Re: [PATCH] highmem I/O for ide-pmac.c

[Paul Mackerras <paulus@samba.org>]

Jens Axboe writes:

> Doesn't look like it's needed at all, at least you never turn on highmem
> I/O with ide_toggle_bounce() :-)

Foo, neither I do. :(

> BTW, it would be ok to export that from ide-dma.c instead of duplicating
> it in ide-pmac.

Looking at it again, both ide_build_sglist and ide_raw_build_sglist do
*almost* what we want.  If ide-pmac used hwif->sg_table instead of
pmif->sg_table, and if ide_[raw_]build_sglist were exported and took
the maximum number of entries as a parameter instead of using the
PRD_ENTRIES constant, then ide-pmac wouldn't need to have its own
versions of those routines.  Would those changes be OK?

Ben, any reason why we have to use pmif->sg_table rather than
hwif->sg_table?

> Also, can you grow sg segments indefinitely?

Each DBDMA (descriptor-based DMA) command has a 16-byte length field,
so is limited to 65535 bytes.  There is code in pmac_ide_build_dmatable
to create multiple DBDMA commands from a single scatterlist element if
necessary.  We should limit the lengths of the segments instead, then
we could have a 1-1 correspondence of scatterlist elements to DBDMA
command blocks.

Paul.