cvs commit: src/sys/dev/mpt mpt_pci.c

[mjacob at freebsd.org]

On Wed, 15 Nov 2006, Scott Long wrote:

> mjacob at freebsd.org wrote:
>> 
>> see http://people.freebsd.org/~mjacob for 2 bootverbose files.
>> 
>> 
>
> Ah, it looks like the MPT chips support 8 messages.  I think that
> John's patch assumes only 1 message for it.

No, John's patch will only enable MSI for a count of one.

The PCI cards I have that support MSI are:

AMD:

QLogic 6312 (MSI 8)
LSI-Logic 1030 (MSI 1)
LSI-Logic 1030 (MSI 1)
[ STUPID ME- The FC929 doesn't have MSI ]
Broadcom NetXtreme (MSI 8)
Broadcom NetXtreme (MSI 8)

SuperMicro

LSI-Logic 4Gb (MSI-1 MSI-X 1)
LSI-Logic 4Gb (MSI-1 MSI-X 1)
LSI-Logic SAS (MSI-1 MSI-X 1)
QLogic 2422 4Gb (MSI 8)
QLogic 2422 4Gb (MSI 8)
em (MSI 1)

I would guess you were looking at the one of the other devs.



>  For some devices,
> requesting 1 message when N are available will work, but it really is
> undefined behaviour and is dependent on the device to DTRT.  So in your
> case, some of the MPT chips handled it well, some did not.  Just out of
> curiosity, do you have any MPT docs that describe its MSI capabilities?
> In any case, it should probably be disabled in the MPT driver until we
> have a better understanding of how it works there.
>
> For your if_em problem, I'm not seeing what the problem is.  It looks
> like em0 was successful at getting an IP address over DHCP, so I would
> assume that interrupts are working on it.

Nope. That's why I included the change to disable msi in the boot 
verbose (there are two boot sessions there). You can have a *mostly* 
functional ethernet w/o interrupts functioning- you just get watchdog 
resets.