LinuxLists.cc - [tip:sched/core 3/19] kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util

2022-11-12 10:26:07

Subject: [tip:sched/core 3/19] kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.

tree:
.kernel.org/pub/scm/linux/kernel/git/tip/tip.git">https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/core
c61873739d90dd0a16f
e55ae5188fab4f428cb [3/19] sched/uclamp: Fix fits_capacity() check in feec()
/> (GCC) 12.1.0
kindly add following tag where applicable
test robot <[email protected]>
<[email protected]>
find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.
find_energy_efficient_cpu() error: uninitialized symbol 'util_max'.
kernel/sched/fair.c
2018-12-03 7168 static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
2018-12-03 7169 {
2022-06-21 7170 struct cpumask *cpus = this_cpu_cpumask_var_ptr(select_rq_mask);
2019-09-12 7171 unsigned long prev_delta = ULONG_MAX, best_delta = ULONG_MAX;
2022-08-04 7172 unsigned long p_util_min = uclamp_is_used() ? uclamp_eff_value(p, UCLAMP_MIN) : 0;
2022-08-04 7173 unsigned long p_util_max = uclamp_is_used() ? uclamp_eff_value(p, UCLAMP_MAX) : 1024;
2022-06-21 7174 struct root_domain *rd = this_rq()->rd;
2022-06-21 7175 int cpu, best_energy_cpu, target = -1;
2018-12-03 7176 struct sched_domain *sd;
2019-09-12 7177 struct perf_domain *pd;
2022-06-21 7178 struct energy_env eenv;
2018-12-03 7179
2018-12-03 7180 rcu_read_lock();
2018-12-03 7181 pd = rcu_dereference(rd->pd);
2018-12-03 7182 if (!pd || READ_ONCE(rd->overutilized))
2021-05-04 7183 goto unlock;
2018-12-03 7184
2018-12-03 7185 /*
2018-12-03 7186 * Energy-aware wake-up happens on the lowest sched_domain starting
2018-12-03 7187 * from sd_asym_cpucapacity spanning over this_cpu and prev_cpu.
2018-12-03 7188 */
2018-12-03 7189 sd = rcu_dereference(*this_cpu_ptr(&sd_asym_cpucapacity));
2018-12-03 7190 while (sd && !cpumask_test_cpu(prev_cpu, sched_domain_span(sd)))
2018-12-03 7191 sd = sd->parent;
2018-12-03 7192 if (!sd)
2021-05-04 7193 goto unlock;
2021-05-04 7194
2021-05-04 7195 target = prev_cpu;
2018-12-03 7196
2018-12-03 7197 sync_entity_load_avg(&p->se);
2018-12-03 7198 if (!task_util_est(p))
2018-12-03 7199 goto unlock;
2018-12-03 7200
2022-06-21 7201 eenv_task_busy_time(&eenv, p, prev_cpu);
2022-06-21 7202
2018-12-03 7203 for (; pd; pd = pd->next) {
2022-06-21 7204 unsigned long cpu_cap, cpu_thermal_cap, util;
2022-06-21 7205 unsigned long cur_delta, max_spare_cap = 0;
2022-08-04 7206 unsigned long rq_util_min, rq_util_max;
2022-08-04 7207 unsigned long util_min, util_max;
2021-05-04 7208 bool compute_prev_delta = false;
2018-12-03 7209 int max_spare_cap_cpu = -1;
2022-06-21 7210 unsigned long base_energy;
2018-12-03 7211
2022-06-21 7212 cpumask_and(cpus, perf_domain_span(pd), cpu_online_mask);
2022-06-21 7213
2022-06-21 7214 if (cpumask_empty(cpus))
2022-06-21 7215 continue;
2022-06-21 7216
2022-06-21 7217 /* Account thermal pressure for the energy estimation */
2022-06-21 7218 cpu = cpumask_first(cpus);
2022-06-21 7219 cpu_thermal_cap = arch_scale_cpu_capacity(cpu);
2022-06-21 7220 cpu_thermal_cap -= arch_scale_thermal_pressure(cpu);
2022-06-21 7221
2022-06-21 7222 eenv.cpu_cap = cpu_thermal_cap;
2022-06-21 7223 eenv.pd_cap = 0;
2022-06-21 7224
2022-06-21 7225 for_each_cpu(cpu, cpus) {
2022-06-21 7226 eenv.pd_cap += cpu_thermal_cap;
2022-06-21 7227
2022-06-21 7228 if (!cpumask_test_cpu(cpu, sched_domain_span(sd)))
2022-06-21 7229 continue;
2022-06-21 7230
Andrzej Siewior 2019-04-23 7231 if (!cpumask_test_cpu(cpu, p->cpus_ptr))
2018-12-03 7232 continue;
2018-12-03 7233
2018-12-03 7234 util = cpu_util_next(cpu, p, cpu);
2018-12-03 7235 cpu_cap = capacity_of(cpu);
