On Wed, 2023-07-05 at 22:35 -0400, Martin K. Petersen wrote: > > Laurence, > > > In the current code, If a device does not have protection, qla2xx > > will > > land up defaulting to a BUG() and will panic the system when > > sg_write_same is sent.This is because SCSI_PROT_NORMAL is matched > > and > > falls through to the BUG() call. The write_same command to a device > > without protection is not handled safely. > > I would like to understand why the driver PI code path was chosen for > a > SCSI_PROT_NORMAL cmnd. That doesn't seem right. > > > + case SCSI_PROT_NORMAL: > > + total_bytes = data_bytes; > > + break; > > case SCSI_PROT_READ_INSERT: > > case SCSI_PROT_WRITE_STRIP: > > total_bytes = data_bytes; > > All this transfer size wrangling in the driver should be removed and > replaced with a call to scsi_transfer_length() which takes the PI > size > into account. Martin, good questions I am waiting on Marvell to decide what to do about all this. The patch was focused only on avoiding the nasty effect of what you mention happens in that code path, causing customers to have system panics. The patch is to avoid the panic at this point with minimal impact to prior code functionality. Regards Laurence >
