Re: [Xen-devel] no-carrier on qlogic 8242 10gig with linux 3.x running xen

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 28/03/12 10:29, Kristoffer Harthing Egefelt wrote:
>> How about pci=nomsi on kernel (i.e. vmlinuz) command line?
> That worked! The card is now working with legacy interrupts:
>
> [  143.767091] QLogic 1/10 GbE Converged/Intelligent Ethernet Driver v5.0.25
> [  143.767222] xen: registering gsi 38 triggering 0 polarity 1
> [  143.767238] xen_map_pirq_gsi: returning irq 38 for gsi 38
> [  143.767242] xen: --> pirq=38 -> irq=38 (gsi=38)
> [  143.767245] Already setup the GSI :38
> [  143.767250] qlcnic 0000:03:00.0: PCI INT A -> GSI 38 (level, low) -> IRQ
> 38
> [  143.767267] qlcnic 0000:03:00.0: setting latency timer to 64
> [  143.768057] qlcnic 0000:03:00.0: 2MB memory map
> [  144.403025] qlcnic 0000:03:00.0: phy port: 0 switch_mode: 0,
> [  144.403027]        max_tx_q: 1 max_rx_q: 8 min_tx_bw: 0x0,
> [  144.403029]        max_tx_bw: 0x64 max_mtu:0x2580, capabilities: 0x6affae
> [  144.411026] qlcnic 0000:03:00.0: failed card response code:0x10
> [  144.411099] qlcnic 0000:03:00.0: Can't get template size 16
> [  144.411102] qlcnic 0000:03:00.0: firmware v4.7.83
> [  144.419028] qlcnic: 24:b6:fd:64:1e:45: QME8242-k 10GbE Dual Port
> Mezzanine Card Board Chip rev 0x54
> [  144.419075] qlcnic 0000:03:00.0: using legacy interrupts
> [  144.419842] qlcnic 0000:03:00.0: eth0: XGbE port initialized
>
> Irq 38 is now visible in /proc/interrupts
>
> But what might the consequences be of this? 
> The NIC currently share irq with the raid card - should I expect a
> performance hit?
> The dom0 is going to have 1 core dedicated.

Out of interest, which version of Xen are you using? (if you have
already said, I apologize - i cant find it from a quick scan of this thread)

There is most certainly a performance hit.  From my bugfixing of
hardware interrupts in the past, line level interrupts take ~30ms to be
services, ranging between extremes of 10ms to 150ms.  XenServer sees an
order of magnitude hard drive performance increase by simply switching
from legacy line interrupts to MSI interrupts for the host disks.

There is no good reason why legacy line interrupts would work while MSIs
don't, which means you have probably found a bug (and indeed, there is a
reason why line level interrupts are legacy and Message Signaled
Interrupts were introduced into the PCI spec in 1995).

~Andrew

> --
> View this message in context: 
> http://xen.1045712.n5.nabble.com/no-carrier-on-qlogic-8242-10gig-with-linux-3-x-running-xen-tp5597283p5599935.html
> Sent from the Xen - Dev mailing list archive at Nabble.com.
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxx
> http://lists.xen.org/xen-devel

-- 
Andrew Cooper - Dom0 Kernel Engineer, Citrix XenServer
T: +44 (0)1223 225 900, http://www.citrix.com


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel