2024-06-12 03:33:04

by PoShao Chen

[permalink] [raw]
Subject: [PATCH v2] cpufreq: Fix per-policy boost behavior after CPU hotplug

This patch fixes the behavior of the cpufreq boost when the
global boost flag is toggled during CPU hotplug offline. This action
previously led to incorrect scaling_max_freq values when the CPU was
brought back online. The issue also manifested as incorrect
scaling_cur_freq under the performance governor.

For example, after the following operations, even if the global boost
is disabled, the resulting scaling_max_freq and scaling_cur_freq
will still reflect the settings of an enabled boost.

$ echo performance > /sys/devices/system/cpu/cpufreq/policy7/scaling_governor
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
3200000
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
3200000

$ echo 1 > /sys/devices/system/cpu/cpufreq/boost
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
3250000
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
3250000

$ echo 0 > /sys/devices/system/cpu/cpu7/online
$ echo 0 > /sys/devices/system/cpu/cpufreq/boost
$ echo 1 > /sys/devices/system/cpu/cpu7/online
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
3250000
$ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
3250000

Signed-off-by: PoShao Chen <[email protected]>

---
V1 -> V2: Adjusted log messages and fixed build error.

Link:
https://lore.kernel.org/all/[email protected]/
---
drivers/cpufreq/cpufreq.c | 29 +++++++++++++++++++++++++++++
1 file changed, 29 insertions(+)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index a45aac17c20f..faadae05bc8a 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1495,6 +1495,35 @@ static int cpufreq_online(unsigned int cpu)

blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
CPUFREQ_CREATE_POLICY, policy);
+ } else {
+ /*
+ * Call freq_qos_update_request() for the per-policy boost flag mirror
+ * the cpufreq_driver boost during hotplug online.
+ * Register an online callback if the default mirroring of the global
+ * boost setting is not intended.
+ */
+ if (!cpufreq_driver->online) {
+ ret = freq_qos_update_request(policy->max_freq_req, policy->max);
+ if (ret)
+ pr_err("%s: freq qos update failed\n", __func__);
+ } else {
+ /*
+ * Let the per-policy boost flag mirror the cpufreq_driver
+ * boost if an illegal state occurs after hotplug
+ */
+ if (policy->boost_enabled && !cpufreq_driver->boost_enabled) {
+ pr_info("%s: local boost flag mirror the global boost\n",
+ __func__);
+ policy->boost_enabled = cpufreq_driver->boost_enabled;
+ ret = cpufreq_driver->set_boost(policy,
+ cpufreq_driver->boost_enabled);
+ if (ret) {
+ policy->boost_enabled = !policy->boost_enabled;
+ pr_err("%s: Failed to mirror the global boost flag\n",
+ __func__);
+ }
+ }
+ }
}

if (cpufreq_driver->get && has_target()) {
--
2.18.0



2024-06-13 09:21:10

by Viresh Kumar

[permalink] [raw]
Subject: Re: [PATCH v2] cpufreq: Fix per-policy boost behavior after CPU hotplug

On 12-06-24, 11:31, PoShao Chen wrote:
> This patch fixes the behavior of the cpufreq boost when the
> global boost flag is toggled during CPU hotplug offline. This action
> previously led to incorrect scaling_max_freq values when the CPU was
> brought back online. The issue also manifested as incorrect
> scaling_cur_freq under the performance governor.
>
> For example, after the following operations, even if the global boost
> is disabled, the resulting scaling_max_freq and scaling_cur_freq
> will still reflect the settings of an enabled boost.
>
> $ echo performance > /sys/devices/system/cpu/cpufreq/policy7/scaling_governor
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
> 3200000
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
> 3200000
>
> $ echo 1 > /sys/devices/system/cpu/cpufreq/boost
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
> 3250000
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
> 3250000
>
> $ echo 0 > /sys/devices/system/cpu/cpu7/online
> $ echo 0 > /sys/devices/system/cpu/cpufreq/boost
> $ echo 1 > /sys/devices/system/cpu/cpu7/online
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_cur_freq
> 3250000
> $ cat /sys/devices/system/cpu/cpufreq/policy7/scaling_max_freq
> 3250000

Please try this instead:

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 7c6879efe9ef..bd9fe2b0f032 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -43,6 +43,9 @@ static LIST_HEAD(cpufreq_policy_list);
#define for_each_inactive_policy(__policy) \
for_each_suitable_policy(__policy, false)

+#define for_each_policy(__policy) \
+ list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
+
/* Iterate over governors */
static LIST_HEAD(cpufreq_governor_list);
#define for_each_governor(__governor) \
@@ -2815,7 +2818,7 @@ int cpufreq_boost_trigger_state(int state)
write_unlock_irqrestore(&cpufreq_driver_lock, flags);

cpus_read_lock();
- for_each_active_policy(policy) {
+ for_each_policy(policy) {
policy->boost_enabled = state;
ret = cpufreq_driver->set_boost(policy, state);
if (ret) {

--
viresh