Age | Commit message (Collapse) | Author | Files | Lines |
|
Too much print may lead to a panic in kernel. Change ibdev_err() to
ibdev_err_ratelimited(), and change the printing level of cqe dump
to debug level.
Fixes: 7c044adca272 ("RDMA/hns: Simplify the cqe code of poll cq")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-11-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Add mutex_destroy().
Signed-off-by: wenglianfa <wenglianfa@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-9-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
GMV's BA table only supports 4K pages. Currently, PAGESIZE is used to
calculate gmv_bt_num, which will cause an abnormal number of gmv_bt_num
in a 64K OS.
Fixes: d6d91e46210f ("RDMA/hns: Add support for configuring GMV table")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-8-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Add max_ah and cq moderation capacities to hns_roce_query_device().
Fixes: 9a4435375cd1 ("IB/hns: Add driver files for hns RoCE driver")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Remove unused parameters and variables.
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-3-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Use macro instead of magic number.
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240412091616.370789-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Add support for DSCP configuration. For DSCP, get dscp-prio mapping
via hns3 nic driver api .get_dscp_prio() and fill the SL (in WQE for
UD or in QPC for RC) with the priority value. The prio-tc mapping is
configured to HW by hns3 nic driver. HW will select a corresponding
TC according to SL and the prio-tc mapping.
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240315093551.1650088-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
SCCC (SCC Context) is a context with QP granularity that contains
information about congestion control. Dump SCCC and QPC together
to improve troubleshooting.
When dumping raw QPC with rdmatool, there will be a total of 576 bytes
data output, where the first 512 bytes is QPC and the last 64 bytes is
SCCC. When congestion control is disabled, the 64 byte SCCC will be all 0.
Example:
$rdma res show qp -jpr
[ {
"ifindex": 0,
"ifname": "hns_0",
"data": [ 67,0,0,0... 512bytes
4,0,2... 64bytes]
},...
} ]
Signed-off-by: wenglianfa <wenglianfa@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240305055257.823513-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
granularity
Currently, congestion control algorithm is statically configured in
FW, and all QPs use the same algorithm(except UD which has a fixed
configuration of DCQCN). This is not flexible enough.
Support userspace configuring congestion control algorithm with QP
granularity while creating QPs. If the algorithm is not specified in
userspace, use the default one.
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240301104845.1141083-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Commit 27c5fd271d8b ("RDMA/hns: The UD mode can only be configured
with DCQCN") adds a check of congest control alorithm for UD. But
that patch causes a problem: hr_dev->caps.congest_type is global,
used by all QPs, so modifying this field to DCQCN for UD QPs causes
other QPs unable to use any other algorithm except DCQCN.
Revert the modification in commit 27c5fd271d8b ("RDMA/hns: The UD
mode can only be configured with DCQCN"). Add a new field cong_type
to struct hns_roce_qp and configure DCQCN for UD QPs.
Fixes: 27c5fd271d8b ("RDMA/hns: The UD mode can only be configured with DCQCN")
Fixes: f91696f2f053 ("RDMA/hns: Support congestion control type selection according to the FW")
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240219061805.668170-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
'struct hns_roce_hem' is used to refer to the last level of
dma buffer managed by the hw, pointed by a single BA(base
address) in the previous level of BT(base table), so the dma
buffer in 'struct hns_roce_hem' must be contiguous.
Right now the size of dma buffer in 'struct hns_roce_hem' is
decided by mhop->buf_chunk_size in get_hem_table_config(),
which ensure the mhop->buf_chunk_size is power of two of
PAGE_SIZE, so there will be only one contiguous dma buffer
allocated in hns_roce_alloc_hem(), which means hem->chunk_list
and chunk->mem for linking multi dma buffers is unnecessary.
This patch removes the hem->chunk_list and chunk->mem and other
related macro and function accordingly.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240113085935.2838701-7-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
hns_roce_mtr_find() is a collection of multiple functions, and the
return value is also difficult to understand, which is not conducive
to modification and maintenance.
Separate the function of obtaining MTR root BA from this function.
And some adjustments has been made to improve readability.
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20240113085935.2838701-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
These two series from Mark and Shun extend RDMA mlx5 API.
Mark's series provides c0 register used to match egress
traffic sent by local device.
Shun's series adds new type for ICM area.
Link: https://lore.kernel.org/all/cover.1701871118.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
When a reserved QP fails to be created, the memory of the remaining
created reserved QPs is leaked.
