lkml.org 
[lkml]   [2006]   [Oct]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH 6/13] KVM: memory slot management
Date
On Thursday 26 October 2006 19:27, Avi Kivity wrote:
> kvm defines memory in "slots", more or less corresponding to the DIMM
> slots.
>
> this allows us to:
>  - avoid the VGA hole at 640K
>  - add a pci framebuffer at runtime
>  - hotplug memory
>
> Signed-off-by: Yaniv Kamay <yaniv@qumranet.com>
> Signed-off-by: Avi Kivity <avi@qumranet.com>

To bring up the discussion about guest memory allocation again,
I'd like to make a case for using defining guest real memory
as host user, not a special in-kernel address space.

You're probably aware of many of these points, but I'd like
to list all that I can think of, in case you haven't thought
of them:

- no need to preallocate memory that the guest doesn't actually use.
- guest memory can be paged to disk.
- you can mmap files into multiple guest for fast communication
- you can use mmap host files as backing store for guest blockdevices,
including ext2 with the -o xip mount option to avoid double paging
- you can mmap packet sockets or similar to provide virtual networking
devices.
- you can use hugetlbfs inside of guests
- you can mmap simple devices (e.g. frame buffer) directly into
trusted guests without HW emulation.
- you can use gdb to debug the running guest address space
- no need for ioctl to access or allocate guest memory
- you can mmap a kernel image (MAP_PRIVATE) into multiple guests
and share instruction cache lines.
- the kernel code doesn't need special accessors, but can use
asm/uaccess.h.
- may be able to avoid a bunch of TLB flushes with nested page tables.

On the downside, I can see these points:

- As you mentioned guest size on 32 bit hosts is limited to around 1G.
- you probably have to rewrite your virtual MMU from scratch
- for optimal performance, pageable guests need something like the s390
pagex/pfault mechanism in the guest kernel.
- if you want a guest not to be paged out, you need privileges to do mlock.
- you can't use swap space in the guest if you want to avoid the
double paging problem (host needs to read a page from disk for the guest
to swap it out), or you'd have to implement a mechanism like Martin
Schwidefsky's page hints (cmm2) for s390.

Arnd <><
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-10-27 00:47    [W:0.075 / U:0.096 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site