On Tue, May 19, 2020 at 12:58:55PM +0200, Oleksij Rempel wrote: > On Tue, May 19, 2020 at 10:55:20AM +0200, Michal Kubecek wrote: > > On Tue, May 19, 2020 at 09:51:59AM +0200, Oleksij Rempel wrote:
> > I'm also a bit worried about hardcoding the 0-7 value range. While I > > understand that it's defined by standard for 100base-T1, we my want to > > provide such information for other devices in the future. I tried to > > search if there is something like that for 1000base-T1 and found this: > > > > The screenshot on page 10 shows "SQI Value: 00015". > > Nice, screenshot based reverse engineering :) > > > It's probably not > > standardized (yet?) but it seems to indicate we may expect other devices > > providing SQI information with different value range. > > what maximal range do we wont to export? u8, u16 or u32? > > > Would it make sense to add ETHTOOL_A_LINKSTATE_SQI_MAX attribute telling > > userspace what the range is? > > sounds plausible. SQI is not yet standardized for 1000BASE-T1 to my knowledge. But looking at the definition of SQI register for 100BASE-T1 gives some good clues. [0] reserved [3:1] current SQI value [4] reserved [7:5] worst case SQI value since last read So, the registers are implemented in a way that another LSB can be added. This would increase the resolution of SQI measurement, not the range. I.e. the resulting value range could be seen as [0, 0.5, 1, ..., 7.5] or [0, 1, ..., 15]. For the screenshot, I wouldn’t assume the reserved bit is implemented, it could just be that the SQI value was not shifted. Regards, Christian
