cpufreq: Specify the default governor on command line
Subsystems (5)
v2 cover cpufreq: Specify the default governor on command line
This series enables users of prebuilt kernels (e.g. distro kernels) to specify their CPUfreq governor of choice using the kernel command line, instead of having to wait for the system to fully boot to userspace to switch using the sysfs interface. This is helpful for 2 reasons:
1. users get to choose the governor that runs during the actual boot;
2. it simplifies the userspace boot procedure a bit (one less thing to worry about).
To enable this, the first patch moves all governor init calls to core_initcall, to make sure they are registered by the time the drivers probe. This should be relatively low impact as registering a governor is a simple procedure (it gets added to a llist), and all governors already load at core_initcall anyway when they're set as the default in Kconfig. This also allows to clean-up the governors' init/exit code, and reduces boilerplate.
The second patch introduces the new command line parameter, inspired by its cpuidle counterpart. More details can be found in the respective patch headers.
Changes in v2:
- added Viresh's ack to patch 01
- moved the assignment of 'default_governor' in patch 02 to the governor registration path instead of the driver registration (Viresh)
Thanks,
Quentin
Quentin Perret (2):
cpufreq: Register governors at core_initcall
cpufreq: Specify default governor on command line
.../admin-guide/kernel-parameters.txt | 5 ++++ Documentation/admin-guide/pm/cpufreq.rst | 6 ++--- .../platforms/cell/cpufreq_spudemand.c | 26 ++----------------- drivers/cpufreq/cpufreq.c | 23 ++++++++++++---- drivers/cpufreq/cpufreq_conservative.c | 22 ++++------------ drivers/cpufreq/cpufreq_ondemand.c | 24 +++++------------ drivers/cpufreq/cpufreq_performance.c | 14 ++-------- drivers/cpufreq/cpufreq_powersave.c | 18 +++---------- drivers/cpufreq/cpufreq_userspace.c | 18 +++---------- include/linux/cpufreq.h | 14 ++++++++++ kernel/sched/cpufreq_schedutil.c | 6 +---- 11 files changed, 62 insertions(+), 114 deletions(-)
Signature · 1 line
2.27.0.111.gc72c7da667-goog
PATCH v2 1/2 cpufreq: Register governors at core_initcall
Currently, most CPUFreq governors are registered at core_initcall time when used as default, and module_init otherwise. In preparation for letting users specify the default governor on the kernel command line, change all of them to use core_initcall unconditionally, as is already the case for schedutil and performance. This will enable us to assume builtin governors have been registered before the builtin CPUFreq drivers probe.
And since all governors now have similar init/exit patterns, introduce two new macros cpufreq_governor_{init,exit}() to factorize the code.
---
Note: I couldn't boot-test the change to spudemand, by lack of hardware. But I can confirm cell_defconfig compiles just fine.
--- .../platforms/cell/cpufreq_spudemand.c | 26 ++----------------- drivers/cpufreq/cpufreq_conservative.c | 22 ++++------------ drivers/cpufreq/cpufreq_ondemand.c | 24 +++++------------ drivers/cpufreq/cpufreq_performance.c | 14 ++-------- drivers/cpufreq/cpufreq_powersave.c | 18 +++---------- drivers/cpufreq/cpufreq_userspace.c | 18 +++---------- include/linux/cpufreq.h | 14 ++++++++++ kernel/sched/cpufreq_schedutil.c | 6 +---- 8 files changed, 36 insertions(+), 106 deletions(-)
arch/powerpc/platforms/cell/cpufreq_spudemand.c +2 -24
@@ -126,30 +126,8 @@ static struct cpufreq_governor spu_governor = { .stop = spu_gov_stop, .owner = THIS_MODULE, }; - -/* -* module init and destoy -*/ - -static int __init spu_gov_init (void ) -{ -int ret; - - ret =cpufreq_register_governor (&spu_governor); -if (ret) -printk (KERN_ERR "registration of governor failed\n" ); -return ret; -} - -static void __exit spu_gov_exit(void) -{ -cpufreq_unregister_governor (&spu_governor); -} - - -module_init (spu_gov_init); -module_exit (spu_gov_exit); +cpufreq_governor_init (spu_governor); +cpufreq_governor_exit (spu_governor);MODULE_LICENSE ("GPL" );MODULE_AUTHOR ("Christian Krafft <[email protected]>" ); -
drivers/cpufreq/cpufreq_conservative.c +5 -17
