FreeBSD-6 amr and ahd trouble

[Joerg Pulz]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Content-ID: <20051117145924.K7025 at hades.admin.frm2>

On Wed, 16 Nov 2005, Scott Long wrote:

> Joerg Pulz wrote:
> 
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>> 
>> 
>> Hi guys,
>> 
>> I'm running an Fujitsu-Siemens Primergy RX300 dual-XEON hyperthreading
>> enabled server with an onboard LSI MegaRAID controller and an Adaptec
>> 39320A Ultra320 dual channel SCSI adapter. The LSI MegaRAID controller is
>> configured to RAID1 with two disk and one hotspare. On this array FreeBSD
>> is installed.
>> Up to now, the system was running fine with FreeBSD-5.3 first and
>> FreeBSD-5.4 now.
>> I tried to upgrade this beast to FreeBSD-6.0-RELEASE without success. The
>> kernel is booting and detects all devices correctly but when it comes to
>> read from the amr(4) the last thing i see is "GEOM: new disk amrd0" after
>> that the system "hangs" and its nearly impossible to scroll the kernel
>> messages up or down (Scroll lock pressed). then after a while there are a
>> lot of SCSI error messages about SCB timeouts coming from the ahd(4).
>> I decided to boot the old RELENG_5_4 kernel and cvsup'ed the sources to
>> RELENG_6 but i got the same results. booting from a FreeBSD-6.0-RELEASE
>> bootonly CDRom got again the same results.
>> I searched google about this, and found something about a tuneable
>> sysctl/loader setting called hw.pci.do_powerstate and tried it, but the
>> same result. later i saw, that in RELENG_6 this tuneable is renamed and set
>> to 0 anyway.
>> the next step was removing the Adaptec card to make sure this one is not
>> interrupting the amr(4) but the only thing that happened was the SCSI error
>> messages going away so this was not the problem.
>> I decided to give CURRENT from today a try, and it was working without any
>> problems. I have tested CURRENT some steps back until i hit 700003 dated to
>> "Sun Sep 18 05:12:39 2005 UTC" which is exactly the same time the RELENG_6
>> branch was marked for 6.0-BETA5 and CURRENT was working with every point i
>> checked out from cvs. Unfortunately 6.0-BETA5 is NOT working.
>> I checked out the sources for 6.0-BETA4 and it is working again. So
>> somewhere between 6.0-BETA4 and 6.0-BETA5 the whole thing is broken, at
>> least for me and my hardware.
>> I've seen some differences in sys/cam/cam_xpt.c, maybe these cause the
>> trouble i have, but I'm not so deep in the FreeBSD kernel code to make this
>> sure.
>> 
>> It would be nice if someone can take a look at this to get this fixed in
>> RELENG_6.
>> Any patches to test are welcome.
>> 
>> regards
>> Joerg
>> 
> 
> This is almost certainly an interrupt routing bug.  Can you try booting with
> ACPI disabled?  Can you try building a 6.0 kernel without SMP and
> the 'apic' devices?  From 5.4, can you send your system information?

Hi Scott,

i've attached the kernel messages of the different tries.
Here is a short description.

RELENG-5.4_SMP-APIC-ACPI_verbose:
 	- selfmade kernel with SMP, apic and acpi enabled
 	- this one is working since 5.4 is out
 	- the only thing to mention is the wrong order of the serial ports
 	  sio0 == COM2, sio1 == COM1

RELENG-6_UP-APIC-ACPI_verbose:
 	- GENERIC kernel, NO SMP, apic and acpi enabled
 	- this one isn't working
 	- the order of the serial ports is wrong too

RELENG-6_UP-APIC-NOACPI_verbose:
 	- GENERIC kernel, NO SMP, apic enabled and acpi disabled
 	- this one isn't working
 	- the order of the serial ports is correct
 	  sio0 == COM1, sio1 == COM2

RELENG-6_UP-NOAPIC-ACPI_verbose:
 	- GENERIC kernel, NO SMP, apic disabled and acpi enabled
 	- this one isn't working,
 	  it hangs at "start_init: trying /sbin/init"
 	  but walking through amr(4) and ahd(4) seems to work
 	- the order of the serial ports is wrong too

RELENG-6_UP-NOAPIC-NOACPI_verbose:
 	- GENERIC kernel, NO SMP, apic nad acpi disabled
 	- this on IS WORKING, i can login and use all devices (NIC, DISK)
 	- the order of the serial ports is correct

Only to mention, that the order of the serial ports is wrong in an recent
CURRENT too, but the recent CURRENT is working with SMP, apic and acpi
enabled.
Any hints/patches to make this working with SMP again are welcome.

regards
Joerg

- --
The beginning is the most important part of the work.
 				-Plato
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (FreeBSD)

iD8DBQFDfI0jSPOsGF+KA+MRAoBtAJ9eqzKu00JBCqZpp+0dYyEhJfJoIQCZAWOn
uafR1CcJyYUCXCXvUqGzrLY=
=XS6w
-----END PGP SIGNATURE-----
-------------- next part --------------
A non-text attachment was scrubbed...
Name: logs.tgz
Type: application/octet-stream
Size: 44500 bytes
Desc: 
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20051117/923a252d/logs.obj