Fixes: 70f92521584f ("RDMA/hns: Use the reserved loopback QPs to free MR before destroying MPT")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231207114231.2872104-6-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
IB-core and rdma-core restrict the sgid_index specified by users,
which is uint8_t/u8 data type, to only be within the range of 0-255,
so it's meaningless to support excessively large gid_table_len.
On the other hand, ib-core creates as many sysfs gid files as
gid_table_len, most of which are not only useless because of the
reason above, but also greatly increase the traversal time of
the sysfs gid files for applications.
This patch limits the maximum length of gid table to 256.
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231207114231.2872104-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Now, different devices may have the same interrupt name, which
makes it difficult for users to distinguish between these
interrupts.
Modify the naming style to be consistent with our network devices.
Before:
"hns-aeq-0"
"hns-ceq-0"
...
Now:
"hns-0000:35:00.0-aeq-0"
"hns-0000:35:00.0-ceq-0"
...
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231207114231.2872104-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Support SW stats with debugfs.
Query output:
$ cat /sys/kernel/debug/hns_roce/hns_0/sw_stat/sw_stat
aeqe --- 3341
ceqe --- 0
cmds --- 6764
cmds_err --- 0
posted_mbx --- 3344
polled_mbx --- 3
mbx_event --- 3341
qp_create_err --- 0
qp_modify_err --- 0
cq_create_err --- 0
cq_modify_err --- 0
srq_create_err --- 0
srq_modify_err --- 0
xrcd_alloc_err --- 0
mr_reg_err --- 0
mr_rereg_err --- 0
ah_create_err --- 0
mmap_err --- 0
uctx_alloc_err --- 0
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Link: https://lore.kernel.org/r/20231114123449.1106162-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Add debugfs to hns RoCE. This patch only adds an empty directory
"hns_roce" to debugs root directory.
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231114123449.1106162-3-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
EOPNOTSUPP is more situable than EINVAL for allocating XRCD while XRC
is not supported and unsupported resizing SRQ.
Fixes: 32548870d438 ("RDMA/hns: Add support for XRC on HIP09")
Fixes: 221109e64316 ("RDMA/hns: Add interception for resizing SRQs")
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231114123449.1106162-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
algorithm
Add a default congest control algorithm so that driver won't return
an error when the configured algorithm is invalid.
Fixes: f91696f2f053 ("RDMA/hns: Support congestion control type selection according to the FW")
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231028093242.670325-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Due to hardware limitations, only DCQCN is supported for UD. Therefore, the
default algorithm for UD is set to DCQCN.
Fixes: f91696f2f053 ("RDMA/hns: Support congestion control type selection according to the FW")
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231017125239.164455-6-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
SL set by users may exceed the capability of devices. So add check
for this situation.
Fixes: fba429fcf9a5 ("RDMA/hns: Fix missing fields in address vector")
Fixes: 70f92521584f ("RDMA/hns: Use the reserved loopback QPs to free MR before destroying MPT")
Fixes: f0cb411aad23 ("RDMA/hns: Use new interface to modify QP context")
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231017125239.164455-5-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
The ib_mtu_enum_to_int() and uverbs_attr_get_len() may returns a negative
value. In this case, mixed comparisons of signed and unsigned types will
throw wrong results.
This patch adds judgement for this situation.
Fixes: 30b707886aeb ("RDMA/hns: Support inline data in extented sge space for RC")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231017125239.164455-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
The current driver will print all asynchronous events. Some of the
print levels are set improperly, e.g. SRQ limit reach and SRQ last
wqe reach, which may also occur during normal operation of the software.
Currently, the information of these event is printed as a warning,
which causes a large amount of printing even during normal use of the
application. As a result, the service performance deteriorates.
This patch fixes the printing storms by modifying the print level.
Fixes: b00a92c8f2ca ("RDMA/hns: Move all prints out of irq handle")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20231017125239.164455-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Compared with normal doorbell, using record doorbell can shorten the
process of ringing the doorbell and reduce the latency.
Add a flag HNS_ROCE_CAP_FLAG_SRQ_RECORD_DB to allow FW to
enable/disable SRQ record doorbell.
If the flag above is set, allocate the dma buffer for SRQ record
doorbell and write the buffer address into SRQC during SRQ creation.
For userspace SRQ, add a flag HNS_ROCE_RSP_SRQ_CAP_RECORD_DB to notify
userspace whether the SRQ record doorbell is enabled.
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230926130026.583088-1-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
The SRQ restrack attributes come from the context maintained by ROCEE.