@@ -322,17 +322,7 @@ static struct dbs_governor cs_governor = { .start = cs_start, }; -#define CPU_FREQ_GOV_CONSERVATIVE (&cs_governor.gov) - -static int __init cpufreq_gov_dbs_init (void ) -{ -return cpufreq_register_governor (CPU_FREQ_GOV_CONSERVATIVE ); -} - -static void __exit cpufreq_gov_dbs_exit(void) -{ -cpufreq_unregister_governor (CPU_FREQ_GOV_CONSERVATIVE ); -} +#define CPU_FREQ_GOV_CONSERVATIVE (cs_governor.gov)MODULE_AUTHOR ("Alexander Clouter <[email protected]>" ); MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for "
@@ -343,11 +333,9 @@ MODULE_LICENSE("GPL");#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE struct cpufreq_governor *cpufreq_default_governor (void ) { -return CPU_FREQ_GOV_CONSERVATIVE ; +return &CPU_FREQ_GOV_CONSERVATIVE ; } - -core_initcall (cpufreq_gov_dbs_init); -#else -module_init (cpufreq_gov_dbs_init);#endif -module_exit (cpufreq_gov_dbs_exit); + +cpufreq_governor_init (CPU_FREQ_GOV_CONSERVATIVE ); +cpufreq_governor_exit (CPU_FREQ_GOV_CONSERVATIVE );
drivers/cpufreq/cpufreq_ondemand.c +6 -18
@@ -408,7 +408,7 @@ static struct dbs_governor od_dbs_gov = { .start = od_start, }; -#define CPU_FREQ_GOV_ONDEMAND (&od_dbs_gov.gov) +#define CPU_FREQ_GOV_ONDEMAND (od_dbs_gov.gov)static void od_set_powersave_bias (unsigned int powersave_bias) {
@@ -429,7 +429,7 @@ static void od_set_powersave_bias(unsigned int powersave_bias)continue ; policy =cpufreq_cpu_get_raw (cpu); -if (!policy || policy->governor !=CPU_FREQ_GOV_ONDEMAND ) +if (!policy || policy->governor != &CPU_FREQ_GOV_ONDEMAND )continue ; policy_dbs = policy->governor_data;
@@ -461,16 +461,6 @@ void od_unregister_powersave_bias_handler(void) }EXPORT_SYMBOL_GPL (od_unregister_powersave_bias_handler); -static int __init cpufreq_gov_dbs_init (void ) -{ -return cpufreq_register_governor (CPU_FREQ_GOV_ONDEMAND ); -} - -static void __exit cpufreq_gov_dbs_exit(void) -{ -cpufreq_unregister_governor (CPU_FREQ_GOV_ONDEMAND ); -} -MODULE_AUTHOR ("Venkatesh Pallipadi <[email protected]>" );MODULE_AUTHOR ("Alexey Starikovskiy <[email protected]>" ); MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
@@ -480,11 +470,9 @@ MODULE_LICENSE("GPL");#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND struct cpufreq_governor *cpufreq_default_governor (void ) { -return CPU_FREQ_GOV_ONDEMAND ; +return &CPU_FREQ_GOV_ONDEMAND ; } - -core_initcall (cpufreq_gov_dbs_init); -#else -module_init (cpufreq_gov_dbs_init);#endif -module_exit (cpufreq_gov_dbs_exit); + +cpufreq_governor_init (CPU_FREQ_GOV_ONDEMAND ); +cpufreq_governor_exit (CPU_FREQ_GOV_ONDEMAND );
drivers/cpufreq/cpufreq_performance.c +2 -12
@@ -23,16 +23,6 @@ static struct cpufreq_governor cpufreq_gov_performance = { .limits = cpufreq_gov_performance_limits, }; -static int __init cpufreq_gov_performance_init (void ) -{ -return cpufreq_register_governor (&cpufreq_gov_performance); -} - -static void __exit cpufreq_gov_performance_exit(void) -{ -cpufreq_unregister_governor (&cpufreq_gov_performance); -} -#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE struct cpufreq_governor *cpufreq_default_governor (void ) {
@@ -50,5 +40,5 @@ MODULE_AUTHOR("Dominik Brodowski <[email protected]>");MODULE_DESCRIPTION ("CPUfreq policy governor 'performance'" );MODULE_LICENSE ("GPL" ); -core_initcall (cpufreq_gov_performance_init); -module_exit (cpufreq_gov_performance_exit); +cpufreq_governor_init (cpufreq_gov_performance); +cpufreq_governor_exit (cpufreq_gov_performance);
drivers/cpufreq/cpufreq_powersave.c +3 -15
@@ -23,16 +23,6 @@ static struct cpufreq_governor cpufreq_gov_powersave = { .owner = THIS_MODULE, }; -static int __init cpufreq_gov_powersave_init (void ) -{ -return cpufreq_register_governor (&cpufreq_gov_powersave); -} - -static void __exit cpufreq_gov_powersave_exit(void) -{ -cpufreq_unregister_governor (&cpufreq_gov_powersave); -} -MODULE_AUTHOR ("Dominik Brodowski <[email protected]>" );MODULE_DESCRIPTION ("CPUfreq policy governor 'powersave'" );MODULE_LICENSE ("GPL" );
@@ -42,9 +32,7 @@ struct cpufreq_governor *cpufreq_default_governor(void) {return &cpufreq_gov_powersave; } - -core_initcall (cpufreq_gov_powersave_init); -#else -module_init (cpufreq_gov_powersave_init);#endif -module_exit (cpufreq_gov_powersave_exit); + +cpufreq_governor_init (cpufreq_gov_powersave); +cpufreq_governor_exit (cpufreq_gov_powersave);
drivers/cpufreq/cpufreq_userspace.c +3 -15
