]> git.baikalelectronics.ru Git - kernel.git/commitdiff
Revert "cpufreq: CPPC: Add support for frequency invariance"
authorViresh Kumar <viresh.kumar@linaro.org>
Fri, 11 Jun 2021 03:18:02 +0000 (08:48 +0530)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Mon, 14 Jun 2021 13:55:02 +0000 (15:55 +0200)
This reverts commit 3bb11fbdcf8a225ce9e56c6601bcf15975d409ea.

There are few races in the frequency invariance support for CPPC driver,
namely the driver doesn't stop the kthread_work and irq_work on policy
exit during suspend/resume or CPU hotplug.

A proper fix won't be possible for the 5.13-rc, as it requires a lot of
changes. Lets revert the patch instead for now.

Fixes: 3bb11fbdcf8a ("cpufreq: CPPC: Add support for frequency invariance")
Reported-by: Qian Cai <quic_qiancai@quicinc.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/cpufreq/Kconfig.arm
drivers/cpufreq/cppc_cpufreq.c
include/linux/arch_topology.h
kernel/sched/core.c

index a5c5f70acfc9ef0f7634736926e77bed75bd1614..e65e0a43be6449b655e2126ce133a786c1554a02 100644 (file)
@@ -19,16 +19,6 @@ config ACPI_CPPC_CPUFREQ
 
          If in doubt, say N.
 
-config ACPI_CPPC_CPUFREQ_FIE
-       bool "Frequency Invariance support for CPPC cpufreq driver"
-       depends on ACPI_CPPC_CPUFREQ && GENERIC_ARCH_TOPOLOGY
-       default y
-       help
-         This extends frequency invariance support in the CPPC cpufreq driver,
-         by using CPPC delivered and reference performance counters.
-
-         If in doubt, say N.
-
 config ARM_ALLWINNER_SUN50I_CPUFREQ_NVMEM
        tristate "Allwinner nvmem based SUN50I CPUFreq driver"
        depends on ARCH_SUNXI
index 3848b4c222e1332cae44f19503c03c2177c95fa6..2f769b1630c5714d1cc03cb82f420baffb8c82d7 100644 (file)
 
 #define pr_fmt(fmt)    "CPPC Cpufreq:" fmt
 
-#include <linux/arch_topology.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/delay.h>
 #include <linux/cpu.h>
 #include <linux/cpufreq.h>
 #include <linux/dmi.h>
-#include <linux/irq_work.h>
-#include <linux/kthread.h>
 #include <linux/time.h>
 #include <linux/vmalloc.h>
-#include <uapi/linux/sched/types.h>
 
 #include <asm/unaligned.h>
 
@@ -61,204 +57,6 @@ static struct cppc_workaround_oem_info wa_info[] = {
        }
 };
 
