cpufreq: intel_pstate: Get rid of unnecessary READ_ONCE() annotations
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Thu, 28 Mar 2024 18:52:06 +0000 (19:52 +0100)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 2 Apr 2024 10:56:40 +0000 (12:56 +0200)
Drop two redundant checks involving READ_ONCE() from notify_hwp_interrupt()
and make it check hwp_active without READ_ONCE() which is not necessary,
because that variable is only set once during the early initialization of
the driver.

In order to make that clear, annotate hwp_active with __ro_after_init.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/cpufreq/intel_pstate.c

index b702430dac29f179946a649c93cde99f42a8b2c4..fa707a207c8e5c6efb74eeeaa32fe4784a364d1c 100644 (file)
@@ -292,7 +292,7 @@ struct pstate_funcs {
 
 static struct pstate_funcs pstate_funcs __read_mostly;
 
-static int hwp_active __read_mostly;
+static bool hwp_active __ro_after_init;
 static int hwp_mode_bdw __read_mostly;
 static bool per_cpu_limits __read_mostly;
 static bool hwp_boost __read_mostly;
@@ -1636,11 +1636,10 @@ static cpumask_t hwp_intr_enable_mask;
 void notify_hwp_interrupt(void)
 {
        unsigned int this_cpu = smp_processor_id();
-       struct cpudata *cpudata;
        unsigned long flags;
        u64 value;
 
-       if (!READ_ONCE(hwp_active) || !boot_cpu_has(X86_FEATURE_HWP_NOTIFY))
+       if (!hwp_active || !boot_cpu_has(X86_FEATURE_HWP_NOTIFY))
                return;
 
        rdmsrl_safe(MSR_HWP_STATUS, &value);
@@ -1652,24 +1651,8 @@ void notify_hwp_interrupt(void)
        if (!cpumask_test_cpu(this_cpu, &hwp_intr_enable_mask))
                goto ack_intr;
 
-       /*
-        * Currently we never free all_cpu_data. And we can't reach here
-        * without this allocated. But for safety for future changes, added
-        * check.
-        */
-       if (unlikely(!READ_ONCE(all_cpu_data)))
-               goto ack_intr;
-
-       /*
-        * The free is done during cleanup, when cpufreq registry is failed.
-        * We wouldn't be here if it fails on init or switch status. But for
-        * future changes, added check.
-        */
-       cpudata = READ_ONCE(all_cpu_data[this_cpu]);
-       if (unlikely(!cpudata))
-               goto ack_intr;
-
-       schedule_delayed_work(&cpudata->hwp_notify_work, msecs_to_jiffies(10));
+       schedule_delayed_work(&all_cpu_data[this_cpu]->hwp_notify_work,
+                             msecs_to_jiffies(10));
 
        spin_unlock_irqrestore(&hwp_notify_lock, flags);
 
@@ -3464,7 +3447,7 @@ static int __init intel_pstate_init(void)
                 * deal with it.
                 */
                if ((!no_hwp && boot_cpu_has(X86_FEATURE_HWP_EPP)) || hwp_forced) {
-                       WRITE_ONCE(hwp_active, 1);
+                       hwp_active = true;
                        hwp_mode_bdw = id->driver_data;
                        intel_pstate.attr = hwp_cpufreq_attrs;
                        intel_cpufreq.attr = hwp_cpufreq_attrs;