@@ -126,16 +126,6 @@ static struct cpufreq_governor cpufreq_gov_userspace = { .owner = THIS_MODULE, }; -static int __init cpufreq_gov_userspace_init (void ) -{ -return cpufreq_register_governor (&cpufreq_gov_userspace); -} - -static void __exit cpufreq_gov_userspace_exit(void) -{ -cpufreq_unregister_governor (&cpufreq_gov_userspace); -} -MODULE_AUTHOR ("Dominik Brodowski <[email protected]>, " "Russell King <[email protected]>" );MODULE_DESCRIPTION ("CPUfreq policy governor 'userspace'" );
@@ -146,9 +136,7 @@ struct cpufreq_governor *cpufreq_default_governor(void) {return &cpufreq_gov_userspace; } - -core_initcall (cpufreq_gov_userspace_init); -#else -module_init (cpufreq_gov_userspace_init);#endif -module_exit (cpufreq_gov_userspace_exit); + +cpufreq_governor_init (cpufreq_gov_userspace); +cpufreq_governor_exit (cpufreq_gov_userspace);
include/linux/cpufreq.h +14 -0
@@ -577,6 +577,20 @@ unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy);int cpufreq_register_governor (struct cpufreq_governor *governor);void cpufreq_unregister_governor (struct cpufreq_governor *governor); +#define cpufreq_governor_init (__governor) \ +static int __init __governor##_init(void) \ +{ \ + return cpufreq_register_governor(&__governor); \ +} \ +core_initcall(__governor##_init) + +#define cpufreq_governor_exit (__governor) \ +static void __exit __governor##_exit(void) \ +{ \ + return cpufreq_unregister_governor(&__governor); \ +} \ +module_exit(__governor##_exit) +struct cpufreq_governor *cpufreq_default_governor (void );struct cpufreq_governor *cpufreq_fallback_governor (void );
kernel/sched/cpufreq_schedutil.c +1 -5
@@ -909,11 +909,7 @@ struct cpufreq_governor *cpufreq_default_governor(void) }#endif -static int __init sugov_register (void ) -{ -return cpufreq_register_governor (&schedutil_gov); -} -core_initcall (sugov_register); +cpufreq_governor_init (schedutil_gov);#ifdef CONFIG_ENERGY_MODEL extern bool sched_energy_update;
Signature · 1 line
2.27.0.111.gc72c7da667-goog
PATCH v2 2/2 cpufreq: Specify default governor on command line
In reply to Quentin PerretCurrently, the only way to specify the default CPUfreq governor is via Kconfig options, which suits users who can build the kernel themselves perfectly.
However, for those who use a distro-like kernel (such as Android, with the Generic Kernel Image project), the only way to use a different default is to boot to userspace, and to then switch using the sysfs interface. Being able to specify the default governor on the command line, like is the case for cpuidle, would enable those users to specify their governor of choice earlier on, and to simplify slighlty the userspace boot procedure.
To support this use-case, add a kernel command line parameter enabling to specify a default governor for CPUfreq, which takes precedence over the builtin default.
This implementation has one notable limitation: the default governor must be registered before the driver. This is solved for builtin governors and drivers using appropriate *_initcall() functions. And in the modular case, this must be reflected as a constraint on the module loading order.
--- .../admin-guide/kernel-parameters.txt | 5 ++++ Documentation/admin-guide/pm/cpufreq.rst | 6 ++--- drivers/cpufreq/cpufreq.c | 23 +++++++++++++++---- 3 files changed, 26 insertions(+), 8 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +5 -0
@@ -703,6 +703,11 @@ cpufreq.off=1 [CPU_FREQ] disable the cpufreq sub-system + cpufreq.default_governor= + [CPU_FREQ] Name of the default cpufreq governor to use. + This governor must be registered in the kernel before + the cpufreq driver probes. + cpu_init_udelay=N [X86] Delay for N microsec between assert and de-assert of APIC INIT to start processors. This delay occurs
Documentation/admin-guide/pm/cpufreq.rst +3 -3
@@ -147,9 +147,9 @@ CPUs in it. The next major initialization step for a new policy object is to attach a scaling governor to it (to begin with, that is the default scaling governor -determined by the kernel configuration, but it may be changed later -via ``sysfs``). First, a pointer to the new policy object is passed to the -governor's ``->init()`` callback which is expected to initialize all of the +determined by the kernel command line or configuration, but it may be changed +later via ``sysfs``). First, a pointer to the new policy object is passed to +the governor's ``->init()`` callback which is expected to initialize all of the data structures necessary to handle the given policy and, possibly, to add a governor ``sysfs`` interface to it. Next, the governor is started by invoking its ``->start()`` callback.
