Leon Romanovsky [Tue, 13 Aug 2019 10:28:14 +0000 (13:28 +0300)]
RDMA/mlx5: Annotate lock dependency in bind/unbind slave port
NULL-ing notifier_call is performed under protection
of mlx5_ib_multiport_mutex lock. Such protection is
not easily spotted and better to be guarded by lockdep
annotation.
Weihang Li [Thu, 8 Aug 2019 14:53:53 +0000 (22:53 +0800)]
RDMA/hns: Disable alw_lcl_lpbk of SSU
If we enabled alw_lcl_lpbk in promiscuous mode, packet whose source
and destination mac address is equal will be handled in both inner
loopback and outer loopback. This will halve performance of roce in
promiscuous mode.
Kamal Heib [Wed, 7 Aug 2019 10:31:37 +0000 (13:31 +0300)]
RDMA/core: Add common iWARP query port
Add support for a common iWARP query port function, the new function
includes a common code that is used by the iWARP devices to update the
port attributes like max_mtu, active_mtu, state, and phys_state, the
function also includes a call for the driver-specific query_port callback
to query the device-specific port attributes.
Chuhong Yuan [Tue, 6 Aug 2019 01:59:50 +0000 (09:59 +0800)]
mlx5: Use refcount_t for refcount
Reference counters are preferred to use refcount_t instead of
atomic_t.
This is because the implementation of refcount_t can prevent
overflows and detect possible use-after-free.
So convert atomic_t ref counters to refcount_t.
Colin Ian King [Fri, 2 Aug 2019 15:13:16 +0000 (16:13 +0100)]
net/mlx5: remove self-assignment on esw->dev
There is a self assignment of esw->dev to itself, clean this up by
removing it. Also make dev a const pointer.
Addresses-Coverity: ("Self assignment") Fixes: 6cedde451399 ("net/mlx5: E-Switch, Verify support QoS element type") Signed-off-by: Colin Ian King <colin.king@canonical.com> Reviewed-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Gal Pressman [Thu, 1 Aug 2019 17:14:47 +0000 (20:14 +0300)]
RDMA/efa: Rate limit admin queue error prints
Admin queue error prints should never happen unless something wrong
happened to the device. However, in the unfortunate case that it does,
we should take extra care not to flood the log with error messages.
Leon Romanovsky [Thu, 1 Aug 2019 11:48:27 +0000 (14:48 +0300)]
RDMA/hns: Remove not used UAR assignment
UAR in CQ is not used and generates the following compilation
warning, clean the code by removing uar assignment.
drivers/infiniband/hw/hns/hns_roce_cq.c: In function _create_user_cq_:
drivers/infiniband/hw/hns/hns_roce_cq.c:305:27: warning: parameter _uar_ set but not used [-Wunused-but-set-parameter]
305 | struct hns_roce_uar *uar,
| ~~~~~~~~~~~~~~~~~~~~~^~~
Chuck Lever [Mon, 29 Jul 2019 17:22:09 +0000 (13:22 -0400)]
rdma: Enable ib_alloc_cq to spread work over a device's comp_vectors
Send and Receive completion is handled on a single CPU selected at
the time each Completion Queue is allocated. Typically this is when
an initiator instantiates an RDMA transport, or when a target
accepts an RDMA connection.
Some ULPs cannot open a connection per CPU to spread completion
workload across available CPUs and MSI vectors. For such ULPs,
provide an API that allows the RDMA core to select a completion
vector based on the device's complement of available comp_vecs.
ULPs that invoke ib_alloc_cq() with only comp_vector 0 are converted
to use the new API so that their completion workloads interfere less
with each other.
Mark Zhang [Wed, 31 Jul 2019 11:40:14 +0000 (14:40 +0300)]
IB/mlx5: Support MLX5_CMD_OP_QUERY_LAG as a DEVX general command
The "MLX5_CMD_OP_QUERY_LAG" is one of the DEVX general commands, add it.
Fixes: 8aa8c95ce4cc ("IB/mlx5: Add support for DEVX general command") Signed-off-by: Mark Zhang <markz@mellanox.com> Reviewed-by: Yishai Hadas <yishaih@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
However for VFs, vports are initialized before creating their
respective netdev represnetors in event handling context.
Similarly while disabling eswitch, first hardware vports are disabled,
followed by destroying their representors.
Here while underlying vports gets destroyed but its respective user
facing netdevice can still exist on which user can continue to perform
more offload operations.