Example:
$ rdma res show srq -jp -dd
[ {
"ifindex": 0,
"ifname": "hns_0",
"srqn": 0,
"type": "BASIC",
"lqpn": [ "14-15","22-23" ],
"pdn": 2,
"pid": 1224,
"comm": "ib_send_bw",{
"drv_srqn": 0,
"drv_wqe_cnt": 512,
"drv_max_gs": 2,
"drv_xrcdn": 0
}
} ]
$ rdma res show srq link hns_0 -jpr
[ {
"ifindex": 0,
"ifname": "hns_0",
"data": [ 149,0,0,0,0,0,0,0,0,0,0,0,119,101,120,99,0,
46,62,31,0,0,0,0,3,0,0,1,0,58,62,31,0,0,0,0,
30,159,15,0,0,0,64,5,0,0,0,0,0,0,0,0,0,0,0,
9,0,0,0,0,0,0,0,0 ]
} ]
Signed-off-by: wenglianfa <wenglianfa@huawei.com>
Link: https://lore.kernel.org/r/20230918131110.3987498-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Support query hns HW stats for rdma-tool to help debugging.
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230816091812.2899366-3-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
This patch fixes inaccurate error label name in init instance.
Fixes: 70f92521584f ("RDMA/hns: Use the reserved loopback QPs to free MR before destroying MPT")
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230804012711.808069-4-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Currently, direct wqe is not supported for wr-list. RoCE driver excludes
direct wqe for wr-list by judging whether the number of wr is 1.
For a wr-list where the second wr is a length-error atomic wr, the
post-send driver handles the first wr and adds 1 to the wr number counter
firstly. While handling the second wr, the driver finds out a length error
and terminates the wr handle process, remaining the counter at 1. This
causes the driver mistakenly judges there is only 1 wr and thus enters
the direct wqe process, carrying the current length-error atomic wqe.
This patch fixes the error by adding a judgement whether the current wr
is a bad wr. If so, use the normal doorbell process but not direct wqe
despite the wr number is 1.
Fixes: 01584a5edcc4 ("RDMA/hns: Add support of direct wqe")
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230804012711.808069-3-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Remove VF extend configuration since the relative registers are
configured in firmware currently.
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230721025146.450831-3-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Support driver get the num of XRCD from firmware.
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://lore.kernel.org/r/20230721025146.450831-2-huangjunxian6@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Linux 6.4
Resolve conflicts between rdma rc and next in rxe_cq matching linux-next:
drivers/infiniband/sw/rxe/rxe_cq.c:
https://lore.kernel.org/r/20230622115246.365d30ad@canb.auug.org.au
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
It is not necessary to check the type of the queue on IO path because
unsupported QP type cannot be created.
Link: https://lore.kernel.org/r/20230523121641.3132102-2-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
Long message loopback slice is used for achieving traffic balance between
QPs. It prevents the problem that QPs with large traffic occupying the
hardware pipeline for a long time and QPs with small traffic cannot be
scheduled.
Currently, its maximum value is set to 16K, which means only after a QP
sends 16K will the second QP be scheduled. This value is too large, which
will lead to unbalanced traffic scheduling, and thus it needs to be
modified.
The setting range of the long message loopback slice is modified to be
from 1024 (the lower limit supported by hardware) to mtu. Actual testing
shows that this value can significantly reduce error in hardware traffic
scheduling.
This solution is compatible with both HIP08 and HIP09. The modified
lp_pktn_ini has a maximum value of 2 (when mtu is 256), so the range
checking code for lp_pktn_ini is no longer necessary and needs to be
deleted.
Fixes: 0e60778efb07 ("RDMA/hns: Modify the value of MAX_LP_MSG_LEN to meet hardware compatibility")
Link: https://lore.kernel.org/r/20230512092245.344442-4-huangjunxian6@hisilicon.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
On HIP08, the queried timeout attr is different from the timeout attr
configured by the user.
It is found by rdma-core testcase test_rdmacm_async_traffic:
======================================================================
FAIL: test_rdmacm_async_traffic (tests.test_rdmacm.CMTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
File "./tests/test_rdmacm.py", line 33, in test_rdmacm_async_traffic
self.two_nodes_rdmacm_traffic(CMAsyncConnection, self.rdmacm_traffic,
File "./tests/base.py", line 382, in two_nodes_rdmacm_traffic
raise(res)
AssertionError
Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/20230512092245.344442-2-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The current resource query for vf caps is driven by the driver,
which is unreasonable.
This patch adds a new command HNS_ROCE_OPC_QUERY_VF_CAPS_NUM
to support obtaining vf caps information from firmware.