drivers/cpufreq/cpufreq.c +18 -5
@@ -50,6 +50,9 @@ static LIST_HEAD(cpufreq_governor_list);#define for_each_governor (__governor) \ list_for_each_entry(__governor, &cpufreq_governor_list, governor_list) +static char cpufreq_param_governor[CPUFREQ_NAME_LEN ]; +static struct cpufreq_governor *default_governor; +/** * The "cpufreq driver" - the arch- or hardware-dependent low * level driver of CPUFreq support, and its spinlock. This lock
@@ -1055,7 +1058,6 @@ __weak struct cpufreq_governor *cpufreq_default_governor(void)static int cpufreq_init_policy (struct cpufreq_policy *policy) { -struct cpufreq_governor *def_gov =cpufreq_default_governor ();struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ;
@@ -1065,8 +1067,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)if (gov) {pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); - }else if (def_gov) { - gov = def_gov; + }else if (default_governor) { + gov = default_governor; }else {return -ENODATA ; }
@@ -1074,8 +1076,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (def_gov) { - pol =cpufreq_parse_policy (def_gov->name); + }else if (default_governor) { + pol =cpufreq_parse_policy (default_governor->name);/* * In case the default governor is neiter "performance" * nor "powersave", fall back to the initial policy
@@ -2320,6 +2322,9 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)list_add (&governor->governor_list, &cpufreq_governor_list); } +if (!strncasecmp (cpufreq_param_governor, governor->name,CPUFREQ_NAME_LEN )) + default_governor = governor; +mutex_unlock (&cpufreq_governor_mutex);return err; }
@@ -2348,6 +2353,8 @@ void cpufreq_unregister_governor(struct cpufreq_governor *governor)mutex_lock (&cpufreq_governor_mutex);list_del (&governor->governor_list); +if (governor == default_governor) + default_governor =cpufreq_default_governor ();mutex_unlock (&cpufreq_governor_mutex); }EXPORT_SYMBOL_GPL (cpufreq_unregister_governor);
@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject); +mutex_lock (&cpufreq_governor_mutex); +if (!default_governor) + default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex); +return 0 ; }module_param (off, int,0444 ); +module_param_string (default_governor, cpufreq_param_governor,CPUFREQ_NAME_LEN ,0444 );core_initcall (cpufreq_core_init);
Signature · 1 line
2.27.0.111.gc72c7da667-goog
Re: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 126 lines · Quentin Perret
Currently, the only way to specify the default CPUfreq governor is via
Kconfigoptions, which suits users who can build the kernel themselves perfectly.However, for those who use a distro-like kernel (such as Android, with the Generic Kernel Image project), the only way to use a different default is to boot to userspace, and to then switch using the sysfs interface. Being able to specify the default governor on the command line, like is the case for cpuidle, would enable those users to specify their governor of choice earlier on, and to simplify slighlty the userspace boot procedure.
To support this use-case, add a kernel command line parameter enabling to specify a default governor for CPUfreq, which takes precedence over the builtin default.
This implementation has one notable limitation: the default governor must be registered before the driver. This is solved for builtin governors and drivers using appropriate
*_initcall()functions. And in the modular case, this must be reflected as a constraint on the module loading order.Signed-off-by Quentin Perret <[email protected]>--- .../admin-guide/kernel-parameters.txt | 5 ++++ Documentation/admin-guide/pm/cpufreq.rst | 6 ++--- drivers/cpufreq/cpufreq.c | 23 +++++++++++++++---- 3 files changed, 26 insertions(+), 8 deletions(-)Documentation/admin-guide/kernel-parameters.txt +5
-0@@ -703,6 +703,11 @@ cpufreq.off=1 [CPU_FREQ] disable the cpufreq sub-system + cpufreq.default_governor= + [CPU_FREQ] Name of the default cpufreq governor to use. + This governor must be registered in the kernel before + the cpufreq driver probes. + cpu_init_udelay=N [X86] Delay for N microsec between assert and de-assert of APIC INIT to start processors. This delay occursDocumentation/admin-guide/pm/cpufreq.rst +3
