[fm-discuss] Will the PCI diagnosis engine detect faulty PCI adaptors?

[Matty]

On Mon, 26 Jun 2006, Gavin Maltby wrote:

> On 06/26/06 01:04, Matty wrote:
>
>> I see that FMA was recently extended to diagnose problems on PCI, PCI-X and 
>> PCI-E buses. Does anyone happen to know if the FMA PCI code will be able to 
>> identify faulty PCI adaptors? The company I work for has on one more than 
>> one occassion spent endless hours trying to diagnose periodic
>> hardware glitches, which turned out to be the result of a flaky NIC,
>> HBA or SCSI controller. FMA is really really neat, and I am hopeful that
>> the FMA PCI-* support will be able to diagnose faulty adaptors on a PCI 
>> bus.
>
> From one of the engineers who did the recent PCIE work you mention:
>
> 	Yes, the FMA PCI code will be able to identify faulty PCI adapters,
> 	though with a couple of limitations:
>
> 	 - on x86 systems we currently only support fault diagnosis for
> 	   systems with PCI-E hostbridges (ie. this includes PCI-E cards
> 	   and PCI/PCI-X cards behind a PCI-E to PCI/PCI-X bridge, but not
> 	   PCI/PCI-X cards behind a native PCI or PCI-X hostbridge)
>
> 	 - The FMA PCI code is able to detact and diagnose for any card
> 	   those faults that are detected or reported on the bus (eg parity
> 	   errors, master/target aborts). In addition there is also a
> 	   relatively rare class of internal faults within a device
> 	   (eg corrupted ring buffer) that can only be detected by the
> 	   device's driver. If that driver is hardened such that it generates
> 	   the appropriate error reports, then FMA will also be able to
> 	   diagnose these faults.

Thanks for the information. Is anyone currently working on hardening the 
SCSI, FC and network device drivers that are currently in Nevada? Having 
hardened drivers and SMART support would greatly assist with diagnosing 
hardware problems. Also -- will the Solaris SMART implementation come with 
a utility to view the SMART attributes on a device (something similar to 
smartmontool's smartctl)?

Thanks,
- Ryan
--
UNIX Administrator
http://daemons.net/~matty