Instead, its more accurate to do
enable_eswitch switchdev mode:
1. perform FDB tables initialization
2. initialize hw vport
3. create and publish representor for this vport
disable_eswitch switchdev mode:
1. destroy user facing representor for the vport
2. disable hw vport
3. perform FDB tables cleanup
mc_promisc pointer points to an instance of struct esw_mc_addr allocated
as part of the esw structure.
Hence it cannot be NULL.
Removed such redundant check and assign where it is actually used.
While at it, add comment around legacy mode fields and move mc_promisc
close to other legacy mode structures to improve code redability.
We don't need to handle error flow of esw_create_legacy_table() in the
same branch, it is already being handled directly after the if statement,
for both legacy and switchdev modes in one place.
net/mlx5: E-switch, Introduce helper function to enable/disable vports
vports needs to be enabled in switchdev and legacy mode.
In switchdev mode, vports should be enabled after initializing
the FDB tables and before creating their represntors so that
representor works on an initialized vport object.
Prepare a helper function which can be called when enabling either of
the eswitch modes.
net/mlx5: E-switch, Initialize TSAR Qos hardware block before its user vports
First enable TSAR Qos hardware block in device before enabling its
user vports.
This refactor is needed so that vports can be enabled before their
representor netdevice can be created.
While at it, esw_create_tsar() returns error code which was used only to
print error. However esw_create_tsar() already prints warning if it hits
an error.
Hence, remove the redundant warning.
Except bit toggling code, rest of the code is same to enable/disable
metadata passing functionality.
Hence, combine them to single function and control using enable flag.
Also instead of checking metadata supported at multiple places,
fold into the helper function.
Eli Cohen [Mon, 29 Jul 2019 21:13:00 +0000 (21:13 +0000)]
net/mlx5: E-Switch, Verify support QoS element type
Check if firmware supports the requested element type before
attempting to create the element type.
In addition, explicitly specify the request element type and tsar type.
Signed-off-by: Eli Cohen <eli@mellanox.com> Reviewed-by: Paul Blakey <paulb@mellanox.com> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
net/mlx5: Refactor and optimize flow counter bulk query
Towards introducing the ability to allocate bulks of flow counters,
refactor the flow counter bulk query process, removing functions and
structs whose names indicated being used for flow counter bulk
allocation FW commands, despite them actually only being used to
support bulk querying, and migrate their functionality to correctly
named functions in their natural location, fs_counters.c.
Additionally, optimize the bulk query process by:
* Extracting the memory used for the query to mlx5_fc_stats so
that it is only allocated once, and not for each bulk query.
* Querying all the counters in one function call.
RDMA/hns: remove set but not used variable 'irq_num'
Fixes gcc '-Wunused-but-set-variable' warning:
drivers/infiniband/hw/hns/hns_roce_hw_v2.c: In function hns_roce_v2_cleanup_eq_table:
drivers/infiniband/hw/hns/hns_roce_hw_v2.c:5920:6:
warning: variable irq_num set but not used [-Wunused-but-set-variable]
It is not used since
commit 33db6f94847c ("RDMA/hns: Refactor eq table init for hip08")
Stephen Boyd [Tue, 30 Jul 2019 18:15:20 +0000 (11:15 -0700)]
infiniband: Remove dev_err() usage after platform_get_irq()
We don't need dev_err() messages when platform_get_irq() fails now that
platform_get_irq() prints an error message itself when something goes
wrong. Let's remove these prints with a simple semantic patch.
Qian Cai [Mon, 22 Jul 2019 18:34:42 +0000 (14:34 -0400)]
net/mlx5: fix -Wtype-limits compilation warnings
The commit b9a7ba556207 ("net/mlx5: Use event mask based on device
capabilities") introduced a few compilation warnings due to it bumps
MLX5_EVENT_TYPE_MAX from 0x27 to 0x100 which is always greater than
an "struct {mlx5_eqe|mlx5_nb}.type" that is an "u8".
drivers/net/ethernet/mellanox/mlx5/core/eq.c: In function
'mlx5_eq_notifier_register':
drivers/net/ethernet/mellanox/mlx5/core/eq.c:948:21: warning: comparison
is always false due to limited range of data type [-Wtype-limits]
if (nb->event_type >= MLX5_EVENT_TYPE_MAX)
^~
drivers/net/ethernet/mellanox/mlx5/core/eq.c: In function
'mlx5_eq_notifier_unregister':
drivers/net/ethernet/mellanox/mlx5/core/eq.c:959:21: warning: comparison
is always false due to limited range of data type [-Wtype-limits]
if (nb->event_type >= MLX5_EVENT_TYPE_MAX)
Fix them by removing unnecessary checkings.
