I should reiterate that pressing the left arrow key 3 times to get back to the home screen never results in a hang.

feipoa wrote on 2020-04-30, 08:33:

I should reiterate that pressing the left arrow key 3 times to get back to the home screen never results in a hang.

I thought that that was your initial assumption, but later you said it's completely random.

Pressing the left arrow has never resulted in a hung system. it was the pressing of the ESC key that is difficult to reproduce. Sometimes it doesn't yield a hang, sometimes it does. Just like the 3Com sometimes shows up under ISA/VLB, sometimes it does not. The fault is around the ESC key, just don't know why it doesn't always cause the hang.

So far it hasn't hanged with ESC and the new version you just uploaded.

Spoke too soon!

EDIT: I'm going to disable L2 cache and slow down the RAM read/write speed to see if my timings are too aggressive. I doubt it though because usually Win95 will crash on boot if that's the case.

feipoa wrote on 2020-04-30, 08:41:

So far it hasn't hanged with ESC and the new version you just uploaded.

OK, so the new version I posted a few minutes ago doesn't hang during start anymore, correct?
Now the situation with Esc, this is really very strange as the handling of Esc/Left arroy keys is absolutely same. So many surprises in the DOS world...

I tried to reproduce the Esc issue several times on my system, but can't get it to hung.. Must be something pretty specific to that system, perhaps not worth spending more time on it...

I think I'm getting close. The error is only happening now for the instance in which the 3Com is not found. And the arrows are now causing the hang. Sorry for the misleading info - it is very difficult to troubleshoot this. Each hang requires a reboot, etc. I think if you're able to figure out why the 3Com is not always detected, that would solve the issue. ANd now, if I disable check for 3Com, there is no hang, even when "No Devices Found". But with check for 3Com, if the card is not found, HWiNFO will hang on exit (even with left arrow), but if the 3Com is found, there is no hang. This is the trend I'm seeing anyway. I've tested for this numerous times now.

feipoa wrote on 2020-04-30, 08:59:

I think I'm getting close. The error is only happening now for the instance in which the 3Com is not found. And the arrows are now causing the hang. Sorry for the misleading info - it is very difficult to troubleshoot this. Each hang requires a reboot, etc. I think if you're able to figure out why the 3Com is not always detected, that would solve the issue. ANd now, if I disable check for 3Com, there is no hang, even when "No Devices Found". But with check for 3Com, if the card is not found, HWiNFO will hang on exit (even with left arrow), but if the 3Com is found, there is no hang. This is the trend I'm seeing anyway. I've tested for this numerous times now.

Thanks! Yeah, I well understand how difficult it is to debug issues here, so your feedback (no matter what) is very appreciated. Not sure yet how to solve this problem, I will need to check the documents and resources in more detail to see if I can come up with something. So if you'd like to switch to a different system now, go ahead and we'll come back to this later if possible.

Plan to put the DRx2 in there now.

I think the issue with HWiNFO not detecting the 3Com some of the time is related to system speed. If I disable L2 cache and set the RAM timings slow, HWiNFO always found the 3Com card.

Maybe.. the 3Com card has a pretty specific protocol for accessing it. Initially it requires a special ISA Activation procedure via other ports, only then it becomes active...

Moving on now. The DLC, DRx2, SXL2, and BL3 are probably the most difficult CPUs I have to test. They all have some special quirk with setting up the L1 cache, non-cacheable range, etc. Some only work with certain FPU's when in 2x mode, some need to use the BARB software method to flush the cache, but sometimes only if the 1540 SCSI controller is installed - 1520 is a lot easier to test with. The BL3 is without a doubt the most difficult of the group and I will need to research several dozens of pages of notes to get it working optimally again. That CPU was particular about what order drivers were loaded in. It also needs to have the cache flush on every I/O in Windows, which makes it slow, but the 75 MHz BL3 I have can be clocked up to 110 MHz to compensate somewhat.

Wow, then I guess it will be a miracle if HWiNFO will work there with all its low-level access methods and tests...

Here's the log for the DRx2-66 running at 50 MHz.

L1 cache shows up at N/A. Should be 1 KB.

feipoa wrote on 2020-04-30, 09:51:

Here's the log for the DRx2-66 running at 50 MHz.

L1 cache shows up at N/A. Should be 1 KB.

Thanks! If you open the Mainboard info screen multiple times, will it perhaps show the L1 size sometimes?

Here a special build that will show an additional "[Cache Debug]:" line that will help me to tune the L1 cache size reporting.
It would be great if you could open the Mainboard info screen a few times and each time log the output as I believe those numbers might fluctuate..
[build pulled, newer available]

No, the L1 cache size does not appear. The good news is that under ISA, when it doesn't find the 3Com, the system does not hang upon exiting that submenu.

68.3 130.1 253.6 500.9 995.3
68.3 130.1 253.8 501.0 995.5
68.3 130.2 253.8 501.0 995.5
68.3 130.1 253.8 500.9 995.3
68.3 130.2 253.8 501.1 995.3
68.3 130.1 253.8 500.9 995.5

Thanks! Those are relative numbers for memory access latency. 68.3 is for 1K block, 130.1 for 2K. The difference in access latency is almost none (130.1 / 68.3 = 1.90x ratio for 2K). I'd expect something much >>2.0x for 2K.
So I'm wondering is L1 enabled there?

Most programs don't pickup the 1K L1 cache of the DRx2 or DLC unless you set the L2 timings to really slow. Usually the write speed (or maybe it was move speed) is easier to see a difference in. I'll run Doom with L1 enabled and disabled to ensure that its on.