From owner-freebsd-hackers  Mon Jul  8 14:17:00 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id OAA09319
          for hackers-outgoing; Mon, 8 Jul 1996 14:17:00 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id OAA09298
          for <hackers@FreeBSD.ORG>; Mon, 8 Jul 1996 14:16:55 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id HAA17933; Tue, 9 Jul 1996 07:12:05 +1000
Date: Tue, 9 Jul 1996 07:12:05 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199607082112.HAA17933@godzilla.zeta.org.au>
To: Nicolas.Souchu@prism.uvsq.fr, hackers@FreeBSD.ORG
Subject: Re: polling driver
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>Surely, doing a tsleep() in a driver is a bad idea... I think
>resources are not well locked before the drive-dependent scsi_cmd()
>routine is called because the process is supposed to keep the CPU
>during the call and return without being interrupted.

Sleeping works (only) in process context (not in interrupt handlers
including timeout handlers) .  The SCSI_NOSLEEP flag is supposed to
tell SCSI drivers when it is safe to sleep.

>Timeouts are not good either because they interrupt the CPU at a high
>level (>splbio) and the callback is executed at a very low level.

Timeout handlers are always called at the same low level (splsoftclock)
(<splbio).  See the fd driver for how to turn a timeout handler into a
normal disk interrupt handler.  At worst, a polled disk driver can work
much like an interrupt driven driver, except slower, by scheduling a
timeout every clock tick and turning the timeout into a normal disk
interrupt.

Bruce