Subject: Re: amap memory allocation
To: YAMAMOTO Takashi <firstname.lastname@example.org>
From: Garrett D'Amore <email@example.com>
Date: 06/11/2006 20:43:03
YAMAMOTO Takashi wrote:
>> On Sun, Jun 11, 2006 at 02:45:45PM +0900, YAMAMOTO Takashi wrote:
>>> which includes:
>>> 1. implement solaris-like vmem.
>> Could you elaborate a bit what this is and why you are changing things?
>> I have no idea what solaris "vmem" is.
> for "what's vmem", see:
> as i wrote in the previous mail, my primary purpose at this point is
> to make amap use kernel_map rather than kmem_map.
> (the latter is more restricted and there's no point for amap to use it.)
> for long term, i'd like to replace extent(9) and malloc(9),
> but my implementation is still far from it.
> YAMAMOTO Takashi
The Solaris kernel memory allocator is *very* robust, and quite
efficient. I used it to replace a busted allocator in thin-client, and
much joy was held at suddenly new-found memory that gained by an
efficient allocator. (Efficient in both space and time.)
On the other hand, I've not looked seriously at our versions. Is there a
"need" to replace extent and malloc? I think extent is still needed at
least, because I don't think the Solaris vmem system really addresses
this need (at least Solaris 9 didn't -- they have separate code built on
top of kmem_alloc for that.)
The other thing is that the cost of free/alloc with a cache allocation
goes *way* down, but it requires more intelligence on the part of
drivers. (You have to specify a size with kmem_free().)
Finally, is there a reason that you couldn't make KM_NOSLEEP style
allocations work from interrupt context ala the Solaris kmem_alloc
(perhaps by going to a different allocation pool?)
I should probably look at your source in more detail. What would be
really nice would be to expose kmem_cache_alloc() - like implementation.
Garrett D'Amore, Principal Software Engineer
Tadpole Computer / Computing Technologies Division,
General Dynamics C4 Systems
Phone: 951 325-2134 Fax: 951 325-2191