Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 27 4月, 2023 19 次提交
    • J
      RDMA/hns: Support dispatching IB event for RoCE bonding · 49f27d3e
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZAIM

---------------------------------------------------------------

Support dispatching IB event for RoCE bonding. After setting bond,
IB_EVENT_PORT_ERR is dispatched in the following situation:
	1. bond0 becomes link down;
	2. all slaves become link down (as it will lead to a link-down
           of the upper device).

IB_EVENT_PORT_ACTIVE is dispatched in the following situation:
	1. bond0 becomes link up;
	2. one slave becomes link up when all slaves were link down (as
           it will lead to a link-up of the upper device).
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      49f27d3e
    • J
      RDMA/hns: Set IB port state depending on upper device for RoCE bonding · c65db67a
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZAIM

---------------------------------------------------------------

For RoCE bonding, the IB port state should depend on the link
status of upper device. When the upper device is link up, the
IB port state should be IB_PORT_ATIVE; otherwise, the state
should be IB_PORT_DOWN.

Particularly, when all slaves are link down, the upper device will become
link down automatically, and when at least one slave become link up, the
upper device will become link up. In these situations the IB port will
also change accordingly.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      c65db67a
    • J
      RDMA/hns: Get real-time port state of bonding slave · e4ad37ea
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZAIM

---------------------------------------------------------------

Currently, when a change-lowerstate event occurs, the bonding driver
gets the slave's port state from the event notifier and stores it in
the bond_grp struct. Whenever the state needs to be configured to HW
or is queried, the driver reads from bond_grp.

But actually, the state read in this way is not real-time data, as there
is always a delay before the newly changed state is stored. So it is
uneccessary and also inaccurate to store the state. Whenever the state
is needed, just read it from kernel in real time.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      e4ad37ea
    • J
      RDMA/hns: Fix the counting error of bonding with more than 2 slaves · ab97048c
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZACP

---------------------------------------------------------------

When changeupper event happens involving bond group with more than 2
slaves, the counting of slave num is incorrect for the reason that there
are no corresponding hr_dev for uninited slaves, which leads to the
failure in adding 3 or more slaves to a bond group.

Applying this patch, the counting begins from the previous value but not
0, and updates slave_map and slave_map_diff depending on whether slave
num increases or decreases. Besides, since bond_grp->slave_num is no
longer used, remove it.

Fixes: e62a2027 ("RDMA/hns: support RoCE bonding")
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      ab97048c
    • J
      RDMA/hns: Fix the repetitive workqueue mission in RoCE Bonding · 483ccd44
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZACP

---------------------------------------------------------------

When the third slave is going to be added to a bond group, a
ChangeLowerState event happens firstly and notifies the driver
to queue a ChangeState mission. However, the hr_dev of the third
slave and bond group will both be notified and queue a repetitive
mission separately. As the mission is triggered twice, there will
be also repetitive mission printing, which is misleading and
incorrect.

Applying this patch, only the hr_dev of the bond group will queue and
trigger the mission.

Fixes: e62a2027 ("RDMA/hns: support RoCE bonding")
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      483ccd44
    • J
      RDMA/hns: Rename hns_roce_bond_info_record() to make sense · b5db3025
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ZACP

---------------------------------------------------------------

The "record" in "hns_roce_bond_info_record()" is not accurate.
Change the name to "hns_roce_bond_info_update()".
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      b5db3025
    • J
      RDMA/hns: Support reset recovery for RoCE bonding · b0f80ad2
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

Currently, a RoCE bond device cannot be recovered to a bond
device after reset.

Applying this patch, RoCE bonding device 'hns_bond_xx' can be recovered
after reset, including the following changes:
   1. modify the condition for juding whether bond_grp is active, as the
      bond_grp may be also holding HNS_ROCE_CHANGE_BOND during reset init.
      Thus, as long as the bond_grp's state is not
      HNS_ROCE_BOND_NOT_BONDED, it should be considered active.
   2. update the link status of slave in bond_grp from NIC bonding driver
      right before sending command to firmware, as RoCE driver is uninited
      for a while in reset process, and during this period bond_grp cannot
      update the information.
   3. After the reset, re-config the bond_grp information to firmware,
      as the firmware is also reset and the previous configuration is
      cleared.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      b0f80ad2
    • J
      RDMA/hns: Fix the counting error of slave number · 01c810c8
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

Currently, the driver will count the number of slave hr_dev,
but during bond clearing event and slave decreasing event, only
one slave has a corresponding hr_dev, and the number will always
be 1. It causes that all slave decrease events are redirected to
a bond clearing operation.

