This colocates some hot fields in "struct rq" to be on the same cache line
as others that are often accessed at the same time or in similar ways.
Using data from a Google-internal fleet-scale profiler, I found three
distinct groups of hot fields in struct rq:
- (1) The runqueue lock: __lock.
- (2) Those accessed from hot code in pick_next_task_fair():
nr_running, nr_numa_running, nr_preferred_running,
ttwu_pending, cpu_capacity, curr, idle.
- (3) Those accessed from some other hot codepaths, e.g.
update_curr(), update_rq_clock(), and scheduler_tick():
clock_task, clock_pelt, clock, lost_idle_time,
clock_update_flags, clock_pelt_idle, clock_idle.
The cycles spent on accessing these different groups of fields broke down
roughly as follows:
- 50% on group (1) (the runqueue lock, always read-write)
- 39% on group (2) (load:store ratio around 38:1)
- 8% on group (3) (load:store ratio around 5:1)
- 3% on all the other fields
Most of the fields in group (3) are already in a cache line grouping; this
patch just adds "clock" and "clock_update_flags" to that group. The fields
in group (2) are scattered across several cache lines; the main effect of
this patch is to group them together, on a single line at the beginning of
the structure. A few other less performance-critical fields (nr_switches,
numa_migrate_on, has_blocked_load, nohz_csd, last_blocked_load_update_tick)
were also reordered to reduce holes in the data structure.
Since the runqueue lock is acquired from so many different contexts, and is
basically always accessed using an atomic operation, putting it on either
of the cache lines for groups (2) or (3) would slow down accesses to those
fields dramatically, since those groups are read-mostly accesses.
To test this, I wrote a focused load test that would put load on the
pick_next_task_fair() path. A parent process would fork many child
processes, and each child would nanosleep() for 1 msec many times in a
loop. The load test was monitored with "perf", and I looked at the amount
of cycles that were spent with sched_balance_rq() on the stack. The test
was reliably spending ~5% of all of its cycles there. I ran it 100 times
on a pair of 2-socket Intel Haswell machines (72 vCPUs per machine) - one
running the tip of sched/core, the other running this change - using 360
child processes and 8192 1-msec sleeps per child. The mean cycle count
dropped from 5.14B to 4.91B, or a *4.6% decrease* in relevant scheduler
cycles.
Given that this change reduces cache misses in a very hot kernel codepath,
there's likely to be additional application performance improvement due to
reduced cache conflicts from kernel data structures.
On a Power11 system with 128-byte cache lines, my test showed a ~5%
decrease in relevant scheduler cycles, along with a slight increase in user
time - both positive indicators. This data comes from
https://lore.kernel.org/lkml/
affdc6b1-9980-44d1-89db-
d90730c1e384@linux.ibm.com/
This is the case even though the additional "____cacheline_aligned" that
puts the runqueue lock on the next cache line adds an additional 64 bytes
of padding on those machines. This patch does not change the size of
"struct rq" on machines with 64-byte cache lines.
I also ran "hackbench" to try to test this change, but it didn't show
conclusive results. Looking at a CPU cycle profile of the hackbench run,
it was spending 95% of its cycles inside __alloc_skb(), __kfree_skb(), or
kmem_cache_free() - almost all of which was spent updating memcg counters
or contending on the list_lock in kmem_cache_node. And it spent less than
0.5% of its cycles inside either schedule() or try_to_wake_up(). So it's
not surprising that it didn't show useful results here.
The "__no_randomize_layout" was added to reflect the fact that performance
of code that references this data structure is unusually sensitive to
placement of its members.
Signed-off-by: Blake Jones <blakejones@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Madadi Vineeth Reddy <vineethr@linux.ibm.com>
Reviewed-by: Josh Don <joshdon@google.com>
Tested-by: Madadi Vineeth Reddy <vineethr@linux.ibm.com>
Link: https://patch.msgid.link/20251202023743.1524247-1-blakejones@google.com
* acquire operations must be ordered by ascending &runqueue.
