network hang fixed for me

network hang fixed for me

[James Harper]

Whatever the changes were in the last 24 hours, they appear to have
fixed all of my network problems. Somewhere recently the problem of 'xm
destroy' not always destroying a domain was fixed too.

That puts an end to my two biggest annoyances with xen. Woohoo!

Just out of interest, what were the problems causing the network hang?

Thanks

James


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: network hang fixed for me

[Steven Hand]

>Whatever the changes were in the last 24 hours, they appear to have
>fixed all of my network problems. Somewhere recently the problem of 'xm
>destroy' not always destroying a domain was fixed too.
>
>That puts an end to my two biggest annoyances with xen. Woohoo!
>
>Just out of interest, what were the problems causing the network hang?

See cset 1.1315 -- basically the frontend/backend synchronization was
being rather too agreesive at 'batching' (aka holding off sending events 
to avoid taking too many hypercalls). In some circumstances (e.g. IP 
fragments) this could lead to the backend being unable to make any 
progress ('waiting' for the remaining fragments) while the frontend 
would happily fail to notify it of these. 

This only affects you when communicating domU<->dom0 ; for the default
(for us!) case where domU packets are bridged at L2 onto the wire, no
reassembly happens and hence no hang... if people use this configuration
exclusively they can enable the old behaviour (one of the new build 
options), but it's probably not worth it (we see very little performance
effect using the 'pessimistic' strategy all the time). 

A second bug has also just been fixed -- this was actually a linux 
bug, or at least a bug which arose due to linux code assuming something
which is generally true but not guaranteed to be (and in fact, isn't 
true when using xen virtual devices). 

Anyway: hopefully we're now moving closer to stabilising 2.0 beta and 
should be able to do the official 2.0 'release' soon. 


cheers,

S.



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

How to mmap xen pages?

[Rob Gardner]

I am try to convey a large data collection buffer from xen to a 
userspace program running in domain 0. I examined the xentrace facility 
and copied the code to mmap pages. My mmap fails, and upon trying out 
the xentrace program, I found that it fails in exactly the same way, ie, 
returns errno 11, "Resource Temporarily Unavailable". I have allocated 
my memory buffers just like common/trace.c does it, etc. Is this trace 
stuff not working anymore, did it ever work, and/or what is the 
recommended way to do what I want?

Thanks,

Rob Gardner
HP





-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: How to mmap xen pages?

[Keir Fraser]

For each page you wish to share, you must 
SHARE_PFN_WITH_DOMAIN(page, dom0).

Then, from dom0 userspace, you must map the appropriate pages using 
libxc's xc_map_foreign_range(). Grep around in libxc itself for some
example uses.

 -- Keir

> I am try to convey a large data collection buffer from xen to a 
> userspace program running in domain 0. I examined the xentrace facility 
> and copied the code to mmap pages. My mmap fails, and upon trying out 
> the xentrace program, I found that it fails in exactly the same way, ie, 
> returns errno 11, "Resource Temporarily Unavailable". I have allocated 
> my memory buffers just like common/trace.c does it, etc. Is this trace 
> stuff not working anymore, did it ever work, and/or what is the 
> recommended way to do what I want?
> 
> Thanks,
> 
> Rob Gardner
> HP
> 
> 
> 
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
> Project Admins to receive an Apple iPod Mini FREE for your judgement on
> who ports your project to Linux PPC the best. Sponsored by IBM.
> Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: How to mmap xen pages?

[Mark A. Williamson]

> I am try to convey a large data collection buffer from xen to a
> userspace program running in domain 0. I examined the xentrace facility
> and copied the code to mmap pages. My mmap fails, and upon trying out
> the xentrace program, I found that it fails in exactly the same way, ie,
> returns errno 11, "Resource Temporarily Unavailable".

Oh dear :-/

> I have allocated 
> my memory buffers just like common/trace.c does it, etc. Is this trace
> stuff not working anymore, did it ever work, and/or what is the
> recommended way to do what I want?

Used to work but it's apparently rotted.  It does SHARE_PFN_WITH_DOMAIN an Xen 
but the userspace code doesn't use the xc_map_foreign_range() that Keir 
described.

I'll check in a fix, assuming nobody beats me to it.

Cheers,
Mark


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php

Re: How to mmap xen pages?

[Keir Fraser]

Yes, I'm afraid that mmap'ing via /dev/mem doesn't work any more --
for safety I restrict that device to normally only map I/O memory.

 -- Keir

> > I am try to convey a large data collection buffer from xen to a
> > userspace program running in domain 0. I examined the xentrace facility
> > and copied the code to mmap pages. My mmap fails, and upon trying out
> > the xentrace program, I found that it fails in exactly the same way, ie,
> > returns errno 11, "Resource Temporarily Unavailable".
> 
> Oh dear :-/
> 
> > I have allocated 
> > my memory buffers just like common/trace.c does it, etc. Is this trace
> > stuff not working anymore, did it ever work, and/or what is the
> > recommended way to do what I want?
> 
> Used to work but it's apparently rotted.  It does SHARE_PFN_WITH_DOMAIN an Xen 
> but the userspace code doesn't use the xc_map_foreign_range() that Keir 
> described.
> 
> I'll check in a fix, assuming nobody beats me to it.
> 
> Cheers,
> Mark
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
> Project Admins to receive an Apple iPod Mini FREE for your judgement on
> who ports your project to Linux PPC the best. Sponsored by IBM.
> Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php