Hello, Paul I think the next patch is better. Because ULONG_CMP_GE could cause double write, which has risk that write back old value. Please help review. I don't test it. If you agree, we will test it. Thanks!
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 0b760c1..c00f34e 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -1849,7 +1849,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, zero_cpu_stall_ticks(rdp); } rdp->gp_seq = rnp->gp_seq; /* Remember new grace-period state. */ - if (ULONG_CMP_GE(rnp->gp_seq_needed, rdp->gp_seq_needed) || rdp->gpwrap) + if (ULONG_CMP_LT(rdp->gp_seq_needed, rnp->gp_seq_needed) || rdp->gpwrap) rdp->gp_seq_needed = rnp->gp_seq_needed; WRITE_ONCE(rdp->gpwrap, false); rcu_gpnum_ovf(rnp, rdp); -----Original Message----- From: Paul E. McKenney [mailto:paul...@linux.ibm.com] Sent: Thursday, December 13, 2018 08:12 To: He, Bo <bo...@intel.com> Cc: Steven Rostedt <rost...@goodmis.org>; linux-kernel@vger.kernel.org; j...@joshtriplett.org; mathieu.desnoy...@efficios.com; jiangshan...@gmail.com; Zhang, Jun <jun.zh...@intel.com>; Xiao, Jin <jin.x...@intel.com>; Zhang, Yanmin <yanmin.zh...@intel.com>; Bai, Jie A <jie.a....@intel.com>; Sun, Yi J <yi.j....@intel.com> Subject: Re: rcu_preempt caused oom On Wed, Dec 12, 2018 at 11:13:22PM +0000, He, Bo wrote: > I don't see the rcutree.sysrq_rcu parameter in v4.19 kernel, I also checked > the latest kernel and the latest tag v4.20-rc6, not see the sysrq_rcu. > Please correct me if I have something wrong. That would be because I sent you the wrong patch, apologies! :-/ Please instead see the one below, which does add sysrq_rcu. Thanx, Paul > -----Original Message----- > From: Paul E. McKenney <paul...@linux.ibm.com> > Sent: Thursday, December 13, 2018 5:03 AM > To: He, Bo <bo...@intel.com> > Cc: Steven Rostedt <rost...@goodmis.org>; > linux-kernel@vger.kernel.org; j...@joshtriplett.org; > mathieu.desnoy...@efficios.com; jiangshan...@gmail.com; Zhang, Jun > <jun.zh...@intel.com>; Xiao, Jin <jin.x...@intel.com>; Zhang, Yanmin > <yanmin.zh...@intel.com>; Bai, Jie A <jie.a....@intel.com> > Subject: Re: rcu_preempt caused oom > > On Wed, Dec 12, 2018 at 07:42:24AM -0800, Paul E. McKenney wrote: > > On Wed, Dec 12, 2018 at 01:21:33PM +0000, He, Bo wrote: > > > we reproduce on two boards, but I still not see the > > > show_rcu_gp_kthreads() dump logs, it seems the patch can't catch the > > > scenario. > > > I double confirmed the CONFIG_PROVE_RCU=y is enabled in the config as > > > it's extracted from the /proc/config.gz. > > > > Strange. > > > > Are the systems responsive to sysrq keys once failure occurs? If > > so, I will provide you a sysrq-R or some such to dump out the RCU state. > > Or, as it turns out, sysrq-y if booting with rcutree.sysrq_rcu=1 using the > patch below. Only lightly tested. ------------------------------------------------------------------------ commit 04b6245c8458e8725f4169e62912c1fadfdf8141 Author: Paul E. McKenney <paul...@linux.ibm.com> Date: Wed Dec 12 16:10:09 2018 -0800 rcu: Add sysrq rcu_node-dump capability Backported from v4.21/v5.0 Life is hard if RCU manages to get stuck without triggering RCU CPU stall warnings or triggering the rcu_check_gp_start_stall() checks for failing to start a grace period. This commit therefore adds a boot-time-selectable sysrq key (commandeering "y") that allows manually dumping Tree RCU state. The new rcutree.sysrq_rcu kernel boot parameter must be set for this sysrq to be available. Signed-off-by: Paul E. McKenney <paul...@linux.ibm.com> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 0b760c1369f7..e9392a9d6291 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -61,6 +61,7 @@ #include <linux/trace_events.h> #include <linux/suspend.h> #include <linux/ftrace.h> +#include <linux/sysrq.h> #include "tree.h" #include "rcu.h" @@ -128,6 +129,9 @@ int num_rcu_lvl[] = NUM_RCU_LVL_INIT; int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ /* panic() on RCU Stall sysctl. */ int sysctl_panic_on_rcu_stall __read_mostly; +/* Commandeer a sysrq key to dump RCU's tree. */ static bool sysrq_rcu; +module_param(sysrq_rcu, bool, 0444); /* * The rcu_scheduler_active variable is initialized to the value @@ -662,6 +666,27 @@ void show_rcu_gp_kthreads(void) } EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); +/* Dump grace-period-request information due to commandeered sysrq. */ +static void sysrq_show_rcu(int key) { + show_rcu_gp_kthreads(); +} + +static struct sysrq_key_op sysrq_rcudump_op = { + .handler = sysrq_show_rcu, + .help_msg = "show-rcu(y)", + .action_msg = "Show RCU tree", + .enable_mask = SYSRQ_ENABLE_DUMP, +}; + +static int __init rcu_sysrq_init(void) +{ + if (sysrq_rcu) + return register_sysrq_key('y', &sysrq_rcudump_op); + return 0; +} +early_initcall(rcu_sysrq_init); + /* * Send along grace-period-related data for rcutorture diagnostics. */