Re: DPC vs tasklet

[Arjan van de Ven <arjanv@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

On Fri, 2004-05-21 at 02:36, Bryan Henderson wrote:
> > Some
> > of the LINUX documentation I've read states "tasklets are the preferred
> > mechanism with which to implement your bottom half for a normal hardware
> > device".
> 
> In all the great answers about why other things are better than tasklets, 
> the basic question of why this documentation doesn't agree isn't 
> addressed.  Let me try:  I believe the context of this statement is a 
> comparison to the more traditional Unix method of implementing the bottom 
> half in the interrupt handler.  The statement is saying that a tasklet is 
> normally preferred to putting the same processing right in the interrupt 
> handler.
> 
> But in at least some cases, a kernel thread is better than either of 
> those.

yup. And the scsi driver case is special because the SCSI layer
implements the tasklet/softirq *for you* already, so instead of each
driver having to do it's own, just use the scsi one (by completing
directly in the ISR, which queues for the generic scsi softirq)


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]