Fixes: b9a7ba556207 ("net/mlx5: Use event mask based on device capabilities") Signed-off-by: Qian Cai <cai@lca.pw> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
GID table entry operations such as add/remove/modify are triggered
by the IB core for RoCE ports.
Hence, remove GID change notification from hw driver.
Michal Kalderon [Sun, 28 Jul 2019 11:13:38 +0000 (14:13 +0300)]
RDMA/qedr: Fix the hca_type and hca_rev returned in device attributes
There was a place holder for hca_type and vendor was returned
in hca_rev. Fix the hca_rev to return the hw revision and fix
the hca_type to return an informative string representing the
hca.
If INFINIBAND_HNS_HIP08 is selected and HNS3 is m,
but INFINIBAND_HNS is y, building fails:
drivers/infiniband/hw/hns/hns_roce_hw_v2.o: In function `hns_roce_hw_v2_exit':
hns_roce_hw_v2.c:(.exit.text+0xd): undefined reference to `hnae3_unregister_client'
drivers/infiniband/hw/hns/hns_roce_hw_v2.o: In function `hns_roce_hw_v2_init':
hns_roce_hw_v2.c:(.init.text+0xd): undefined reference to `hnae3_register_client'
Also if INFINIBAND_HNS_HIP06 is selected and HNS_DSAF
is m, but INFINIBAND_HNS is y, building fails:
drivers/infiniband/hw/hns/hns_roce_hw_v1.o: In function `hns_roce_v1_reset':
hns_roce_hw_v1.c:(.text+0x39fa): undefined reference to `hns_dsaf_roce_reset'
hns_roce_hw_v1.c:(.text+0x3a25): undefined reference to `hns_dsaf_roce_reset'
Reported-by: Hulk Robot <hulkci@huawei.com> Fixes: dd74282df573 ("RDMA/hns: Initialize the PCI device for hip08 RoCE") Fixes: 08805fdbeb2d ("RDMA/hns: Split hw v1 driver from hns roce driver") Signed-off-by: YueHaibing <yuehaibing@huawei.com> Link: https://lore.kernel.org/r/20190724065443.53068-1-yuehaibing@huawei.com Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/mlx5: Support per device q counters in switchdev mode
When parent mlx5_core_dev is in switchdev mode, q_counters are not
applicable to multiple non uplink vports.
Hence, have make them limited to device level.
While at it, correct __mlx5_ib_qp_set_counter() and
__mlx5_ib_modify_qp() to use u16 set_id as defined by the device.
To support per device counters in switchdev mode (instead of
per port counter), refactor query routines to work on mlx5_ib_counter
structure instead of mlx5_ib_port structure.
Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Link: https://lore.kernel.org/r/20190723073117.7175-2-leon@kernel.org Signed-off-by: Doug Ledford <dledford@redhat.com>
Michal Kalderon [Tue, 9 Jul 2019 14:17:33 +0000 (17:17 +0300)]
qed*: Change dpi_addr to be denoted with __iomem
Several casts were required around dpi_addr parameter in qed_rdma_if.h
This is an address on the doorbell bar and should therefore be marked with
__iomem.
Max Gurtovoy [Tue, 23 Jul 2019 07:04:12 +0000 (10:04 +0300)]
IB/mlx5: Add CREATE_PSV/DESTROY_PSV for devx interface
Limit the number of PSV's created through devx to 1, to create a symmetry
between create/destroy cmds. In the kernel, one can create up to 4 PSV's
using CREATE_PSV cmd but the destruction is one by one. Add a protection
for this a-symmetric definition for devx.
RDMA/core: Support netlink commands in non init_net net namespaces
Now that IB core supports RDMA device binding with specific net namespace,
enable IB core to accept netlink commands in non init_net namespaces.
This is done by having per net namespace netlink socket.
At present only netlink device handling client RDMA_NL_NLDEV supports
device handling in multiple net namespaces. Hence do not accept netlink
messages for other clients in non init_net net namespaces.
Link: https://lore.kernel.org/r/20190723070205.6247-1-leon@kernel.org Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Leon Romanovsky [Thu, 4 Jul 2019 13:09:35 +0000 (16:09 +0300)]
RDMA/mlx4: Separate creation of RWQ and QP
The mlx4 WQ is implemented with HW QP without special HW object. Current
implementation which tried to reuse the code did it with common QP
creation flows. Such decision caused to the absence of mlx4_ib_wq struct,
which is needed to ensure proper allocation of ib_wq inside of IB/core.
