hrm... on sparc panicbuf is kept at a fixed address.. I suspect this is how you're able to see the panic string on the next boot (regardless of your dump config)...

i wonder how hard it would be to do something similar on x86 -- with UEFI at least, you can tell the firmware 'don't clear this chunk of memory between resets'

that could be a useful fallback if there's a problem with a dump (or dumps aren't configured)

i'm guessing right now since panicbuf is defined in machdep.c, it might not always be at the same address across different builds

jbk: I feel like you've asked this before haha

yeah, but it was more how it worked on sparc, but I think I figured out what was actually going on

and, yes, if EFI has a mechanism like SPARC does to avoid zeroing some pages on reset, I expect one could do the same thing

at first i thought it was maybe something with the log_ctl_t records...

i guess the trick is how to get it at a fixed address.. maybe something with a linker map (though not sure what flag(s) would tell it to not clobber the contents)

Presumably you have to instruct it in terms of PAs, not VAs.

yeah -- you have to give the PA to UEFI, but presumably we also need to do something with the VA of the symbol so it always maps to the explicit PA

don't know if giving it a specific VA would make that easier or not..

also i'm not sure if there's any PAs available that should be guaranteed to be usable...

hrm...

building latest smartos, illumos-joyent is failing in libjedec.. seems like it was building before the header files were installed (from the error msg)

github.com/illumos/illumos-gate/blo…r/usr/src/lib/libjedec/Makefile#L32 -- does that imply that $(SUBDIRS) should maybe depend on install_h ?

am, I would guess so, yes. If we need to make sure the headers are installed (other than outside of setup phase), then yes.

but then again, the insall_h should happen before we get to run make install anyhow, so in that sense, it is just redundant there.

did you had clobber done in build tree? because I dont recall having seen such error while building smartos...

yep, no error.