More than 10 IDE interfaces

More than 10 IDE interfaces

[Baldur Norddahl]

Hi,

I have a machine with the following configuration:

2 on board IDE interfaces (AMD chipset)
2 Promise Technology UltraDMA100 controllers with each 2 IDE interfaces.
4 Promise Technology UltraDMA133 controllers with each 2 IDE interfaces.

This adds up to 14 IDE interfaces. And I just discovered that the kernel
only supports 10 IDE interfaces :-(

So I tried to hack the kernel, and I was partially successfull. I changed
MAX_HWIF from 10 to 14. I made up some major numbers for the extra
interfaces (115, 116, 117 and 118).

drivers/ide/ide.c and fs/partitions/check.c were modified to know about
IDE10_MAJOR to IDE13_MAJOR.

With there changes the kernel detects the extra interfaces and the disks on
them. They get some strange names like IDE< and the last disk is named hd{,
but I guess I can live with that :-)

But when it tries to detect the partitions on the extra interfaces, it locks
up. The last lines it writes is:

Partition check:
 hda: hda1
 hde: hde1
 hdg: hdg1
 hdi: hdi1
 hdk: hdk1
 hdm: hdm1
 hdo: hdo1
 hdq: hdq1
 hds: hds1
 hdu:

I am looking for any clues about how to do this.

Some information about the system:

Linux kernel 2.4.18 with ide.2.4.18-rc1.02152002.patch IDE patches applied.
Dual Athlon MP 1800+ 1.53 GHz. Tyan Tiger MPX motherboard, AMD 760 MPX
chipset. 8 x 80 GB western digital disks in a raid5, 4 x 160 GB maxtor in a
raid5, the two raids are merged with LVM into one filesystem with reiserfs.

Thanks, 
  Baldur

Re: More than 10 IDE interfaces

[Martin Dalecki]

Baldur Norddahl wrote:
> Hi,
> 
> I have a machine with the following configuration:
> 
> 2 on board IDE interfaces (AMD chipset)
> 2 Promise Technology UltraDMA100 controllers with each 2 IDE interfaces.
> 4 Promise Technology UltraDMA133 controllers with each 2 IDE interfaces.
> 
> This adds up to 14 IDE interfaces. And I just discovered that the kernel
> only supports 10 IDE interfaces :-(
> 
> So I tried to hack the kernel, and I was partially successfull. I changed
> MAX_HWIF from 10 to 14. I made up some major numbers for the extra

In your case if should be changed to 15 there is an off by one error here in the
interpretation of this constant.

> interfaces (115, 116, 117 and 118).
> 
> drivers/ide/ide.c and fs/partitions/check.c were modified to know about
> IDE10_MAJOR to IDE13_MAJOR.
> 
> With there changes the kernel detects the extra interfaces and the disks on
> them. They get some strange names like IDE< and the last disk is named hd{,
> but I guess I can live with that :-)

The cause of those funny names is well known in the 2.5.xx series.
The solution to it will first involve a complete rewrite of the kernel
parameter parsing in ide.c

> 
> But when it tries to detect the partitions on the extra interfaces, it locks
> up. The last lines it writes is:
> 
> Partition check:
>  hda: hda1
>  hde: hde1
>  hdg: hdg1
>  hdi: hdi1
>  hdk: hdk1
>  hdm: hdm1
>  hdo: hdo1
>  hdq: hdq1
>  hds: hds1
>  hdu:

See above + make MAX_HWIFS 15 and you should have more luck. (Not tested
actually).

Re: More than 10 IDE interfaces

[Bill Davidsen]

On Thu, 11 Apr 2002, Martin Dalecki wrote:

> Baldur Norddahl wrote:
> > Hi,
> > 
> > I have a machine with the following configuration:
> > 
> > 2 on board IDE interfaces (AMD chipset)
> > 2 Promise Technology UltraDMA100 controllers with each 2 IDE interfaces.
> > 4 Promise Technology UltraDMA133 controllers with each 2 IDE interfaces.
> > 
> > This adds up to 14 IDE interfaces. And I just discovered that the kernel
> > only supports 10 IDE interfaces :-(
> > 
> > So I tried to hack the kernel, and I was partially successfull. I changed
> > MAX_HWIF from 10 to 14. I made up some major numbers for the extra
> 
> In your case if should be changed to 15 there is an off by one error here in the
> interpretation of this constant.

  ??? If the current value is 10, and supports 10 interfaces, and I
believe that is the case, why should he need a value of 15 to get 14?
Doesn't the off by one error happen on smaller values, or what?