Separate create_qp_common() to pure QP flow and to create_rq() for RWQ.
Lijun Ou [Mon, 8 Jul 2019 13:41:24 +0000 (21:41 +0800)]
RDMA/hns: Refactor hem table mhop check and calculation
The calculation of mhop for hem is duplicated in hns_roce_init_hem_table
and hns_roce_calc_hem_mhop, extracting it from them to a separate
function. Moreover, this patch refactors hns_roce_check_whether_mhop to
reduce complexity.
Xi Wang [Mon, 8 Jul 2019 13:41:21 +0000 (21:41 +0800)]
RDMA/hns: optimize the duplicated code for qpc setting flow
Currently, more than 20 lines of duplicate code exist in function
'modify_qp_init_to_init' and function 'modify_qp_reset_to_init', which
affects the readability of the code. Consolidate them.
Lijun Ou [Mon, 8 Jul 2019 13:41:19 +0000 (21:41 +0800)]
RDMA/hns: Refactor for hns_roce_v2_modify_qp function
Move some lines which exist hns_roce_v2_modify_qp function into a new
function. The code refactored mainly includes some absolute fields of qp
context and some optional fields of qp context.
RDMA/core: Annotate destroy of mutex to ensure that it is released as unlocked
While compiled with CONFIG_DEBUG_MUTEXES, the kernel ensures that mutex is
not held during destroy. Hence add mutex_destroy() for mutexes used in
RDMA modules.
Link: https://lore.kernel.org/r/20190723065733.4899-2-leon@kernel.org Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
IB/mlx5: Fix RSS Toeplitz setup to be aligned with the HW specification
The specification for the Toeplitz function doesn't require to set the key
explicitly to be symmetric. In case a symmetric functionality is required
a symmetric key can be simply used.
Wrongly forcing the algorithm to symmetric causes the wrong packet
distribution and a performance degradation.
Link: https://lore.kernel.org/r/20190723065733.4899-7-leon@kernel.org Cc: <stable@vger.kernel.org> # 4.7 Fixes: 28d6137008b2 ("IB/mlx5: Add RSS QP support") Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Alex Vainman <alexv@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
IB/counters: Always initialize the port counter object
Port counter objects should be initialized even if alloc_stats is
unsupported, otherwise QP bind operations in user space can trigger a NULL
pointer deference if they try to bind QP on RDMA device which doesn't
support counters.
Fixes: f34a55e497e8 ("RDMA/core: Get sum value of all counters when perform a sysfs stat read") Link: https://lore.kernel.org/r/20190723065733.4899-11-leon@kernel.org Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Mark Zhang <markz@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fixes: f34a55e497e8 ("RDMA/core: Get sum value of all counters when perform a sysfs stat read") Link: https://lore.kernel.org/r/20190723065733.4899-10-leon@kernel.org Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Mark Zhang <markz@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Moni Shoua [Tue, 23 Jul 2019 06:57:30 +0000 (09:57 +0300)]
IB/mlx5: Prevent concurrent MR updates during invalidation
The device requires that memory registration work requests that update the
address translation table of a MR will be fenced if posted together. This
scenario can happen when address ranges are invalidated by the mmu in
separate concurrent calls to the invalidation callback.
We prefer to block concurrent address updates for a single MR over fencing
since making the decision if a WQE needs fencing will be more expensive
and fencing all WQEs is a too radical choice.
Further, it isn't clear that this code can even run safely concurrently,
so a lock is a safer choice.
Fixes: b4cfe447d47b ("IB/mlx5: Implement on demand paging by adding support for MMU notifiers") Link: https://lore.kernel.org/r/20190723065733.4899-8-leon@kernel.org Signed-off-by: Moni Shoua <monis@mellanox.com> Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
IB/mlx5: Move MRs to a kernel PD when freeing them to the MR cache
Fix unreg_umr to move the MR to a kernel owned PD (i.e. the UMR PD) which
can't be accessed by userspace.
This ensures that nothing can continue to access the MR once it has been
placed in the kernels cache for reuse.
MRs in the cache continue to have their HW state, including DMA tables,
present. Even though the MR has been invalidated, changing the PD provides
an additional layer of protection against use of the MR.