-3@@ -147,9 +147,9 @@ CPUs in it. The next major initialization step for a new policy object is to attach a scaling governor to it (to begin with, that is the default scaling governor -determined by the kernel configuration, but it may be changed later -via ``sysfs``). First, a pointer to the new policy object is passed to the -governor's ``->init()`` callback which is expected to initialize all of the +determined by the kernel command line or configuration, but it may be changed +later via ``sysfs``). First, a pointer to the new policy object is passed to +the governor's ``->init()`` callback which is expected to initialize all of the data structures necessary to handle the given policy and, possibly, to add a governor ``sysfs`` interface to it. Next, the governor is started by invoking its ``->start()`` callback.drivers/cpufreq/cpufreq.c +12
-5@@ -50,6 +50,9 @@ static LIST_HEAD(cpufreq_governor_list);#define for_each_governor (__governor) \ list_for_each_entry(__governor, &cpufreq_governor_list, governor_list) +static char cpufreq_param_governor[CPUFREQ_NAME_LEN ]; +static struct cpufreq_governor *default_governor; +/** * The "cpufreq driver" - the arch- or hardware-dependent low * level driver of CPUFreq support, and its spinlock. This lock @@ -1055,7 +1058,6 @@ __weak struct cpufreq_governor *cpufreq_default_governor(void)static int cpufreq_init_policy (struct cpufreq_policy *policy) { -struct cpufreq_governor *def_gov =cpufreq_default_governor ();struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ;@@ -1065,8 +1067,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)if (gov) {pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); - }else if (def_gov) { - gov = def_gov; + }else if (default_governor) { + gov = default_governor; }else {return -ENODATA ; }@@ -1074,8 +1076,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (def_gov) { - pol =cpufreq_parse_policy (def_gov->name); + }else if (default_governor) { + pol =cpufreq_parse_policy (default_governor->name);/* * In case the default governor is neiter "performance" * nor "powersave", fall back to the initial policy @@ -2320,6 +2322,9 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)list_add (&governor->governor_list, &cpufreq_governor_list); } +if (!strncasecmp (cpufreq_param_governor, governor->name,CPUFREQ_NAME_LEN )) + default_governor = governor; +mutex_unlock (&cpufreq_governor_mutex);return err; }@@ -2348,6 +2353,8 @@ void cpufreq_unregister_governor(struct cpufreq_governor *governor)mutex_lock (&cpufreq_governor_mutex);list_del (&governor->governor_list); +if (governor == default_governor) + default_governor =cpufreq_default_governor ();mutex_unlock (&cpufreq_governor_mutex); }EXPORT_SYMBOL_GPL (cpufreq_unregister_governor);
drivers/cpufreq/cpufreq.c static int __init cpufreq_core_init(void) · line 2802
Quoted 7 lines
drivers/cpufreq/cpufreq.c +4
-0@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject); +mutex_lock (&cpufreq_governor_mutex); +if (!default_governor) + default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex);
I don't think locking is required here at core-initcall level. Apart from that:
Quoted 8 lines
drivers/cpufreq/cpufreq.c +2
-0+return 0 ; }module_param (off, int,0444 ); +module_param_string (default_governor, cpufreq_param_governor,CPUFREQ_NAME_LEN ,0444 );core_initcall (cpufreq_core_init);Signature · 1 line
2.27.0.111.gc72c7da667-goog
Signature · 1 line
viresh
Rafael J. Wysocki
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 131 lines · Viresh Kumar
Quoted 129 lines · Quentin Perret
Currently, the only way to specify the default CPUfreq governor is via
Kconfigoptions, which suits users who can build the kernel themselves perfectly.However, for those who use a distro-like kernel (such as Android, with the Generic Kernel Image project), the only way to use a different default is to boot to userspace, and to then switch using the sysfs interface. Being able to specify the default governor on the command line, like is the case for cpuidle, would enable those users to specify their governor of choice earlier on, and to simplify slighlty the userspace boot procedure.
To support this use-case, add a kernel command line parameter enabling to specify a default governor for CPUfreq, which takes precedence over the builtin default.