-#ifdef CONFIG_ACPI_CPPC_CPUFREQ_FIE
-
-/* Frequency invariance support */
-struct cppc_freq_invariance {
-       int cpu;
-       struct irq_work irq_work;
-       struct kthread_work work;
-       struct cppc_perf_fb_ctrs prev_perf_fb_ctrs;
-       struct cppc_cpudata *cpu_data;
-};
-
-static DEFINE_PER_CPU(struct cppc_freq_invariance, cppc_freq_inv);
-static struct kthread_worker *kworker_fie;
-static bool fie_disabled;
-
-static struct cpufreq_driver cppc_cpufreq_driver;
-static unsigned int hisi_cppc_cpufreq_get_rate(unsigned int cpu);
-static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
-                                struct cppc_perf_fb_ctrs fb_ctrs_t0,
-                                struct cppc_perf_fb_ctrs fb_ctrs_t1);
-
-/**
- * cppc_scale_freq_workfn - CPPC arch_freq_scale updater for frequency invariance
- * @work: The work item.
- *
- * The CPPC driver register itself with the topology core to provide its own
- * implementation (cppc_scale_freq_tick()) of topology_scale_freq_tick() which
- * gets called by the scheduler on every tick.
- *
- * Note that the arch specific counters have higher priority than CPPC counters,
- * if available, though the CPPC driver doesn't need to have any special
- * handling for that.
- *
- * On an invocation of cppc_scale_freq_tick(), we schedule an irq work (since we
- * reach here from hard-irq context), which then schedules a normal work item
- * and cppc_scale_freq_workfn() updates the per_cpu arch_freq_scale variable
- * based on the counter updates since the last tick.
- */
-static void cppc_scale_freq_workfn(struct kthread_work *work)
-{
-       struct cppc_freq_invariance *cppc_fi;
-       struct cppc_perf_fb_ctrs fb_ctrs = {0};
-       struct cppc_cpudata *cpu_data;
-       unsigned long local_freq_scale;
-       u64 perf;
-
-       cppc_fi = container_of(work, struct cppc_freq_invariance, work);
-       cpu_data = cppc_fi->cpu_data;
-
-       if (cppc_get_perf_ctrs(cppc_fi->cpu, &fb_ctrs)) {
-               pr_warn("%s: failed to read perf counters\n", __func__);
-               return;
-       }
-
-       cppc_fi->prev_perf_fb_ctrs = fb_ctrs;
-       perf = cppc_perf_from_fbctrs(cpu_data, cppc_fi->prev_perf_fb_ctrs,
-                                    fb_ctrs);
-
-       perf <<= SCHED_CAPACITY_SHIFT;
-       local_freq_scale = div64_u64(perf, cpu_data->perf_caps.highest_perf);
-       if (WARN_ON(local_freq_scale > 1024))
-               local_freq_scale = 1024;
-
-       per_cpu(arch_freq_scale, cppc_fi->cpu) = local_freq_scale;
-}
-
-static void cppc_irq_work(struct irq_work *irq_work)
-{
-       struct cppc_freq_invariance *cppc_fi;
-
-       cppc_fi = container_of(irq_work, struct cppc_freq_invariance, irq_work);
-       kthread_queue_work(kworker_fie, &cppc_fi->work);
-}
-
-static void cppc_scale_freq_tick(void)
-{
-       struct cppc_freq_invariance *cppc_fi = &per_cpu(cppc_freq_inv, smp_processor_id());
-
-       /*
-        * cppc_get_perf_ctrs() can potentially sleep, call that from the right
-        * context.
-        */
-       irq_work_queue(&cppc_fi->irq_work);
-}
-
-static struct scale_freq_data cppc_sftd = {
-       .source = SCALE_FREQ_SOURCE_CPPC,
-       .set_freq_scale = cppc_scale_freq_tick,
-};
-
-static void cppc_freq_invariance_policy_init(struct cpufreq_policy *policy,
-                                            struct cppc_cpudata *cpu_data)
-{
-       struct cppc_perf_fb_ctrs fb_ctrs = {0};
-       struct cppc_freq_invariance *cppc_fi;
-       int i, ret;
-
-       if (cppc_cpufreq_driver.get == hisi_cppc_cpufreq_get_rate)
-               return;
-
-       if (fie_disabled)
-               return;
-
-       for_each_cpu(i, policy->cpus) {
-               cppc_fi = &per_cpu(cppc_freq_inv, i);
-               cppc_fi->cpu = i;
-               cppc_fi->cpu_data = cpu_data;
-               kthread_init_work(&cppc_fi->work, cppc_scale_freq_workfn);
-               init_irq_work(&cppc_fi->irq_work, cppc_irq_work);
-
-               ret = cppc_get_perf_ctrs(i, &fb_ctrs);
-               if (ret) {
-                       pr_warn("%s: failed to read perf counters: %d\n",
-                               __func__, ret);
-                       fie_disabled = true;
-               } else {
-                       cppc_fi->prev_perf_fb_ctrs = fb_ctrs;
-               }
-       }
-}
-
-static void __init cppc_freq_invariance_init(void)
-{
-       struct sched_attr attr = {
-               .size           = sizeof(struct sched_attr),
-               .sched_policy   = SCHED_DEADLINE,
-               .sched_nice     = 0,
-               .sched_priority = 0,
-               /*
-                * Fake (unused) bandwidth; workaround to "fix"
-                * priority inheritance.
-                */
-               .sched_runtime  = 1000000,
-               .sched_deadline = 10000000,
-               .sched_period   = 10000000,
-       };
-       int ret;
-
-       if (cppc_cpufreq_driver.get == hisi_cppc_cpufreq_get_rate)
-               return;
-
-       if (fie_disabled)
-               return;
-
-       kworker_fie = kthread_create_worker(0, "cppc_fie");
-       if (IS_ERR(kworker_fie))
-               return;
-
-       ret = sched_setattr_nocheck(kworker_fie->task, &attr);
-       if (ret) {
-               pr_warn("%s: failed to set SCHED_DEADLINE: %d\n", __func__,
-                       ret);
-               kthread_destroy_worker(kworker_fie);
-               return;
-       }
-
-       /* Register for freq-invariance */
-       topology_set_scale_freq_source(&cppc_sftd, cpu_present_mask);
-}
-
-static void cppc_freq_invariance_exit(void)
-{
-       struct cppc_freq_invariance *cppc_fi;
-       int i;
-
-       if (cppc_cpufreq_driver.get == hisi_cppc_cpufreq_get_rate)
-               return;
-
-       if (fie_disabled)
-               return;
-
-       topology_clear_scale_freq_source(SCALE_FREQ_SOURCE_CPPC, cpu_present_mask);
-
-       for_each_possible_cpu(i) {
-               cppc_fi = &per_cpu(cppc_freq_inv, i);
-               irq_work_sync(&cppc_fi->irq_work);
-       }
-
-       kthread_destroy_worker(kworker_fie);
-       kworker_fie = NULL;
-}
-
-#else
-static inline void
-cppc_freq_invariance_policy_init(struct cpufreq_policy *policy,
-                                struct cppc_cpudata *cpu_data)
-{
-}
-
-static inline void cppc_freq_invariance_init(void)
-{
-}
-
-static inline void cppc_freq_invariance_exit(void)
-{
-}
-#endif /* CONFIG_ACPI_CPPC_CPUFREQ_FIE */
-
 /* Callback function used to retrieve the max frequency from DMI */
 static void cppc_find_dmi_mhz(const struct dmi_header *dm, void *private)
 {
@@ -547,12 +345,9 @@ static int cppc_cpufreq_cpu_init(struct cpufreq_policy *policy)
        cpu_data->perf_ctrls.desired_perf =  caps->highest_perf;
 
