bug in hpecc() ?

[Dave Martindale]

In article <672 at bbncc5.UUCP> cspencer at bbncc5.UUCP (Clifford Spencer) writes:
>In 4.2 and ULTRIX hp.c(hpecc) I see the following happening under 
>the ECC case:
>
>	npf--;		/* because block in error is previous block */
>	*
>	mpte = mbp->mba_map[npf+btop(byte)];
>	* (correct error here)
>	*
>	npf++;

npf is calculated as the number of sectors of data that have already
been transferred.  In the case of a BSE (bad sector) error, this
does not include the block with the problem since the transfer is aborted
before moving any data.  But for an ECC error, the data in the block
*has* been transferred to memory, and the controller's ECC registers
tell you how to correct that last sector of data.  Thus "npf" is really
pointing to the *following* sector, and is decremented to point to the
sector with the error.

>Why the fiddling with npf? What if npf is initially zero and byte < 512? 
>If the error occurs in the first page of a transfer, isn't npf going to 
>go negative and we will reference mba_map[-1] ?  Can someone explain this?  

In the case of an ECC error, npf cannot be zero since at least one sector
has already been transferred into memory.

	Dave Martindale