This implementation has one notable limitation: the default governor must be registered before the driver. This is solved for builtin governors and drivers using appropriate
*_initcall()functions. And in the modular case, this must be reflected as a constraint on the module loading order.Signed-off-by Quentin Perret <[email protected]>--- .../admin-guide/kernel-parameters.txt | 5 ++++ Documentation/admin-guide/pm/cpufreq.rst | 6 ++--- drivers/cpufreq/cpufreq.c | 23 +++++++++++++++---- 3 files changed, 26 insertions(+), 8 deletions(-)Documentation/admin-guide/kernel-parameters.txt +5
-0@@ -703,6 +703,11 @@ cpufreq.off=1 [CPU_FREQ] disable the cpufreq sub-system + cpufreq.default_governor= + [CPU_FREQ] Name of the default cpufreq governor to use. + This governor must be registered in the kernel before + the cpufreq driver probes. + cpu_init_udelay=N [X86] Delay for N microsec between assert and de-assert of APIC INIT to start processors. This delay occursDocumentation/admin-guide/pm/cpufreq.rst +3
-3@@ -147,9 +147,9 @@ CPUs in it. The next major initialization step for a new policy object is to attach a scaling governor to it (to begin with, that is the default scaling governor -determined by the kernel configuration, but it may be changed later -via ``sysfs``). First, a pointer to the new policy object is passed to the -governor's ``->init()`` callback which is expected to initialize all of the +determined by the kernel command line or configuration, but it may be changed +later via ``sysfs``). First, a pointer to the new policy object is passed to +the governor's ``->init()`` callback which is expected to initialize all of the data structures necessary to handle the given policy and, possibly, to add a governor ``sysfs`` interface to it. Next, the governor is started by invoking its ``->start()`` callback.drivers/cpufreq/cpufreq.c +12
-5@@ -50,6 +50,9 @@ static LIST_HEAD(cpufreq_governor_list);#define for_each_governor (__governor) \ list_for_each_entry(__governor, &cpufreq_governor_list, governor_list) +static char cpufreq_param_governor[CPUFREQ_NAME_LEN ]; +static struct cpufreq_governor *default_governor; +/** * The "cpufreq driver" - the arch- or hardware-dependent low * level driver of CPUFreq support, and its spinlock. This lock @@ -1055,7 +1058,6 @@ __weak struct cpufreq_governor *cpufreq_default_governor(void)static int cpufreq_init_policy (struct cpufreq_policy *policy) { -struct cpufreq_governor *def_gov =cpufreq_default_governor ();struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ;@@ -1065,8 +1067,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)if (gov) {pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); - }else if (def_gov) { - gov = def_gov; + }else if (default_governor) { + gov = default_governor; }else {return -ENODATA ; }@@ -1074,8 +1076,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (def_gov) { - pol =cpufreq_parse_policy (def_gov->name); + }else if (default_governor) { + pol =cpufreq_parse_policy (default_governor->name);/* * In case the default governor is neiter "performance" * nor "powersave", fall back to the initial policy @@ -2320,6 +2322,9 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)list_add (&governor->governor_list, &cpufreq_governor_list); } +if (!strncasecmp (cpufreq_param_governor, governor->name,CPUFREQ_NAME_LEN )) + default_governor = governor; +mutex_unlock (&cpufreq_governor_mutex);return err; }@@ -2348,6 +2353,8 @@ void cpufreq_unregister_governor(struct cpufreq_governor *governor)mutex_lock (&cpufreq_governor_mutex);list_del (&governor->governor_list); +if (governor == default_governor) + default_governor =cpufreq_default_governor ();mutex_unlock (&cpufreq_governor_mutex); }EXPORT_SYMBOL_GPL (cpufreq_unregister_governor);@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject);
Quoted 6 lines
drivers/cpufreq/cpufreq.c static int __init cpufreq_core_init(void) · line 2802
Quoted 4 lines
drivers/cpufreq/cpufreq.c +4
-0+mutex_lock (&cpufreq_governor_mutex); +if (!default_governor) + default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex);I don't think locking is required here at core-initcall level.
It isn't necessary AFAICS, but it may as well be regarded as annotation (kind of instead of having a comment explaining why it need not be used).
Re: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 12 lines · Rafael J. Wysocki
Quoted 10 lines · Viresh Kumar
drivers/cpufreq/cpufreq.c static int __init cpufreq_core_init(void) · line 2802
Quoted 8 lines
drivers/cpufreq/cpufreq.c +4
-0@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject); +mutex_lock (&cpufreq_governor_mutex); +if (!default_governor) + default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex);I don't think locking is required here at core-initcall level.
Quoted 3 lines
It isn't necessary AFAICS, but it may as well be regarded as annotation (kind of instead of having a comment explaining why it need not be used).
Right, but I must admit that, looking at this more, I'm getting a bit confused with the overall locking for governors :/
When in cpufreq_init_policy() we find a governor using find_governor(policy->last_governor), what guarantees this governor is not concurrently unregistered? That is, what guarantees this governor doesn't go away between that find_governor() call, and the subsequent call to try_module_get() in cpufreq_set_policy() down the line?
Can we somewhat assume that whatever governor is referred to by policy->last_governor will have a non-null refcount? Or are the cpufreq_online() and cpufreq_unregister_governor() path mutually exclusive? Or is there something else?
Thanks,
Quentin
Viresh Kumar
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 9 lines · Quentin Perret
Right, but I must admit that, looking at this more, I'm getting a bit confused with the overall locking for governors :/
When in
cpufreq_init_policy()we find a governor usingfind_governor(policy->last_governor), what guarantees this governor is not concurrently unregistered? That is, what guarantees this governor doesn't go away between thatfind_governor()call, and the subsequent call totry_module_get()incpufreq_set_policy()down the line?
Quoted 4 lines
Can we somewhat assume that whatever governor is referred to by
policy->last_governorwill have a non-null refcount? Or are thecpufreq_online()andcpufreq_unregister_governor()path mutually exclusive? Or is there something else?
