assertion failed: lso_max > 0 (0x0 > 0x0), file: ../../common/inet/tcp/tcp.c, line: 3364

has anyone else been getting this recently?

havent seen, sorry.

Woodstock: That's a first for me. IPv4 or 6? underlying link?

i see LSO and I shiver having dealt with it in i40e

(though that's mostly intel's fault)

which reminds me... I need to see if I ever upstreamed the bits for jumbo frames+LSO for i40e

(it limits the size of a DMA buffer to <= 4k and has the card do more segmentation/scatter gather so adding a vnic after the system is running doesn't take the upwards of 30 minutes to complete)

we've been running it for a couple of years and haven't had any problems since.. and I know I gave a copy to Dan, not sure how much he's been able to do with it

Not much jbk, and someone else is asking for a common-code update because honestly, we're behind on it.

rzezeski: ipv6, i presume

rzezeski: on ixgbe

Odd.

A quick follow-the-bouncing-values and

usr/src/uts/common/io/ixgbe/ixgbe_gld.c: cap_lso->lso_basic_tcp_ipv6.lso_max = IXGBE_LSO_MAXLEN;

I wonder if that's not getting set?!

if (ixgbe->lso_enable) {

THat's what sets the other line above. That percolates up.

i'm bisecting it, i should be able to point to a changeset causing it really soon

Oh... it's something new? Ping me for review, please.

I need to see if FiOS has removed their heads from their rear-ends about co-existence of static-IPv4 and IPv6 co-existing on their business class.

not exactly new, but it must be something recent

(Also an opportunity to bring > GigE to my-house aka. kebe.com )

at least you get static IPv4 :P

i have CGN for IPv4 (but seem to have a static-ish at least IPv6)

Sorry for whining.

and periodically, it seems like it loses state for long-running TCP connections

Lily Tomlin's Ernestine-from-The-Phone-Company enters my head whenever I think of VZ, AT&T...

NAT LOSING STATE? I'm SHOCKED.

this is the local power co-op that's run fiber along all the power lines

but being small, I'm guessing they can't get many public IPv4 IPs

yeah, probably not

just need more things to adopt IPv6 :P

(including work's VPN)

would that IPv4 died outside of retrocomputing already

All of the networks I'm responsible for have static IPv6 configurations.

I wish I could say that about the networks I was tangentially responsible for. <sigh>

hrm.. we don't really have a standard way of 'inventoring' things... e.g. gettign things like serial#s of components, do we?

if it's in fmtopo, then that's the standard way, I guess :)

yeah, but for like PCI devices, there's no standard way (AFAIK), so you'd need a driver specific topo module

(i wonder if that was part of what libfru was meant to help with on sparc, but since it's almost completely undocumented, it's hard to know if there might still be bits that are useful outside of sparc or not)

or could be useful w/ some work

Hello lovely Illumos people! Does someone know why physmem and pagestotal differ, or point me in the right direction? The mdb output suggests it's the other way around, which seems odd to me. Keep in mind: My knowledge of operating system kernels is quite limited.

[root@host01 ~]# kstat -n system_pages -p | grep -E "physmem|pagestotal"

unix:0:system_pages:pagestotal 33513938

unix:0:system_pages:physmem 33513939

[root@host01 ~]# echo ::memstat | mdb -k | grep -E "Total|Physical"

Total 33513939 130913

Physical 33513938 130913

jacuzzibob: if you look at the mdb source, you'll see that it (confusingly) prints out the kernel's total_pages as "Physical", and physmem as 'Total'.

Thank you!

it's not clear to me why they differ by one page; there's a comment in mdb's memory.c about "if physmem != total_pages" but that appears to not be relevant on x86

off by one bug?:)

one of the three hardest problems in CS (along with naming)

:)

:D

alanc: I had mentioned somewhere during the break about pkg.summary v. pkg.description and what the right thing to do if they both basically matched was. I think you know the old answer?

I have the memory that only one should exist if they're both basically the same? but I guess that wasn't it

sommerfeld: if you're in that mdb stuff, and rudely, could you confirm for me that "::report_maps" doesn't even nearly work?

sommerfeld: the code seems clear it expects miracles, but I haven't been in a position to file a bug

richlowe: only to the extent that I looked at files that turned up on a "git grep" of pagestotal, physmem, and total_pages

ah, it'll stay on my todo list until it's todone then

(if I'm reading it correctly it intends to walk all page tables looking for mappings to the requested page number)

it does, but then it just does something else entirely

reading the code, at least

I was convinced from code reading, but then I saw they mention it -- with examples -- in solaris internals 2nd ed.

github.com/oracle/solaris-ips/blob/…ter/doc/tags-and-attributes.txt#L68 - summary is supposed to be a few words, description up to a paragraph - if they're both the same and just a few words, I'd probably drop the description - but if they're much longer, I'd leave the description and replace the summary with something short

I forgot ips/doc contained things that were still relevant :\