[SRU][J/N/R][PATCH 0/1] HMC error triggers cascading PF resets and tears down RDMA connections

[Seyeong Kim]

BugLink: https://bugs.launchpad.net/bugs/2167995

[Impact]

A single HMC error triggers cascading PF resets and tears down RDMA connections on the device:
kernel: ice 0000:17:00.1 irdma1: HMC Error
kernel: ice 0000:17:00.1 irdma1: Requesting a reset
kernel: DMAR: DRHD: handling fault status reg 2
kernel: DMAR: [DMA Read NO_PASID] Request device [17:00.1] fault addr 0xfddce000 [fault reason 0x06] PTE Read access is not set
kernel: bond1: (slave enp23s0f1np1): link status definitely down, disabling slave
kernel: bond1: active interface up!
kernel: ice 0000:17:00.1: PTP reset successful
ice 0000:17:00.1: VSI rebuilt. VSI index 0, type ICE_VSI_PF
ice 0000:17:00.1: VSI rebuilt. VSI index 1, type ICE_VSI_CTRL
bond1: (slave enp23s0f1np1): link status definitely up, 25000 Mbps full duplex

This causes a short network outage on intel ice NIC, but not i40e which already has a fix.
upstream commit:
https://github.com/torvalds/linux/commit/4dc9c884c0aafac1e8d37536d88d485135e26883
applies the same fix to ice.

[Fix]

Backport upstream commit 4dc9c884c0aa ("RDMA/irdma: Suppress PF reset on HMC error")
https://github.com/torvalds/linux/commit/4dc9c884c0aafac1e8d37536d88d485135e26883

[Test Plan]

Make sure machine with intel ice works fine by running basic network traffic test
and then reproduce the issue in a lab environment.
Confirm the issue can not be reproduced with the above commit.

[Where problems could occur]

The same fix has been applied to intel i40e, the commit does the same thing.
If something goes wrong in the path, and PF is not reset, it may hit the same network outage when using intel ice

Seyeong Kim (1):
  RDMA/irdma: Suppress PF reset on HMC error

 drivers/infiniband/hw/irdma/i40iw_hw.c  | 4 +++-
 drivers/infiniband/hw/irdma/icrdma_hw.c | 2 ++
 drivers/infiniband/hw/irdma/icrdma_hw.h | 2 ++
 drivers/infiniband/hw/irdma/irdma.h     | 2 ++
 drivers/infiniband/hw/irdma/main.c      | 8 ++++++--
 5 files changed, 15 insertions(+), 3 deletions(-)

-- 
2.43.0