Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 29 Aug 2012 17:16:51 +0300
From:      Andriy Gapon <avg@FreeBSD.org>
To:        =?UTF-8?B?R3VzdGF1IFDDqXJleiBpIFF1ZXJvbA==?= <gperez@entel.upc.edu>, John Baldwin <jhb@FreeBSD.org>
Cc:        freebsd-stable@FreeBSD.org
Subject:   Re: Problem adding more than 8 network adapters
Message-ID:  <503E2453.9040407@FreeBSD.org>
In-Reply-To: <503E1F02.6050505@entel.upc.edu>
References:  <503C930C.3010405@entel.upc.edu> <503DE1BC.4050907@entel.upc.edu> <20120829101202.GA74970@server.rulingia.com> <201208290813.55855.jhb@freebsd.org> <503E1961.80402@FreeBSD.org> <503E1F02.6050505@entel.upc.edu>

next in thread | previous in thread | raw e-mail | index | archive | help
on 29/08/2012 16:54 Gustau PĂ©rez i Querol said the following:
> Al 29/08/2012 15:30, En/na Andriy Gapon ha escrit:
>> I wonder where the discrepancy could come from.
>> Why would VirtualBox emulate the bridge differently for different OSes?
>> And I do not see any quirks related to bus numbers for this PCI ID in either
>> Linux, FreeBSD or lspci code...
>>
>> I think that output of lspci on FreeBSD could be interesting too (it's available
>> via sysutils/pciutils port).
>>
> 
>    The output of lspci gives the same info as pciconf. I'm attaching it however.
[snip]
> 00:18.0 PCI bridge: Intel Corporation 82801 Mobile PCI Bridge (rev f2) (prog-if 01 [Subtractive decode])
> 	Flags: bus master, 66MHz, fast devsel, latency 0
> 	Bus: primary=01, secondary=01, subordinate=02, sec-latency=0
> 	!!! Unknown I/O range types e0/df
> 	!!! Unknown memory range types f100/f0ff
> 
> 00:19.0 PCI bridge: Intel Corporation 82801 Mobile PCI Bridge (rev f2) (prog-if 01 [Subtractive decode])
> 	Flags: bus master, 66MHz, fast devsel, latency 0
> 	Bus: primary=02, secondary=02, subordinate=03, sec-latency=0
> 	!!! Unknown I/O range types e0/df
> 	!!! Unknown memory range types f100/f0ff

I think that I was wrong with regard to Linux.  I see that it does extensive
bridge reconfiguring if it notices any insanity.  And I'd say that VirtualBix does
create an insane config here.
I believe that primary should be 0, secondary should be 1 and 2 respectively (as
they are) and subordinate should be equal to secondary.  So primary bus numbers
and subordinate bus numbers are insane here.
I am not sure how much the incorrect bus numbers actually affect FreeBSD PCI-PCI
driver as it does not seem to use primary and subordinate numbers for anything
important.

Memory and I/O misconfiguration are most likely much more important here.

In any case, here is a link to the broken VirtualBox code:
http://www.virtualbox.org/svn/vbox/trunk/src/VBox/Devices/Bus/DevPciIch9.cpp
See function ich9pciInitBridgeTopology, which sets primary bus and secondary bus
to X and subordinate bus to X+1.

And here a link to Linux code that re-configures those bus numbers:
http://lxr.linux.no/#linux+v3.5.3/drivers/pci/probe.c#L663

I bet that was "bus configuration invalid, reconfiguring" message during Linux boot.
-- 
Andriy Gapon



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?503E2453.9040407>