2019-12-11 7236
2019-12-11 7237 /*
2019-12-11 7238 * Skip CPUs that cannot satisfy the capacity request.
2019-12-11 7239 * IOW, placing the task there would make the CPU
2019-12-11 7240 * overutilized. Take uclamp into account to see how
2019-12-11 7241 * much capacity we can get out of the CPU; this is
2020-12-08 7242 * aligned with sched_cpu_util().
2019-12-11 7243 */
2022-08-04 7244 if (uclamp_is_used()) {
2022-08-04 7245 if (uclamp_rq_is_idle(cpu_rq(cpu))) {
2022-08-04 7246 util_min = p_util_min;
2022-08-04 7247 util_max = p_util_max;
2022-08-04 7248 } else {
2022-08-04 7249 /*
2022-08-04 7250 * Open code uclamp_rq_util_with() except for
2022-08-04 7251 * the clamp() part. Ie: apply max aggregation
2022-08-04 7252 * only. util_fits_cpu() logic requires to
2022-08-04 7253 * operate on non clamped util but must use the
2022-08-04 7254 * max-aggregated uclamp_{min, max}.
2022-08-04 7255 */
2022-08-04 7256 rq_util_min = uclamp_rq_get(cpu_rq(cpu), UCLAMP_MIN);
2022-08-04 7257 rq_util_max = uclamp_rq_get(cpu_rq(cpu), UCLAMP_MAX);
2022-08-04 7258
2022-08-04 7259 util_min = max(rq_util_min, p_util_min);
2022-08-04 7260 util_max = max(rq_util_max, p_util_max);
2022-08-04 7261 }
2022-08-04 7262 }
if uclamp_is_used() is false. (I thought
earlier but I don't see it on vger).
2022-08-04 @7263 if (!util_fits_cpu(util, util_min, util_max, cpu))
2018-12-03 7264 continue;
2018-12-03 7265
2022-06-21 7266 lsub_positive(&cpu_cap, util);
2022-06-21 7267
2018-12-03 7268 if (cpu == prev_cpu) {
2021-05-04 7269 /* Always use prev_cpu as a candidate. */
2021-05-04 7270 compute_prev_delta = true;
2022-06-21 7271 } else if (cpu_cap > max_spare_cap) {
2018-12-03 7272 /*
2021-05-04 7273 * Find the CPU with the maximum spare capacity
2021-05-04 7274 * in the performance domain.
2018-12-03 7275 */
2022-06-21 7276 max_spare_cap = cpu_cap;
2018-12-03 7277 max_spare_cap_cpu = cpu;
2018-12-03 7278 }
2018-12-03 7279 }
2018-12-03 7280
2021-05-04 7281 if (max_spare_cap_cpu < 0 && !compute_prev_delta)
2021-05-04 7282 continue;
2021-05-04 7283
2022-06-21 7284 eenv_pd_busy_time(&eenv, cpus, p);
2021-05-04 7285 /* Compute the 'base' energy of the pd, without @p */
2022-06-21 7286 base_energy = compute_energy(&eenv, pd, cpus, p, -1);
2021-05-04 7287
2021-05-04 7288 /* Evaluate the energy impact of using prev_cpu. */
2021-05-04 7289 if (compute_prev_delta) {
2022-06-21 7290 prev_delta = compute_energy(&eenv, pd, cpus, p,
2022-06-21 7291 prev_cpu);
2022-06-21 7292 /* CPU utilization has changed */
2022-06-21 7293 if (prev_delta < base_energy)
2021-05-04 7294 goto unlock;
2022-06-21 7295 prev_delta -= base_energy;
2021-05-04 7296 best_delta = min(best_delta, prev_delta);
2021-05-04 7297 }
2021-05-04 7298
2021-05-04 7299 /* Evaluate the energy impact of using max_spare_cap_cpu. */
2021-05-04 7300 if (max_spare_cap_cpu >= 0) {
2022-06-21 7301 cur_delta = compute_energy(&eenv, pd, cpus, p,
2022-06-21 7302 max_spare_cap_cpu);
2022-06-21 7303 /* CPU utilization has changed */
2022-06-21 7304 if (cur_delta < base_energy)
2021-05-04 7305 goto unlock;
2022-06-21 7306 cur_delta -= base_energy;
2019-09-12 7307 if (cur_delta < best_delta) {
2019-09-12 7308 best_delta = cur_delta;
2018-12-03 7309 best_energy_cpu = max_spare_cap_cpu;
2018-12-03 7310 }
2018-12-03 7311 }
2018-12-03 7312 }
2018-12-03 7313 rcu_read_unlock();
2018-12-03 7314
2022-06-21 7315 if (best_delta < prev_delta)
2021-05-04 7316 target = best_energy_cpu;
2018-12-03 7317
2021-05-04 7318 return target;
2018-12-03 7319
2021-05-04 7320 unlock:
2018-12-03 7321 rcu_read_unlock();
2018-12-03 7322
2021-05-04 7323 return target;
2018-12-03 7324 }
>https://01.org/lkp

