Hi Tony, are you familiar with how the xencomm stuff works? It's pretty
straightforward: the kernel constructs a scatter/gather list containing
physical addresses to map virtual-contiguous userspace buffers. Those
lists are created in arch/powerpc/platforms/xen/hcall.c .
At first glance, it appears the DOM0_GETDOMAININFOLIST dom0 op is being
handled correctly, so it looks like you'll need to dig in to that a
On Tue, 2006-09-05 at 14:23 +1000, Tony Breeds wrote:
> Hi All,
> I've been looking at the "xm top" command. Currently this fails
> with "Failed to retrieve statistics from libxenstat". libxenstat makes
> a getdomaininfo_list hypercall. Doing a little digging it seems the
> copy_to_guest_offset() call (dom0_ops.c:437) is returning -EFAULT.
> If I compare getdomaininfo with getdomaininfo_list it looks to me that
> the main difference is the destination for of the results.
> getdomaininfo(), overwrites the u_dom0_op of the hypercall
> getdomaininfolist(), places the results into a userspace buffer, whose
> address comes in via the u_dom0_op.
> I've poked around in xencomm_copy_to_guest and it seems that all the
> addresses are XENCOMM_INVALID. When makes be think that the address
> we're trying to write to is bogus for some reason.
*All* of them? You might see how many bytes xencomm_create() is
handling, since it only marks *unused* addresses as INVALID.
> I'm not really sure I know what my question is other than to ask for
> pointers on how narrow down what's missing on the ppc side to make this
> type of hypercall work.
> I'm not sure I'm on the right track here but I don't think that we're
> passing copy_to_guest_offset(), a pointer that can successfully be cast
> as a xencomm_desc.
Try this to start:
diff -r 199b457c4325 drivers/xen/core/xencomm.c
--- a/drivers/xen/core/xencomm.c Mon Aug 28 18:43:52 2006 -0500
+++ b/drivers/xen/core/xencomm.c Tue Sep 05 10:51:02 2006 -0500
@@ -21,7 +21,7 @@
+int xencomm_debug = 1;
/* translate virtual address to physical address */
static unsigned long xen_vaddr_to_paddr(unsigned long vaddr)
(Hmm, looks like all those printks are missing log levels. Patch would
be happily accepted. :)
IBM Linux Technology Center
Xen-ppc-devel mailing list