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Link: https://lore.kernel.org/r/20230304091555.2241298-2-xuhaoyue1@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
Enable the CQEIE field and configure the CQEIS field of QPC. And add
compatibility handling.
Link: https://lore.kernel.org/r/20221224102201.3114536-4-xuhaoyue1@hisilicon.com
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The rq inline makes some changes as follows, Firstly, it is only used in
user space. Secondly, it should notify hardware in QP RTR status. Thirdly,
Add compatibility processing between different user space and kernel
space.
Link: https://lore.kernel.org/r/20221224102201.3114536-3-xuhaoyue1@hisilicon.com
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The roce driver kernel space will no longer provide support for the rq
inline feature. This patch deletes the code related to the rq inline
feature in the kernel space.
Link: https://lore.kernel.org/r/20221224102201.3114536-2-xuhaoyue1@hisilicon.com
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
For dependencies in following patches
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
XRC caps has been set by default. But in fact, XRC is not supported in
HIP08.
Fixes: 32548870d438 ("RDMA/hns: Add support for XRC on HIP09")
Link: https://lore.kernel.org/r/20221126102911.2921820-7-xuhaoyue1@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The error code is fixed to EIO when CMD fails to excute. This patch
converts the error status reported by firmware to linux errno.
Fixes: a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
Link: https://lore.kernel.org/r/20221126102911.2921820-6-xuhaoyue1@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
Add verification to make sure the roce page size cap is supported by the
system page size.
Fixes: ba6bb7e97421 ("RDMA/hns: Add interfaces to get pf capabilities from firmware")
Link: https://lore.kernel.org/r/20221126102911.2921820-5-xuhaoyue1@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
Now, The address of the first two pages in the MR will be searched, which
use to speed up the lookup of the pbl table for hardware. An exception
will occur when there is only one page in this MR. This patch fix the
number of page to search.
Fixes: 9b2cf76c9f05 ("RDMA/hns: Optimize PBL buffer allocation process")
Link: https://lore.kernel.org/r/20221126102911.2921820-4-xuhaoyue1@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The queried AH attr is invalid. This patch fix it.
This problem is found by rdma-core test test_mr_rereg_pd
ERROR: test_mr_rereg_pd (tests.test_mr.MRTest)
Test that cover rereg MR's PD with this flow:
----------------------------------------------------------------------
Traceback (most recent call last):
File "./tests/test_mr.py", line 157, in test_mr_rereg_pd
self.restate_qps()
File "./tests/test_mr.py", line 113, in restate_qps
self.server.qp.to_rts(self.server_qp_attr)
File "qp.pyx", line 1137, in pyverbs.qp.QP.to_rts
File "qp.pyx", line 1123, in pyverbs.qp.QP.to_rtr
pyverbs.pyverbs_error.PyverbsRDMAError: Failed to modify QP state to RTR.
Errno: 22, Invalid argument
Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/20221126102911.2921820-3-xuhaoyue1@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
After the hns roce driver is loaded, if you modify the mac address of the
network port, the following error will appear:
__ib_cache_gid_add: unable to add gid fe80:0000:0000:0000:4600:4dff:fe22:abb5 error=-28
hns3 0000:7d:00.0 hns_0: attr path_mtu(1) invalid while modify qp
The reason for the error is that the gid being occupied will cause the
failure to modify the gid. The gid is occupied by the loopback QP used by
free mr. When the mac address is modified, the gid will change. If there
is a busy QP at this time, the gid will not be released and the
modification will fail. The QP of free mr is created using the ib
interface. The ib interface will add a reference count to the gid,
resulting in this error scenario.
Considering that free mr is solving a bug in HIP08, not an actual
business, it is not necessary to use ib interfaces.
Fixes: 70f92521584f ("RDMA/hns: Use the reserved loopback QPs to free MR before destroying MPT")
Link: https://lore.kernel.org/r/20221126102911.2921820-2-xuhaoyue1@hisilicon.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
The user usually configures the number of sge through the max_send_sge
parameter when creating qp, and configures the maximum size of inline data
that can be sent through max_inline_data. Inline uses sge to fill data to
send. Expect the following:
1) When the sge space cannot hold inline data, the sge space needs to be
expanded to accommodate all inline data
2) When the sge space is enough to accommodate inline data, the upper
limit of inline data can be increased so that users can send larger
inline data
Currently case one is not implemented. When the inline data is larger than
the sge space, an error of insufficient sge space occurs. This part of
the code needs to be reimplemented according to the expected rules. The
calculation method of sge num is modified to take the maximum value of
max_send_sge and the sge for max_inline_data to solve this problem.
