PRINS%HLERUL5.BITNET@cunyvm.cuny.edu (Ryko Prins, CRI, Leiden, Netherlands) (02/16/89)
Problem:
My Mega 2 can only access the Megafile 60 when I have just powered up,
after some minutes I get unrecoverable disk errors.
If I recompile my system using the TRACE option for the driver it does some
disk access, but after that a kernel panic is generated and everyting stops.
Hardware problems are ruled out; both TOS and Aladin work fine on the same
partitions.
Hypothesis:
There is a timing problem in the Minix disk driver.
The Megafile 60 is the only drive in the Atari line with an average seek time
of 61 Ms compared to 65 Ms on all other Atari drives, so I suspect some diffe-
rences in timing.
If the drive is cold, timing could be slightly different, (just within the
margins in which Minix can operate) than if the drive is warm.
By the way, did you know the hard disk driver p o l l s for interrupts? Which
would take account for the many system CPU seconds you see being used in the
CNTRL-ALT-PF1 display for process HARDWARE (if I am correct). Anyway, I guess it
costs a lot of system performance the way it is implemented.
Well, I don't have much time for it, but I think I should take a serious look
at the Minix-ST hard disk driver now, which will take some time, unless there
are other people with the same problems, so we could share some work.
Any comments? Dou you think I am thinking along the right line?
Greetings, Ryko Prins.johan@nlgvax.UUCP (Johan Stevenson) (02/17/89)
In article <8899@louie.udel.EDU> PRINS%HLERUL5.BITNET@cunyvm.cuny.edu (Ryko Prins, CRI, Leiden, Netherlands) writes: >My Mega 2 can only access the Megafile 60 when I have just powered up, >after some minutes I get unrecoverable disk errors. Are more people experiencing problems with Megafile 60? >By the way, did you know the hard disk driver p o l l s for interrupts? Yes, each disk request generates 7 (8 for the Supra I believe) interrupts. 6 interrupts come normally immediately, so I wait a few cycles. A normal interrupt costs hundreds/thousands of instructions. This way it takes only 10 or so. >Which would take account for the many system CPU seconds you see being used in the >CNTRL-ALT-PF1 display for process HARDWARE (if I am correct). No. Since that counter increases as well if you have no hard disk at all. >Anyway, I guess it >costs a lot of system performance the way it is implemented. No. This is by far the cheapest way. >Well, I don't have much time for it, but I think I should take a serious look >at the Minix-ST hard disk driver now, which will take some time, unless there >are other people with the same problems, so we could share some work. I always like improvements. -- Johan W. Stevenson johan@pcg.philips.nl Philips Research
PRINS%HLERUL5.BITNET@cunyvm.cuny.edu (Ryko Prins, CRI, Leiden, Netherlands) (02/20/89)
> >CNTRL-ALT-PF1 display for process HARDWARE (if I am correct). > No. Since that counter increases as well if you have no hard disk > at all. Sorry, I was wrong, I meant the process WINCHESTER. The process HARDWARE will be the null process or whatever it is called in Unix terminology. Sorry, I am a VMS man, and have not very much experience with Unix or Minix, so I look at it from a VMS point of view. In VMS one problably would use AST's (Asynchronous System Traps, I don't know yet if there is a Minix homologue). > No. This is by far the cheapest way. To get a better understanding of the ATARI ST hard disk handling I am currently disassembling (and interpreting) AHDI.PRG to compare it to the Minix disk driver (AHDI has no problems with my Megafile 60). It seems AHDI does the same kind of polling of the drive status, so it really seems the cheapest way. Greetings, Ryko Prins.