Link: https://lore.kernel.org/r/20190723065733.4899-5-leon@kernel.org Cc: <stable@vger.kernel.org> # 3.10 Fixes: e126ba97dba9 ("mlx5: Add driver for Mellanox Connect-IB adapters") Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fix unreg_umr to ignore the mkey state and do not fail if was freed. This
prevents a case that a user space application already changed the mkey
state to free and then the UMR operation will fail leaving the mkey in an
inappropriate state.
Link: https://lore.kernel.org/r/20190723065733.4899-3-leon@kernel.org Cc: <stable@vger.kernel.org> # 3.19 Fixes: 968e78dd9644 ("IB/mlx5: Enhance UMR support to allow partial page table update") Signed-off-by: Yishai Hadas <yishaih@mellanox.com> Reviewed-by: Artemy Kovalyov <artemyko@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Jason Gunthorpe <jgg@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Colin Ian King [Fri, 31 May 2019 09:21:00 +0000 (10:21 +0100)]
RDMA/hns: Fix comparison of unsigned long variable 'end' with less than zero
Currently the comparison of end with less than zero is always false
because end is an unsigned long. Also, replace checks of end with
non-zero with end > 0 as it is possible that the #defined decrement may be
changed in the future causing end to step over zero and go negative.
The initialization of end with 0 is also redundant as this value is never
read and is later set to HW_SYNC_TIMEOUT_MSECS, so fix this by
initializing it with this value to begin with.
Link: https://lore.kernel.org/r/20190531092101.28772-1-colin.king@canonical.com
Addresses-Coverity: ("Unsigned compared against 0") Fixes: 669cefb654cb ("RDMA/hns: Remove jiffies operation in disable interrupt context") Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
RDMA/mlx4: Untag user pointers in mlx4_get_umem_mr
This patch is a part of a series that extends kernel ABI to allow to pass
tagged user pointers (with the top byte set to something else other than
0x00) as syscall arguments.
mlx4_get_umem_mr() uses provided user pointers for vma lookups, which can
only by done with untagged pointers.
Kaike Wan [Mon, 15 Jul 2019 16:45:46 +0000 (12:45 -0400)]
IB/hfi1: Do not update hcrc for a KDETH packet during fault injection
When a KDETH packet is subject to fault injection during transmission,
HCRC is supposed to be omitted from the packet so that the hardware on the
receiver side would drop the packet. When creating pbc, the PbcInsertHcrc
field is set to be PBC_IHCRC_NONE if the KDETH packet is subject to fault
injection, but overwritten with PBC_IHCRC_LKDETH when update_hcrc() is
called later.
This problem is fixed by not calling update_hcrc() when the packet is
subject to fault injection.
Fixes: 6b6cf9357f78 ("IB/hfi1: Set PbcInsertHcrc for TID RDMA packets") Cc: <stable@vger.kernel.org> Link: https://lore.kernel.org/r/20190715164546.74174.99296.stgit@awfm-01.aw.intel.com Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Mao Wenan [Fri, 19 Jul 2019 01:29:38 +0000 (09:29 +0800)]
RDMA/siw: Remove set but not used variables 'rv'
Fixes gcc '-Wunused-but-set-variable' warning:
drivers/infiniband/sw/siw/siw_cm.c: In function siw_cep_set_inuse:
drivers/infiniband/sw/siw/siw_cm.c:223:6: warning: variable rv set but not used [-Wunused-but-set-variable]
Fixes: 6c52fdc244b5 ("rdma/siw: connection management") Link: https://lore.kernel.org/r/20190719012938.100628-1-maowenan@huawei.com Signed-off-by: Mao Wenan <maowenan@huawei.com> Reviewed-by: Bernard Metzler <bmt@zurich.ibm.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
RDMA/bnxt_re: Honor vlan_id in GID entry comparison
A GID entry consists of GID, vlan, netdev and smac. Extend GID duplicate
check comparisons to consider vlan_id as well to support IPv6 VLAN based
link local addresses. Introduce a new structure (bnxt_qplib_gid_info) to
hold gid and vlan_id information.
