linux-mips-fnet
[Top] [All Lists]

Re: The bad news ...

To: linux-mips@fnet.fr
Subject: Re: The bad news ...
From: Systemkennung Linux <linux@mailhost.uni-koblenz.de>
Date: Wed, 8 May 1996 21:54:49 +0200 (MET DST)
In-reply-to: <199605081432.IAA10590@rover.village.org> from "Warner Losh" at May 8, 96 08:32:36 am
Hi,

> : The good news is that I now have two of 'em running :-)  So
> : far EATA DMA and EATA PIO are running.  I'll now swap
> : the hostadapters of my Acer and my 486 and make that Adaptec 1542
> : running ...
> 
> Cool!  How hard was this.  What did you need to do.  I'll be facing
> the same problem with the BusTech driver here in a little bit.

Well, basically this was pretty easy.  The PIO driver tried directly
to access the memory of the VGA card for scrolling up a line and
printing the word "reset".  I threw that shit away and it worked.

The DMA driver was a bit harder.  A virt_to_phys call was missing, a
zero pointer was being dereferenced and the primary caches weren't
correctly being flushed and the interrupt code was broken.

That bad thing is that the SCSI code isn't that broken as I already
knew.  It is a complete mess.  For a low-mem MIPS like my 8mb machine
sometimes after booting the machine and adding 48mb of swap (while
just init, five /sbin/getty, one ash and fsck are running) not enough
resources are left to continue for the SCSI code.

Ok, I hear you say why does this stupid guy have just 8mb of RAM in
box (Because I don't print my money myself ...) - if he' have more RAM
this wouldn't happen.  Wrong.  Completly wrong.  The problem is that
the SCSI code (and the network code and ...) run into heavy problems
when the physical memory is fragmented enough.  When this happens
usually at some hundred kilobytes memory are free plus some megs in
the buffercache (not dirty buffers!).

And now you know with what I spend the rest of time with; so there is
still no 1542 support yet...

   Ralf

<Prev in Thread] Current Thread [Next in Thread>