MFI and passthrough

[Scott Long]

Doug Ambrisko wrote:
> Scott Long writes:
> | The passthrough interface is really only meant for doing management
> | tasks like SMART monitoring and firmware flashes.  I've also seen it
> | used for low-duty devices like tape drives.  I do not recommend using it
> | to directly control disks in a primary fashion.  However, since this is
> | open source, I won't prevent you from trying =-)  Try the following
> | patch:
> | 
> | --- mfi_cam.c   12 Oct 2007 16:52:55 -0000      1.3
> | +++ mfi_cam.c   31 Oct 2007 03:42:25 -0000
> | @@ -344,9 +344,11 @@
> |                          command = ccb->csio.cdb_io.cdb_bytes[0];
> |                  if (command == INQUIRY) {
> |                          device = ccb->csio.data_ptr[0] & 0x1f;
> | +#if 0
> |                          if ((device == T_DIRECT) || (device == 
> | T_PROCESSOR))
> |                                  csio->data_ptr[0] =
> |                                       (device & 0xe0) | T_NODEVICE;
> | +#endif
> |                  }
> |                  break;
> |          }
> 
> BTW, it works great in this mode if you know what you are doing :-)
> 

Can you explain what that means?  I recommend against it because it's
not a well-tested configuration either in FreeBSD or in Dell.  It's not
clear, at least to me, how basic things like i/o errors get handled; 
does SCSI sense data get consumed by the controller firmware, or is it
passed through to the OS without problem?

> | I do believe that Dell does sell a direct attached disk option for
> | the 2950/1950 called the PERC5/e.  It's essentially an LSI MPT-SAS
> | controller that directly replaces the PERC5/i card that you have now.
> | It should be able to control all 6 disk slots, and can do both SAS
> | and SATA.
> 
> I've been told the PERC5/e and PERC5/i are the same except for PCI
> sub-device ID and are both the mfi(4) RAID controllers.  They do
> have a mpt(4) based card but it only supports 4 bays.  I'm not sure
> what it's real name is but we have some lying around for random
> testing.  I don't leave them in machines.
> 

We should get a definitive answer on this.

Scott