This patch fixes this problem by counting slave netdev directly instead
of getting its hr_dev during bond clearing event and slave decreasing
event.

Fixes: 646b97db ("RDMA/hns: adjust the structure of RoCE bonding driver")
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      01c810c8
    • J
      RDMA/hns: Fix the driver uninit order during bond setting · 8d3ee0ae
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

In the slave list of bond_grp, the first slave is always the main_hr_dev,
and the main_hr_dev is responsible for bonding netdev event handling. If
the main_hr_dev is uninit too early, some CHANGELOWERSTATE events may be
missed before the main_hr_dev is re-register, which may cause the HW is
configured with an outdated slave state.

This patch reverses the order of uniniting slave hr_dev to ensure
main_hr_dev is the last one being uninited.

Fixes: e62a2027 ("RDMA/hns: support RoCE bonding")
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      8d3ee0ae
    • J
      RDMA/hns: Simplify the slave uninit logic of RoCE bonding operations · bc80b728
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

Currently, bond group resource is always associated to a main_hr_dev
whether bond device has been set or not. When setting bond, all slaves,
including the main_hr_dev, will be uninited. Normally, during main_hr_dev
uninit, the bond group resource is also released. However, the bond group
resource should not be released in the case of setting bond.

To avoid releasing bond group resource when setting bond, the code logic
is like:
    1. uninit a not-main_hr_dev slave
    2. register bond device, with the slave in step 1 becoming the new
       main_hr_dev
    3. uninit the original main_hr_dev.
In step 3, the slave is not current main_hr_dev any more, so during its
uninit, bond group resource will not be released. But in general, the
logic is not simple enough to understand.

This patch adds an argument 'bond_cleanup' to
__hns_roce_hw_v2_uninit_instance() in order to indicate whether bond
group resources should be released during RoCE driver uninit. During
driver uninit when setting bond or clearing bond, 'bond_cleanup' is false;
when RoCE driver kernel object is being removed , 'bond_cleanup' is true.

With this change and the previous patch, bonding operations including
bond setting, bond clearing and decreasing slave do not have to concern
about avoiding releasing bond_grp when uniniting hr_dev. The uninit
order is simplified to a more intuitive one: simply uninit all slaves and
then register the bond device.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      bc80b728
    • J
      RDMA/hns: Remove the struct member 'bond_grp' from hns_roce_dev · 9003ac2a
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

Currently, bond_grp is attached to only one hr_dev corresponding to a
slave in RoCE bonding, which is called main_hr_dev in the driver. When
a non-main_hr_dev try to obtain its bond_grp, the driver has to find
the main_hr_dev, and then obtain the bond_grp, which leads to a
complicated code.

Applying this patch, bond_grp is removed from struct hns_roce_dev. hr_dev
can obtain bond_grp by XArray, where die_info and bond_grp are stored
according to bus number. With this change, hr_dev can get its bond_grp
directly without depending on main_hr_dev, and the code logic can be
simplified.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      9003ac2a
    • J
      RDMA/hns: Initial value assignment cleanup for RoCE Bonding variables · 96064446
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

This patch assigns initial value when variable is defined in
HNS RoCE Bonding driver, instead of doing so on a new line.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      96064446
    • J
      RDMA/hns: Delete a useless assignment to bond_state · 41adb38e
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

In hns_roce_slave_dec(), bond_state will be changed to
HNS_ROCE_BOND_REGISTERING right before the current main_hr_dev
is being removed from bond group. When the slave decrease
operation is over, bond_state will be changed to
HNS_ROCE_BOND_IS_BONDED in the end of this function. So the
assignment to bond_state in the beginning of the function is
useless and should be deleted.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      41adb38e
    • J
      RDMA/hns: Apply XArray for Bond ID allocation · 82ee5d30
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

