DAC960 on IA/64

Date: Tue, 24 Jul 2001 14:58:14 +0200 (CEST)
From: Martin Wilck <[email protected]>

Hi Leonard,

I have made a very small step forward in debugging the problems. After reading this message from Linux to the LK mailinglist
<http://www.uwsg.indiana.edu/hypermail/linux/kernel/0107.2/0533.html>
I suspected that SMP may have something to do with it.

So I repeated our test with the stock RedHat UP kernel (2.4.3), and with much less catastrophic results. I got no more hanging processes, no system crash, "just" some data errors that suggest that 18 512-byte blocks in a 2GB file were pointing at a wrong disk area (there were valid bit patterns there, but patterns that should appear elsewhere in the file).

This may look like a VFS or ext2-related problem, but similar tests on other controllers have revealed nothing of the kind.

Interestingly, the driver also reported
no "vendor-specific" or other errors during the test.

I cannot exclude 100% that this test went better than the previous tests simply by "luck", but it seems unlikely, because the system survived 12 hours of heavy copy-compare whereas it always showed uninterruptibly sleeping processes or even total crashes after ca. 5 hours with SMP.

Hope that helps a bit, regards,
Martin

Hmmm. That's certainly a possibility, though I don't fully undertand yet the form of the failure. I posted a patch for Linux kernel 2.4.7 on my web page last night that updates the 2.4.10 DAC960 driver to use the new completion code. Please try 2.4.7 with that patch and see if it works in UP or SMP.

Leonard