  I am NOT disagreeing with you, I just don't see how to code an off by
one and have it work some of the time and not others.
 
> > interfaces (115, 116, 117 and 118).
> > 
> > drivers/ide/ide.c and fs/partitions/check.c were modified to know about
> > IDE10_MAJOR to IDE13_MAJOR.
> > 
> > With there changes the kernel detects the extra interfaces and the disks on
> > them. They get some strange names like IDE< and the last disk is named hd{,
> > but I guess I can live with that :-)
> 
> The cause of those funny names is well known in the 2.5.xx series.
> The solution to it will first involve a complete rewrite of the kernel
> parameter parsing in ide.c

  I thought devfs was supposed to eliminate all this stuff and replace
names with long hierarchical names. In hindsight it probably would have
been better to call drives something like hdNNX, where NN is the interface
and X is m or s. Hum, I wonder how hard that would be as a retrofit?

> > But when it tries to detect the partitions on the extra interfaces, it locks
> > up. The last lines it writes is:
> > 
> > Partition check:
> >  hda: hda1
> >  hde: hde1
> >  hdg: hdg1
> >  hdi: hdi1
> >  hdk: hdk1
> >  hdm: hdm1
> >  hdo: hdo1
> >  hdq: hdq1
> >  hds: hds1
> >  hdu:
> 
> See above + make MAX_HWIFS 15 and you should have more luck. (Not tested
> actually).

  Right. I would be interested in this one myself, but I don't have a good
way of getting that many drives attached. I have a pile of old Promise
ATA33 cards, and I could come up with a hundred 400-500MB drives, but the
case and cabling is a serious issue.

  Anyway, if you have a moment to hint why an off by one error is not
biting us on ten drives I'd be interested.

-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: More than 10 IDE interfaces

[Martin Dalecki]

Bill Davidsen wrote:
> On Thu, 11 Apr 2002, Martin Dalecki wrote:
> 
> 
>>Baldur Norddahl wrote:
>>
>>>Hi,
>>>
>>>I have a machine with the following configuration:
>>>
>>>2 on board IDE interfaces (AMD chipset)
>>>2 Promise Technology UltraDMA100 controllers with each 2 IDE interfaces.
>>>4 Promise Technology UltraDMA133 controllers with each 2 IDE interfaces.
>>>
>>>This adds up to 14 IDE interfaces. And I just discovered that the kernel
>>>only supports 10 IDE interfaces :-(
>>>
>>>So I tried to hack the kernel, and I was partially successfull. I changed
>>>MAX_HWIF from 10 to 14. I made up some major numbers for the extra
>>
>>In your case if should be changed to 15 there is an off by one error here in the
>>interpretation of this constant.

>   Anyway, if you have a moment to hint why an off by one error is not
> biting us on ten drives I'd be interested.

I think simply (not tested) that MAX_HWIF provides only 9 possible interfaces 
right now.

Re: More than 10 IDE interfaces

[Martin Dalecki]

Bill Davidsen wrote:
> On Thu, 11 Apr 2002, Martin Dalecki wrote:
> 
> 
>>Baldur Norddahl wrote:
>>
>>>Hi,
>>>
>>>I have a machine with the following configuration:
>>>
>>>2 on board IDE interfaces (AMD chipset)
>>>2 Promise Technology UltraDMA100 controllers with each 2 IDE interfaces.
>>>4 Promise Technology UltraDMA133 controllers with each 2 IDE interfaces.
>>>
>>>This adds up to 14 IDE interfaces. And I just discovered that the kernel
>>>only supports 10 IDE interfaces :-(
>>>
>>>So I tried to hack the kernel, and I was partially successfull. I changed
>>>MAX_HWIF from 10 to 14. I made up some major numbers for the extra
>>
>>In your case if should be changed to 15 there is an off by one error here in the
>>interpretation of this constant.
> 
> 
>   ??? If the current value is 10, and supports 10 interfaces, and I
> believe that is the case, why should he need a value of 15 to get 14?
> Doesn't the off by one error happen on smaller values, or what?
> 
>   I am NOT disagreeing with you, I just don't see how to code an off by
> one and have it work some of the time and not others.

You can have luck due to alignment issues.

Re: More than 10 IDE interfaces

[Stephen Samuel]

It might be trowing bad money after good to do it, but why
not just put together a simple translation table.

rather than
drivenum=(driveletter - 'a')
and
driveletter=(drivenum+'a')

have translation tables, so that
	drivenum= chartodrivenum[driveletter]
    and
	driveletter= drivenumtoletter[drivenum]
If you're willing to map (almost) all of the printable
characters, you could get 46 controllers and 92 drives
(I'd refuse to map  ', ", \  or space)


You'd still be limited to one character, but it would, at least
make it easy to have 26 controllers and 52 drives  (and that's
just using upper and lower case characters!)

