pastebin.com/3d0uf2Mn has some mostly trivial things you could fix in the illumos man pages, mostly in section 4M (kernel modules that aren't drivers, filesystems, network protocols, etc.)

[illumos-gate] 17884 loader.efi: tss_pa setup seems to be flawed in trap.c -- Toomas Soome <tsoome⊙mc>

[illumos-gate] 17885 loader.efi: free_tables() appears to free tss_pa twice. -- Toomas Soome <tsoome⊙mc>

[illumos-gate] 17886 loader.efi: efi_redirect_exceptions does use uninitialized pointer -- Toomas Soome <tsoome⊙mc>

most trivial fix in a while (single-character deletion fixes the bug): code.illumos.org/c/illumos-gate/+/4582

→ CODE REVIEW 4582: 17888 pargs botches composite locale naming (NEW) | illumos.org/issues/17888

I still like my one fix which was almost as trivial that fixed the random lockups during boot on my home server

I think it was there was some locking of the CPU structs right below a comment and statement that said 'don't do any further locking below this line'

and sure enough, the fix was to move the offending line so it wasn't below it :)

Really drives home the "all the details matter" aspects of what we do.

oh i misremebered it slightly

→ GitHub commit af1da5d: 9760 mp_startup_common races itself (committed)

we were creating the device node after what was supposed to be the last thing for initializing the cpu

jbk: oof, yes. boot cpu sees the bit flip, starts looking at the new cpu's device node and it's only half built...

but so thoughtful for the original commenter to point out the problem right by it :)

'don't do X' 'ok' <problem fixed>

yeah, for me that would trigger the "how did this ever work" investigation...

yeah, i was surprised more people didn't hit it... there was something with my system though that managed to get the timing right to hit it frequently

[illumos-gate] 17882 lpsched confuses RO and RW memory, and core dumps -- Andy Fiddaman <illumos⊙fn>

tsoome: am I correct in thinking we use multiboot2_loadfile() to load everything into memory (at least initially before we copy or such)?

I just went through some stuff in .../copy.c that does that, with toomas' help, merging

[illumos-gate] 17888 pargs botches composite locale naming -- Bill Sommerfeld <sommerfeld⊙ho>

I think that's what loads unix, but I'm less clear what loads the boot archive into memory..

danmcd: do you recall if the mlxcx work you did included cx-7 support?

do any of you recall whether page relocation is exclusively a kcage artifact?

I don't *think* so but I'm not 100% sure. Easiest way to check is consult the mlxcx driver aliases info in the .p5m file.

unfortunately, I don't have the PCI ids, just vendor model #s at this point

does the pci db not tie that together?

You mean like admin.pci-ids.ucw.cz/read/PC/15b3 ?

It *should* but sometimes the user-reported data is incomplete or even erroneous.

ahh if only

you forget.. every OEM has to have at least 4 different names for the same part

i suspect if we don't have 'general' cx-7 support, than just the fact it says it's some sort of cx-7 card is enough that it won't work

I have a PRM that includes CX-7 but it's shared with us by a buys-a-lot-of nVidia customer.

i have to ask.. github.com/illumos/illumos-gate/blo…rc/uts/i86pc/vm/kboot_mmu.c#L52-L56 who's Joe? :)

the hg/git logs point to a large, multi-person putback, but when I see "Joe" in a file under i86pc/vm my mind first goes to Bonasera

richlowe: would that be used to retire a DIMM as well?

(assuming we had all the bits plumbed up to do it... i'm not sure we're there for some systems since they'd need more ACPI integration IIRC)