[haiku-development] Re: ATA vs. IDE resolved?
- From: Niels Reedijk <niels.reedijk@xxxxxxxxx>
- To: haiku-development@xxxxxxxxxxxxx
- Date: Mon, 17 Aug 2009 12:09:15 +0200
Hi,
2009/8/17 Axel Dörfler <axeld@xxxxxxxxxxxxxxxx>:
> Niels Reedijk <niels.reedijk@xxxxxxxxx> wrote:
>> I might have missed it, but up to now there hasn't been a decision on
>> ATA vs. IDE. I doubt that this is a matter to be voted over, but
>> rather one that requires a technical discussion between those that
>> know.
>>
>> What I do suggest is to resolve the question which one to use ASAP,
>> especially if it turns out to be ATA, in which case the nightlies can
>> be switched so that it can be tested by a larger group of people.
>
> Both have their pros and cons:
> ATA:
> - not as well tested
> - does make my EeePC hang (during ATA reset)
> + works on my desktop, where IDE does not (crashes)
> + solved an interrupt race condition in ata_adapter
> + cleaner/simpler code
>
> IDE:
> - does crash on some hardware
> + supposed to be faster than ATA
> + is well tested on a variety of hardware
>
> So our options would be: a) fix interrupt race, and crashing bug in
> IDE, or b) have ATA more tested, and resolve issues on EeePC (and
> probably others that show up during testing).
> Oh, and we would need to find someone actually doing the work :-)
Actually, it seems to me that the choice is more or less guesswork on
which one will work on most machines.
To me personally it seems that ATA should be the default because that
is what it will be anyway. I hope that we can have a middle way by
making sure both the stacks can be in the kernel, and that using the
boot menu people can choose to use the IDE.
Of course I know little about the internal technicalities, but I guess
it would be impossible to have this be configurable by a kernel
settings file since you probably need ata/ide to get this file.
However, perhaps we can gather information during the coming weeks to
build in some automated guesswork based on what hardware we know is
working.
Would that be implementable?
Kind regards,
Niels
Other related posts:
