[Devel] [PATCH RH8 8/8] mm/page_alloc: use sched_clock() instead of jiffies to measure latency
Andrey Zhadchenko
andrey.zhadchenko at virtuozzo.com
Tue May 18 20:48:32 MSK 2021
From: Andrey Ryabinin <aryabinin at virtuozzo.com>
sched_clock() (which is rdtsc() on x86) gives us more precise result
than jiffies.
Q: Why do we need greater accuracy?
A: Because if we target to, say, 10000 IOPS (per cpu) then
1 ms memory allocation latency is too much and we need
to achieve less alloc latency and thus measure it.
https://pmc.acronis.com/browse/VSTOR-19040
Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
(cherry-picked from 99407f6d6f504d00aa3cd3ca87782ab32b0ec364)
Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
---
mm/page_alloc.c | 14 ++++++++------
1 file changed, 8 insertions(+), 6 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index d3222ac..b5afa2a 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -4556,9 +4556,11 @@ static void __alloc_collect_stats(gfp_t gfp_mask, unsigned int order,
{
#ifdef CONFIG_VE
unsigned long flags;
+ u64 current_clock, delta;
int ind, cpu;
- time = jiffies_to_usecs(jiffies - time) * 1000;
+ current_clock = sched_clock();
+ delta = current_clock - time;
if (!(gfp_mask & __GFP_RECLAIM))
ind = KSTAT_ALLOCSTAT_ATOMIC;
else
@@ -4569,12 +4571,12 @@ static void __alloc_collect_stats(gfp_t gfp_mask, unsigned int order,
local_irq_save(flags);
cpu = smp_processor_id();
- KSTAT_LAT_PCPU_ADD(&kstat_glob.alloc_lat[ind], time);
+ KSTAT_LAT_PCPU_ADD(&kstat_glob.alloc_lat[ind], delta);
if (in_task()) {
- current->alloc_lat[ind].totlat += time;
+ current->alloc_lat[ind].totlat += delta;
current->alloc_lat[ind].count++;
- update_maxlat(¤t->alloc_lat[ind], time, jiffies);
+ update_maxlat(¤t->alloc_lat[ind], delta, current_clock);
}
if (!page)
@@ -4594,7 +4596,7 @@ struct page *
unsigned int alloc_flags = ALLOC_WMARK_LOW;
gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
struct alloc_context ac = { };
- cycles_t start;
+ u64 start;
gfp_mask &= gfp_allowed_mask;
alloc_mask = gfp_mask;
@@ -4606,7 +4608,7 @@ struct page *
finalise_ac(gfp_mask, &ac);
- start = jiffies;
+ start = sched_clock();
/* First allocation attempt */
page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
if (likely(page))
--
1.8.3.1
More information about the Devel
mailing list