- 03 Oct, 2016 36 commits
-
-
Salil authored
This patch removes the redundant code lines present in the functions get_send_wqe() and get_recv_wqe(). This also fixes the error in calculating the SQ WQE. Signed-off-by:
Lijun Ou <oulijun@huawei.com> Reviewed-by:
Wei Hu (Xavier) <xavier.huwei@huawei.com> Signed-off-by:
Salil Mehta <salil.mehta@huawei.com> Signed-off-by:
Doug Ledford <dledford@redhat.com>
-
Lijun Ou authored
It doesn't need to assign for the filed of qp state in qpc separately when qp happen to migrate state which supported in RoCE engine v1. Signed-off-by:
-
Lijun Ou authored
This patch mainly fixes the bug with platform_get_resource(). It should return NULL when platform_get_resource() exec fail. Signed-off-by:
-
Lijun Ou authored
The rq head in qpc was zero will miss the rq wqes which have be sent, so here we should take the real value. Signed-off-by:
-
Lijun Ou authored
Cq has not been freed when fail to ib_copy_to_udata, so need to free it. Signed-off-by:
Peter Chen <luck.chen@huawei.com> Reviewed-by:
-
Lijun Ou authored
The mtu should be validated when modify qp,so we check it. Signed-off-by:
-
Lijun Ou authored
Some items of qpc need to take user param when modified qp state. Signed-off-by:
Dongdong Huang(Donald) <hdd.huang@huawei.com> Reviewed-by:
-
Lijun Ou authored
The Ack timeout of qpc need a lower limit value,otherwise the read performance will be very lower. Signed-off-by:
-
Lijun Ou authored
While post failed, hns roce should return the wr failed to user. We omitted this while qp type is wrong and fixed it. Signed-off-by:
-
Lijun Ou authored
While the page size attribute of umem is illegal, we should release umem that get by ib_umem_get interface. Also, we should return a non-zero value while pbl number is wrong. Signed-off-by:
-
Lijun Ou authored
This lock will be used in query port interface, and will be called while IB device was registered to OFED framework/IB Core. So, the lock of iboe must be initiated before IB device was registered. Signed-off-by:
-
Wei Hu (Xavier) authored
This patch optimized the codes of aeq and ceq interrupt handle and fixed the bug in the calculation of qpn. For the special qp(GSI or SMI), calculated the qp number according to physical port and the qpn reported in the event of async event queue. Signed-off-by:
-
Wei Hu (Xavier) authored
This patch deleted the sqp_start from the structure hns_roce_caps, and modified the calculation of the qp number. Signed-off-by:
-
Wei Hu (Xavier) authored
In hip06, there's no interface to release hem memory. So, hardware can't identify whether hem memory released or not. If all context in a hem memory released, the related hem memory will be released by driver and reused by others. But, hardware don't know that this memory can't be used already. In order to fix this bug, hns roce driver reserved 128K memory for each type of hem(QPC/CQC/MTPT). While unmap hem memory, hns roce driver will write base address of reserved memory according to hem type. Signed-off-by:
-
Lijun Ou authored
This patch removes the qp_type parameter in hns_roce_set_kernel_sq_size(). Signed-off-by:
Ping Zhang <zhangping5@huawei.com> Reviewed-by:
-
Lijun Ou authored
Hns_roce_pd_alloc and hns_roce_reserve_range_qp use unnecessary transformation of parameters. This patch simplify these two functions. Signed-off-by:
-
Lijun Ou authored
In current version, it uses uninitialized parameters named refcount and free in hns_roce_cq_event. This patch initializes these parameter in cq alloc and add correspond process in cq free. Signed-off-by:
-
Lijun Ou authored
In old version of RoCE, it doesn't support to resize cq. So, we remove parameters related to resize cq. Signed-off-by:
-
Lijun Ou authored
The parameter named collapsed unused in hns_roce_cq_alloc. Also, parameter named doorbell_lock unsed in hns_roce_v1_cq_set_ci. This patch optimize these parameters. Signed-off-by:
-
Lijun Ou authored
This patch adds node_guid definition in bindings document. The value of node_guid will be used during RDMA connection. Signed-off-by:
-
Lijun Ou authored
This patch mainly modify the value of HNS_ROCE_SL_SHIFT and delete the lines for assigning for the field of local_enable_e2e_credit in QP1C. Signed-off-by:
-
Lijun Ou authored
Fix bug of modify qp from init to init on user mode. Otherwise, it will oops when rmda cm established. Signed-off-by:
-
Lijun Ou authored
This patch mainly modifies the logic for allocating uar registers. In HiP06 SoC, HW has 8 group of uar registers for kernel and user space application. The uar index is assigned as follows: 0 ------ for kernel 1~7 ------ for user space application Signed-off-by: -
Lijun Ou authored
This patch mainly adds phy_port to HNS RoCE QP. This shall be used in calculating the GSI QPN for the port. Initally when RDMA is being established, all IB ports share a QPN which later needs to be re-assigned to a particular GSI/QPN and which is per-port. This also fixes a bug in base driver where iboe port was being used instead of phy_port at some places. This values might not be same always. Signed-off-by:
-
Lijun Ou authored
Fix the length of wqe that maybe lead to an error and write the end bytes of QP1C into the register. Signed-off-by:
-
Lijun Ou authored
In the latest IB core version, it has some known issues with memory registration using the local_dma_lkey. Thus RoCE don't expose support for it, and remove device->local_dma_lkey which is introduced to working systems. Signed-off-by:
