link: https://lore.kernel.org/r/20200706122716.647338-3-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Allocating an MR flow can only be initiated by kernel users, and not from
userspace so a udata parameter is redundant.
Link: https://lore.kernel.org/r/20200706120343.10816-4-galpress@amazon.com
Signed-off-by: Gal Pressman <galpress@amazon.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Allocating an MR flow can only be initiated by kernel users, and not from
userspace. As a result, the udata parameter is always being passed as
NULL. Rename ib_alloc_mr_user function to ib_alloc_mr and remove the udata
parameter.
Link: https://lore.kernel.org/r/20200706120343.10816-3-galpress@amazon.com
Signed-off-by: Gal Pressman <galpress@amazon.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
For debugging purpose it will be easier to understand if prefetch works
okay if it has its own counter. Introduce ODP prefetch counter and count
per MR the total number of prefetched pages.
In addition remove comment which is not relevant anymore and anyway not in
the correct place.
Link: https://lore.kernel.org/r/20200621104147.53795-1-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
The RWQ table is used for RSS uverbs and not in used for the kernel
consumers, delete ib_create_rwq_ind_table() routine that is not
called at all.
Link: https://lore.kernel.org/r/20200624105422.1452290-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Add support to get resource dump in raw format. It enable drivers to
return the entire device specific QP/CQ/MR context without a need from the
driver to set each field separately.
The raw query returns only the device specific data, general data is still
returned by using the existing queries.
Example:
$ rdma res show mr dev mlx5_1 mrn 2 -r -j
[{"ifindex":7,"ifname":"mlx5_1",
"data":[0,4,255,254,0,0,0,0,0,0,0,0,16,28,0,216,...]}]
Link: https://lore.kernel.org/r/20200623113043.1228482-9-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In order to avoid double multiplexing of the resource when it is a cm id,
add a dedicated callback function. In addition remove fill_res_entry which
is not used anymore.
Link: https://lore.kernel.org/r/20200623113043.1228482-8-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In order to avoid double multiplexing of the resource when it is a QP, add
a dedicated callback function.
Link: https://lore.kernel.org/r/20200623113043.1228482-7-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In order to avoid double multiplexing of the resource when it is a CQ, add
a dedicated callback function.
Link: https://lore.kernel.org/r/20200623113043.1228482-6-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In order to avoid double multiplexing of the resource when it is a MR, add
a dedicated callback function.
Link: https://lore.kernel.org/r/20200623113043.1228482-5-leon@kernel.org
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Instead of enabling dynamic debug globally with CONFIG_DYNAMIC_DEBUG,
CONFIG_DYNAMIC_DEBUG_CORE will only enable core function of dynamic
debug. With the DYNAMIC_DEBUG_MODULE defined for any modules, dynamic
debug will be tied to them.
This is useful for people who only want to enable dynamic debug for
kernel modules without worrying about kernel image size and memory
consumption is increasing too much.
[orson.zhai@unisoc.com: v2]
Link: http://lkml.kernel.org/r/1587408228-10861-1-git-send-email-orson.unisoc@gmail.com
Signed-off-by: Orson Zhai <orson.zhai@unisoc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Petr Mladek <pmladek@suse.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Jason Baron <jbaron@akamai.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Link: http://lkml.kernel.org/r/1586521984-5890-1-git-send-email-orson.unisoc@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A few large, long discussed works this time. The RNBD block driver has
been posted for nearly two years now, and the removal of FMR has been a
recurring discussion theme for a long time. The usual smattering of
features and bug fixes.
- Various small driver bugs fixes in rxe, mlx5, hfi1, and efa
- Continuing driver cleanups in bnxt_re, hns
- Big cleanup of mlx5 QP creation flows
- More consistent use of src port and flow label when LAG is used and a
mlx5 implementation
- Additional set of cleanups for IB CM
- 'RNBD' network block driver and target. This is a network block RDMA
device specific to ionos's cloud environment. It brings strong multipath
and resiliency capabilities.
