[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] ne-2000 ethernet support

> On Tue, Feb 03, 2004 at 10:51:11AM -0500, Barry Silverman wrote:
> > Does the Xen hypervisor support the ne2k? The comments in SUPPORTED-CARDS
> > suggests that it does, but I don't see the code. 
> It doesn't seem to.

I think there was a ne2k driver at some point, but it turned out
not to work and got blown away. 

It should be pretty straight forward to get the stock Linux
driver going, but since its a PIO card the tx/rx routines will
have to be wrapped in a (un)map_domain_mem() to get the buffer
mapped into Xen's address space.

> > I am using bochs with the gdb-stub to symbolically debug the hypervisor. It
> > all works, except for the ethernet interface (Bochs only supports an ne2k).
> If you want a bit of speed up, it also works in Qemu although with the
> same lack of ne2k problem.  I've not played with the gdb-stub in Qemu
> for Xen purposes yet.

Alex is about to checkin a GDB serial stub for Xen that enables
remote debugging of Xen and guest OSes. Keep an eye on the
xeno-unstable.bk tree...

Can I ask what you're trying to debug in the hypervisor?

I notice that a previous email regarding support for the Native
POSIX Thread Library (NPTL) went unanswered.  I haven't looked in
to this, but I suspect that it may be using the new 'set thread
local storage' system call, which isn't to my knowledge in
standard 2.4.24 Linux.  Porting the patch to XenoLinux should be
fairly straight forward since Xen has LDT/GDT support.

*However* it's possible that it still won't work: I've heard
unconfirmed rumours that at least one of the new threads
libraries horribly abuses use of segmentation for thread local
storage.  Rumour has it that it creates and loads a 4GB segment
into gs, and then accesses location gs:0xfffffffc (-4).  It
relies on the segment wrapping at the top of the linear address
space! Barf... :-(

This interacts badly with Xen because we clip segments to prevent
access to the top 64MB of the virtual address space (where Xen
lives). Other VMMs e.g. (VMWare) almost certainly suffer the same

I'd be very interested to hear if the rumour is correct. If so,
we'll have to work out what the best way of tackling this is to
retain user space ABI compatibility. Of course, the best
solution is probably to beat some sense into whoever thought this
was a good idea...  I hope the rumour isn't true.


The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.