Paul Durrant wrote:
On 3/6/07, Rao Shoaib <[EMAIL PROTECTED]> wrote:
We are in the process of designing the kernel socket api so it would be
very helpful if you and other folks can provide input on what they want
to see. Should the api be similar to the user land socket api or should
it be different ? . Do we need polling in the kernel or do we need event
ports or both or something else ? Any input will be helpful.
Polling would be highly useful and not too hard to support in a nice way:
Change the pollhead struct to contain a callback function and a
context arg. pollwakeup() could be modified to simply call this
function passing the arg. This would allow a common interface for a
kernel client and the code in uts/common/syscall/poll.c.
The current definition of pollhead/pollwakeup() makes this very hard;
believe me, I've had to hack around it to use sockets from within the
kernel in the past.
We have prototyped all the approaches. However we are leaning towards
providing event ports in the kernel as event ports scale nicely.
Rao.
Paul
_______________________________________________
networking-discuss mailing list
[email protected]
