Anthony Liguori
2005-Oct-04 04:43 UTC
[Xen-devel] [PATCH 0/2] Change xs_read_watch interfaces to support variable arguments
The following set of changes modify the read_watch interface in userspace and in the kernel to support variable arguments. This is in preparation for adding domid to @{introduce,release}Domain watches. This will allow backends to automatically reap on domain destruction which should eliminate a large number of the zombie problems we''re having on domain destruction since Xend no longer plays a role in freeing that memory. The interface for read_watch is now consistent between userspace and kernel space. It''s modelled after xs_read_directory. The returned array is indexable via XS_WATCH_* macros. While slightly awkward, IMHO it seems like the cleanest and most flexible option of the proposed interfaces. Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> Regards, Anthony Liguori _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Anthony Liguori
2005-Oct-04 04:48 UTC
Re: [Xen-devel] [PATCH 0/2] Change xs_read_watch interfaces to support variable arguments
I should mention that I ran the xenstore testsuite (which rocks btw) and tested the block devices and network devices. I haven''t tested blktap and vtpm but the changes are identical (and I don''t have any test cases for those). Regards, Anthony Liguori Anthony Liguori wrote:> The following set of changes modify the read_watch interface in > userspace and in the kernel to support variable arguments. This is in > preparation for adding domid to @{introduce,release}Domain watches. > This will allow backends to automatically reap on domain destruction > which should eliminate a large number of the zombie problems we''re > having on domain destruction since Xend no longer plays a role in > freeing that memory. > > The interface for read_watch is now consistent between userspace and > kernel space. It''s modelled after xs_read_directory. The returned > array is indexable via XS_WATCH_* macros. > > While slightly awkward, IMHO it seems like the cleanest and most > flexible option of the proposed interfaces. > > Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> > > Regards, > > Anthony Liguori > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Ewan Mellor
2005-Oct-04 06:55 UTC
Re: [Xen-devel] [PATCH 0/2] Change xs_read_watch interfaces to support variable arguments
On Mon, Oct 03, 2005 at 11:43:43PM -0500, Anthony Liguori wrote:> The following set of changes modify the read_watch interface in > userspace and in the kernel to support variable arguments. This is in > preparation for adding domid to @{introduce,release}Domain watches. > This will allow backends to automatically reap on domain destruction > which should eliminate a large number of the zombie problems we''re > having on domain destruction since Xend no longer plays a role in > freeing that memory.Hi Anthony, What you''ve done in your patch stands as a sensible thing on its own, I think, so I''m not objecting to that, but I am interested in why you think it will help with the zombies. On domain destruction, xend is _supposed_ to remove all the relevant entries of the store, and the backend drivers are _supposed_ to notice this and close themselves down. If you have evidence that either of those things aren''t happening, then we should regard that as a bug, independent of your patch. Have you seen that happening? Ewan. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Anthony Liguori
2005-Oct-04 14:34 UTC
Re: [Xen-devel] [PATCH 0/2] Change xs_read_watch interfaces to support variable arguments
Ewan Mellor wrote:>On domain destruction, xend is _supposed_ to remove >all the relevant entries of the store, and the backend drivers are _supposed_ >to notice this and close themselves down. If you have evidence that either of >those things aren''t happening, then we should regard that as a bug, >independent of your patch. Have you seen that happening? > >Hi Ewan, I should have been more specific. Zombie domains is a bug that frequently appears. I''ve not seen it recently but it''s one of those things that frequently creeps into the tree. Regards, Anthony Liguori>Ewan. > >_______________________________________________ >Xen-devel mailing list >Xen-devel@lists.xensource.com >http://lists.xensource.com/xen-devel > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Ewan Mellor
2005-Oct-04 14:51 UTC
Re: [Xen-devel] [PATCH 0/2] Change xs_read_watch interfaces to support variable arguments
On Tue, Oct 04, 2005 at 09:34:43AM -0500, Anthony Liguori wrote:> I should have been more specific. Zombie domains is a bug that > frequently appears. I''ve not seen it recently but it''s one of those > things that frequently creeps into the tree.Good, that''s fine. Hopefully there''ll be fewer creeping zombies and a lot more slaying from now on ;-) Ewan. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel