[Linux-aus] strange NVMe errors when USB is busy

[Russell Coker]

[N:0:8224:1] disk    PC SN810 NVMe WDC 2048GB__1                /dev/nvme0n1  
2.04TB
[N:1:0:1]    disk    KXG70PNV2T04 NVMe KIOXIA 2048GB__1         /dev/nvme1n1  
2.04TB

I have a HP ML110 Gen9 workstation with the above NVMe devices.  A problem I 
recently discovered is when writing to two SATA disks in a USB enclosure at 
maximum speed with dd (overwriting old disks before reuse in a new system) I 
get NVMe errors on the WD NVMe device but never on the Kioxia device.  It's a 
combined write speed of over 300MB/s which shouldn't be a problem.

Why would USB access cause NVMe issues?  Why would it be on only one device?  
Would it be some difference between the devices or a difference between the 
PCIe slots they are in?

Any suggestions on how to investigate this? The google results for those nvme 
errors were all referring to different situations like power saving mode on 
laptops and I couldn't find any hits for NVMe when USB is busy.  I can do 3D 
gaming on the system without any issues like that so presumably my video card 
is doing more than 300MB/s of transfers.


Jan 26 13:04:25 xev kernel: nvme nvme0: Abort status: 0x0
Jan 26 13:04:25 xev kernel: nvme nvme0: I/O tag 11 (800b) opcode 0x1 (I/O Cmd) 
QID 15 timeout, aborting req_op:WRITE(1) size:4096
Jan 26 13:04:25 xev kernel: nvme nvme0: Abort status: 0x0
Jan 26 13:04:25 xev kernel: nvme nvme0: Abort status: 0x0
Jan 26 13:04:25 xev kernel: nvme nvme0: I/O tag 10 (800a) opcode 0x1 (I/O Cmd) 
QID 15 timeout, aborting req_op:WRITE(1) size:4096
Jan 26 13:04:25 xev kernel: nvme nvme0: Abort status: 0x0
Jan 26 13:04:25 xev kernel: nvme nvme0: I/O tag 4 (b004) opcode 0x1 (I/O Cmd) 
QID 15 timeout, aborting req_op:WRITE(1) size:4096

-- 
My Main Blog         http://etbe.coker.com.au/
My Documents Blog    http://doc.coker.com.au/