Re: systematic panic on an SMP machine for 5.1-Release

[taking one random Terry mail to get back to the list]

Hello,

I have attached some more info about the machine :
* script.crash : simple gdb output from the vmcore (with symbols) - one 
surprising point - for me - is the message :
---Can't read userspace from dump, or kernel process---
the core has ben taken as said in the handbook ????
* SMP.plus : kernel config file with PSE and PG_G work-around options
* vmic.dmesg.boot : verbose dmesg for the machine

the main point is : the crash also happens when the "work-around" kernel 
options are selected.

[some more questions inline in the Terry message]

	Thierry

Le Wednesday 09 July 2003 10:10, Terry Lambert a écrit :

> This is really strange; I'd like to try and help debug this one
> (off-list, so we don't bother anyone else with a lot of back and
> forth).  There's actually nothing I could think of that could
> cause this, unless it's the BIOS accessing what it thinks is

As the crash happens late after FreeBSD has started (in the crash above, 
during a "make buildworld", or before, when decompressing the contrib src 
files out of all src install distributions, over NFS and after the install 
step), does the BIOS still has something to do ? Is the BIOS even used at 
this time ?

> a linear access to low-core memory, and it not having a page
> mapping without the 4M page mapped.  If this is happening, it's
> most likely spamming something on you without DISABLE_PSE, and
> you just don't know it because it's not important.
>
> I could give you a snippet of code, or walk you through writing
> one, that would make sure to establish 4K page mappings for the
> whole area that's normally covered by a 4M mapping; if this was
> able to fix the problem, then it's likely you are being spammed
> and that your BIOS is doing Very Bad Things(tm).

If I may, I would like to ask once more : how can the BIOS be involved, when 
FreeBSD has taken the machine (running its IDE driver, for example)

> My gut feeling on this is that the APM is spamming memory,
> though, and you won't be happy with a workaround that doesn't
> save you from this.

This board is very simple, without any APM or ACPI (it could use IPMI instead, 
but I think IPMI is not enabled in the BIOS, and I have not added any IPMI 
driver to the stock FreeBSD kernel I have used)

> Is there any way you could try ACPI on this board, instead of
> using APM?  Or see if the vendor has a BIOS update (making sure
> you keep an old copy, just in case you want to go back)?

The BIOS seems to be up to date (I still have to check back with the support 
here in France)

> -- Terry