kern/109936: SMP kernel performance problem on FSC TX600
Ludger Bolmerg
lbolmerg at web.de
Tue Mar 6 19:30:09 UTC 2007
The following reply was made to PR kern/109936; it has been noted by GNATS.
From: Ludger Bolmerg <lbolmerg at web.de>
To: remko at freebsd.org
Cc: freebsd-gnats-submit at freebsd.org
Subject: Re: kern/109936: SMP kernel performance problem on FSC TX600
Date: Tue, 06 Mar 2007 20:18:51 +0100
Here is some more information that may be interesting. I ran the Linux
firmware kit from http://www.linuxfirmwarekit.org/ and got the following
results:
- the CPUs have version 8 microcode, the tool reports this to be
outdated, version 9 seems to be available
- Fail: HPET configuration test: failed to locate HPET base
- Fail: MTRR validation gives the following output
Memory range 0x0 to 0x9fbff (System RAM) has incorrect attribute default
Memory range 0x0 to 0x9fbff (System RAM) is lacking attribute write-back
Memory range 0x100000 to 0xbf891fff (System RAM) has incorrect attribute
default
Memory range 0x100000 to 0xbf891fff (System RAM) is lacking attribute
write-back
Memory range 0xbf912000 to 0xbf941fff (System RAM) has incorrect
attribute default
Memory range 0xbf912000 to 0xbf941fff (System RAM) is lacking attribute
write-back
Memory range 0xbfa12000 to 0xbfa6afff (System RAM) has incorrect
attribute default
Memory range 0xbfa12000 to 0xbfa6afff (System RAM) is lacking attribute
write-back
Memory range 0xbfa92000 to 0xbfb05fff (System RAM) has incorrect
attribute default
Memory range 0xbfa92000 to 0xbfb05fff (System RAM) is lacking attribute
write-back
Memory range 0xbfb12000 to 0xbfb1ffff (System RAM) has incorrect
attribute default
Memory range 0xbfb12000 to 0xbfb1ffff (System RAM) is lacking attribute
write-back
Memory range 0xbfb33000 to 0xbfb79fff (System RAM) has incorrect
attribute default
0Memory range xbfb33000 to 0xbfb79fff (System RAM) is lacking attribute
write-back
Memory range xbfbfa000 to 0xbfbfffff (System RAM) has incorrect
attribute default
Memory range 0xbfbfa000 to 0xbfbfffff (System RAM) is lacking attribute
write-back
Memory range 0xd0000000 to 0xd7ffffff (0000:11:00.0) has incorrect
attribute uncached
Memory range 0xe3000000 to 0xe300ffff (000:0f:0e.0) has incorrect
attribute uncached
- Fail: MCFG PCI Express* memory mapped config space
E820: MCFG mmio config space at 0xc0000000 is not reserved in ..
detailed list of E820 memory layout follows (12 lines) 5 memory ranges
are marked as 'OS used ram', 5 ranges are marked as 'reserved', 2 ranges
are marked as 'ACPI use'
- Fail: General ACPI information
PCI Express port driver reports an invalid IRQ (multiple lines same output)
_PSD domain information is corrupt (multiple lines same output)
I don't know if this is of any value. Thought it might be.
Ludger Bolmerg
Remko Lodder wrote:
> Ludger Bolmerg wrote:
>>> Number: 109936
>>> Category: kern
>>> Synopsis: SMP kernel performance problem on FSC TX600
>>> Confidential: no
>>> Severity: serious
>>> Priority: high
>>> Responsible: freebsd-bugs
>>> State: open
>>> Quarter: Keywords: Date-Required:
>>> Class: sw-bug
>>> Submitter-Id: current-users
>>> Arrival-Date: Mon Mar 05 18:20:03 GMT 2007
>>> Closed-Date:
>>> Last-Modified:
>>> Originator: Ludger Bolmerg
>>> Release: 6.2
>>> Organization:
>>> Environment:
>> FreeBSD bsd60.myname.de 6.2-RELEASE-p2 FreeBSD 6.2-RELEASE-p2 #1
>>> Description:
>> I installed FreeBSD 6.2 on a Fujitsu Siemens TX600 S3 with 2 dual core
>> Xeon processors and 4G RAM.
>> Here is the link to the data sheet of the server on Fujitsu-Siemens'
>> Web page. It may be helpful
>>
>> http://extranet.fujitsu-siemens.com/vil/pc/vil/primergy/model_line_400/rack_server/ds_prim-rx600-s3.pdf
>>
>>
>> The SMP kernel is awfully slow. make buildkernel takes about 6 hours
>> to complete. When I boot a UP kernel the build completes within 12
>> minutes. I don't see any unusual message in any log file.
>>
>> Booting the SMP kernel with kern.smp.disabled=1 shows good performance
>> values.
>>
>> When booting the SMP kernel with
>> hint.lapic.10.disabled=1
>> hint.lapic.12.disabled=1
>> hint.lapic.14.disabled=1
>>
>> also results in good performance. Disabling single CPUs didn't work.
>> I had to apply a patch provided by John Baldwin.
>> http://lists.freebsd.org/pipermail/freebsd-smp/2007-February/001320.html
>>
>> Disabling all but one CPU also gives good performance. As soon as the
>> second CPU gets enabled it becomes slow again.
>>
>>
>>
>>> How-To-Repeat:
>> Install on same hardware
>>> Fix:
>>
>>> Release-Note:
>>> Audit-Trail:
>>> Unformatted:
>> _______________________________________________
>> freebsd-bugs at freebsd.org mailing list
>> http://lists.freebsd.org/mailman/listinfo/freebsd-bugs
>> To unsubscribe, send any mail to "freebsd-bugs-unsubscribe at freebsd.org"
>
> Can you tell us what your kernel configuration is?
> What kind of scheduler do you use etc..
>
> Thanks!
>
More information about the freebsd-bugs
mailing list