On Fri, Jan 19, 2018 at 7:21 AM, Majd Dibbiny <majd@xxxxxxxxxxxx> wrote: > >> On Jan 19, 2018, at 1:09 PM, Leon Romanovsky <leon@xxxxxxxxxx> wrote: >> >>> On Thu, Jan 18, 2018 at 11:13:08AM -0500, Olga Kornievskaia wrote: >>>> On Wed, Jan 17, 2018 at 5:03 PM, Olga Kornievskaia <aglo@xxxxxxxxx> wrote: >>>>> On Wed, Jan 17, 2018 at 4:03 PM, Doug Ledford <dledford@xxxxxxxxxx> wrote: >>>>>> On Tue, 2018-01-16 at 16:14 -0500, Olga Kornievskaia wrote: >>>>>>> On Tue, Jan 16, 2018 at 2:50 PM, Olga Kornievskaia <aglo@xxxxxxxxx> wrote: >>>>>>> On Fri, Jan 12, 2018 at 7:07 PM, Steve Wise <swise@xxxxxxxxxxxxxxxxxxxxx> wrote: >>>>>>>>>> Ok. The memory probably doesn't matter. Maybe run krping client and >>>>>>>>> >>>>>>>>> server on the same host (to use hw-loopback), and see if it works on both, >>>>>>>>> one, or neither systems when they are both the client and server. >>>>>>>>> >>>>>>>>> Loopback on the original "server" machine produces the same failure. >>>>>>>>> Jan 12 17:05:40 localhost kernel: mlx5_0:dump_cqe:277:(pid 0): dump error >>>>>>>>> cqe >>>>>>>>> Jan 12 17:05:40 localhost kernel: 00000000 00000000 00000000 00000000 >>>>>>>>> Jan 12 17:05:40 localhost kernel: 00000000 00000000 00000000 00000000 >>>>>>>>> Jan 12 17:05:40 localhost kernel: 00000000 00000000 00000000 00000000 >>>>>>>>> Jan 12 17:05:40 localhost kernel: 00000000 93003204 1000017c 0005e1d2 >>>>>>>>> Jan 12 17:05:40 localhost kernel: krping: cq completion failed with >>>>>>>>> wr_id 0 status 4 opcode 0 vender_err 32 >>>>>>>> >>>>>>>> Can someone from Mellanox comment more on the above CQE error? What exactly is it tell us? >>>>>>>> >>>>>>>>> >>>>>>>>> What does this means? >>>>>>>> >>>>>>>> Not sure. But it does seem to be tied to that specific machine. Question: Is an IOMMU enabled on that system? >>>>>>> >>>>>>> IOMMU (Inter's VT-d) is enabled in BIOS (on both machines). >>>>>>> >>>>>>>> Perhaps that is exposing a dma mapping problem with krping? >>>>>> >>>>>> I have replaces the CX-5 card with another one and I no longer see the >>>>>> krping problem. I think it speaks that it's a card issue... >>>>> >>>>> Check the firmware on the bad card. Lots of issues disappear if you >>>>> have older firmware and update to the latest. >>>> >>>> That's a valid point. A check of firmware versions is needed. At the >>>> time of the problem, I believe I had two machines that each had same >>>> firmware versions. After card replacement, the replacement card >>>> displays newer firmware. >>> >>> I have upgraded the firmware on both machines involved to the latest >>> available firmware for the card and now I'm in the situation where >>> krping does not work on either machine --- when either of them is a >>> server it fails with the same information in the var log messages: >> >> Doesn't it mean that the issue in FW? > Did you do cold reboot after FW upgrade? No I have done so. Firmware update instruction were to either mlxfwreset or reboot (which i assumed would be warm). I will try a cold reboot. >> >>> >>> Jan 18 11:05:54 localhost kernel: mlx5_0:dump_cqe:277:(pid 0): dump error cqe >>> Jan 18 11:05:54 localhost kernel: 00000000 00000000 00000000 00000000 >>> Jan 18 11:05:54 localhost kernel: 00000000 00000000 00000000 00000000 >>> Jan 18 11:05:54 localhost kernel: 00000000 00000000 00000000 00000000 >>> Jan 18 11:05:54 localhost kernel: 00000000 93003204 10000122 0005bfd2 >>> Jan 18 11:05:54 localhost kernel: krping: cq completion failed with >>> wr_id 0 status 4 opcode 128 vender_err 32 >>> Jan 18 11:05:54 localhost kernel: krping: cq completion in ERROR state >>> Jan 18 11:05:54 localhost kernel: krping: wait for RDMA_READ_COMPLETE state 10 >>> >>> client side logs: >>> Jan 18 11:14:30 localhost kernel: krping: DISCONNECT EVENT... >>> Jan 18 11:14:30 localhost kernel: krping: wait for RDMA_WRITE_ADV state 10 >>> Jan 18 11:14:30 localhost kernel: krping: cq completion in ERROR state -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
