search

D-TACQ / AFHBA404

Linux host device driver for AFHBA404 and MTCA PCIe
GNU General Public License v2.0
3 stars 8 forks source link

MGT fiber offload issues. #48

Closed zack-vii closed 4 years ago

zack-vii commented 4 years ago

After having trouble getting acq480 to work with rtm-t-stream I noticed that even for 425 the offload breaks after rebooting the carrier. this was not a problem before (~ a year ago). this might be related to my 480 issues. i will revert the AFHBA404 and see if i can find a version where it still works

zack-vii commented 4 years ago

might have been caused by a faulty insmod parameters: what parameters are required/recommended have impact on performance/functionality. Is there a more detailed documentation about them other than

modinfo afhba.ko
zack-vii commented 4 years ago

works now.. seems like some options are bad for communication

aurora_monitor should be on .. I now have only aurora_monitor=1 afhba4_stream=4 afhba_nports=4 nbuffers=60 buffer_len=0x400000 assume_stuck_buffers_are_ok=1 and it works for 480 as well. a detailed documentation would be good..

zack-vii commented 4 years ago

no looks like thiese option must be given for it to work although the descriptions is written in a way that suggests this is the default. but it does not seem to be the case.

aurora_monitor=1 dma_descriptor_ram=1

petermilne commented 4 years ago

dma_descriptor_ram=1 has been default since: 40faf9c1 (Peter Milne 2018-07-05 08:06:17 +0100 105) int dma_descriptor_ram = 1; It's definitely recommended for high throughput (lower bus access overhead).

aurora_monitor is still default disable in the driver, but has been enabled in ./scripts/loadNIRQ since 7cf07839 (dt100 2015-02-25 22:53:57 +0000 7) AMON=${AMON:-1}

It could be made default in the device driver, but there's also an ambition to make it interrupt driven.

afhba4_stream is a boolean. If you want to stream, set it.