Blizzard 1230 + PFS-AIO Unexpected Phase Change

[GraveR]

I'm not sure if this is WinUAE-only (since I lack the real hardware to confirm).

I tried to make a SCSI-only Amiga setup in WinUAE 3.6.0 and during the installation of AsimCDFS (AsimCDFS v3.9 (1998-01-30)(Asimware Innovations)[h Digital Corruption][HD]) I found that the mentioned error always caused an error at the exact same step of extracting an archive.

Steps to reproduce:
1) Create a basic A1200 + Blizzard 1230 IV config
2) Activate the Blizzard IV SCSI kit and attach a hardfile
3) Partition the drive with PFS AIO and format it
4) Insert the AsimCDFS installdisk and use the supplied lx-command to extract asimcdfs.lha from the l/ directory on the floppy (df0:c/lx x df0:l/asimcdfs.lha)

I think there could be an issue with the SCSI kit emulation, because I can't seem to mount the emulated CD-ROM regardless of this issue. (Meaning, install on an IDE-disk and just having a CD-ROM on the SCSI kit); it hangs indefinitely. Changing the SCSI-controller to, for example, A4091 works without issues.

[Toni Wilen]

Add -scsilog -scsiemulog command line parameters and then recreate winuaelog.txt, thanks.

[GraveR]

Here it is. Thanks for looking into it.

[Toni Wilen]

Thanks. I had to duplicate it myself because it wasn't trivial issue. (and unexpectedly I managed to duplicate it instantly and reliably! Thats rare..)

Fixed. This write operation wrote two bytes to SCSI chip's FIFO and then started DMA (with size - 2 of total transfer size). Data in FIFO was ignored which caused phase mismatch.

I am not sure why SCSI Kit does this, possibly DMA requires long word alignment? But on the other hand DMA IO address and length is in words, not in longs.

[GraveR]

Nice to hear you've found what caused the issue. Also, I don't know why lx triggered that, because during normal installation of Workbench and such the problem did not occur. Unpacking the same archive with lzx just worked.

I wouldn't be surprised if there is an broader issue with this kit. I recall on real hardware, recording track-at-once to a CD writer would sometimes cause unexpected phase changes between writing tracks as well. They (Phase5/Ralph Schmidt(?)) blamed it on bad termination and ignored the issue.

Again, thanks for looking into it. I appreciate it!

[Toni Wilen]

Originally Posted by GraveR

Nice to hear you've found what caused the issue. Also, I don't know why lx triggered that, because during normal installation of Workbench and such the problem did not occur. Unpacking the same archive with lzx just worked.

lzx probably writes in smaller pieces or has different data aligment (if it really depends on long/word aligment)

I wouldn't be surprised if there is an broader issue with this kit. I recall on real hardware, recording track-at-once to a CD writer would sometimes cause unexpected phase changes between writing tracks as well. They (Phase5/Ralph Schmidt(?)) blamed it on bad termination and ignored the issue.

This problem is something different. Random, non-repeatable problems are generally always hardware problems (termination etc..)

SCSI "phase mismatch" only means that device wants something else than what host expected. (for example device wants more data but host expected transfer to end etc)