Skip to content

Commit

Permalink
credit2: avoid vCPUs to ever reach lower credits than idle
Browse files Browse the repository at this point in the history 
There have been report of stalls of guest vCPUs, when Credit2 was used.
It seemed like these vCPUs were not getting scheduled for very long
time, even under light load conditions (e.g., during dom0 boot).

Investigations led to the discovery that --although rarely-- it can
happen that a vCPU manages to run for very long timeslices. In Credit2,
this means that, when runtime accounting happens, the vCPU will lose a
large quantity of credits. This in turn may lead to the vCPU having less
credits than the idle vCPUs (-2^30). At this point, the scheduler will
pick the idle vCPU, instead of the ready to run vCPU, for a few
"epochs", which often times is enough for the guest kernel to think the
vCPU is not responding and crashing.

An example of this situation is shown here. In fact, we can see d0v1
sitting in the runqueue while all the CPUs are idle, as it has
-1254238270 credits, which is smaller than -2^30 = −1073741824:

    (XEN) Runqueue 0:
    (XEN)   ncpus              = 28
    (XEN)   cpus               = 0-27
    (XEN)   max_weight         = 256
    (XEN)   pick_bias          = 22
    (XEN)   instload           = 1
    (XEN)   aveload            = 293391 (~111%)
    (XEN)   idlers: 00,00000000,00000000,00000000,00000000,00000000,0fffffff
    (XEN)   tickled: 00,00000000,00000000,00000000,00000000,00000000,00000000
    (XEN)   fully idle cores: 00,00000000,00000000,00000000,00000000,00000000,0fffffff
    [...]
    (XEN) Runqueue 0:
    (XEN) CPU[00] runq=0, sibling=00,..., core=00,...
    (XEN) CPU[01] runq=0, sibling=00,..., core=00,...
    [...]
    (XEN) CPU[26] runq=0, sibling=00,..., core=00,...
    (XEN) CPU[27] runq=0, sibling=00,..., core=00,...
    (XEN) RUNQ:
    (XEN)     0: [0.1] flags=0 cpu=5 credit=-1254238270 [w=256] load=262144 (~100%)

We certainly don't want, under any circumstance, this to happen.
Let's, therefore, define a minimum amount of credits a vCPU can have.
During accounting, we make sure that, for however long the vCPU has
run, it will never get to have less than such minimum amount of
credits. Then, we set the credits of the idle vCPU to an even
smaller value.

NOTE: investigations have been done about _how_ it is possible for a
vCPU to execute for so much time that its credits becomes so low. While
still not completely clear, there are evidence that:
- it only happens very rarely,
- it appears to be both machine and workload specific,
- it does not look to be a Credit2 (e.g., as it happens when
  running with Credit1 as well) issue, or a scheduler issue.

This patch makes Credit2 more robust to events like this, whatever
the cause is, and should hence be backported (as far as possible).

Reported-by: Glen <[email protected]>
Reported-by: Tomas Mozes <[email protected]>
Signed-off-by: Dario Faggioli <[email protected]>
Reviewed-by: George Dunlap <[email protected]>
  • Loading branch information
@dfaggioli @jbeulich
dfaggioli authored and jbeulich committed Apr 3, 2020
1 parent 4de936a commit 36f3662
Show file tree
Hide file tree
Showing 3 changed files with 31 additions and 29 deletions.
2 changes: 1 addition & 1 deletion tools/xentrace/formats
View file
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@
0x00022204 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:credit_add
0x00022205 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:tickle_check [ dom:vcpu = 0x%(1)08x, credit = %(2)d, score = %(3)d ]
0x00022206 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:tickle [ cpu = %(1)d ]
0x00022207 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:credit_reset [ dom:vcpu = 0x%(1)08x, cr_start = %(2)d, cr_end = %(3)d, mult = %(4)d ]
0x00022207 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:credit_reset [ dom:vcpu = 0x%(1)08x, cr_start = %(2)d, cr_end = %(3)d ]
0x00022208 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:sched_tasklet
0x00022209 CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:update_load
0x0002220a CPU%(cpu)d %(tsc)d (+%(reltsc)8d) csched2:runq_assign [ dom:vcpu = 0x%(1)08x, rq_id = %(2)d ]
Expand Down
5 changes: 2 additions & 3 deletions tools/xentrace/xenalyze.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7718,13 +7718,12 @@ void sched_process(struct pcpu_info *p)
struct {
unsigned int vcpuid:16, domid:16;
int credit_start, credit_end;
unsigned int multiplier;
} *r = (typeof(r))ri->d;