Martin Dalecki wrote:
> Baldur Norddahl wrote:
>> With there changes the kernel detects the extra interfaces and the 
>> disks on
>> them. They get some strange names like IDE< and the last disk is named 
>> hd{,
>> but I guess I can live with that :-)
> 
> 
> The cause of those funny names is well known in the 2.5.xx series.
> The solution to it will first involve a complete rewrite of the kernel
> parameter parsing in ide.


Geh... Here's even a perl script to generate the mapping..
Change the characters if you wish.

#!/bin/perl -w
# Copyright (c) 2002 Stephen Samuel
# Can be re-licensed under either GPL or LGPL
#
my (%drivenum, %idenum);
my (@drivenames, @idenames);

@drivenames= qw' a b c d e f g h i j k l m n o p q r s t u v w x y z A B C D E F
		G H I J K L M N O P Q R S T U V W X Y Z 0 1 2 3 4 5 6 7 8 9 ';


@idenames= qw' 0 1 2 3 4 5 6 7 8 9 A B C D E F G H I J K L M N O P Q R S T U ' ;
$drivenames=scalar(@drivenames);
$idenames=scalar(@idenames);

print <<EOF1

/*
         defines the ide drive and controller mappings

*/

#define MAX_IDE_DEFINABLE $idenames
#define MAX_IDE_DRIVES_DEFINABLE $drivenames


EOF1
;
print "\n signed char drivenumtochar[$drivenames] ={";
$drivecount=0;
foreach $drive ( @drivenames ){
         printf "%s%s\t'%s'",
                 $drivecount?",":"",
                 ($drivecount & 7) ?"":"\n/*$drivecount*/" ,
                 $drive;
         $drivenum{$drive} = $drivecount++;
};
print "};\n\n";

print "signed char idenumtochar[$idenames] ={";
$idecount=0;
foreach $ide ( @idenames ){
         printf "%s%s\t'%s'",
                 $idecount?",":"",
                 ($idecount & 7) ?"":"\n/*$idecount*/" ,
                 $ide;
         $idenum{$ide} = $idecount++;
};
print "};\n\n";

$idecount=0;
foreach $ide ( @idenames ){
         $idenum{$ide} = $idecount++;
};

print "\n/* can't be 128 because of EBCDIC */\nsigned char chartodrivenum[256] ={";
for ($char=0; $char<256; $char++){
         printf "%s%s\t%s",
                 $char?",":"",
                 ($char & 7) ?"":"\n/*$char*/" ,
                 defined($drivenum{chr($char)})?$drivenum{chr($char)}:-1;
};
print "}; \n\n";


print " signed char chartoidenum[256] ={";
for ($char=0; $char<256; $char++){
         printf "%s%s\t%s",
                 $char?",":"",
                 ($char & 7) ?"":"\n/*$char*/" ,
                 defined($idenum{chr($char)})?$idenum{chr($char)}:-1;
}
print "}; \n\n"
-- 
Stephen Samuel +1(604)876-0426                samuel@bcgreen.com
		   http://www.bcgreen.com/~samuel/
Powerful committed communication, reaching through fear, uncertainty and
doubt to touch the jewel within each person and bring it to life.

Re: More than 10 IDE interfaces

[Martin Dalecki]

Stephen Samuel wrote:
> It might be trowing bad money after good to do it, but why
> not just put together a simple translation table.
> 
> rather than
> drivenum=(driveletter - 'a')
> and
> driveletter=(drivenum+'a')
> 
> have translation tables, so that
>     drivenum= chartodrivenum[driveletter]
>    and
>     driveletter= drivenumtoletter[drivenum]
> If you're willing to map (almost) all of the printable
> characters, you could get 46 controllers and 92 drives
> (I'd refuse to map  ', ", \  or space)
> 
> 
> You'd still be limited to one character, but it would, at least
> make it easy to have 26 controllers and 52 drives  (and that's
> just using upper and lower case characters!)

Hmm this propasal is not without reaons but the main obstacle
is currently just the option parsing code in ide. At some
point in time I just intend to rewrite it and to support
namings along the way of /dev/hdXX, where XX are simply digits.
The old names will be preserved as "backwards compatibility".

OK?
If you wish you could of course look in to this yourself, since
it's not *that* difficult. I would be really glad if someone did this.