diff options
| author | Paul E. McKenney <paulmck@linux.vnet.ibm.com> | 2016-11-30 09:24:30 -0500 |
|---|---|---|
| committer | Paul E. McKenney <paulmck@linux.vnet.ibm.com> | 2017-10-09 17:25:17 -0400 |
| commit | f79c3ad6189624c3de0ad5521610c9e22a1c33cf (patch) | |
| tree | 82bfe610ef11d2dbaf120f7ddbbc250ec4f917fb /kernel | |
| parent | 8a5776a5f49812d29fe4b2d0a2d71675c3facf3f (diff) | |
sched,rcu: Make cond_resched() provide RCU quiescent state
There is some confusion as to which of cond_resched() or
cond_resched_rcu_qs() should be added to long in-kernel loops.
This commit therefore eliminates the decision by adding RCU quiescent
states to cond_resched(). This commit also simplifies the code that
used to interact with cond_resched_rcu_qs(), and that now interacts with
cond_resched(), to reduce its overhead. This reduction is necessary to
allow the heavier-weight cond_resched_rcu_qs() mechanism to be invoked
everywhere that cond_resched() is invoked.
Part of that reduction in overhead converts the jiffies_till_sched_qs
kernel parameter to read-only at runtime, thus eliminating the need for
bounds checking.
Reported-by: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
[ paulmck: Keep PREEMPT=n cond_resched a no-op, per Peter Zijlstra. ]
Diffstat (limited to 'kernel')
| -rw-r--r-- | kernel/rcu/tree.c | 25 | ||||
| -rw-r--r-- | kernel/sched/core.c | 1 |
2 files changed, 6 insertions, 20 deletions
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index b0ad62b0e7b8..0dda57a28276 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c | |||
| @@ -534,8 +534,8 @@ module_param(rcu_kick_kthreads, bool, 0644); | |||
| 534 | * How long the grace period must be before we start recruiting | 534 | * How long the grace period must be before we start recruiting |
| 535 | * quiescent-state help from rcu_note_context_switch(). | 535 | * quiescent-state help from rcu_note_context_switch(). |
| 536 | */ | 536 | */ |
| 537 | static ulong jiffies_till_sched_qs = HZ / 20; | 537 | static ulong jiffies_till_sched_qs = HZ / 10; |
| 538 | module_param(jiffies_till_sched_qs, ulong, 0644); | 538 | module_param(jiffies_till_sched_qs, ulong, 0444); |
| 539 | 539 | ||
| 540 | static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, | 540 | static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
| 541 | struct rcu_data *rdp); | 541 | struct rcu_data *rdp); |
| @@ -1235,7 +1235,6 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |||
| 1235 | unsigned long jtsq; | 1235 | unsigned long jtsq; |
| 1236 | bool *rnhqp; | 1236 | bool *rnhqp; |
| 1237 | bool *ruqp; | 1237 | bool *ruqp; |
| 1238 | unsigned long rjtsc; | ||
| 1239 | struct rcu_node *rnp; | 1238 | struct rcu_node *rnp; |
| 1240 | 1239 | ||
| 1241 | /* | 1240 | /* |
| @@ -1252,23 +1251,13 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |||
| 1252 | return 1; | 1251 | return 1; |
| 1253 | } | 1252 | } |
| 1254 | 1253 | ||
| 1255 | /* Compute and saturate jiffies_till_sched_qs. */ | ||
| 1256 | jtsq = jiffies_till_sched_qs; | ||
| 1257 | rjtsc = rcu_jiffies_till_stall_check(); | ||
| 1258 | if (jtsq > rjtsc / 2) { | ||
| 1259 | WRITE_ONCE(jiffies_till_sched_qs, rjtsc); | ||
| 1260 | jtsq = rjtsc / 2; | ||
| 1261 | } else if (jtsq < 1) { | ||
| 1262 | WRITE_ONCE(jiffies_till_sched_qs, 1); | ||
| 1263 | jtsq = 1; | ||
| 1264 | } | ||
| 1265 | |||
| 1266 | /* | 1254 | /* |
| 1267 | * Has this CPU encountered a cond_resched_rcu_qs() since the | 1255 | * Has this CPU encountered a cond_resched_rcu_qs() since the |