printf(" %s csched2:reset_credits d%uv%u, "
"credit_start = %d, credit_end = %d, mult = %u\n",
"credit_start = %d, credit_end = %d\n",
ri->dump_header, r->domid, r->vcpuid,
r->credit_start, r->credit_end, r->multiplier);
r->credit_start, r->credit_end);
}
break;
case TRC_SCHED_CLASS_EVT(CSCHED2, 8): /* SCHED_TASKLET */
Expand Down
53 changes: 28 additions & 25 deletions xen/common/sched/credit2.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -229,12 +229,22 @@
* before a reset event.
*/
#define CSCHED2_CREDIT_INIT MILLISECS(10)
/*
* Minimum amount of credits VMs can have. Ideally, no VM would get
* close to this (unless a vCPU manages to execute for really long
* time uninterrupted). In case it happens, it makes no sense to
* track even deeper undershoots.
*
* NOTE: If making this smaller than -CSCHED2_CREDIT_INIT, adjust
* reset_credit() accordingly.
*/
#define CSCHED2_CREDIT_MIN (-CSCHED2_CREDIT_INIT)
/*
* Amount of credit the idle units have. It never changes, as idle
* units does not consume credits, and it must be lower than whatever
* amount of credit 'regular' unit would end up with.
*/
#define CSCHED2_IDLE_CREDIT (-(1U<<30))
#define CSCHED2_IDLE_CREDIT (CSCHED2_CREDIT_MIN-1)
/*
* Carryover: How much "extra" credit may be carried over after
* a reset.
Expand Down Expand Up @@ -781,10 +791,15 @@ static int get_fallback_cpu(struct csched2_unit *svc)
static void t2c_update(const struct csched2_runqueue_data *rqd, s_time_t time,
struct csched2_unit *svc)
{
uint64_t val = time * rqd->max_weight + svc->residual;
int64_t val = time * rqd->max_weight + svc->residual;

svc->residual = do_div(val, svc->weight);
svc->credit -= val;
/* Getting to lower credit than CSCHED2_CREDIT_MIN makes no sense. */
val = svc->credit - val;
if ( unlikely(val < CSCHED2_CREDIT_MIN) )
svc->credit = CSCHED2_CREDIT_MIN;
else
svc->credit = val;
}

static s_time_t c2t(const struct csched2_runqueue_data *rqd, s_time_t credit,
Expand Down Expand Up @@ -1616,28 +1631,25 @@ static void reset_credit(int cpu, s_time_t now, struct csched2_unit *snext)
{
struct csched2_runqueue_data *rqd = c2rqd(cpu);
struct list_head *iter;
int m;
int reset = CSCHED2_CREDIT_INIT;

/*
* Under normal circumstances, snext->credit should never be less
* than -CSCHED2_MIN_TIMER. However, under some circumstances, an
* unit with low credits may be allowed to run long enough that
* its credits are actually less than -CSCHED2_CREDIT_INIT.
* its credits are actually much lower than that.
* (Instances have been observed, for example, where an unit with
* 200us of credit was allowed to run for 11ms, giving it -10.8ms
* of credit. Thus it was still negative even after the reset.)
*
* If this is the case for snext, we simply want to keep moving
* everyone up until it is in the black again. This fair because
* none of the other units want to run at the moment.
*
* Rather than looping, however, we just calculate a multiplier,
* avoiding an integer division and multiplication in the common
* case.
* everyone up until it is in the black again. This means that,
* since CSCHED2_CREDIT_MIN is -CSCHED2_CREDIT_INIT, we need to
* actually add 2*CSCHED2_CREDIT_INIT.
*/
m = 1;
if ( snext->credit < -CSCHED2_CREDIT_INIT )
m += (-snext->credit) / CSCHED2_CREDIT_INIT;
ASSERT(snext->credit >= CSCHED2_CREDIT_MIN);
if ( unlikely(snext->credit == CSCHED2_CREDIT_MIN) )
reset += CSCHED2_CREDIT_INIT;

list_for_each( iter, &rqd->svc )
{
Expand Down Expand Up @@ -1668,15 +1680,7 @@ static void reset_credit(int cpu, s_time_t now, struct csched2_unit *snext)
}

start_credit = svc->credit;

/*
* Add INIT * m, avoiding integer multiplication in the common case.
*/
if ( likely(m==1) )
svc->credit += CSCHED2_CREDIT_INIT;
else
svc->credit += m * CSCHED2_CREDIT_INIT;

svc->credit += reset;
/* "Clip" credits to max carryover */
if ( svc->credit > CSCHED2_CREDIT_INIT + CSCHED2_CARRYOVER_MAX )
svc->credit = CSCHED2_CREDIT_INIT + CSCHED2_CARRYOVER_MAX;
Expand All @@ -1688,19 +1692,18 @@ static void reset_credit(int cpu, s_time_t now, struct csched2_unit *snext)
struct {
unsigned unit:16, dom:16;
int credit_start, credit_end;
unsigned multiplier;
} d;
d.dom = svc->unit->domain->domain_id;
d.unit = svc->unit->unit_id;
d.credit_start = start_credit;
d.credit_end = svc->credit;
d.multiplier = m;
__trace_var(TRC_CSCHED2_CREDIT_RESET, 1,
sizeof(d),
(unsigned char *)&d);
}
}

ASSERT(snext->credit > 0);
SCHED_STAT_CRANK(credit_reset);

/* No need to resort runqueue, as everyone's order should be the same. */
Expand Down

0 comments on commit 36f3662

Please sign in to comment.