davejiang / linux Goto Github PK

Hello, I'm using ntb_transport and ntb_netdev drivers to enable communication between two hosts using Switchtec PFX PCIe Switch.

After assigning IP addresses to the virtual eth interfaces of both hosts and successfully pinging each other I tried testing the connection using iperf3. However, when using TCP as a transport protocol, only certain random number of packets gets sent after which the connection breaks. Wireshark states that a few spurious retransmission happens in those moments followed by Dup ACKs from the destination. After that ARP from the destination keeps asking where the source is and there are some keep-alive messages sent from the source address to destination address, it is a mess really. I am attaching a screenshot of the Wireshark log.

Has this kind of behavior been observed before? Note that this does not happen with UDP as a transport protocol. Yes, there are some packet drops here and there but the connection is pretty much stable in that case.

Thank you in advance.

Hi davejiang,
I try to test the virtual Ethernet over NTB function in our system by iPerf.
But when I run "modporbe ntb_transport dyndbg=+p" command, dmesg will shows "Unable to set mw1 translation" error.
This error will cause ntb link disable, so I can't ping to each other node in our system.
I trace your code and found the mw->dma_addr will get 0x176700000 after running dma_alloc_coherent() at line 686 of ntb_transport.c.
But the input parameter “addr” of intel_ntb_mw_set_trans() is 0x176700000, so it will return –EINVAL error when runs to line 904 of ntb_hw_intel.c.
Do you know how to solve this issue to let me use iPerf to check the function?
Thank you.

We have a Red Hat 7.7 system where the IOATDMA driver is not quite built for this particular IOAT DMA engine. We managed to get it to build and run, but we have an issue where transfers are not completing and the COMPLETION TIMEOUT is being invoked. The transfer is happening, but the DMA engine doesn't think so.

We're suspicious of a caching issue, but I haven't found the magic twiddle to get past it. If our data buffer is >8MB, then the transfer completes successfully with good transfer speeds.

We discovered your page (actually was here once chasing down NTB/IOAT stuff), and decided to grab a late-breaking snapshot. However, it isn't working either, and it's not working in a different way. You had made a change with regard to completion in this class of DMA engine. It seems like the point where the other code is not getting a completion, your code is complaining about errors but not errors...

This is the device in question:

03:00.0 System peripheral: Intel(R) Xeon Processor D Family QuickData Technology Register DMA Channel 0
03:00.0 0880: 8086:6f50

And once a channel is allocated and a DMA_MEMCPY transfer is submitted, we get this spew on the console:

[ 6138.674756] ioatdma 0000:03:00.0: Errors:
[ 6138.778480] ioatdma 0000:03:00.0: CHANSTS: 0x80701d941 CHANERR: 0x0
[ 6138.784761] ioatdma 0000:03:00.0: Errors:
[ 6138.888486] ioatdma 0000:03:00.0: CHANSTS: 0x80701d981 CHANERR: 0x0
[ 6138.894767] ioatdma 0000:03:00.0: Errors:
[ 6138.998491] ioatdma 0000:03:00.0: CHANSTS: 0x80701d9c1 CHANERR: 0x0
[ 6139.004772] ioatdma 0000:03:00.0: Errors:
[ 6139.108496] ioatdma 0000:03:00.0: CHANSTS: 0x80701da01 CHANERR: 0x0
[ 6139.114778] ioatdma 0000:03:00.0: Errors:
[ 6139.218501] ioatdma 0000:03:00.0: CHANSTS: 0x80701da41 CHANERR: 0x0
[ 6139.224782] ioatdma 0000:03:00.0: Errors:
[ 6139.328506] ioatdma 0000:03:00.0: CHANSTS: 0x80701da81 CHANERR: 0x0
[ 6139.334787] ioatdma 0000:03:00.0: Errors:
[ 6139.438511] ioatdma 0000:03:00.0: CHANSTS: 0x80701dac1 CHANERR: 0x0
[ 6139.444792] ioatdma 0000:03:00.0: Errors:
[ 6139.548516] ioatdma 0000:03:00.0: CHANSTS: 0x80701db01 CHANERR: 0x0
[ 6139.554798] ioatdma 0000:03:00.0: Errors:
<>

This is happening with any of your code after about Jun 2017...

Can you give us a hint where to look next?

vendor_id : GenuineIntel
cpu family : 6
model : 86
model name : Intel(R) Xeon(R) CPU D-1559 @ 1.50GHz
stepping : 4
microcode : 0xf000015
cpu MHz : 1500.000
cache size : 18432 KB
physical id : 0
siblings : 24
core id : 13
cpu cores : 12
apicid : 27
initial apicid : 27
fpu : yes
fpu_exception : yes
cpuid level : 20
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 cdp_l3 intel_ppin intel_pt ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a rdseed adx smap xsaveopt cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm arat pln pts md_clear spec_ctrl intel_stibp flush_l1d
bogomips : 2999.85
clflush size : 64
cache_alignment : 64
address sizes : 46 bits physical, 48 bits virtual
power management: