[v2] xen: credit2: avoid using cpumask_weight() in hot-paths
diff mbox series

Message ID 156439734950.9656.3257482583234913248.stgit@Palanthas
State New, archived
Headers show
Series
  • [v2] xen: credit2: avoid using cpumask_weight() in hot-paths
Related show

Commit Message

Dario Faggioli July 29, 2019, 10:49 a.m. UTC
cpumask_weight() is known to be expensive. In Credit2, we use it in
load-balancing, but only for knowing how many CPUs are active in a
runqueue.

Keeping such count in an integer field of the per-runqueue data
structure we have, completely avoids the need for cpumask_weight().

While there, remove as much other uses of it as we can, even if not in
hot-paths.

Signed-off-by: Dario Faggioli <dfaggioli@suse.com>
---
Cc: George Dunlap <george.dunlap@eu.citrix.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: Andrii Anisov <andrii.anisov@gmail.com>
---
I just realized this patch fell through the cracks! :-O

I sent it quite a while ago, I got some comments from Andrew and
Andrii, but then never really followed up.

So here we hare!

Changes from v1:
* nr_cpus is now unsigned
* coding style (no hard tabs)
* killed a couple of redundant ASSERT()s
---
 xen/common/sched_credit2.c |   19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

Comments

Andrii Anisov July 29, 2019, 12:12 p.m. UTC | #1
On 29.07.19 13:49, Dario Faggioli wrote:
> cpumask_weight() is known to be expensive. In Credit2, we use it in
> load-balancing, but only for knowing how many CPUs are active in a
> runqueue.
> 
> Keeping such count in an integer field of the per-runqueue data
> structure we have, completely avoids the need for cpumask_weight().
> 
> While there, remove as much other uses of it as we can, even if not in
> hot-paths.
> 
> Signed-off-by: Dario Faggioli <dfaggioli@suse.com>
> ---
> Cc: George Dunlap <george.dunlap@eu.citrix.com>
> Cc: Andrew Cooper <andrew.cooper3@citrix.com>
> Cc: Andrii Anisov <andrii.anisov@gmail.com>
> ---
> I just realized this patch fell through the cracks! :-O
> 
> I sent it quite a while ago, I got some comments from Andrew and
> Andrii, but then never really followed up.
> 
> So here we hare!
> 
> Changes from v1:
> * nr_cpus is now unsigned
> * coding style (no hard tabs)
> * killed a couple of redundant ASSERT()s
> ---

Reviewed-by: Andrii Anisov <andrii_anisov@epam.com>
George Dunlap July 30, 2019, 11:22 a.m. UTC | #2
On 7/29/19 11:49 AM, Dario Faggioli wrote:
> cpumask_weight() is known to be expensive. In Credit2, we use it in
> load-balancing, but only for knowing how many CPUs are active in a
> runqueue.
> 
> Keeping such count in an integer field of the per-runqueue data
> structure we have, completely avoids the need for cpumask_weight().
> 
> While there, remove as much other uses of it as we can, even if not in
> hot-paths.
> 
> Signed-off-by: Dario Faggioli <dfaggioli@suse.com>

Looks good, thanks.

Reviewed-by: George Dunlap <george.dunlap@citrix.com>

Patch
diff mbox series

diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
index 8e4381d8a7..fbdc4618cb 100644
--- a/xen/common/sched_credit2.c
+++ b/xen/common/sched_credit2.c
@@ -466,6 +466,7 @@  struct csched2_runqueue_data {
     spinlock_t lock;           /* Lock for this runqueue                     */
 
     struct list_head runq;     /* Ordered list of runnable vms               */
+    unsigned int nr_cpus;      /* How many CPUs are sharing this runqueue    */
     int id;                    /* ID of this runqueue (-1 if invalid)        */
 
     int load;                  /* Instantaneous load (num of non-idle vcpus) */
@@ -2613,8 +2614,8 @@  retry:
         if ( st.orqd->b_avgload > load_max )
             load_max = st.orqd->b_avgload;
 
-        cpus_max = cpumask_weight(&st.lrqd->active);
-        i = cpumask_weight(&st.orqd->active);
+        cpus_max = st.lrqd->nr_cpus;
+        i = st.orqd->nr_cpus;
         if ( i > cpus_max )
             cpus_max = i;
 
@@ -3697,7 +3698,7 @@  csched2_dump(const struct scheduler *ops)
                "\tinstload           = %d\n"
                "\taveload            = %"PRI_stime" (~%"PRI_stime"%%)\n",
                i,
-               cpumask_weight(&prv->rqd[i].active),
+               prv->rqd[i].nr_cpus,
                nr_cpu_ids, cpumask_bits(&prv->rqd[i].active),
                prv->rqd[i].max_weight,
                prv->rqd[i].pick_bias,
@@ -3815,7 +3816,7 @@  init_pdata(struct csched2_private *prv, struct csched2_pcpu *spc,
 
     __cpumask_set_cpu(cpu, &spc->sibling_mask);
 
-    if ( cpumask_weight(&rqd->active) > 0 )
+    if ( rqd->nr_cpus > 0 )
         for_each_cpu ( rcpu, per_cpu(cpu_sibling_mask, cpu) )
             if ( cpumask_test_cpu(rcpu, &rqd->active) )
             {
@@ -3828,7 +3829,10 @@  init_pdata(struct csched2_private *prv, struct csched2_pcpu *spc,
     __cpumask_set_cpu(cpu, &prv->initialized);
     __cpumask_set_cpu(cpu, &rqd->smt_idle);
 
-    if ( cpumask_weight(&rqd->active) == 1 )
+    rqd->nr_cpus++;
+    ASSERT(cpumask_weight(&rqd->active) == rqd->nr_cpus);
+
+    if ( rqd->nr_cpus == 1 )
         rqd->pick_bias = cpu;
 
     return spc->runq_id;
@@ -3934,7 +3938,10 @@  csched2_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
     for_each_cpu ( rcpu, &rqd->active )
         __cpumask_clear_cpu(cpu, &csched2_pcpu(rcpu)->sibling_mask);
 
-    if ( cpumask_empty(&rqd->active) )
+    rqd->nr_cpus--;
+    ASSERT(cpumask_weight(&rqd->active) == rqd->nr_cpus);
+
+    if ( rqd->nr_cpus == 0 )
     {
         printk(XENLOG_INFO " No cpus left on runqueue, disabling\n");
         deactivate_runqueue(prv, spc->runq_id);