The issue is discussed in the following thread
https://lore.kernel.org/r/AM0PR05MB4866CFEDCDF3CDA1D7D18AA5D1F20@AM0PR05MB4866.eurprd05.prod.outlook.com
Fixes: 823b23da7113 ("IB/core: Allow vlan link local address based RoCE GIDs") Cc: <stable@vger.kernel.org> # v5.2+ Link: https://lore.kernel.org/r/20190715091913.15726-1-selvin.xavier@broadcom.com Reported-by: Yi Zhang <yi.zhang@redhat.com> Co-developed-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com> Tested-by: Yi Zhang <yi.zhang@redhat.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Kaike Wan [Mon, 15 Jul 2019 16:45:40 +0000 (12:45 -0400)]
IB/hfi1: Drop all TID RDMA READ RESP packets after r_next_psn
When a TID sequence error occurs while receiving TID RDMA READ RESP
packets, all packets after flow->flow_state.r_next_psn should be dropped,
including those response packets for subsequent segments.
The current implementation will drop the subsequent response packets for
the segment to complete next, but may accept packets for subsequent
segments and therefore mistakenly advance the r_next_psn fields for the
corresponding software flows. This may result in failures to complete
subsequent segments after the current segment is completed.
The fix is to only use the flow pointed by req->clear_tail for checking
KDETH PSN instead of finding a flow from the request's flow array.
Fixes: b885d5be9ca1 ("IB/hfi1: Unify the software PSN check for TID RDMA READ/WRITE") Cc: <stable@vger.kernel.org> Link: https://lore.kernel.org/r/20190715164540.74174.54702.stgit@awfm-01.aw.intel.com Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Kaike Wan [Mon, 15 Jul 2019 16:45:34 +0000 (12:45 -0400)]
IB/hfi1: Field not zero-ed when allocating TID flow memory
The field flow->resync_npkts is added for TID RDMA WRITE request and
zero-ed when a TID RDMA WRITE RESP packet is received by the requester.
This field is used to rewind a request during retry in the function
hfi1_tid_rdma_restart_req() shared by both TID RDMA WRITE and TID RDMA
READ requests. Therefore, when a TID RDMA READ request is retried, this
field may not be initialized at all, which causes the retry to start at an
incorrect psn, leading to the drop of the retry request by the responder.
This patch fixes the problem by zeroing out the field when the flow memory
is allocated.
Fixes: 838b6fd2d9ca ("IB/hfi1: TID RDMA RcvArray programming and TID allocation") Cc: <stable@vger.kernel.org> Link: https://lore.kernel.org/r/20190715164534.74174.6177.stgit@awfm-01.aw.intel.com Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Kaike Wan [Mon, 15 Jul 2019 16:45:28 +0000 (12:45 -0400)]
IB/hfi1: Unreserve a flushed OPFN request
When an OPFN request is flushed, the request is completed without
unreserving itself from the send queue. Subsequently, when a new
request is post sent, the following warning will be triggered:
This patch fixes the problem by moving rvt_qp_wqe_unreserve() into
rvt_qp_complete_swqe() to simplify the code and make it less
error-prone.
Fixes: ca95f802ef51 ("IB/hfi1: Unreserve a reserved request when it is completed") Link: https://lore.kernel.org/r/20190715164528.74174.31364.stgit@awfm-01.aw.intel.com Cc: <stable@vger.kernel.org> Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Xi Wang [Thu, 11 Jul 2019 01:32:17 +0000 (09:32 +0800)]
RDMA/hns: Fix sg offset non-zero issue
When run perftest in many times, the system will report a BUG as follows:
BUG: Bad rss-counter state mm:(____ptrval____) idx:0 val:-1
BUG: Bad rss-counter state mm:(____ptrval____) idx:1 val:1
We tested with different kernel version and found it started from the the
following commit:
commit d10bcf947a3e ("RDMA/umem: Combine contiguous PAGE_SIZE regions in
SGEs")
In this commit, the sg->offset is always 0 when sg_set_page() is called in
ib_umem_get() and the drivers are not allowed to change the sgl, otherwise
it will get bad page descriptor when unfolding SGEs in __ib_umem_release()
as sg_page_count() will get wrong result while sgl->offset is not 0.
However, there is a weird sgl usage in the current hns driver, the driver
modified sg->offset after calling ib_umem_get(), which caused we iterate
past the wrong number of pages in for_each_sg_page iterator.
This patch fixes it by correcting the non-standard sgl usage found in the
hns_roce_db_map_user() function.
Fixes: d10bcf947a3e ("RDMA/umem: Combine contiguous PAGE_SIZE regions in SGEs") Fixes: 0425e3e6e0c7 ("RDMA/hns: Support flush cqe for hip08 in kernel space") Link: https://lore.kernel.org/r/1562808737-45723-1-git-send-email-oulijun@huawei.com Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
projects / users / jedix / linux-maple.git / log