2022-11-12 18:16:20

by Qais Yousef

[permalink] [raw]

Subject: Re: [tip:sched/core 3/19] kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.

On 11/12/22 13:16, Dan Carpenter wrote:
> tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/core
> head: 52b33d87b9197c51e8ffdc61873739d90dd0a16f
> commit: 244226035a1f9b2b6c326e55ae5188fab4f428cb [3/19] sched/uclamp: Fix fits_capacity() check in feec()
> config: riscv-randconfig-m031-20221111
> compiler: riscv64-linux-gcc (GCC) 12.1.0
>
> If you fix the issue, kindly add following tag where applicable
> | Reported-by: kernel test robot <[email protected]>
> | Reported-by: Dan Carpenter <[email protected]>
>
> smatch warnings:
> kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.
> kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_max'.
>
> vim +/util_min +7263 kernel/sched/fair.c

[...]

> util_min/max not initialized if uclamp_is_used() is false. (I thought
> I had reported this earlier but I don't see it on vger).

Thanks Dan!

It's harmless since util_min/max are ignored in util_fits_cpu() too;
uclamp_is_used() is checked there as well.

I couldn't reproduce, I need to get GCC 12.1.0, but the below ought to cure it.
Let me test it a bit more and send a patch.

Thanks

--
Qais Yousef

--->8---

idiff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 4cc56c91e06e..89dadaafc1ec 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -7217,10 +7217,10 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
eenv_task_busy_time(&eenv, p, prev_cpu);

for (; pd; pd = pd->next) {
+ unsigned long util_min = p_util_min, util_max = p_util_max;
unsigned long cpu_cap, cpu_thermal_cap, util;
unsigned long cur_delta, max_spare_cap = 0;
unsigned long rq_util_min, rq_util_max;
- unsigned long util_min, util_max;
unsigned long prev_spare_cap = 0;
int max_spare_cap_cpu = -1;
unsigned long base_energy;
@@ -7258,10 +7258,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
* aligned with sched_cpu_util().
*/
if (uclamp_is_used()) {
- if (uclamp_rq_is_idle(cpu_rq(cpu))) {
- util_min = p_util_min;
- util_max = p_util_max;
- } else {
+ if (!uclamp_rq_is_idle(cpu_rq(cpu))) {
/*
* Open code uclamp_rq_util_with() except for
* the clamp() part. Ie: apply max aggregation

2022-11-14 08:53:53

by Dan Carpenter

[permalink] [raw]

Subject: Re: [tip:sched/core 3/19] kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.

On Sat, Nov 12, 2022 at 05:47:05PM +0000, Qais Yousef wrote:
> On 11/12/22 13:16, Dan Carpenter wrote:
> > tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/core
> > head: 52b33d87b9197c51e8ffdc61873739d90dd0a16f
> > commit: 244226035a1f9b2b6c326e55ae5188fab4f428cb [3/19] sched/uclamp: Fix fits_capacity() check in feec()
> > config: riscv-randconfig-m031-20221111
> > compiler: riscv64-linux-gcc (GCC) 12.1.0
> >
> > If you fix the issue, kindly add following tag where applicable
> > | Reported-by: kernel test robot <[email protected]>
> > | Reported-by: Dan Carpenter <[email protected]>
> >
> > smatch warnings:
> > kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.
> > kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_max'.
> >
> > vim +/util_min +7263 kernel/sched/fair.c
>
> [...]
>
> > util_min/max not initialized if uclamp_is_used() is false. (I thought
> > I had reported this earlier but I don't see it on vger).
>
> Thanks Dan!
>
> It's harmless since util_min/max are ignored in util_fits_cpu() too;
> uclamp_is_used() is checked there as well.
>
> I couldn't reproduce, I need to get GCC 12.1.0, but the below ought to cure it.
> Let me test it a bit more and send a patch.