*/
struct rq {
- /* runqueue lock: */
- raw_spinlock_t __lock;
-
+ /*
+ * The following members are loaded together, without holding the
+ * rq->lock, in an extremely hot loop in update_sg_lb_stats()
+ * (called from pick_next_task()). To reduce cache pollution from
+ * this operation, they are placed together on this dedicated cache
+ * line. Even though some of them are frequently modified, they are
+ * loaded much more frequently than they are stored.
+ */
unsigned int nr_running;
#ifdef CONFIG_NUMA_BALANCING
unsigned int nr_numa_running;
unsigned int nr_preferred_running;
- unsigned int numa_migrate_on;
#endif
+ unsigned int ttwu_pending;
+ unsigned long cpu_capacity;
+#ifdef CONFIG_SCHED_PROXY_EXEC
+ struct task_struct __rcu *donor; /* Scheduling context */
+ struct task_struct __rcu *curr; /* Execution context */
+#else
+ union {
+ struct task_struct __rcu *donor; /* Scheduler context */
+ struct task_struct __rcu *curr; /* Execution context */
+ };
+#endif
+ struct task_struct *idle;
+ /* padding left here deliberately */
+
+ /*
+ * The next cacheline holds the (hot) runqueue lock, as well as
+ * some other less performance-critical fields.
+ */
+ u64 nr_switches ____cacheline_aligned;
+
+ /* runqueue lock: */
+ raw_spinlock_t __lock;
+
#ifdef CONFIG_NO_HZ_COMMON
- unsigned long last_blocked_load_update_tick;
- unsigned int has_blocked_load;
- call_single_data_t nohz_csd;
unsigned int nohz_tick_stopped;
atomic_t nohz_flags;
+ unsigned int has_blocked_load;
+ unsigned long last_blocked_load_update_tick;
+ call_single_data_t nohz_csd;
#endif /* CONFIG_NO_HZ_COMMON */
- unsigned int ttwu_pending;
- u64 nr_switches;
-
#ifdef CONFIG_UCLAMP_TASK
/* Utilization clamp values based on CPU's RUNNABLE tasks */
struct uclamp_rq uclamp[UCLAMP_CNT] ____cacheline_aligned;
struct list_head *tmp_alone_branch;
#endif /* CONFIG_FAIR_GROUP_SCHED */
+#ifdef CONFIG_NUMA_BALANCING
+ unsigned int numa_migrate_on;
+#endif
/*
* This is part of a global counter where only the total sum
* over all CPUs matters. A task can increase this counter on
*/
unsigned long nr_uninterruptible;
-#ifdef CONFIG_SCHED_PROXY_EXEC
- struct task_struct __rcu *donor; /* Scheduling context */
- struct task_struct __rcu *curr; /* Execution context */
-#else
- union {
- struct task_struct __rcu *donor; /* Scheduler context */
- struct task_struct __rcu *curr; /* Execution context */
- };
-#endif
struct sched_dl_entity *dl_server;
- struct task_struct *idle;
struct task_struct *stop;
const struct sched_class *next_class;
unsigned long next_balance;
struct mm_struct *prev_mm;
- unsigned int clock_update_flags;
- u64 clock;
- /* Ensure that all clocks are in the same cache line */
+ /*
+ * The following fields of clock data are frequently referenced
+ * and updated together, and should go on their own cache line.
+ */
u64 clock_task ____cacheline_aligned;
u64 clock_pelt;
+ u64 clock;
unsigned long lost_idle_time;
+ unsigned int clock_update_flags;
u64 clock_pelt_idle;
u64 clock_idle;
+
#ifndef CONFIG_64BIT
u64 clock_pelt_idle_copy;
u64 clock_idle_copy;
#endif
- atomic_t nr_iowait;
-
u64 last_seen_need_resched_ns;
int ticks_without_resched;
struct root_domain *rd;
struct sched_domain __rcu *sd;
- unsigned long cpu_capacity;
-
struct balance_callback *balance_callback;
unsigned char nohz_idle_balance;
call_single_data_t cfsb_csd;
struct list_head cfsb_csd_list;
#endif
-};
+
+ atomic_t nr_iowait;
+} __no_randomize_layout;
#ifdef CONFIG_FAIR_GROUP_SCHED
projects / thirdparty / kernel / linux.git / commitdiff