-
Lijun Ou authored
According to the Infiniband spec, NodeGUID uniquely identifies a node. This must be initialized to some unique value. This patch adds the support to the HNS RoCE driver to fetch the NodeGUID value from DT or ACPI and then use this value to initialize the node_guid parameter of IB device. This value shall be used by RDMA CM. Signed-off-by:
-
Lijun Ou authored
This patch adds get_netdev() function to the IB device. This shall be used to fetch netdev corresponding to the port number. This function would be called by IB core(Generic CM Agent) for example, when the RDMA connection is being established. Signed-off-by:
-
Daode Huang authored
The broadcast packets is filtered in the hardware now, so this process is no need in the driver, just delete it. Signed-off-by:
Daode Huang <huangdaode@hisilicon.com> Reviewed-by:
Yisen Zhuang <yisen.zhuang@huawei.com> Signed-off-by:
-
Daode Huang authored
When the dsaf mode receives a broadcast packet, it will filter the packet by comparing the received queue number and destination queue number(get from forwarding table), if they are the same, the packet will be filtered. Otherwise, the packet will be loopback. So this patch select queue 0 to send broadcast and multicast packets. Signed-off-by:
-
Daode Huang authored
As the sub queue id in the broadcast forwarding table is always set to absolute queue 0 rather than the interface's relative queue 0, this will cause the received broadcast packets loopback to rcb. This patch sets the sub queue id to relative queue 0 of each port. Signed-off-by:
-
Kejian Yan authored
After running command "ethtool -t eth0", eth0 can not be connected to network. It is caused by the changing the inner loopback register and this register cannot be changed when hns connected to network. The routine of setting this register needs to be removed and using promisc mode to let the packet looped back pass by dsaf mode. Reported-by:
Jun He <hjat2005@huawei.com> Signed-off-by:
Kejian Yan <yankejian@huawei.com> Reviewed-by:
-
Daode Huang authored
Because we handle the received packets after napi, so delete the checking before submitting. It delete the code of read the fbd number register, which reduces the cpu usages while receiving packets Signed-off-by:
-
Daode Huang authored
This patch adds fini_process for v2, it handles the packets recevied by the hardware in the napi porcess. With this patch, the hardware irq numbers will drop 50% per sec. Signed-off-by:
-
Daode Huang authored
When set rx/tx coalesce usecs to 0, the interrupt coalesce will be disabled, but there is a interrupt rate limit which set to 1us, it will cause no interrupt occurs. This patch disable interrupt limit when sets coalsecs usecs to 0, and restores it to 1 in other case. Signed-off-by:
-
lipeng authored
When hnae_reserve_buffer_map fail, it will break cycle and some buffer description has no available memory, therefore the port will be unavailable. Signed-off-by:
Peng Li <lipeng321@huawei.com> Reviewed-by:
-
- 25 Aug, 2016 2 commits
-
-
Salil authored
This patch is meant to add support of ACPI to the Hisilicon RoCE driver. Changes done are primarily meant to detect the type and then either use DT specific or ACPI spcific functions. Where ever possible, this patch tries to make use of Unified Device Property Interface APIs to support both DT and ACPI through single interface. This patch depends upon HNS ethernet driver to Reset RoCE. This function within HNS ethernet driver has also been enhanced to support ACPI and is part of other accompanying patch with this patch-set. NOTE: The changes in this patch are done over below branch, https://github.com/dledford/linux/tree/hns-roceSigned-off-by:
-
Salil authored
In the Hip06 SoC, the RoCE Engine is part of the Hisilicon Network Subsystem and is dependent upon DSAF module. Therefore, certain functions like RESET are exposed through the common registers of HNS DSAF module which are memory-mapped by the HNS driver and currently can only be accessed through DT/syscon interface. This patch adds the support of ACPI to the existing RoCE reset function in the HNS driver(please refer NOTE 2). Hisilicon RoCE driver (please refer NOTE 1) shall call this reset function during probe time to reset the RoCE Engine. The HNS Reset function indirectly ends up in calling the _DSM() function part of the DSDT ACPI Table. Actual reset functionality for ACPI is implemented within the ACPI DSDT Table which also has been enhanced to support this change. Support of ACPI in the HNS RoCE driver shall be pushed through a different accompanying below patch: "IB/hns: Add support of ACPI to the Hisilicon RoCE Driver" NOTE 1: HNS RoCE driver has already been accepted by its maintainer Doug Ledford<dledford@redhat.com>. Please refer below link: https://www.spinics.net/lists/linux-rdma/msg38850.html NOTE 2: RoCE reset function patch has been accepted and now is part of the net-next: https://www.mail-archive.com/netdev@vger.kernel.org/msg123867.htmlSigned-off-by:
-
- 22 Aug, 2016 2 commits
-
-
Wei Yongjun authored
In case of error, the function devm_ioremap_resource() returns ERR_PTR() and never returns NULL. The NULL test in the return value check should be replaced with IS_ERR(). Signed-off-by:
Wei Yongjun <weiyj.lk@gmail.com> Signed-off-by:
-
oulijun authored
This patch added maintainers for RoCE driver. Signed-off-by:
-
