Subject: Re: RFC: patch to change the disk I/O scheduler via dkctl(8)
To: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
From: Juan RP <juan@xtrarom.org>
List: tech-kern
Date: 10/10/2005 16:51:17
On Thu, 29 Sep 2005 19:32:51 +0900
YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp> wrote:

> - is there any point to change "sortmethod" on the fly?
>   it's characteristics of the device.

I removed this, and I modified bufq_set_strat to accept
an int argument to set the sort method in the driver
and not via dkctl(8).

> - if you create a new file for bufq_ routines,
>   it should contain bufq_alloc etc as well.

Done.

> - string -> BUFQ_* conversion shouldn't be done in bufq_set_strat.
>   it's better to make bufq_alloc itsself take a string.

I could do that, but I think it's better to not modify the API
(and a bunch of drivers), anyway I don't have any problem with
modifying all drivers if you think it's better.

> - bufq_set_strat's argument shouldn't be a void pointer.

Fixed. I moved the bufq_*_strat prototypes to sys/bufq.h.

> - bufq_set_strat seems broken.
>   have you read bufq_alloc/free?> 

I believe this is fixed in my latest patch.

New patch for review with all those things (hopefully) fixed at:

http://www.xtrarom.org/bufq_on_the_fly.diff

Thanks for reviewing the patch and/or comments about this.