Messages in this thread Patch in this message | | | Date | Thu, 17 Nov 2016 13:54:27 +0000 (UTC) | From | Mathieu Desnoyers <> | Subject | Re: [PATCH] Fix: disable sys_membarrier when nohz_full is enabled |
| |
----- On Nov 17, 2016, at 8:40 AM, Paul E. McKenney paulmck@linux.vnet.ibm.com wrote:
> On Thu, Nov 17, 2016 at 11:46:34AM +0000, Mathieu Desnoyers wrote: >> ----- On Nov 17, 2016, at 1:51 AM, Lai Jiangshan jiangshanlai@gmail.com wrote: >> >> > On Fri, Nov 4, 2016 at 12:29 AM, Mathieu Desnoyers >> > <mathieu.desnoyers@efficios.com> wrote: >> >> Userspace applications should be allowed to expect the membarrier system >> >> call with MEMBARRIER_CMD_SHARED command to issue memory barriers on >> >> nohz_full CPUs, but synchronize_sched() does not take those into >> >> account. >> >> >> >> Given that we do not want unrelated processes to be able to affect >> >> real-time sensitive nohz_full CPUs, simply return ENOSYS when membarrier >> >> is invoked on a kernel with enabled nohz_full CPUs. >> >> >> >> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> >> >> CC: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> >> >> CC: Josh Triplett <josh@joshtriplett.org> >> >> CC: Steven Rostedt <rostedt@goodmis.org> >> >> CC: Lai Jiangshan <jiangshanlai@gmail.com> >> >> CC: <stable@vger.kernel.org> [3.10+] >> >> --- >> >> kernel/membarrier.c | 4 ++++ >> >> 1 file changed, 4 insertions(+) >> >> >> >> diff --git a/kernel/membarrier.c b/kernel/membarrier.c >> >> index 536c727..9f9284f 100644 >> >> --- a/kernel/membarrier.c >> >> +++ b/kernel/membarrier.c >> >> @@ -16,6 +16,7 @@ >> >> >> >> #include <linux/syscalls.h> >> >> #include <linux/membarrier.h> >> >> +#include <linux/tick.h> >> >> >> >> /* >> >> * Bitmask made from a "or" of all commands within enum membarrier_cmd, >> >> @@ -51,6 +52,9 @@ >> >> */ >> >> SYSCALL_DEFINE2(membarrier, int, cmd, int, flags) >> >> { >> >> + /* MEMBARRIER_CMD_SHARED is not compatible with nohz_full. */ >> >> + if (tick_nohz_full_enabled()) >> >> + return -ENOSYS; >> > >> > I guess this code needs to be moved down into the branch of >> > "case MEMBARRIER_CMD_SHARED" to match its comment. >> >> No, that would be unexpected from user-space. Either a system >> call is implemented or not, not "implemented for some parameters". >> >> We also want MEMBARRIER_CMD_QUERY to return -ENOSYS in this case, >> and all other parameter values to also return -ENOSYS (rather than >> -EINVAL). >> >> If a system call that returns successfully on CMD_QUERY or EINVAL, >> user-space may assume it will not have to handle ENOSYS in the >> next calls. >> >> >> > >> > Acked-by: Lai Jiangshan <jiangshanlai@gmail.com> >> > >> > But I'm afraid, in the future, tick_nohz_full will become a default y >> > feature. thus it makes sys_membarrier() always disabled. we might >> > need a new MEMBARRIER_CMD_XXX to handle it? >> >> This may require that we send an IPI to nohz_full CPUs, which will >> disturb them real-time wise. Any better ideas ? > > Restrict the IPIs to CPUs running the process executing the > sys_membarrier() system call. This would mean that CPUs only > are interrupted by their own application's request.
This would break use-cases of cross-process shared memory. :-(
Mathieu
> > Thanx, Paul
-- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com
| |