kernel: sioxx: silo overflow with 8 ports pci MOXA cards

[Bruce Evans]

On Wed, 19 Jan 2005, Roberto Pereyra wrote:

> I have a 8 port pci Moxa C168U/PCI and use it like ppp server (with mgetty and
> radius support).
>
> The server not works well because I have many "sio overflow" errors. The modem
> drop the link or when the link is active (some minutes later) the link don't
> send packets and drop.

Large numbers of ports might cause this problem under versions of FreeBSD
newer than FreeBSD-4, due to poor interrupt latency in these versions.
However, "large" is hopefully more than 8.

> Many times the modems not answer the call.
>
> Can anybody help me ? (excuse my english)
>
> My system is Freebsd 5.2 RELEASE
>
> My kernel is GENERIC plus this:
>
> # Serial (COM) ports
> device  sio  # 8250, 16[45]50 based serial ports
> options         COM_MULTIPORT
> device         puc

COM_MULTIPORT is not needed for puc devices.  Using it when it is not needed
(for other devices) just increases interrupt latency and reduces efficiency.
(The driver doesn't handle mixtures of devices (with only some devices
needing it) very well.)

> My dmesg show:
>
> ioapic0 <Version 2.0> irqs 0-23 on motherboard
> pcibios: BIOS version 2.10
> usb0: USB revision 1.0
> usb1: USB revision 1.0
> usb2: USB revision 1.0
> sio4: <Moxa Technologies, C168H/PCI> on puc0
> sio4: type 16550A
> sio4: unable to activate interrupt in fast mode - using normal mode
> sio5: <Moxa Technologies, C168H/PCI> on puc0
> sio5: type 16550A
> sio5: unable to activate interrupt in fast mode - using normal mode
> sio6: <Moxa Technologies, C168H/PCI> on puc0
> sio6: type 16550A
> sio6: unable to activate interrupt in fast mode - using normal mode
> sio7: <Moxa Technologies, C168H/PCI> on puc0
> sio7: type 16550A
> sio7: unable to activate interrupt in fast mode - using normal mode
> sio8: <Moxa Technologies, C168H/PCI> on puc0
> sio8: type 16550A
> sio8: unable to activate interrupt in fast mode - using normal mode
> sio9: <Moxa Technologies, C168H/PCI> on puc0
> sio9: type 16550A
> sio9: unable to activate interrupt in fast mode - using normal mode
> sio10: <Moxa Technologies, C168H/PCI> on puc0
> sio10: type 16550A
> sio10: unable to activate interrupt in fast mode - using normal mode
> sio11: <Moxa Technologies, C168H/PCI> on puc0
> sio11: type 16550A
> sio11: unable to activate interrupt in fast mode - using normal mode
> sio0 port 0x3f8-0x3ff irq 4 on acpi0
> sio0: type 16550A
> sio1 port 0x2f8-0x2ff irq 3 on acpi0
> sio1: type 16550A

Try using PUC_FASTINTR to reduce interrupt latency.  This may require
juggling irqs so that the puc interrupt is not shared (hopefully not
since apic mode is used).  This is not the default because if the
interrupt ends up shared then it may do more than just fail to activate
the interrupt in fast mode as above -- it may break other devices that
want the interrupt in normal mode, depending on whether it is activated
in fast mode first.

Bryce