- Accelerated IPoIB for HFI1
- QP/WQ/SRQ ioctl migration for uverbs, and support for multiple async fds
- Support for exchanging the new IBTA defiend ECE data during RDMA CM
exchanges
- Removal of the very old and insecure FMR interface from all ULPs and
drivers. FRWR should be preferred for at least a decade now.
-----BEGIN PGP SIGNATURE-----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=9zTe
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"A more active cycle than most of the recent past, with a few large,
long discussed works this time.
The RNBD block driver has been posted for nearly two years now, and
flowing through RDMA due to it also introducing a new ULP.
The removal of FMR has been a recurring discussion theme for a long
time.
And the usual smattering of features and bug fixes.
Summary:
- Various small driver bugs fixes in rxe, mlx5, hfi1, and efa
- Continuing driver cleanups in bnxt_re, hns
- Big cleanup of mlx5 QP creation flows
- More consistent use of src port and flow label when LAG is used and
a mlx5 implementation
- Additional set of cleanups for IB CM
- 'RNBD' network block driver and target. This is a network block
RDMA device specific to ionos's cloud environment. It brings strong
multipath and resiliency capabilities.
- Accelerated IPoIB for HFI1
- QP/WQ/SRQ ioctl migration for uverbs, and support for multiple
async fds
- Support for exchanging the new IBTA defiend ECE data during RDMA CM
exchanges
- Removal of the very old and insecure FMR interface from all ULPs
and drivers. FRWR should be preferred for at least a decade now"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (247 commits)
RDMA/cm: Spurious WARNING triggered in cm_destroy_id()
RDMA/mlx5: Return ECE DC support
RDMA/mlx5: Don't rely on FW to set zeros in ECE response
RDMA/mlx5: Return an error if copy_to_user fails
IB/hfi1: Use free_netdev() in hfi1_netdev_free()
RDMA/hns: Uninitialized variable in modify_qp_init_to_rtr()
RDMA/core: Move and rename trace_cm_id_create()
IB/hfi1: Fix hfi1_netdev_rx_init() error handling
RDMA: Remove 'max_map_per_fmr'
RDMA: Remove 'max_fmr'
RDMA/core: Remove FMR device ops
RDMA/rdmavt: Remove FMR memory registration
RDMA/mthca: Remove FMR support for memory registration
RDMA/mlx4: Remove FMR support for memory registration
RDMA/i40iw: Remove FMR leftovers
RDMA/bnxt_re: Remove FMR leftovers
RDMA/mlx5: Remove FMR leftovers
RDMA/core: Remove FMR pool API
RDMA/rds: Remove FMR support for memory registration
RDMA/srp: Remove support for FMR memory registration
...
This ancient and unsafe method for memory registration is no longer used
by any RDMA based ULP. Remove the FMR pool API from the core driver.
Link: https://lore.kernel.org/r/4-v3-f58e6669d5d3+2cf-fmr_removal_jgg@mellanox.com
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Allow a ULP to ask the core to provide a completion queue based on a
least-used search on a per-device CQ pools. The device CQ pools grow in a
lazy fashion when more CQs are requested.
This feature reduces the amount of interrupts when using many QPs. Using
shared CQs allows for more effcient completion handling. It also reduces
the amount of overhead needed for CQ contexts.
Test setup:
Intel(R) Xeon(R) Platinum 8176M CPU @ 2.10GHz servers.
Running NVMeoF 4KB read IOs over ConnectX-5EX across Spectrum switch.
TX-depth = 32. The patch was applied in the nvme driver on both the target
and initiator. Four controllers are accessed from each core. In the
current test case we have exposed sixteen NVMe namespaces using four
different subsystems (four namespaces per subsystem) from one NVM port.
Each controller allocated X queues (RDMA QPs) and attached to Y CQs.
Before this series we had X == Y, i.e for four controllers we've created
total of 4X QPs and 4X CQs. In the shared case, we've created 4X QPs and
only X CQs which means that we have four controllers that share a
completion queue per core. Until fourteen cores there is no significant
change in performance and the number of interrupts per second is less than
a million in the current case.
