feipoa wrote on 2023-02-19, 07:51:
Thanks for your reply. It was not clear to me from the PS/2-HOT433 thread that mkarcher fixed the IRQ routing table in his BIOS patch of 2A4X5H21. I could not locate any mention of it in the dialogue. Maybe mkarcher only set you the BIOS with IRQ routing table fixes?
I'm sorry for any confusion about the IRQ routing. The 2A4X5H21 BIOS does not need a patched IRQ routing table. The talk about the fixing the IRQ rouing table was at the time we were using a BIOS image meant for a different UMC8881 board, with the ID 2A4X5G03. As the IRQ routing table is board specific, it is no surprise that the IRQ routing table from the "foreign" BIOS did not match the HOT-433. When we added the PS/2 mouse port, we found the updated HOT-433 BIOS (most likely a unofficial community build?) with the ID 2A4X5H21. This BIOS already contains the correct IRQ routing table, so no patch is required.
Yet, even with the BIOS with the correct IRQ routing table, the board turned out to be very picky about UDMA IDE controllers. We tried a Promise FastTrack controller, a Silicon Image controller and the Highpoint controller. The only controller that managed to run reliably was the HighPoint controller, and even that controller doesn't do busmaster transfers reliably at PCI@20. You don't need a BIOS patch to fix that issue, because no IDE busmaster transfers are performed before the CMOS settings are applied. Furthermore, even with the HighPoint controller, we experienced intermittent data corruption when L2 writeback is enabled. As L2WT is required, we are operating with 256MB of RAM, which is fully cached by the 1MB L2 cache.
Our HOT-433 uses an early revision of the UMC chipset (without EDO support), so maybe some of the difficulties are resolved with later revisions. Unfortunately, I don't remember whether we tested all those issues on the Biostar MB-8433UUD we have. That one has a newer UMC chipset revision.