This should be sufficient to fix pending issues I believe. Based over your patches.
Signature · 17 lines
viresh -------------------------8<------------------------- From: Viresh Kumar <[email protected]> Date: Thu, 25 Jun 2020 13:15:23 +0530 Subject: [PATCH] cpufreq: Fix locking issues with governors The locking around governors handling isn't adequate currently. The list of governors should never be traversed without locking in place. Also we must make sure the governor isn't removed while it is still referenced by code. Reported-by: Quentin Perret <[email protected]> Signed-off-by: Viresh Kumar <[email protected]> --- drivers/cpufreq/cpufreq.c | 59 ++++++++++++++++++++++++--------------- 1 file changed, 36 insertions(+), 23 deletions(-)
drivers/cpufreq/cpufreq.c +36 -23
@@ -624,6 +624,24 @@ static struct cpufreq_governor *find_governor(const char *str_governor)return NULL ; } +static struct cpufreq_governor *get_governor (const char *str_governor) +{ +struct cpufreq_governor *t; + +mutex_lock (&cpufreq_governor_mutex); + t =find_governor (str_governor); +if (!t) + goto unlock; + +if (!try_module_get (t->owner)) + t =NULL ; + +unlock: +mutex_unlock (&cpufreq_governor_mutex); + +return t; +} +static unsigned int cpufreq_parse_policy (char *str_governor) {if (!strncasecmp (str_governor,"performance" ,CPUFREQ_NAME_LEN ))
@@ -643,28 +661,14 @@ static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor) {struct cpufreq_governor *t; -mutex_lock (&cpufreq_governor_mutex); - - t =find_governor (str_governor); -if (!t) { -int ret; - -mutex_unlock (&cpufreq_governor_mutex); - - ret =request_module ("cpufreq_%s" , str_governor); -if (ret) -return NULL ; - -mutex_lock (&cpufreq_governor_mutex); + t =get_governor (str_governor); +if (t) +return t; - t =find_governor (str_governor); - } -if (t && !try_module_get (t->owner)) - t =NULL ; - -mutex_unlock (&cpufreq_governor_mutex); +if (request_module ("cpufreq_%s" , str_governor)) +return NULL ; -return t; +return get_governor (str_governor); }/**
@@ -818,12 +822,14 @@ static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy, goto out; } +mutex_lock (&cpufreq_governor_mutex);for_each_governor (t) {if (i >= (ssize_t ) ((PAGE_SIZE /sizeof (char )) - (CPUFREQ_NAME_LEN +2 ))) - goto out; +break ; i +=scnprintf (&buf[i],CPUFREQ_NAME_PLEN ,"%s " , t->name); } +mutex_unlock (&cpufreq_governor_mutex); out: i +=sprintf (&buf[i],"\n" );return i;
@@ -1060,11 +1066,14 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy) {struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ; +bool put_governor = false; +int ret;if (has_target ()) {/* Update policy governor to the one used before hotplug. */ - gov =find_governor (policy->last_governor); + gov =get_governor (policy->last_governor);if (gov) { + put_governor = true;pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); }else if (default_governor) {
@@ -1091,7 +1100,11 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)return -ENODATA ; } -return cpufreq_set_policy (policy, gov, pol); + ret =cpufreq_set_policy (policy, gov, pol); +if (put_governor) +module_put (gov->owner); + +return ret; }static int cpufreq_add_policy_cpu (struct cpufreq_policy *policy,unsigned int cpu)
Rafael J. Wysocki
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 11 lines · Viresh Kumar
Quoted 9 lines · Quentin Perret
Right, but I must admit that, looking at this more, I'm getting a bit confused with the overall locking for governors :/
When in
cpufreq_init_policy()we find a governor usingfind_governor(policy->last_governor), what guarantees this governor is not concurrently unregistered? That is, what guarantees this governor doesn't go away between thatfind_governor()call, and the subsequent call totry_module_get()incpufreq_set_policy()down the line?
Quoted 7 lines
Quoted 4 lines
Can we somewhat assume that whatever governor is referred to by
policy->last_governorwill have a non-null refcount? Or are thecpufreq_online()andcpufreq_unregister_governor()path mutually exclusive? Or is there something else?This should be sufficient to fix pending issues I believe. Based over your patches.
LGTM, but can you post it in a new thread to let Patchwork pick it up?
Quoted 125 lines
-------------------------8<-------------------------
From: Viresh Kumar <[email protected]>
Date: Thu, 25 Jun 2020 13:15:23 +0530
Subject: [PATCH] cpufreq: Fix locking issues with governorsThe locking around governors handling isn't adequate currently. The list of governors should never be traversed without locking in place. Also we must make sure the governor isn't removed while it is still referenced by code.
