On 3/18/21 1:22 AM, Daniel Jordan wrote:

Andrey Ryabinin [off-list ref] writes:

quoted

cpuacct has 2 different ways of accounting and showing user
and system times.

The first one uses cpuacct_account_field() to account times
and cpuacct.stat file to expose them. And this one seems to work ok.

The second one is uses cpuacct_charge() function for accounting and
set of cpuacct.usage* files to show times. Despite some attempts to
fix it in the past it still doesn't work. E.g. while running KVM
guest the cpuacct_charge() accounts most of the guest time as
system time. This doesn't match with user&system times shown in
cpuacct.stat or proc/<pid>/stat.

I couldn't reproduce this running a cpu bound load in a kvm guest on a
nohz_full cpu on 5.11.  The time is almost entirely in cpuacct.usage and
_user, while _sys stays low.

Could you say more about how you're seeing this?  Don't really doubt
there's a problem, just wondering what you're doing.

Yeah, I it's almost unnoticable if you run some load in guest like qemu.

But more simple case with busy loop in KVM_RUN triggers this:

# git clone https://github.com/aryabinin/kvmsample
# make
# mkdir /sys/fs/cgroup/cpuacct/test
# echo $$ > /sys/fs/cgroup/cpuacct/test/tasks
# ./kvmsample &
# for i in {1..5}; do cat /sys/fs/cgroup/cpuacct/test/cpuacct.usage_sys; sleep 1; done
1976535645
2979839428
3979832704
4983603153
5983604157

quoted

diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
index 941c28cf9738..7eff79faab0d 100644
--- a/kernel/sched/cpuacct.c
+++ b/kernel/sched/cpuacct.c

@@ -29,7 +29,7 @@ struct cpuacct_usage {
 struct cpuacct {
 	struct cgroup_subsys_state	css;
 	/* cpuusage holds pointer to a u64-type object on every CPU */
-	struct cpuacct_usage __percpu	*cpuusage;

Definition of struct cpuacct_usage can go away now.

Done.

quoted

@@ -99,7 +99,8 @@ static void cpuacct_css_free(struct cgroup_subsys_state *css)
 static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu,
 				 enum cpuacct_stat_index index)
 {
-	struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
+	u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
+	u64 *cpustat = per_cpu_ptr(ca->cpustat, cpu)->cpustat;
 	u64 data;

There's a BUG_ON below this that could probably be WARN_ON_ONCE while
you're here

Sure.

quoted

@@ -278,8 +274,8 @@ static int cpuacct_stats_show(struct seq_file *sf, void *v)
 	for_each_possible_cpu(cpu) {
 		u64 *cpustat = per_cpu_ptr(ca->cpustat, cpu)->cpustat;
 
-		val[CPUACCT_STAT_USER]   += cpustat[CPUTIME_USER];
-		val[CPUACCT_STAT_USER]   += cpustat[CPUTIME_NICE];
+		val[CPUACCT_STAT_USER] += cpustat[CPUTIME_USER];
+		val[CPUACCT_STAT_USER] += cpustat[CPUTIME_NICE];

unnecessary whitespace change?

yup