Thanks Yi,
On Nov 9, 2006, at 5:22 PM, geyi wrote:


We made several tests on the save/restore with different guest domain configuration. Here is the problems we meet in the tests:

1. Cannot get the pfn list on large domain (domain memory >= 1GB). The xc_get_pfn_list will fail for large domain, because the xemcomm_create only allocate 1 page for the address mapping table between two domains. That restricts the pfn_list size transferred in one hcall to less than 2MB. And in current implementation, we can't transfer the pfn_list by two hcalls. So to support the large domain save/restore, either the getmemlist hcall should be modified to support partially pfn_list transfer or the xencomm allocate more pages for the address mapping in an hcall. What's you advice?
This is the kernel message we got from domain 0:
        xencomm_init: could only translate 2093048 of 2129920 bytes
        xencomm_init failure: -28
This is a known problem, when the dust settles on out next Xen upstream merge (in progress), I hope to have this interface fixed.

2. We tried to restore a domain with disk root file system. I used Hao's disk image file for the testing. The restored domain will trap into Data Access Exception(300). We found that's because the global pointer "shared" was not NULL before gnttab_resume is called in kernel resuming. The domain can be restored successfully when we clear the value of shared to NULL in gnttab_suspend.

Hmm, x86 does not check for NULL, it is inserted in the existing address space.
I think setting it to NULL is a good idea.
Good catch.
Can you resend the diff below as a "signed-off" patch to the list?

diff -r b03ebb287fa9 drivers/xen/core/gnttab.c
--- a/drivers/xen/core/gnttab.c Fri Nov 03 17:58:22 2006 -0500
+++ b/drivers/xen/core/gnttab.c Thu Nov 09 15:40:35 2006 -0500
@@ -426,7 +426,9 @@ int gnttab_suspend(void)
 int gnttab_suspend(void)
-       return arch_gnttab_suspend(shared);
+       int ret =  arch_gnttab_suspend(shared);
+       shared = NULL;
+       return ret;
 #if !defined(__ia64__)
        apply_to_page_range(&init_mm, (unsigned long)shared,
Xen-ppc-devel mailing list

Xen-ppc-devel mailing list

Reply via email to