[PATCH] Selective attach for ide-scsi

[PATCH] Selective attach for ide-scsi

[Willem Riede]

Today, if you boot 2.6.x with hdd=ide-scsi, ide-scsi will attach to
all your Atapi drives, not just hdd, unless you explicitely specified
another driver for those.

Given that we don't want people to use ide-scsi for cdroms and cd-writers,
that behavior is IMHO suboptimal.

The patch below makes ide-scsi attach ONLY to those drives that you tell
it to. So if you want it to handle hdb and hdd, but not hdc, you boot
with hdb=ide-scsi hdd=ide-scsi.

Regards, Willem Riede.

--- p0/drivers/scsi/ide-scsi.c	2004-01-31 10:29:11.000000000 -0500
+++ a1/drivers/scsi/ide-scsi.c	2004-02-08 16:40:19.000000000 -0500
@@ -955,17 +955,18 @@
 	static int warned;
 	int err;
 
-	if (!warned && drive->media == ide_cdrom) {
-		printk(KERN_WARNING "ide-scsi is deprecated for cd burning! Use ide-cd and give dev=/dev/hdX as device\n");
-		warned = 1;
-	}
-
-	if (!strstr("ide-scsi", drive->driver_req) ||
+	if (!drive->driver_req ||
+	    !strstr(drive->driver_req, "ide-scsi") ||
 	    !drive->present ||
 	    drive->media == ide_disk ||
 	    !(host = scsi_host_alloc(&idescsi_template,sizeof(idescsi_scsi_t))))
 		return 1;
 
+	if (!warned && drive->media == ide_cdrom) {
+		printk(KERN_WARNING "ide-scsi is deprecated for cd burning! Use ide-cd and give dev=/dev/hdX as device\n");
+		warned = 1;
+	}
+
 	host->max_id = 1;
 	host->max_lun = 1;
 	drive->driver_data = host;

Re: [PATCH] Selective attach for ide-scsi

[Mikael Pettersson]

Willem Riede writes:
 > Today, if you boot 2.6.x with hdd=ide-scsi, ide-scsi will attach to
 > all your Atapi drives, not just hdd, unless you explicitely specified
 > another driver for those.
 > 
 > Given that we don't want people to use ide-scsi for cdroms and cd-writers,
 > that behavior is IMHO suboptimal.
 > 
 > The patch below makes ide-scsi attach ONLY to those drives that you tell
 > it to. So if you want it to handle hdb and hdd, but not hdc, you boot
 > with hdb=ide-scsi hdd=ide-scsi.

The patch I posted, which you apparently didn't like, doesn't
require the use of boot-only options: it instead adds a module_param
to ide-scsi which allows for greater flexibility.

Personally I never liked that butt-ugly hdX=ide-scsi hack.

/Mikael

Re: [PATCH] Selective attach for ide-scsi

[Willem Riede]

On 2004.02.09 03:24, Mikael Pettersson wrote:
> Willem Riede writes:
>  > Today, if you boot 2.6.x with hdd=ide-scsi, ide-scsi will attach to
>  > all your Atapi drives, not just hdd, unless you explicitely specified
>  > another driver for those.
>  > 
>  > Given that we don't want people to use ide-scsi for cdroms and cd-writers,
>  > that behavior is IMHO suboptimal.
>  > 
>  > The patch below makes ide-scsi attach ONLY to those drives that you tell
>  > it to. So if you want it to handle hdb and hdd, but not hdc, you boot
>  > with hdb=ide-scsi hdd=ide-scsi.
> 
> The patch I posted, which you apparently didn't like, doesn't
> require the use of boot-only options: it instead adds a module_param
> to ide-scsi which allows for greater flexibility.
> 
> Personally I never liked that butt-ugly hdX=ide-scsi hack.

I hear you. There are certainly advantages to use a module parameter rather
than a boot argument.

However, there should not be two mechanisms to achieve the same goal. For
better or for worse, the hdX=<driver> construction exists, and people are
using it. Its use is not limited to ide-scsi.

Since it can very easily be adjusted to achieve the desired selectivety,
I believe it is the mechanism of choice. 

Does anyone else have an opinion?

Thanks, Willem Riede.

Re: [PATCH] Selective attach for ide-scsi

[Patrick Mansfield]

On Mon, Feb 09, 2004 at 07:02:05PM -0500, Willem Riede wrote:
> On 2004.02.09 03:24, Mikael Pettersson wrote:
> > Willem Riede writes:

> > The patch I posted, which you apparently didn't like, doesn't
> > require the use of boot-only options: it instead adds a module_param
> > to ide-scsi which allows for greater flexibility.
> > 
> > Personally I never liked that butt-ugly hdX=ide-scsi hack.
> 
> I hear you. There are certainly advantages to use a module parameter rather
> than a boot argument.

But module_param allows module arguments when built as a module, and boot
arguments when built into the kernel.

> However, there should not be two mechanisms to achieve the same goal. For
> better or for worse, the hdX=<driver> construction exists, and people are
> using it. Its use is not limited to ide-scsi.

So does module_param not work because the usage is across modules? That
seems odd.