This patch provides the ability to map an integer ID to a bond
group pointer by:
	1. adding a new struct hns_roce_die_info to store the
	   pointers and IDs of bond groups on a specific I/O die.
	2. applying XArray to map the bus number to the die info struct.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      82ee5d30
    • J
      RDMA/hns: Move bond_work from hns_roce_dev to hns_roce_bond_group · 8aeaa671
      由 Junxian Huang 提交于 
      driver inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I6Z4E9

---------------------------------------------------------------

Currently, bond_work, the struct of delayed work for RoCE bonding,
is attached to hns_roce_dev. During setting bond, hns_roce_dev
will be uninited and the pending works will be canceled.

This patch moves bond_work from hns_roce_dev to hns_roce_bond_group so
that the pending works can be executed after setting bond rather than
being cancelled.
Signed-off-by: NJunxian Huang <huangjunxian6@hisilicon.com>
      8aeaa671
    • O
      !381 net: hns3: add support handling tx dhcp packets for ROH · ab866dc7
      由 openeuler-ci-bot 提交于 
      Merge Pull Request from: @chenke1978 
 
[Description]
For ROH distributed scenario, EID is allocated by DHCP mode.
Driver needs to convert the origin MAC address to EID format,
and updates the destination MAC, chaddr and client id(if exists)
when transmit DHCP packets. Meantime, the chaddr field should
follow the source mac address, in order to make the dhcp
server reply to the right client. For the payload of
dhcp packet changed, so the checksum of L4 should be
calculated too.

[Testing]
kernel options:
CONFIG_ROH=m
CONFIG_ROH_HNS=m

Test passed with below step:
1. Load the NIC/ROCE/ROH driver normally on the ROH device.
2. Preparing the DHCP Server and DHCP Client applications
3. Enable the DHCP service on the server node.
4. Execute the DHCP client application on the client node.
5. Check the DHCP process and wait until the DHCP IP address
   allocation is complete.
6. The communication is normal based on the new IP address.

    
 
Link:https://gitee.com/openeuler/kernel/pulls/381 

Signed-off-by: Jialin Zhang <zhangjialin11@huawei.com>
      ab866dc7
    • K
      net: hns3: add support handling tx dhcp packets for ROH · 19f053b7
      由 Ke Chen 提交于 
      driver inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I6BSMN

-----------------------------------------------------------------------

For ROH distributed scenario, EID is allocated by DHCP mode.
Driver needs to convert the origin MAC address to EID format,
and updates the destination MAC, chaddr and client id(if exists)
when transmit DHCP packets. Meantime, the chaddr field should
follow the source mac address, in order to make the dhcp
server reply to the right client. For the payload of
dhcp packet changed, so the checksum of L4 should be
calculated too.
Signed-off-by: NJian Shen <shenjian15@huawei.com>
Signed-off-by: NKe Chen <chenke54@huawei.com>
      19f053b7
    • O
      !617 Support geting xrcd num from firmware · 02362c9c
      由 openeuler-ci-bot 提交于 
      Merge Pull Request from: @stinft 
 
Support driver gets the num_xrcds and reserved_xrcds from firmware.
#I6WAZI 
 
Link:https://gitee.com/openeuler/kernel/pulls/617 

Reviewed-by: Chengchang Tang <tangchengchang@huawei.com> 
Signed-off-by: Jialin Zhang <zhangjialin11@huawei.com>
      02362c9c
    • L
      RDMA/hns: Support getting xrcd num from firmware · bbfeb5d8
      由 Luoyouming 提交于 
      driver inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I6WAZI

---------------------------------------------------------------

Support driver gets the num_xrcds and reserved_xrcds from firmware.
Signed-off-by: NLuoyouming <luoyouming@huawei.com>
Signed-off-by: NChengchang Tang <tangchengchang@huawei.com>
      bbfeb5d8
  3. 26 4月, 2023 21 次提交