[2730] in linux-scsi channel archive
Line of even in SCSI
daemon@ATHENA.MIT.EDU (Marc SCHAEFER)
Sun Nov 2 04:12:31 1997
From: Marc SCHAEFER <schaefer@alphanet.ch>
Date: 2 Nov 1997 09:34:08 +0100
Apparently-To: linux-scsi@vger.rutgers.edu
To: ;@unlisted-recipients (no To-header on input)
Hi,
in a driver where command queuing is disabled, the flow of
command probably looks like this (dd to the /dev/sda device):
user level
-> kernel I/O
-> SCSI subsystem
-> hardware specific driver
-> host adapter hardware
-> disk drive
<- command complete
<-
<-
<-
<-
generation of new input data
-> kernel
[ etc ]
Is that right, or is the kernel allowing to receive e.g. twice
the data that can fit in a SCSI I/O request, and as soon as
the command completes, sends the new command, and unlocks the
user level for one transfer ?
How does that look like in a read ?
I am asking, because on a SCSI analyzer I saw upto 4 ms
delay between commands (command complete/bus free to selection).
It seems to look the same when the drive does not support command queuing
and the driver does (aic7xxx 4.0).