In that case it's fine. This is inlined. I guess, generally, we are
going to only consider passing uninitialized variables to functions as
bugs when the functions are not inlined. I believe that one of the
KMsan things will trigger a warning at runtime for this as well but it's
the same situation, where it's considered a false positive because it's
inlined. (Technically it's undefined behavior either way according to
the C standard, but the standard is sometimes useless).

It's not a GCC warning, it's from Smatch.

GCC uninitialized warnings were disabled by defualt because they have
too many false positives. Use "make W=2" to enable them. GCC will
complain.

regards,
dan carpenter

2022-11-20 21:48:15

by Qais Yousef

[permalink] [raw]

Subject: Re: [tip:sched/core 3/19] kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.

On 11/14/22 11:23, Dan Carpenter wrote:
> On Sat, Nov 12, 2022 at 05:47:05PM +0000, Qais Yousef wrote:
> > On 11/12/22 13:16, Dan Carpenter wrote:
> > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/core
> > > head: 52b33d87b9197c51e8ffdc61873739d90dd0a16f
> > > commit: 244226035a1f9b2b6c326e55ae5188fab4f428cb [3/19] sched/uclamp: Fix fits_capacity() check in feec()
> > > config: riscv-randconfig-m031-20221111
> > > compiler: riscv64-linux-gcc (GCC) 12.1.0
> > >
> > > If you fix the issue, kindly add following tag where applicable
> > > | Reported-by: kernel test robot <[email protected]>
> > > | Reported-by: Dan Carpenter <[email protected]>
> > >
> > > smatch warnings:
> > > kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.
> > > kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_max'.
> > >
> > > vim +/util_min +7263 kernel/sched/fair.c
> >
> > [...]
> >
> > > util_min/max not initialized if uclamp_is_used() is false. (I thought
> > > I had reported this earlier but I don't see it on vger).
> >
> > Thanks Dan!
> >
> > It's harmless since util_min/max are ignored in util_fits_cpu() too;
> > uclamp_is_used() is checked there as well.
> >
> > I couldn't reproduce, I need to get GCC 12.1.0, but the below ought to cure it.
> > Let me test it a bit more and send a patch.
>
> In that case it's fine. This is inlined. I guess, generally, we are
> going to only consider passing uninitialized variables to functions as
> bugs when the functions are not inlined. I believe that one of the
> KMsan things will trigger a warning at runtime for this as well but it's
> the same situation, where it's considered a false positive because it's
> inlined. (Technically it's undefined behavior either way according to
> the C standard, but the standard is sometimes useless).
>
> It's not a GCC warning, it's from Smatch.
>
> GCC uninitialized warnings were disabled by defualt because they have
> too many false positives. Use "make W=2" to enable them. GCC will
> complain.

Thanks Dan. I could reproduce it a while ago and I have the full patch below
- but I couldn't run it on real hardware just to do some sanity check and it
seems I'll have to wait for a while longer.

Putting it down here in case Peter fancies picking it up. I'll be posting it
along with other fixes in this area as soon as I get chance to run them
somewhere.

Cheers

--
Qais Yousef

--->8----

commit 8539b7d7b9e3c3bb6d817cbbd5be002f80f6eb31
Author: Qais Yousef <[email protected]>
Date: Sat Nov 19 23:12:34 2022 +0000

sched/uclamp: Fix a uninitialized variable warnings

Addresses the following warnings:

> config: riscv-randconfig-m031-20221111
> compiler: riscv64-linux-gcc (GCC) 12.1.0
>
> smatch warnings:
> kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_min'.
> kernel/sched/fair.c:7263 find_energy_efficient_cpu() error: uninitialized symbol 'util_max'.

Reported-by: kernel test robot <[email protected]>
Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Qais Yousef (Google) <[email protected]>

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 4cc56c91e06e..89dadaafc1ec 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -7217,10 +7217,10 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
eenv_task_busy_time(&eenv, p, prev_cpu);

for (; pd; pd = pd->next) {
+ unsigned long util_min = p_util_min, util_max = p_util_max;
unsigned long cpu_cap, cpu_thermal_cap, util;
unsigned long cur_delta, max_spare_cap = 0;
unsigned long rq_util_min, rq_util_max;
- unsigned long util_min, util_max;
unsigned long prev_spare_cap = 0;
int max_spare_cap_cpu = -1;
unsigned long base_energy;
@@ -7258,10 +7258,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
* aligned with sched_cpu_util().
*/
if (uclamp_is_used()) {
- if (uclamp_rq_is_idle(cpu_rq(cpu))) {
- util_min = p_util_min;
- util_max = p_util_max;
- } else {
+ if (!uclamp_rq_is_idle(cpu_rq(cpu))) {
/*
* Open code uclamp_rq_util_with() except for
* the clamp() part. Ie: apply max aggregation