[Qemu-devel] Re: [PATCH] Fix sparc booting with no CD in drive

[Paul Brook]

On Tuesday 15 August 2006 16:45, Blue Swirl wrote:
> >Why are we getting reads when no data is available? The command should
> >already
> >have completed.
>
> I don't know. Should all SCSI commands return something with the transfer
> mechanism? Or is it a bug in OpenBIOS, should it get the SCSI error status
> somehow without transferring data?

The scsi_read_data and scsi_write_data routines correspond directly to the 
SCSI DATA IN and DATA OUT phases.

It's up to the HBA (ie. esp) emulation to keep track of the SCSI phase based 
on the return value of scsi_send_command and whether the command completion 
routine has been called.
It's also up to the HBA to implement the SCSI STATUS phase based on the sense 
code passed to the completion routine. Calling completion routine with 
SCSI_REASON_DONE indicates that the command has finished (success is 
determined by the sense code) and the disk is ready for the next command.

Commands can fail at any time, including before any data has been transferred. 
On LSI HBA this causes an invalid phase exception to be raised when data 
transfer is attempted. I don't remember offhand how the ESP controller 
handles this. There are also some commands that complete successfully without 
transfering any data.

> The background here is that OpenBIOS probes for SCSI devices, but when
> probing of capacity for nonexistent CDROM, the command fails.

The read capacity command deliberately fails when there is no media present.

Paul