[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20210521140227.GW4441@paulmck-ThinkPad-P17-Gen-1>
Date: Fri, 21 May 2021 07:02:27 -0700
From: "Paul E. McKenney" <paulmck@...nel.org>
To: Sergey Senozhatsky <senozhatsky@...omium.org>
Cc: Josh Triplett <josh@...htriplett.org>,
Steven Rostedt <rostedt@...dmis.org>,
Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
Lai Jiangshan <jiangshanlai@...il.com>,
Joel Fernandes <joel@...lfernandes.org>,
Suleiman Souhlal <suleiman@...gle.com>, rcu@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] rcu/tree: consider time a VM was suspended
On Fri, May 21, 2021 at 03:42:31PM +0900, Sergey Senozhatsky wrote:
> On (21/05/20 17:14), Paul E. McKenney wrote:
> > > On a side note.
> > >
> > > I wish we didn't have to put kvm_check_and_clear_guest_paused() all
> > > over the place.
> > >
> > > We do load jiffies at the start of check_cpu_stall(). So, in theory,
> > > we can just use that captured jiffies (which can become obsolete but
> > > so will grace period timestamps) value and never read current system
> > > jiffies because they can jump way forward. IOW
> > >
> > > jn = j + 3 * rcu_jiffies_till_stall_check() + 3;
> > >
> > > instead of
> > >
> > > jn = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
> > >
> > > Then we probably can remove kvm_check_and_clear_guest_paused().
> > >
> > > But that "don't load current jiffies" is rather fragile.
> > >
> > > kvm_check_and_clear_guest_paused() is not pretty, but at least it's
> > > explicit.
> >
> > If this works for you, I am very much in favor!
>
> Oh, no. Sorry for the noise. This is racy and won't work.
I had been hoping! ;-)
Thanx, Paul
Powered by blists - more mailing lists