Messages in this thread | | | Date | Tue, 5 Jun 2012 21:27:00 +0300 | From | Gleb Natapov <> | Subject | Re: automount hangs during boot somewhere in rcu_barrier. |
| |
On Tue, Jun 05, 2012 at 08:19:30AM -0700, Paul E. McKenney wrote: > On Tue, Jun 05, 2012 at 12:46:58PM +0300, Gleb Natapov wrote: > > On Mon, Jun 04, 2012 at 07:54:32AM -0700, Paul E. McKenney wrote: > > > On Mon, Jun 04, 2012 at 01:09:25PM +0300, Gleb Natapov wrote: > > > > On Mon, Jun 04, 2012 at 10:48:03AM +0300, Gleb Natapov wrote: > > > > > On Sun, Jun 03, 2012 at 08:15:49PM -0700, Paul E. McKenney wrote: > > > > > > Hmmm... Could you please try disabling CONFIG_RCU_FAST_NO_HZ? If that > > > > > > makes the problem go away, I have some patches 3.6-targeted patches > > > > > > for you. > > > > > Will report back. > > > > > > > > > Do not see any problems with the option disabled. > > > > > > OK, then you are probably being bitten by the same bug that was biting > > > Pascal and Heiko. Could you please apply the following series from -rcu? > > > > > > bf7757c3 (Merge commit 'v3.4-rc6' into HEAD) > > > edc17350 (RCU_FAST_NO_HZ detection of callback adoption) > > > 8e03a88e (Update RCU_FAST_NO_HZ tracing for lazy callbacks) > > > f2298c63 (Move RCU_FAST_NO_HZ per-CPU variables to rcu_dynticks structure) > > > 05cc0fc9 (Precompute RCU_FAST_NO_HZ timer offsets) > > > 477fca01 (Dump only the current CPU's buffers for idle-entry/exit warnings) > > > f29075fb (The rcu_needs_cpu() function is not a quiescent state) > > > > > > Of these, 05cc0fc9 (Precompute RCU_FAST_NO_HZ timer offsets) is likely > > > to be the most relevant. The -rcu tree may be found at: > > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git > > > > > > Please let me know if you have any trouble pulling this series. > > > > > The problem is gone with the series + enabling CONFIG_RCU_FAST_NO_HZ > > back. > > Very good!!! > > Maybe I should be submitting this series to 3.5 rather than 3.6? > As it stands now I cannot use kernels with CONFIG_RCU_FAST_NO_HZ enabled. May be disabling it is a good enough workaround for poor souls like me, but I doubt distros can enable it with known problems like that.
Thanks you for your promptly response!
-- Gleb.
| |