| 1268 | * beginning of the grace period? For this to be the case, | 1256 | * beginning of the grace period? For this to be the case, |
| 1269 | * the CPU has to have noticed the current grace period. This | 1257 | * the CPU has to have noticed the current grace period. This |
| 1270 | * might not be the case for nohz_full CPUs looping in the kernel. | 1258 | * might not be the case for nohz_full CPUs looping in the kernel. |
| 1271 | */ | 1259 | */ |
| 1260 | jtsq = jiffies_till_sched_qs; | ||
| 1272 | rnp = rdp->mynode; | 1261 | rnp = rdp->mynode; |
| 1273 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); | 1262 | ruqp = per_cpu_ptr(&rcu_dynticks.rcu_urgent_qs, rdp->cpu); |
| 1274 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && | 1263 | if (time_after(jiffies, rdp->rsp->gp_start + jtsq) && |
| @@ -1276,7 +1265,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |||
| 1276 | READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) { | 1265 | READ_ONCE(rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) { |
| 1277 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc")); | 1266 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("rqc")); |
| 1278 | return 1; | 1267 | return 1; |
| 1279 | } else { | 1268 | } else if (time_after(jiffies, rdp->rsp->gp_start + jtsq)) { |
| 1280 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ | 1269 | /* Load rcu_qs_ctr before store to rcu_urgent_qs. */ |
| 1281 | smp_store_release(ruqp, true); | 1270 | smp_store_release(ruqp, true); |
| 1282 | } | 1271 | } |
| @@ -1304,10 +1293,6 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |||
| 1304 | * updates are only once every few jiffies, the probability of | 1293 | * updates are only once every few jiffies, the probability of |
| 1305 | * lossage (and thus of slight grace-period extension) is | 1294 | * lossage (and thus of slight grace-period extension) is |
| 1306 | * quite low. | 1295 | * quite low. |
| 1307 | * | ||
| 1308 | * Note that if the jiffies_till_sched_qs boot/sysfs parameter | ||
| 1309 | * is set too high, we override with half of the RCU CPU stall | ||
| 1310 | * warning delay. | ||
| 1311 | */ | 1296 | */ |
| 1312 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); | 1297 | rnhqp = &per_cpu(rcu_dynticks.rcu_need_heavy_qs, rdp->cpu); |
| 1313 | if (!READ_ONCE(*rnhqp) && | 1298 | if (!READ_ONCE(*rnhqp) && |
| @@ -1316,7 +1301,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |||
| 1316 | WRITE_ONCE(*rnhqp, true); | 1301 | WRITE_ONCE(*rnhqp, true); |
| 1317 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ | 1302 | /* Store rcu_need_heavy_qs before rcu_urgent_qs. */ |
| 1318 | smp_store_release(ruqp, true); | 1303 | smp_store_release(ruqp, true); |
| 1319 | rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */ | 1304 | rdp->rsp->jiffies_resched += jtsq; /* Re-enable beating. */ |
| 1320 | } | 1305 | } |
| 1321 | 1306 | ||
| 1322 | /* | 1307 | /* |
diff --git a/kernel/sched/core.c b/kernel/sched/core.c index d17c5da523a0..904d3ab35c83 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c | |||
| @@ -4842,6 +4842,7 @@ int __sched _cond_resched(void) | |||
| 4842 | preempt_schedule_common(); | 4842 | preempt_schedule_common(); |
| 4843 | return 1; | 4843 | return 1; |
| 4844 | } | 4844 | } |
| 4845 | rcu_all_qs(); | ||
| 4845 | return 0; | 4846 | return 0; |
| 4846 | } | 4847 | } |
| 4847 | EXPORT_SYMBOL(_cond_resched); | 4848 | EXPORT_SYMBOL(_cond_resched); |