Fixes: 05201e01be93 ("RDMA/hns: Refactor process of setting extended sge")
Fixes: 30b707886aeb ("RDMA/hns: Support inline data in extented sge space for RC")
Link: https://lore.kernel.org/r/20221108133847.2304539-3-xuhaoyue1@hisilicon.com
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
In the HNS ROCE driver, The sge is divided into standard sge and extended
sge. There are 2 standard sge in RC/XRC, and the UD standard sge is 0.
In the scenario of RC SQ inline, if the data does not exceed 32bytes, the
standard sge will be used. If it exceeds, only the extended sge will be
used to fill the data.
Currently, when filling the extended sge, max_gs is directly used as the
number of the extended sge, which did not subtract the number of standard
sge. There is a logical error. The new algorithm subtracts the number of
standard sge from max_gs to get the actual number of extended sge.
Fixes: 30b707886aeb ("RDMA/hns: Support inline data in extented sge space for RC")
Link: https://lore.kernel.org/r/20221108133847.2304539-2-xuhaoyue1@hisilicon.com
Signed-off-by: Luoyouming <luoyouming@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
|
|
Lock grab occurs in a concurrent scenario, resulting in stepping on a NULL
pointer. It should be init mutex_init() first before use the lock.
Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
Call trace:
__mutex_lock.constprop.0+0xd0/0x5c0
__mutex_lock_slowpath+0x1c/0x2c
mutex_lock+0x44/0x50
free_mr_send_cmd_to_hw+0x7c/0x1c0 [hns_roce_hw_v2]
hns_roce_v2_dereg_mr+0x30/0x40 [hns_roce_hw_v2]
hns_roce_dereg_mr+0x4c/0x130 [hns_roce_hw_v2]
ib_dereg_mr_user+0x54/0x124
uverbs_free_mr+0x24/0x30
destroy_hw_idr_uobject+0x38/0x74
uverbs_destroy_uobject+0x48/0x1c4
uobj_destroy+0x74/0xcc
ib_uverbs_cmd_verbs+0x368/0xbb0
ib_uverbs_ioctl+0xec/0x1a4
__arm64_sys_ioctl+0xb4/0x100
invoke_syscall+0x50/0x120
el0_svc_common.constprop.0+0x58/0x190
do_el0_svc+0x30/0x90
el0_svc+0x2c/0xb4
el0t_64_sync_handler+0x1a4/0x1b0
el0t_64_sync+0x19c/0x1a0
Fixes: 70f92521584f ("RDMA/hns: Use the reserved loopback QPs to free MR before destroying MPT")
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Link: https://lore.kernel.org/r/20221024083814.1089722-3-xuhaoyue1@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|
|
When function reset and local invalidate are mixed, HNS RoCEE may hang.
Before introducing the cause of the problem, two hardware internal
concepts need to be introduced:
1. Execution queue: The queue of hardware execution instructions,
function reset and local invalidate are queued for execution in this
queue.
2.Local queue: A queue that stores local operation instructions. The
instructions in the local queue will be sent to the execution queue
for execution. The instructions in the local queue will not be removed
until the execution is completed.
The reason for the problem is as follows:
1. There is a function reset instruction in the execution queue, which
is currently being executed. A necessary condition for the successful
execution of function reset is: the hardware pipeline needs to empty
the instructions that were not completed before;
2. A local invalidate instruction at the head of the local queue is
sent to the execution queue. Now there are two instructions in the
execution queue, the first is the function reset instruction, and the
second is the local invalidate instruction, which will be executed in
se quence;
3. The user has issued many local invalidate operations, causing the
local queue to be filled up.
4. The user still has a new local operation command and is queuing to
enter the local queue. But the local queue is full and cannot receive
new instructions, this instruction is temporarily stored at the
hardware pipeline.
5. The function reset has been waiting for the instruction before the
hardware pipeline stage is drained. The hardware pipeline stage also
caches a local invalidate instruction, so the function reset cannot be
completed, and the instructions after it cannot be executed.
These factors together cause the execution logic deadlock of the hardware,
and the consequence is that RoCEE will not have any response. Considering
that the local operation command may potentially cause RoCEE to hang, this
feature is no longer supported.
Fixes: e93df0108579 ("RDMA/hns: Support local invalidate for hip08 in kernel space")
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Haoyue Xu <xuhaoyue1@hisilicon.com>
Link: https://lore.kernel.org/r/20221024083814.1089722-2-xuhaoyue1@hisilicon.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
|