        ret = cppc_set_perf(cpu, &cpu_data->perf_ctrls);
-       if (ret) {
+       if (ret)
                pr_debug("Err setting perf value:%d on CPU:%d. ret:%d\n",
                         caps->highest_perf, cpu, ret);
-       } else {
-               cppc_freq_invariance_policy_init(policy, cpu_data);
-       }
 
        return ret;
 }
@@ -565,12 +360,12 @@ static inline u64 get_delta(u64 t1, u64 t0)
        return (u32)t1 - (u32)t0;
 }
 
-static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
-                                struct cppc_perf_fb_ctrs fb_ctrs_t0,
-                                struct cppc_perf_fb_ctrs fb_ctrs_t1)
+static int cppc_get_rate_from_fbctrs(struct cppc_cpudata *cpu_data,
+                                    struct cppc_perf_fb_ctrs fb_ctrs_t0,
+                                    struct cppc_perf_fb_ctrs fb_ctrs_t1)
 {
        u64 delta_reference, delta_delivered;
-       u64 reference_perf;
+       u64 reference_perf, delivered_perf;
 
        reference_perf = fb_ctrs_t0.reference_perf;
 
@@ -579,21 +374,12 @@ static int cppc_perf_from_fbctrs(struct cppc_cpudata *cpu_data,
        delta_delivered = get_delta(fb_ctrs_t1.delivered,
                                    fb_ctrs_t0.delivered);
 
-       /* Check to avoid divide-by zero and invalid delivered_perf */
-       if (!delta_reference || !delta_delivered)
-               return cpu_data->perf_ctrls.desired_perf;
-
-       return (reference_perf * delta_delivered) / delta_reference;
-}
-
-static int cppc_get_rate_from_fbctrs(struct cppc_cpudata *cpu_data,
-                                    struct cppc_perf_fb_ctrs fb_ctrs_t0,
-                                    struct cppc_perf_fb_ctrs fb_ctrs_t1)
-{
-       u64 delivered_perf;
-
-       delivered_perf = cppc_perf_from_fbctrs(cpu_data, fb_ctrs_t0,
-                                              fb_ctrs_t1);
+       /* Check to avoid divide-by zero */
+       if (delta_reference || delta_delivered)
+               delivered_perf = (reference_perf * delta_delivered) /
+                                       delta_reference;
+       else
+               delivered_perf = cpu_data->perf_ctrls.desired_perf;
 
        return cppc_cpufreq_perf_to_khz(cpu_data, delivered_perf);
 }
@@ -718,8 +504,6 @@ static void cppc_check_hisi_workaround(void)
 
 static int __init cppc_cpufreq_init(void)
 {
-       int ret;
-
        if ((acpi_disabled) || !acpi_cpc_valid())
                return -ENODEV;
 
@@ -727,11 +511,7 @@ static int __init cppc_cpufreq_init(void)
 
        cppc_check_hisi_workaround();
 
-       ret = cpufreq_register_driver(&cppc_cpufreq_driver);
-       if (!ret)
-               cppc_freq_invariance_init();
-
-       return ret;
+       return cpufreq_register_driver(&cppc_cpufreq_driver);
 }
 
 static inline void free_cpu_data(void)
@@ -748,7 +528,6 @@ static inline void free_cpu_data(void)
 
 static void __exit cppc_cpufreq_exit(void)
 {
-       cppc_freq_invariance_exit();
        cpufreq_unregister_driver(&cppc_cpufreq_driver);
 
        free_cpu_data();
index f180240dc95f41d4f5992499fc9b1ce0d0f24bb0..11e555cfaecb496f7040c18f1b3b06ad4ec36a2e 100644 (file)
@@ -37,7 +37,6 @@ bool topology_scale_freq_invariant(void);
 enum scale_freq_source {
        SCALE_FREQ_SOURCE_CPUFREQ = 0,
        SCALE_FREQ_SOURCE_ARCH,
-       SCALE_FREQ_SOURCE_CPPC,
 };
 
 struct scale_freq_data {
index 5226cc26a095f427dafa864ff37d39d529e2ba85..4ca80df205ce6a43686ddb993066daf37dd134b0 100644 (file)
@@ -6389,7 +6389,6 @@ int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
 {
        return __sched_setscheduler(p, attr, false, true);
 }
-EXPORT_SYMBOL_GPL(sched_setattr_nocheck);
 
 /**
  * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.