Very Computer

Pawing through what little I understand of the kernel, I see that
it has IDE block mode disabled by default.  What is more interesting
is that it apparently disables block mode if it finds block mode en-
abled on boot-up.  So if I enable block mode via my system's CMOS
settings, then boot Linux, Linux will disable it.

How come?

I tried to get a look at what was going on, by the way, with hdparm.
But when I tested the timing with the -t option, my system locked up
tighter than a drum (Linux 1.2.0 + latest hdparm + VLB controller +
Connor 420 meg model - pretty dull, standard stuff).  I'm a little
shy about doing anything fancy with hdparm now :-).

--

: Pawing through what little I understand of the kernel, I see that
: it has IDE block mode disabled by default.  What is more interesting
: is that it apparently disables block mode if it finds block mode en-
: abled on boot-up.  So if I enable block mode via my system's CMOS
: settings, then boot Linux, Linux will disable it.

: How come?

: I tried to get a look at what was going on, by the way, with hdparm.
: But when I tested the timing with the -t option, my system locked up
: tighter than a drum (Linux 1.2.0 + latest hdparm + VLB controller +
: Connor 420 meg model - pretty dull, standard stuff).  I'm a little
: shy about doing anything fancy with hdparm now :-).

Well, one good reason to turn IDE block mode off is hard disks such as
your own -- or at least, such as most of the Conner IDEs I've seen.  Most
Conner IDEs I've played with do not respond properly to a system's
request for the maximum number of multiple sectors that can be read --
this is the MaxMultSect parameter in the kernel.  Conners may not respond
properly to other parameter requests as well, I don't know.

--
||Jonathan E. Brickman
  Run something sweet!  I don't care what it is, SOMETHING SWEET!
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<

                         ^^^

Well, that 128 is supposedly the maximum number of sectors that can be
read from the drive at once.  Unfortunately, it's rather suspect.  It
_may_ be valid, but I have extreme doubts; I've never heard of 128 sectors
being read from a drive at the same time!  It looks like 8 sectors are
being dealt with at most now;  this is reasonable in my experience (I've
heard of up to 32, 16 mostly).

Quote:> I'm curious about what you say about Conner because I just started pricing
> the Conner 850's and 1.275 gig models to get as a second drive.  Do I need
> to be concerned with either of these two models with Linux?

Only if you want the best performance.  Conners work; there's no doubt of
that -- but you will get better performance out of a Western Digital.
I've experience with W.D.'s, Conners, and Seagates; WD gives me the best
performance out of all of them.  WD has the biggest on-board cache as
well, by the way.  Seagate can be good, but I've seen an awful lot of OEM
Seagates fail after a year.  If you go Conner, be sure that you don't let
the kernel pick the multiple-sector read mode; I think that it did in the
case of your 1.1.81 installation, but it saw 128 and said "riiiiight, I
think we'll go with something rational", and picked 8.  The newest
kernels default to zero, and require the user to either recompile the
kernel for a default value other than zero, or also there is a utility
out (don't remember the name :) which will set it for you on-the-fly.

||Jonathan E. Brickman
  Run something sweet!  I don't care what it is, SOMETHING SWEET!
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<

Quote:>I tried to get a look at what was going on, by the way, with hdparm.
>But when I tested the timing with the -t option, my system locked up
>tighter than a drum (Linux 1.2.0 + latest hdparm + VLB controller +
>Connor 420 meg model - pretty dull, standard stuff).  I'm a little
>shy about doing anything fancy with hdparm now :-).

I had exactly the same lockup but with a Maxtor HD. But I became brave
and tried running hdparm NOT under X, and now it works. It is not
saying that X was the problem, or anything. Now I have -m 4 -u 1 set
with hdparm in /etc/rc.d/rc.local at bootup.

Just my 2 cents,

        Tamas

I believe the "128" value.. my own MC2112A (1Gig w/512KB cache) allows
up to "108", but it is fastest at around "32" or so (larger settings can
incur more latency or setup time, depending on the implementation inside
the drive).

Quote:>> I'm curious about what you say about Conner because I just started pricing
>> the Conner 850's and 1.275 gig models to get as a second drive.  Do I need
>> to be concerned with either of these two models with Linux?

If you already have a Conner, and like it, then stick with them.

I personally would not buy a Conner, due to the master/slave problems
they have with just about all other brands, but if I already had one
I'd get another cuz the only guarantee is that Conners work with Conners
(usually).

For extremely high performance and an awesome 5yr replacement warranty,
buy Micropolis.
--

For the latest Linux kernel: ftp.cs.helsinki.fi:/pub/Software/Linux/Kernel/v1.2
For Linux IDE (big/many) help, see:  linux-1.2.0/drivers/block/README.ide

Pretty dull stuff, or pretty dumb ? The Connor drive seems to be the problem. I have two of these CFS420A from different production series in different systems. Both with the same result: They do only work with slow ISA controllers and/or only with the "old" hd.c driver. After several sleepless nights I exchanged my private drive and put it into an old ISA DX-33 at my lab, now it does work with up to 1.4mbit (It had >=3.2mbit w/VLB).
These drives are really plug&play (plug them in and play with the parameters, jumpers, connectors, BIOS stuff,... all day long without any result :-)
BTW: My "backup" drive, a Conner CFS340 does play without any problems.

Quote:> I had exactly the same lockup but with a Maxtor HD. But I became brave
> and tried running hdparm NOT under X, and now it works. It is not
> saying that X was the problem, or anything. Now I have -m 4 -u 1 set
> with hdparm in /etc/rc.d/rc.local at bootup.

Might work for maxtor and others (even better throughput with -a0 to disable read-ahead). Rule of (my) thumb: multiple = 1/4 of max., usually read-ahead turned off. But if you're not shy try others and see what happens to hdparm -t ... (And don't bother me or any of the ide/hdparm-developers with your crashed drive afterwards)

--
Greetings from Beck'sTown, Germany,

Juergen

| For more information on our departments research activities, lectures, jobs |
| etc. have a look at our new WWW-service:    http://www.fe.uni-bremen.de/    |

Quote:>> >But when I tested the timing with the -t option, my system locked up
>> >tighter than a drum (Linux 1.2.0 + latest hdparm + VLB controller +

A long standing kernel shared memory bug was recently uncovered
as the cause of the "hdparm -t" lockups.. the fix may be in kernel
version 1.2.2 (?).
--

For the latest Linux kernel: ftp.cs.helsinki.fi:/pub/Software/Linux/Kernel/v1.2
For Linux IDE (big/many) help, see:  /usr/src/linux/drivers/block/README.ide