[parisc-linux] Progress - Update

Grant Grundler grundler@cup.hp.com
Wed, 01 Dec 1999 10:55:21 -0800

"John David Anglin" wrote:
> Also, we may need to flush the cache line for &var.  The coding in buffer.c
> which caused the original panic in xchg looks dubiousin this regoard:
>         /* Update the reuse list */
> 		tail->b_next_free = xchg(&reuse_list, NULL);
> 		reuse_list = bh;

If this data struct is only touched by processors, they are coherent
and flushing is not needed. (Except VM aliases...but I have to ignore
this case due to my ignorance).  If any driver DMAs this data and the
platform is I/O coherent (ie has ccio, epic or sba driver), processor
doesn't need to flush either. Processor should only need to flush on
*any* (SMP or not) platform which is not I/O coherent (assume the page
is mapped cacheable).


Grant Grundler
Unix Developement Lab