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

[Xen-devel] [PATCH for 4.7 0/4] Assorted scheduling fixes



Hi,

This small series contains some bugfixes for various schedulers. They're all
bugfixes, so I think all should be considered for 4.7. Here's some more
detailed analysis.

Patch 1 and 3 are for Credit2. Patch 1 is a lot more important, as we have an
ASSERT triggering without it. Patch 2 is behavioral fixing, which I believe it
is important, but at least does not make anything explode.

Patch 2 fixes another ASSERT, in case a pCPU fails to come up. This is what
Julien reported here:

 https://www.mail-archive.com/xen-devel@xxxxxxxxxxxxx/msg65918.html

Julien, the patch is very very similar to the one attached to one of my reply
in that thread, but I had to change some small bits... Can you please re-test
it?

Patch 4 makes the code of RTDS look consistent with what we state in patch 2,
so it's also important. Furthermore, it does fix a bug (although, again, not
one that would splat Xen) as, without it, we may have a timer used by the RTDS
scheduler bound to the pCPU of another cpupool with another scheduler. That
would introduce some unwanted and very difficult to recognize interference
between different schedulers in different pool, and should hence be avoided.

So this was awesomeness; about risks:
 - patch 1 is very small, super-self contained (zero impact outside of Credit2
   code) and it fixes an actual and 100% reproducible bug;
 - patch 2 is also totally self-contained and it can't possibly cause problems
   to anything else than to what it is trying to fix (Credit2's load balancer).
   It doesn't cure any ASSERT or Oops, so it's less interesting, but given the
   low risk --also considering that Credit2 will still be considered
   experimental in 4.7-- I think it can go in;
 - patch 3 is bigger, and a bit more complex. Note, however, that most of its
   content is code comments and ASSERT-s; it is self contained to scheduling
   (in the sense that it impacts all schedulers, but "just" them), and fixes
   a situation that, AFAIUI, is important for ARM;
 - patch 4 may again look not that critical. But, the fact that someone wanting
   to experiment with RTDS in a cpupool would face the kind of interference
   between independent cpupools that the patch cures is, I think, something
   worthwhile trying to avoid. Besides, it is again quite self contained, as
   it's indeed only relevant for RTDS (which is also going to be called
   experimental for 4.7).

Hope this helps. Thanks and Regards,
Dario
---
Dario Faggioli (4):
      xen: sched: avoid spuriously re-enabling IRQs in csched2_switch_sched()
      xen: sched: fix killing an uninitialized timer in free_pdata.
      xen: credit2: fix 2 (minor) issues in load tracking logic
      xen: adopt .deinit_pdata and improve timer handling

 xen/common/sched_credit.c  |   31 ++++++++++++++++--
 xen/common/sched_credit2.c |   26 +++++++++++----
 xen/common/sched_rt.c      |   74 +++++++++++++++++++++++++++++++++-----------
 xen/common/schedule.c      |   38 ++++++++++++++++++++++-
 xen/include/xen/sched-if.h |    1 +
 5 files changed, 138 insertions(+), 32 deletions(-)
--
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel

 


Rackspace

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