Re: [PATCH 7/7] libata: Add 32bit PIO support

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

Hello.

Robert Hancock wrote:

>> This matters for some controllers and in one or two cases almost doubles
>> PIO performance. Add a bmdma32 operations set we can inherit and 
>> activate
>> it for some controllers
>>
>> Signed-off-by: Alan Cox <alan@redhat.com>
>> ---
>>
>>  drivers/ata/ata_piix.c    |    2 +-
>>  drivers/ata/libata-sff.c  |   53 
>> +++++++++++++++++++++++++++++++++++++++++++++
>>  drivers/ata/pata_ali.c    |    6 +++--
>>  drivers/ata/pata_amd.c    |    4 ++-
>>  drivers/ata/pata_mpiix.c  |    3 ++-
>>  drivers/ata/pata_sil680.c |    4 ++-
>>  include/linux/libata.h    |    3 +++
>>  7 files changed, 66 insertions(+), 9 deletions(-)
>
> This can likely be enabled for more controllers (presumably all the 
> SFF-based SATA controllers as well as any other PATA that couldn't be 
> on a physical ISA bus).

   Why are you so sure? Some controllers only support 32-bit accesses 
when prefetch is enabled, some may not support it at all...

> This is a start though.
>
> Don't some of the VLB controller drivers have their own private 32-bit 
> PIO implementation? Those should likely be updated to use this support.

   You mean sequence of 3 reads of some non-datra register?

MBR, Sergei