Re: [Lse-tech] multi-queue scheduler update

[Andi K. <ak...@su...>]

On Tue, Jan 23, 2001 at 05:53:55PM -0200, Marcelo Tosatti wrote:
> The right thing is to make the ext2 functions flush dirty buffers, if
> needed, after unlocking the sb lock. 

Hopefully you can do that without additional memory allocation.

This means that some random process can get blocked when you run out of 
elevator entries, even when it doesn't do any IO anymore.

A lot of this clever stuff IMHO needs a truly asynchronous replacement
for ll_rw_block()


-Andi