[Qemu-devel] Re: [PATCH] virtio-serial: PCI device for simple host <-> guest communication

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 1604 bytes --]

Richard W.M. Jones wrote:
> On Tue, Jun 09, 2009 at 07:15:25PM +0200, Jan Kiszka wrote:
>> Amit Shah wrote:
>>> This interface presents a char device from which bits can be
>>> sent and read.
>>>
>>> Sample uses for such a device can be obtaining info from the
>>> guest like the file systems used, apps installed, etc. for
>>> offline usage and logged-in users, clipboard copy-paste, etc.
>>> for online usage.
>> Out of curiosity: What is the advantage over instantiating a multiport
>> 16550A PCI card? Will those channels be so performance critical?
> 
> 16550A has a couple of problems: The fixed sized FIFO (16 bytes) can
> cause problems if you send data too quickly.  (I've seen this problem
> when sending too much data from host to the guest, but I can't find
> the bug report right now ...).

If you buy a real 16550A card with high-speed support (eg. 1 MBit/s),
you get larger FIFOs (eg. 256 byte). Thresholds are scaled
correspondingly by such cards. So there would be no problem in extending
this in QEMU as well, ie. making it a parameter to the init function.

>  Secondly UARTs are complicated to
> emulate compared to a clean virtio device.  Would we be happy with the
> address space taken up and amount of emulation going on once we had,
> say, 16 emulated ports?

Memory-mapped 16550As do not suffer from such a limitation.

> 
> On the other hand, multiport cards don't require special drivers in
> the guest ...

That's what raised my interest: What benefit of a virtio-serial can
over-compensate this advantage of standard hardware?

Jan



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]