[8113] in linux-scsi channel archive
RE: scsi queue depth problem
daemon@ATHENA.MIT.EDU (Ralston, Steve)
Tue Feb 15 10:00:35 2000
Message-ID: <B2B3C90E45AED111B7B40001FA7E520A03833F52@exw-kansas.ks.lsil.com>
From: "Ralston, Steve" <steve.ralston@lsil.com>
To: "'linux-scsi@vger.rutgers.edu'" <linux-scsi@vger.rutgers.edu>
Date: Tue, 15 Feb 2000 08:53:58 -0600
MIME-Version: 1.0
Content-Type: text/plain
Sorry for lack of information. I'm using a LSI Logic SYM40909
Fibre Channel host adapter - which you've most probably never
heard of since it's not a released product yet. The driver is
under development; it's somewhat of a clone of drivers/i2o/i2o_scsi.c
developed by Alan Cox. For development I always compile it
as a module and use insmod.
> - how were the devices introduced (initial scan or
add-single-device)
I'm not sure. My mpi_scsi_detect() routine does:
shpnt = scsi_register(tpnt, sizeof(MPI_SCSI_HOST));
> - how are you determining the queue length?
In the driver I have,
static atomic_t queue_depth;
STATIC int max_qd = 1;
and in my mpi_scsi_queuecommand() routine, after issueing cmd
to the host adapter, I do:
atomic_inc(&queue_depth);
if ( atomic_read(&queue_depth) > max_qd )
{
max_qd = atomic_read(&queue_depth);
dprintk((KERN_INFO "mpiscsi: Queue depth now %d.\n", max_qd));
}
and in the cmd completion callback, I do:
atomic_dec(&queue_depth);
The Scsi_Host_Template I'm using has:
can_queue: 64, \
sg_tablesize: 69, \
cmd_per_lun: 64, \
Things have been working fine up thru linux-2.3.42; I hit a queue
depth of 64 (eventually) while beating on a single Seagate FC drive.
But with 2.3.44 and 2.3.45 I never see queue depth advance past
ONE (1).
> Perhaps you could load the sg driver and try:
> $ cd /proc/scsi/sg
> $ cat device_hdr devices
> $ cat host_hdr hosts
Sorry, I've tried loading the sg driver, but it is not yielding ANY
/proc/scsi entries, even though I definitely have CONFIG_PROC_FS
set in my kernel/module builds. (I don't even see CONFIG_PROC_FS
anywhere in drivers/scsi/sg.c)
Thanks,
-SteveR
-----Original Message-----
From: Douglas Gilbert [SMTP:dgilbert@interlog.com]
Sent: Tuesday, February 15, 2000 6:46 AM
To: Ralston, Steve
Cc: 'linux-scsi@vger.rutgers.edu'
Subject: Re: scsi queue depth problem
Could you supply a bit more information:
- which adapter
- how were the hosts introduced (built in or insmod)
- how were the devices introduced (initial scan or add-single-device)
- how are you determining the queue length?
Perhaps you could load the sg driver and try:
$ cd /proc/scsi/sg
$ cat device_hdr devices
$ cat host_hdr hosts
and look at "qdepth" and "cpl" (commands per lun). There has
been a problem with the new MQ mid-level and advansys whereby
add-single-device gives you a device with a queue length of
0!
Doug Gilbert
-----Original Message-----
From: Eric Youngdale [SMTP:eric@andante.org]
Sent: Tuesday, February 15, 2000 6:45 AM
To: Ralston, Steve; linux-scsi@vger.rutgers.edu
Subject: Re: scsi queue depth problem
Could you tell me what host adapter you are using, and what you expect
the maximum queue depth to be?
-Eric
> -----Original Message-----
> From: Ralston, Steve
> Sent: Monday, February 14, 2000 9:06 PM
> To: 'linux-scsi@vger.rutgers.edu'
> Subject: RE: scsi queue depth problem
>
> Anyone here experiencing limited queue depth?
>
> Thanks,
> -SteveR
>
> -----Original Message-----
> From: Alan Cox [SMTP:alan@lxorguk.ukuu.org.uk]
> Sent: Monday, February 14, 2000 8:12 PM
> To: steve.ralston@lsil.com
> Subject: Re: scsi queue depth problem
>
> Not off hand. linux-scsi@vger.rutgers.edu is the right place to ask - the
> scsi layer is being changed arouhnd a fair bit right now. You may be
> hitting
> a bug in it
>
> -----Original Message-----
> From: Ralston, Steve
> Sent: Monday, February 14, 2000 7:37 PM
> To: 'Alan Cox'
> Subject: scsi queue depth problem
>
> Hello Alan,
>
> I noticed something peculiar after updating from 2.3.42 kernel;
> I no longer seem able to get past max queue depth of ONE (!)
> now on the scsi IO path.
> I guess I never finished 2.3.43 build, but noticed this in both
> 2.3.44 and 2.3.45.
> Anything you know of that would be newly limiting me here?
>
> Thanks,
> -SteveR
>
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
