Re: 2.6.15-rc1: IDE: fix potential data corruption with SL82C105 interfaces

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

Hello.

Alan Cox wrote:
> On Llu, 2006-05-15 at 18:32 +0400, Sergei Shtylyov wrote:

>>    Heh, this driver also tries to cache the single PCI register per-channel 
>>like hpt366.c... This buglet concerns using fast PIO timings and is probably 
>>harmless though but needs to be fixed -- I'll send a patch soon...
>>    I wonder what is otherwise wrong with using 2 channels concurrently?

> I've not got any dual channel devices to test, and in fact I couldn't
> find anyone with dual channel stuff at all.

    Hm, I thought they're all dual channel, at least from W83C553F docs. We 
have this chip on several embedded boards -- I'll try to gain access to one of 
them when I get to this driver...

> The caching is one bug, the
> fact the reset hits both channels is the other I know about.

    Ah, that register 0x7E reset? Strangely, W83C55[34]F datasheets don't even 
mention it. :-/

> Otherwise the libata driver is fairly similar

    Found it, looking...

> although the timing is
> pre-computed from the documentation for the DMA modes.

    As these chips lack 66 MHz PCI support, this should be quite enough, I 
think... :-)

MBR, Sergei