==================================================
|Cores|Current KIOPs |Shared KIOPs |improvement|
|-----|---------------|--------------|-----------|
|14 |2332 |2723 |16.7% |
|-----|---------------|--------------|-----------|
|20 |2086 |2712 |30% |
|-----|---------------|--------------|-----------|
|28 |1971 |2669 |35.4% |
|=================================================
|Cores|Current avg lat|Shared avg lat|improvement|
|-----|---------------|--------------|-----------|
|14 |767us |657us |14.3% |
|-----|---------------|--------------|-----------|
|20 |1225us |943us |23% |
|-----|---------------|--------------|-----------|
|28 |1816us |1341us |26.1% |
========================================================
|Cores|Current interrupts|Shared interrupts|improvement|
|-----|------------------|-----------------|-----------|
|14 |1.6M/sec |0.4M/sec |72% |
|-----|------------------|-----------------|-----------|
|20 |2.8M/sec |0.6M/sec |72.4% |
|-----|------------------|-----------------|-----------|
|28 |2.9M/sec |0.8M/sec |63.4% |
====================================================================
|Cores|Current 99.99th PCTL lat|Shared 99.99th PCTL lat|improvement|
|-----|------------------------|-----------------------|-----------|
|14 |67ms |6ms |90.9% |
|-----|------------------------|-----------------------|-----------|
|20 |5ms |6ms |-10% |
|-----|------------------------|-----------------------|-----------|
|28 |8.7ms |6ms |25.9% |
|===================================================================
Performance improvement with sixteen disks (sixteen CQs per core) is
comparable.
Link: https://lore.kernel.org/r/1590568495-101621-3-git-send-email-yaminf@mellanox.com
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
A pre-step for adding shared CQs. Add the infrastructure to prevent shared
CQ users from altering the CQ configurations. For now all cqs are marked
as private (non-shared). The core driver should use the new force
functions to perform resize/destroy/moderation changes that are not
allowed for users of shared CQs.
Link: https://lore.kernel.org/r/1590568495-101621-2-git-send-email-yaminf@mellanox.com
Signed-off-by: Yamin Friedman <yaminf@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
IBTA declares "vendor option not supported" reject reason in REJ messages
if passive side doesn't want to accept proposed ECE options.
Due to the fact that ECE is managed by userspace, there is a need to let
users to provide such rejected reason.
Link: https://lore.kernel.org/r/20200526103304.196371-7-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The rdma_accept() is called by both passive and active sides of CMID
connection to mark readiness to start data transfer. For passive side,
this is called explicitly, for active side, it is called implicitly while
receiving REP message.
Provide ECE data to rdma_accept function needed for passive side to send
that REP message.
Link: https://lore.kernel.org/r/20200526103304.196371-6-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
ECE parameters are exchanged through REQ->REP/SIDR_REP messages, this
patch adds the data to provide to other side of CMID communication
channel.
Link: https://lore.kernel.org/r/20200526103304.196371-5-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Passive side of CMID connection receives ECE request through REQ message
and needs to respond with relevant REP message which will be forwarded to
active side.
The UCMA events interface is responsible for such communication with the
user space (librdmacm). Extend it to provide ECE wire data.
Link: https://lore.kernel.org/r/20200526103304.196371-4-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Active side of CMID initiates connection through librdmacm's
rdma_connect() and kernel's ucma_connect(). Extend UCMA interface to
handle those new parameters.
Link: https://lore.kernel.org/r/20200526103304.196371-3-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Extend REQ (request for communications), REP (reply to request for
communication), rejected reason and SIDR_REP (service ID resolution
response) structures with hardware vendor ID bits according to IBTA v1.4.
Link: https://lore.kernel.org/r/20200526103304.196371-2-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Make use of the sizeof_field() helper instead of an open-coded version.
Link: https://lore.kernel.org/r/20200527144152.GA22605@embeddedor
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
These constants are going to be used in the ioctl interface in coming
patches so they are part of the UAPI, place them in the correct header
for clarity.
Link: https://lore.kernel.org/r/20200519072711.257271-5-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
While creating a uobject every create reaches a point where the uobject is
fully initialized. For ioctls that go on to copy_to_user this means they
need to open code the destruction of a fully created uobject - ie the
RDMA_REMOVE_DESTROY sort of flow.