-- Patrick Mansfield

Re: [PATCH] Selective attach for ide-scsi

[Willem Riede]

On 2004.02.11 15:11, Patrick Mansfield wrote:
> On Mon, Feb 09, 2004 at 07:02:05PM -0500, Willem Riede wrote:
> > On 2004.02.09 03:24, Mikael Pettersson wrote:
> > > Willem Riede writes:
> 
> > > The patch I posted, which you apparently didn't like, doesn't
> > > require the use of boot-only options: it instead adds a module_param
> > > to ide-scsi which allows for greater flexibility.
> > > 
> > > Personally I never liked that butt-ugly hdX=ide-scsi hack.
> > 
> > I hear you. There are certainly advantages to use a module parameter rather
> > than a boot argument.
> 
> But module_param allows module arguments when built as a module, and boot
> arguments when built into the kernel.
> 
> > However, there should not be two mechanisms to achieve the same goal. For
> > better or for worse, the hdX=<driver> construction exists, and people are
> > using it. Its use is not limited to ide-scsi.
> 
> So does module_param not work because the usage is across modules? That
> seems odd.

I wasn't making myself clear, it seems.

The hdX= construct applies to the entire ide subsystem, which for the vast
majority of people means it has to be specified at boot time, as ide is
compiled in.

If we were to have an ide-scsi module option to tell it which hdX units to
attach to, that would be more flexible than having to tell ide, since I can
then rmmod/insmod ide-scsi if I want to change my mind, whereas I must reboot
if I need to change what I tell ide.

The advantage of the hdX ide parameter is that it applies to the entire ide 
subsystem, and therefor influences ide-cd, ide-scsi, ide-tape.

The main reason I see for sticking with the hdX= construct is that I think
that introducing competing mechanisms that achieve much the same objective
is a bad thing.

Regards, Willem Riede.

Re: [PATCH] Selective attach for ide-scsi

[Bartlomiej Zolnierkiewicz]

On Saturday 14 of February 2004 23:06, Willem Riede wrote:
> On 2004.02.11 15:11, Patrick Mansfield wrote:
> > On Mon, Feb 09, 2004 at 07:02:05PM -0500, Willem Riede wrote:
> > > On 2004.02.09 03:24, Mikael Pettersson wrote:
> > > > Willem Riede writes:
> > > >
> > > > The patch I posted, which you apparently didn't like, doesn't
> > > > require the use of boot-only options: it instead adds a module_param
> > > > to ide-scsi which allows for greater flexibility.
> > > >
> > > > Personally I never liked that butt-ugly hdX=ide-scsi hack.
> > >
> > > I hear you. There are certainly advantages to use a module parameter
> > > rather than a boot argument.
> >
> > But module_param allows module arguments when built as a module, and boot
> > arguments when built into the kernel.
> >
> > > However, there should not be two mechanisms to achieve the same goal.
> > > For better or for worse, the hdX=<driver> construction exists, and
> > > people are using it. Its use is not limited to ide-scsi.
> >
> > So does module_param not work because the usage is across modules? That
> > seems odd.
>
> I wasn't making myself clear, it seems.
>
> The hdX= construct applies to the entire ide subsystem, which for the vast
> majority of people means it has to be specified at boot time, as ide is
> compiled in.
>
> If we were to have an ide-scsi module option to tell it which hdX units to
> attach to, that would be more flexible than having to tell ide, since I can
> then rmmod/insmod ide-scsi if I want to change my mind, whereas I must
> reboot if I need to change what I tell ide.
>
> The advantage of the hdX ide parameter is that it applies to the entire ide
> subsystem, and therefor influences ide-cd, ide-scsi, ide-tape.
>
> The main reason I see for sticking with the hdX= construct is that I think
> that introducing competing mechanisms that achieve much the same objective
> is a bad thing.

$ echo ide-scsi>/proc/ide/hdX/driver
or
$ echo "ide-scsi:1">/proc/ide/hdX/settings
or
use HDIO_SET_IDE_SCSI ioctl

and you can change driver from ide-{cd,floppy,tape} to ide-scsi in-fly.
You can also use it in reverse direction (ie. from ide-scsi to ide-cd).

What more crap do you need?  There is already one /proc setting too much.

Cheers,
--bart

Re: [PATCH] Selective attach for ide-scsi

[Willem Riede]

On 2004.02.14 17:54, Bartlomiej Zolnierkiewicz wrote:
> >
> > The main reason I see for sticking with the hdX= construct is that I think
> > that introducing competing mechanisms that achieve much the same objective
> > is a bad thing.
> 
> $ echo ide-scsi>/proc/ide/hdX/driver
> or
> $ echo "ide-scsi:1">/proc/ide/hdX/settings
> or
> use HDIO_SET_IDE_SCSI ioctl
> 
> and you can change driver from ide-{cd,floppy,tape} to ide-scsi in-fly.
> You can also use it in reverse direction (ie. from ide-scsi to ide-cd).
> 
> What more crap do you need?  There is already one /proc setting too much.

Nothing.

I was actually arguing that we _don't_ need a new mechanism. And, given
that there is a /proc entry to change it (which I didn't realize), there
is no downside to the current mechanism.

Thanks, Willem Riede.