kern/75603: 5.3 kernel crash

[Phil Oleson]

The following reply was made to PR kern/75603; it has been noted by GNATS.

From: Phil Oleson <oz at nixil.net>
To: DJ Gregor <dgregor at interhack.com>
Cc: freebsd-gnats-submit at FreeBSD.org
Subject: Re: kern/75603: 5.3 kernel crash
Date: Tue, 05 Apr 2005 18:40:20 -0600

 I did try to debug this some, though I got distracted by work related 
 tasks.  In my opinion,
 scsi_calc_syncsrate() has a div by 0 possibility, where the old ifdef'd 
 code explicitly avoided
 this prossibility.   something about this code in bt.c which sets 
 sync_period.  The 4.x tree
 called scsi_calc_syncparam() which is safer in this regard, but it's 
 ifdef'd out of the 5.x+ tree.
 and I dont think it's doing the right thing.  But this is as far as I got.
 
     Phil.
 
 bt.c:
 <snip>
 #ifdef  CAM_NEW_TRAN_CODE
         cts->protocol = PROTO_SCSI;
         cts->protocol_version = SCSI_REV_2;
         cts->transport = XPORT_SPI;
         cts->transport_version = 2;
 
         spi->sync_period = sync_period;
         spi->valid |= CTS_SPI_VALID_SYNC_RATE;
         spi->sync_offset = sync_offset;
         spi->valid |= CTS_SPI_VALID_SYNC_OFFSET;
 
         spi->valid |= CTS_SPI_VALID_BUS_WIDTH;
         spi->bus_width = bus_width;
 
         if (cts->ccb_h.target_lun != CAM_LUN_WILDCARD) {
                 scsi->valid = CTS_SCSI_VALID_TQ;
                 spi->valid |= CTS_SPI_VALID_DISC;
         } else
                 scsi->valid = 0;
 
 #else
         /* Convert ns value to standard SCSI sync rate */
         if (cts->sync_offset != 0)
                 cts->sync_period = scsi_calc_syncparam(sync_period);
         else
                 cts->sync_period = 0;
         cts->sync_offset = sync_offset;
         cts->bus_width = MSG_EXT_WDTR_BUS_8_BIT;
 
         cts->valid = CCB_TRANS_SYNC_RATE_VALID
                    | CCB_TRANS_SYNC_OFFSET_VALID
                    | CCB_TRANS_BUS_WIDTH_VALID;
 
 #endif
 </snip>