Open coding this creates bugs, eg the CQ does not properly flush the
events list when it does its error unwind.
Provide a uverbs_finalize_uobj_create() function which indicates that the
uobject is fully initialized and that abort should call to destroy_hw to
destroy the uobj->object and related.
Methods can call this function if they go on to have error cases after
setting uobj->object. Once done those error cases can simply do return,
without an error unwind.
Link: https://lore.kernel.org/r/20200519072711.257271-2-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Currently the ipoib UD mtu is restricted to 4K bytes. Remove this
limitation so that the IPOIB module can potentially use an MTU (in UD
mode) that is bounded by the MTU of the underlying device. A field is
added to the ib_port_attr structure to indicate the maximum physical
MTU the underlying device supports.
Link: https://lore.kernel.org/r/20200511160618.173205.23053.stgit@awfm-01.aw.intel.com
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sadanand Warrier <sadanand.warrier@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Adds capability to create a qpn to be recognized as an accelerated
UD QP for ipoib.
This is accomplished by reserving 0x81 in byte[0] of the qpn as the
prefix for these qp types and reserving qpns between 0x810000 and
0x81ffff.
The hfi1 capability mask already contained a flag for the VNIC netdev.
This has been renamed and extended to include both VNIC and ipoib.
The rvt code to allocate qps now recognizes this flag and sets 0x81
into byte[0] of the qpn.
The code to allocate qpns is modified to reset the qpn numbering when it
is detected that a value is located in byte[0] for a UD QP and it is a
qpn being requested for net dev use. If it is a regular UD QP then it is
allowable to have bits set in byte[0] of the qpn and provide the
previously normal behavior.
The code to free the qpn now checks for the AIP prefix value of 0x81 and
removes it from the qpn before being freed so that the lower 16 bit
number can be reused.
This patch requires minor changes in the IB core and ipoib to facilitate
the creation of accelerated UP QPs.
Link: https://lore.kernel.org/r/20200511160607.173205.11757.stgit@awfm-01.aw.intel.com
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Gary Leshner <Gary.S.Leshner@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The module parameter for KDETH qpns is being removed in favor
of always using the default value of 0x80 as the qpn prefix.
Defines have been added for various KDETH values including
the prefix of 0x80.
The reserved range now starts at the base value for KDETH
qpns (0x80) and extends up to and including the last qpn for
other reserved QP prefixed types.
Adjust other QP prefixed define names to match KDETH defined
names.
Link: https://lore.kernel.org/r/20200511160600.173205.27508.stgit@awfm-01.aw.intel.com
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Gary Leshner <Gary.S.Leshner@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This patch implements the mechanism to accelerate the transmit side of
a multiple transmit queue RDMA netdev by submitting the packets to
the SDMA engine directly instead of sending through the verbs layer.
This patch also changes the UD/SEND_ONLY op to output the entropy value
in byte 0 of deth[1]. UD/SEND_ONLY_WITH_IMMEDIATE uses the previous
behavior with no entropy value being output.
The code in the ipoib rdma netdev which submits tx requests upon
successful submission will call trace_sdma_output_ibhdr to output
the ibhdr to the trace buffer.
Link: https://lore.kernel.org/r/20200511160548.173205.45616.stgit@awfm-01.aw.intel.com
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Gary Leshner <Gary.S.Leshner@intel.com>
Signed-off-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The uverbs layer largely duplicate the code in ib_create_srq(), with the
slight difference that it passes in a udata. Move all the code together
into ib_create_srq_user() and provide an inline for kernel users, similar
to other create calls.
Link: https://lore.kernel.org/r/20200506082444.14502-6-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by
this change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
sizeof(flexible-array-member) triggers a warning because flexible array
members have incomplete type[1]. There are some instances of code in
which the sizeof operator is being incorrectly/erroneously applied to
zero-length arrays and the result is zero. Such instances may be hiding
some bugs. So, this work (flexible-array member conversions) will also
help to get completely rid of those sorts of issues.
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Link: https://lore.kernel.org/r/20200507185342.GA14476@embeddedor
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
