@jbk --> if I sent you a vmware.log file, could you *quickly and easily* see if there's SATA weirdness?

i don't konw :)

i can at least give it a shot

Mailing...

And by "SATA weirdness" I mean "VMware SATA weirdness that'd make the illumos guest have horrid ZFS errors not explainable by code-changes).

we tend to use pvscsi for people using our stuff in vmware, so not sure we've done much with the sata driver in there...

but given vmware's scsi behavior, it wouldn't surprise me if there's strangeness

You know SATA better. Pardon delay, talking about BaseT stuff elsewhere.

nothing's standing out yet aside from something not getting that a CDROM isn't going to support SCSI reservations or SYNC cache requests, yet it persists

CDROM is physically detached. <sigh> if that's all it is then I"m not sure.

I'm re-running the tests on a DIFFERENT VMware Fusion client.

(This is the gcc14 build of this week's SmartOS.)

AI seems to suggests that 'PxSCTL.DET already 0.' which appears to be for the non-CDROM SATA disks (it looks like there's perhaps several attached?)

means a possible corrupted VM config file... but it's AI, so take it with BP bursting amounts of salt...

(is there a reason to not use pvscsi?)

i'd need to dig a bit to see what that actually means... I _think_ that might be referring to an AHCI register (or a bit in one)

but I don't recall _that_ level of detail for SATA stuff offhand :)

@jbk sloth on my part, Ithink.

We may still have some mdb post-mortem-debugging problems. See fenix illumos#17681

BUG 17681: mdb `::whatis` on kernel dump is SUPER NOISY. (New)

↳ illumos.org/issues/17681

[illumos-gate] 17657 Want SMBIOS TPM (Type 43) decoding -- Robert Mustacchi <rm⊙fo>

[illumos-gate] 17656 print Endian-corrected SMBIOS UUID -- John Levon <john.levon⊙jc>

[illumos-gate] 17655 Add Phoenix 2 amdzen(4D) IDs -- Robert Mustacchi <rm⊙fo>

danmcd: see my msg

on a separate topic... there's this 'ddi_xbuf_t' stuff that only seems to be used by sd.c but seems like it was maybe intended at some point (likely distant past) to be some more general api

what do you wonder specifically?

it seems like it's implementing it's own queue on top of all the other bits

well, I can't find any useful history either. It smells of a hack 'cos I/O was slow though, doesn't it

just curious if anyone's looked at it.. or thought if it's still useful...

yeah, and I'm just getting closer to my pain threshold of dealing with sd.c where I might actually do something :P

i'm probably confusing things, but I vaguely remember a way I think with libumem to force a program to coredump at exit(1) (e.g. I want to force one to run umem)

trying to see if I can force the leak in #17676 (which was just found from looking at the code)

unless advocates are ok with just calling passwd and making sure check_circular() still works before/after or such..

I don't think there's an existing way, but you can set a breakpoint on exit or the return from main and just grab a core.

(Or run it there)

fmtopo has an explicit core on exit flag which may be what you're thinking of?

could be...