-
gitomat
[illumos-gate] 14959 NVMe namespace management support -- Robert Mustacchi <rm⊙fo>
-
rmustacc
sommerfeld, danmcd_: The setting advertised speed functions for copper is not a phy thing, but an i40e related bit.
-
rmustacc
But if it's actually an SFP plugging into it that goes out to the device, then I think that's going to be more about i40e firmware off hand.
-
rmustacc
If we think we're running at 10G hard to say exactly what's going on, but i40e just gets a blob of speed info back from the device firmware.
-
sommerfeld
yep, copper SFP for 10G-baseT which works fine with a switch but which results in 85%+ packet loss when connected to the ONT
-
rmustacc
I'm not sure if the firmware is clever enough to take a speed request and deal with a multi-speed txr.
-
sommerfeld
it shows rw properties for en_5000fdx_cap and en_2500fdx_cap (off) but declines to let me set them on.
-
sommerfeld
other en_*_cap properties are readonly.
-
sommerfeld
10g and 1g are on, others are off.
-
rmustacc
Yeah, it's because the 2.5/5G stuff was missed from the i40e propinfo.
-
rmustacc
So it doesn't say read only correctly
-
rmustacc
Sorry, missed that bit in
illumos.org/issues/13334.
-
fenix
→
BUG 13334: Recognize X710 10GBaseT*L Family (Closed) |
code.illumos.org/c/illumos-gate/+/1178
-
rmustacc
Probably some effort could be spent to make those settable and better deal with firmware there.
-
sommerfeld
more interesting question for me is what's causing the packet loss at 10G. when running a ping, the loss (or, well, the non-loss) comes in waves - long stretches of nothing getting through alternating with shorter periods where it works for a handful of seconds.
-
sommerfeld
sort of behaves like a "beat" between two slightly-mistuned frequencies.
-
sommerfeld
looking more closely at system messages I see: " NOTICE: i40e0: The driver for the device detected an older version of the \
-
sommerfeld
NVM image (1.7) than expected (1.9). Please update the NVM image."
-
sommerfeld
oddly, on another -- older -- system, NOTICE: i40e0: The driver for the device detected a newe\
-
sommerfeld
r version of the NVM image (1.15) than expected (1.10).
-
sommerfeld
and it would appear that this is some sort of API version unreleased to the NVM version numbers intel puts in their documentation.
-
sommerfeld
-
fenix
→
BUG 17479: i40e reports en_5000fdx_cap and en_2500fdx_cap as writable when they are not (New)
-
rmustacc
I can spin a quick change for that if you wouldn't mind testing it sommerfeld.
-
gitomat
[illumos-gate] 15450 qede MAC_PROP_MEDIA support -- Robert Mustacchi <rm⊙fo>