Il 19/11/2012 04:45, Osier Yang ha scritto:
On 2012年11月17日 06:47, Paolo Bonzini wrote:
> Il 16/11/2012 10:59, Osier Yang ha scritto:
>> If "path" for "ivshmem" memory device is specified, it's
going
>> to be setup with interrupts enabled. Otherwise interrupts is
>> disabled. To be more clear, let's see the QEMU command line
>> of both cases:
>>
>> * With interrupts (/tmp/nahanni is the ivshmem server socket path)
>>
>> /* This starts the ivshmem server */
>> % ivshmem_server -m 512 -p /tmp/foo
>>
>> /* qemu command line */
>> -chardev socket,path=/tmp/foo,id=nahanni
>> -device ivshmem,chardev=nahanni,size=512m,vectors=8,ioeventfd=on
>>
>> * Without interrupts (no ivshmem server is needed)
>> -device ivshmem,shm=nahanni,size=512m,vectors=8,ioeventfd=on
I'm not sure that ioeventfd=on adds anything when no chardev is in
effect, actually.
> I think<doorbell/> is a better name than ioeventfd.
That's a magic name from my p.o.v, because it can indicate any
boolean options. How should I get it documented? :-)
The "doorbell" element, if present, lets each guest send an interrupt to
other guests. For the ivshmem model, this requires the path to be
specified and ivshmem_server to be running.
Paolo
> That is somewhat
> kvm-specific, but there's no reason for it to be.
Being general is the rule to introduce new XML tags, but sometimes
the better way is to keep it specific. It's not that bad because
at least we can ensure the specific tag only works for the specific
driver.