Xen users - May 2010 - blktap2, also broken in current pv_ops stable-2.6.32.x?

Hey!

As I''ve already written about two days ago, I''m having major
trouble getting
blktap2 to work properly. After trying xen-sources-2.6.32-xen-r1 from Gentoo
(where Andrew Lyon, the maintainer, has already confirmed that he''s
hitting
the same problem as me, i.e. the blktap2 driver is dysfunctional in the
kernel), I decided to give a pv_ops-Kernel a try, and checked out and
compiled the current xen/stable-2.6.32.x branch.

Blktap2 seemed to be working properly (at least I thought so yesterday
evening, testing it with the two hosts that are currently productively
running on the upgraded Dom0, both of which are PV and i386), but after
having given AMD64 Ubuntu a spin at installing today, I''m hitting a
similar
problem to what I''m seeing with the SLES sources: after a certain
amount/speed of I/O from the (HVM-)DomU, the system becomes unstable.

I''m not getting much sensible debugging info, because sometimes
I''m hitting
a "Kernel BUG, invalid opcode (in block-core.c)" in the Dom0, but the
system
still reacts somewhat (can be rebooted), sometimes nothing shows up on the
Dom0 logs, but the xm tools don''t react anymore and a reboot is
impossible,
sometimes the host completely freezes.

I''ve given up on blktap2 for the moment, and returned to simple
"phy"-devices for now (which seem to be running smoothly) so that my
hosts
(and my mailserver) can actually do some work; anyway, I guess my general
question would be if there''s actually anyone out there who''s
productively
using blktap2, and for whom it works properly. Any answer would matter to
me, as I''m currently evaluating whether I should actually test a little
more
(I haven''t started comparing the kernel config that I''m using
for the pv_ops
kernel against the respective config samples, that''d be a route to go
to
check whether I''m simply building a broken kernel), or rather let
things sit
as they are.

Thanks for any feedback!

--- Heiko.




_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Fri, May 7, 2010 at 12:37 AM, Heiko Wundram <modelnine@modelnine.org>
wrote:
> I guess my general
> question would be if there''s actually anyone out there
who''s productively
> using blktap2, and for whom it works properly.
Seeing lack of response, I guess the answer is "no".

I''m actually more concerned about the fact that the max number of
blktap2 device nodes that can be created are 255, and you have to
reboot dom0 after that. And the fact that legacy tap:aio also uses
blktap2 code, creating the same device nodes, thus subject to the same
limitation.

Let us know if you find a working solution.

-- 
Fajar

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Fri, May 7, 2010 at 12:37 AM, Heiko Wundram <modelnine@modelnine.org>
wrote:
> I decided to give a pv_ops-Kernel a try, and checked out and
> compiled the current xen/stable-2.6.32.x branch.
>
> Blktap2 seemed to be working properly (at least I thought so yesterday
> evening, testing it with the two hosts that are currently productively
> running on the upgraded Dom0, both of which are PV and i386), but after
> having given AMD64 Ubuntu a spin at installing today, I''m hitting
a similar
> problem to what I''m seeing with the SLES sources: after a certain
> amount/speed of I/O from the (HVM-)DomU, the system becomes unstable.
>
> I''m not getting much sensible debugging info, because sometimes
I''m hitting
> a "Kernel BUG, invalid opcode (in block-core.c)" in the Dom0, but
the system
> still reacts somewhat (can be rebooted), sometimes nothing shows up on the
> Dom0 logs, but the xm tools don''t react anymore and a reboot is
impossible,
> sometimes the host completely freezes.
Latest pv_ops based on 2.6.32.13 seems to work fine in my tests
regarding load test. There''s a nasty bug with "xm
block-attach" and
"xm block-detach" (which could lead to kernel panic, in the right
circumstances), but as long as you don''t use those commands it works
fine.

-- 
Fajar

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users