Garrett D'Amore wrote: > Cathy Zhou wrote: >> Sowmini.Varadhan at Sun.COM wrote: >> >>> Hi Cathy, >>> >>>> Hi Sowmini, >>>> >>>> I am still reviewing your code and since I am busy with my moving to >>>> US, I am not sure >>>> whether I can finish the review before March 26th. >>> no problem- Seb mentioned this to me, and has offered to review >>> the code, since you/meem are busy with moving. But any comments >>> that you have are welcome, of course. >>> >>> >>>> So far I have only one question: >>>> >>>> the fix of bug 6672490: I think you could simply change the Why >>>> don't simply change the >>>> second argument of get_linkduplex() and get_linkstate() from B_FALSE >>>> to B_TRUE? >>>> >>> You are right, that also looks like it would work. But would this >>> work for devices that are not renamed? If yes, then why don't >>> we always assume islink? >>> >>> >> Basically, if the given name is a link name, we set "islink" to true, >> and when the given name is a device instance name, we set "islink" to >> false. In the former case, we query the "link" kstats (in the form of >> "link":0:<link name>:<stat name>), and in the latter case, we query >> the old kstats form (<driver>:<instance>:<device instance name>:<stat >> name>). Further, note some network devices do not have the first form >> of kstats, for example, links of media types that are not supported by >> GLDv3. >> > > Actually, shouldn't the kstat for GLDv3 links be > <driver>:<instance>:mac:<stat name> ? > There are two sets of kstats in GLDv3 (pre-Clearview UV), one is mac kstats that you mentioned - which is supported in the form of <driver>:<instance>:mac:<stat name> (see mac_stat.c), and another is the link kstats, which is supported in dls_stat.c.
For now, most of the link kstats are directly derived from the mac kstats, which is actually incorrect, especially for VLANs. See bug 6299825. Thanks - Cathy
