SCSI, Ultra 160 LVD HBA must be forced SE with LVD U320 disks? crazy!

[DiTBho]

I am with a Adaptec-29160 Ultra160 PCI-HBA on a POWER8 server with Linux v5.17, and things are really weird with U320 SCSI disks.

• The PCI-HBA is an  U160-SCSI-HBA with an auto-sense LVD/SE interface.
• The SCSI cable I am using is an LVD one, certified by Ampex. It comes with a term.
• Disks are U320 68pin LVD. Auto-term is disabled. Wide mode is enabled.

Code: [Select]

server.pci.SCSI_HBA ==== cable ===+= disk0, id1
                                  += disk1, id2
                                  += disk2, id3
                                  += disk3, id4
                                  += term
The setup looks LVD ready, however, if I connect things this way .... I get tons of OverRun warnings from the kernel driver, which means the channel is noisy and doesn't properly work; while if I force disks in SE mode, things work perfectly, just slower than expected.

No problem with slower than expected, U160 and U320 require LVD, but why that bloody thing only work with SE if the HBA has an U160  interface?!? That's what I don't understand.

Okay, SCSI is an old topic in 2022, but ... does anyone have an idea about it? 

[DiTBho]

Code: [Select]

(scsi1:A:2:0)
                data overrun detected in Data-in phase.  Tag == 0x1.
                Have seen Data Phase.  Length = 8.  NumSGs = 1.
                sg[0] - Addr 0x02d67a020 : Length 32
(scsi1:A:2:0)
                data overrun detected in Data-in phase.  Tag == 0x2.
(scsi1:A:2:0)
                Have seen Data Phase.  Length = 8.  NumSGs = 1.
                sg[0] - Addr 0x02d67a020 : Length 32
...
Example of messages reported by the kernel with all the disks in LVD mode.

All the disks are the same model, with the same setup: or all in LVD mode, or all in SE mode.

[DiTBho]

Probably a "vintage computer" question 

Today, I tested the same cable and HBA with SCA disks, they worked in LVD mode without any problem.

[madires]

Adaptec SCSI HBAs are sometimes quite picky which can be mitigated by playing with the HBA settings (Ctrl-A). Back then we called them Adapdreck (Dreck is German for dirt, colloquial also junk).

[DiTBho]

Which is the best HBA for SCSI U160/U320?

[madires]

I have good experience with HBAs based on LSI (Symbios Logic, NCR) chips.

[CJay]

Check your cable, make 100% certain you've not got a bent or broken pin in *any* of the connectors, also check the terminator to see if it's an Active or Passive one, work out where the power is coming from if it's active and make sure there's only one device supplying it.

If it's not active then find one that is, repeat the above.

Pull the disks one by one and see if it improves, a dodgy LVD connector on any device/portion of the cable can easily cause problems.

SCSI is a finicky bus if something's not quite right, especially when you get to the higher speed variants.

[madires]

You might need a multimode terminator (also called LVD/SE terminator), not just an active terminator.

[DiTBho]

OK, following your advice, I bought:
- two brand new multi-mode terminators
- two brand new U320-certified cables without terminator
- two LSI PCI-X U320 HBAs

I will try soon 

[nightfire]

Also the cable must not be neatly and nicely folded because of overtalk- simply let it loose and long...