Processor Upgrades And SCSI Woes

Andy Dodd asks: "OK, this problem is quite strange and I don't know what to make of it. I just replaced my K6-2/300 with a 500 MHz chip, and now my system's reliability has gone way down under Linux. My first thought was that I had a re-marked counterfeit chip, but the crashes are not the random ones I'd expect from a chip that can't handle its clock rate. Under Linux, when I do anything that stresses my SCSI controller in any way, my system is prone to locking up. Starting any application runs a risk of crashing the system (culprits so far include gkrellm, Netscape, and sshd, one offense per app)." (Read on for more info..)

"I'm generally willing to tolerate the occasional crash, as they are rare. But CPU-intensive applications (Cyberlink PowerDVD and Quake 3 under Windows, recompiling apps and encoding MP3s under Linux) cause no trouble whatsoever.

Unfortunately, there are two cases where my system will hang with 100% certainty - fsck dies before completing 10% of my drive, and cdparanoia will always hang the system. If I drop the clock rate down to 400 MHz, everything works fine.

Any idea what's up with this? The correlation between SCSI use and crashes makes no sense to me, but it's most definitely there.

My setup includes: Epox MVP3E-M mobo with latest BIOS (VIA MVP3 chipset), Mylex/BusLogic FlashPoint LW SCSI controller also with latest BIOS, and the aforementioned AMD processor. (BTW, bus speed isn't an issue since both the old and new processors were running at 100MHz FSB.)"

Problem fixed. (No thanks to A.S.)

[Andy Dodd]

I can't believe it actually got posted - I gave up on getting posted. (This appeared over 3 weeks after submission.)

Fixed the problem about 3 days after I submitted my question. I found out that some people on linux-kernel had similar problems, the FlashPoint driver has some timing problems. I got a patch from one of the l-k guys, works like a charm.

It hasn't been integrated to the kernel tree, as they're still figuring out how to implement it without reducing performance for those who dont' have the problem.

disconnect set?

[pjc]

Check you've got disconnect set ON in the SCSI BIOS.

I had a similar problem when I put a new SCSI disk on my system - CPU intensive:fine, Disk intensive: hang. It turned out that disconnect was disabled by default for some devices (maybe DOS liked it that way??)

I think Linux expects to be able to do disconnect.

Power supply

[h2odragon]

Try a different power supply, or perhaps lose some devices temporarily, and see if that helps. I've got a motherboard/power supply combination that beomes unstable much as you describe, except it's anything RAM intensive that hangs it up. Running it with less load on the PS makes it rock solid.

Coincidence

[ralphclark]

I've just upgraded my Win98SE client to an AMD500 myself and I am just about to put in a Flashpoint LW too since I'm not too happy with the EIDE performance (Norton Utilities tells me the disk only does physical reads at 2MB/sec!!). I was only waiting till I can figure out how to replace the NCR BIOS in the AOpen AX59Pro motherboard with a suitable BIOS image for the Flashpoint (the SCSI card I have is a BIOSless OEM version).

Now I'm wondering if it's worth the trouble - if it crashes Linux then Win98 doesn't stand a chance.

I'd surely appreciate it if you could post more details about your hardware configuration. You should include everything - the tech specs from the RAM's datasheet, your PSU's power rating, the precise markings on your CPU, the make model and revision of your motherboard etc. etc.

Consciousness is not what it thinks it is
Thought exists only as an abstraction

CPU voltage?

[TallGuy]

Are you sure the board supports the voltage needed for the CPU? I think the 500 MHz K6-III needs a lower voltage than your original cpu.
I highly doubt your scsi chain has anything to do with the problem, since it works at 300 and 400 MHz. (I'm assuming you're keeping the PCI bus pegged at 33 MHz, if not, that may very well be your problem)

Check Low Level Settings

[randombit]

I was having very nasty problems with a (brand new) 810A SCSI card and an old CD-R I got for X-mas: I/O timeouts, things like mount, etc would go into uninteruptable sleeps, etc. I changed my kernel configuration so "default tagged command queue depth", "maximum queued commands", and the scyncronous transfer Mhz to levels significantly lower than the defaults. Also, I enabled normal I/O (vs memory mapped I/O). Probably the exact options are different on your card, but in any case it's probably worth a shot. It may lower performance significantly but it beats crashes.

Ram

[nowindowz]

Three things I would suspect. How good is the ram is it really good name brand 8ns or cheap 10ns crap. Second how good is your heat sink and fan? Third I suppect the via chipset, every via I have had has gave me greif and ever board with a via chipset I have sold in my shop has been returned within 2 weeks. Also you could have a bad cpu unlikely, also as the guy in another post said check your voltage. If nothing else chunk it and go Intel, I used to be Mr AMD, but lately I have nothing but trouble with them, I will not even thouch a computer if it has a Atlon.

Motherboard problem?

[hansendc]

I spend several months helping a friend to diagnose his dual PIII system. We tore our hir out trying to diagnose it. He experienced problems close to yours. He especially had problems compiling kernels.
Anyway, after replacing both CPUs, the SCSI card, and the motherboard twice he finally got a completely new motherboard. It worked(and still does) beautifully.
Moral of the story: when you get wacky errors, check the motherboard.

Active termination

[Mark F. Komarinski]

You may want to make sure that the bus is terminated with an active termination. At least with Adaptec cards, this makes a real difference. Fortunately, many newer drives come with termination on-board, so you may just need to turn it on at the last drive on the chain. Also make sure there's power on the bus for the termination.

If your drive doesn't support it, you can buy a physical terminator for the chain for maybe $50 at any good computer store.

Install the VIA PCI Bridge Patch

[biggaijin]

I recently bought a Micron Millenia with PIII 667 and a VIA chipset with 100 MH FSB. Wow, was I disappointed to see two different Linux distributions (RH6 and SuSE6.3) both repeatedly lock up the display and keyboard when I did the same sorts of things that you describe. Finally, I noticed the special PCI setting for the VIA PCI bridge when I was building a new kernel. This patch tests and splits the timing for DMA in a non-standard way for VIA chips. So far, it seems to work for me. Read the description and follow the installation instructions when you do you make config, and it may solve your timing problems. Good luck.

Stupid Stuf??

[photozz]

I hate to ask, but most of the time, it's the stupid stuff that will do you in. Did you check the interupt and timeout settings? might have been fine at a lower CPU speed, then showed up when you upgraded. Can't tell you how many times upgrading to "Better" hardware often shows up problems that were already there.