Reported-by Quentin Perret <[email protected]>Signed-off-by Viresh Kumar <[email protected]>--- drivers/cpufreq/cpufreq.c | 59 ++++++++++++++++++++++++--------------- 1 file changed, 36 insertions(+), 23 deletions(-)drivers/cpufreq/cpufreq.c +36
-23@@ -624,6 +624,24 @@ static struct cpufreq_governor *find_governor(const char *str_governor)return NULL ; } +static struct cpufreq_governor *get_governor (const char *str_governor) +{ +struct cpufreq_governor *t; + +mutex_lock (&cpufreq_governor_mutex); + t =find_governor (str_governor); +if (!t) + goto unlock; + +if (!try_module_get (t->owner)) + t =NULL ; + +unlock: +mutex_unlock (&cpufreq_governor_mutex); + +return t; +} +static unsigned int cpufreq_parse_policy (char *str_governor) {if (!strncasecmp (str_governor,"performance" ,CPUFREQ_NAME_LEN ))@@ -643,28 +661,14 @@ static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor) {struct cpufreq_governor *t; -mutex_lock (&cpufreq_governor_mutex); - - t =find_governor (str_governor); -if (!t) { -int ret; - -mutex_unlock (&cpufreq_governor_mutex); - - ret =request_module ("cpufreq_%s" , str_governor); -if (ret) -return NULL ; - -mutex_lock (&cpufreq_governor_mutex); + t =get_governor (str_governor); +if (t) +return t; - t =find_governor (str_governor); - } -if (t && !try_module_get (t->owner)) - t =NULL ; - -mutex_unlock (&cpufreq_governor_mutex); +if (request_module ("cpufreq_%s" , str_governor)) +return NULL ; -return t; +return get_governor (str_governor); }/** @@ -818,12 +822,14 @@ static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy, goto out; } +mutex_lock (&cpufreq_governor_mutex);for_each_governor (t) {if (i >= (ssize_t ) ((PAGE_SIZE /sizeof (char )) - (CPUFREQ_NAME_LEN +2 ))) - goto out; +break ; i +=scnprintf (&buf[i],CPUFREQ_NAME_PLEN ,"%s " , t->name); } +mutex_unlock (&cpufreq_governor_mutex); out: i +=sprintf (&buf[i],"\n" );return i;@@ -1060,11 +1066,14 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy) {struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ; +bool put_governor = false; +int ret;if (has_target ()) {/* Update policy governor to the one used before hotplug. */ - gov =find_governor (policy->last_governor); + gov =get_governor (policy->last_governor);if (gov) { + put_governor = true;pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); }else if (default_governor) {@@ -1091,7 +1100,11 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)return -ENODATA ; } -return cpufreq_set_policy (policy, gov, pol); + ret =cpufreq_set_policy (policy, gov, pol); +if (put_governor) +module_put (gov->owner); + +return ret; }static int cpufreq_add_policy_cpu (struct cpufreq_policy *policy,unsigned int cpu)
Viresh Kumar
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
In reply to Quentin PerretAfter your last email (reply to my patch), I noticed a change which isn't required. :)
Quoted 36 lines · Quentin Perret
drivers/cpufreq/cpufreq.c +5
-3@@ -50,6 +50,9 @@ static LIST_HEAD(cpufreq_governor_list);#define for_each_governor (__governor) \ list_for_each_entry(__governor, &cpufreq_governor_list, governor_list) +static char cpufreq_param_governor[CPUFREQ_NAME_LEN ]; +static struct cpufreq_governor *default_governor; +/** * The "cpufreq driver" - the arch- or hardware-dependent low * level driver of CPUFreq support, and its spinlock. This lock @@ -1055,7 +1058,6 @@ __weak struct cpufreq_governor *cpufreq_default_governor(void)static int cpufreq_init_policy (struct cpufreq_policy *policy) { -struct cpufreq_governor *def_gov =cpufreq_default_governor ();struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ;@@ -1065,8 +1067,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)if (gov) {pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); - }else if (def_gov) { - gov = def_gov; + }else if (default_governor) { + gov = default_governor; }else {return -ENODATA ; }
drivers/cpufreq/cpufreq.c static int cpufreq_init_policy(struct cpufreq_policy *policy) · line 1080
Quoted 7 lines
drivers/cpufreq/cpufreq.c +2
-2@@ -1074,8 +1076,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (def_gov) { - pol =cpufreq_parse_policy (def_gov->name); + }else if (default_governor) { + pol =cpufreq_parse_policy (default_governor->name);
This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Signature · 1 line
viresh
Rafael J. Wysocki
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 40 lines · Viresh Kumar
After your last email (reply to my patch), I noticed a change which isn't required. :)
Quoted 33 lines · Quentin Perret
drivers/cpufreq/cpufreq.c +5
-3@@ -50,6 +50,9 @@ static LIST_HEAD(cpufreq_governor_list);#define for_each_governor (__governor) \ list_for_each_entry(__governor, &cpufreq_governor_list, governor_list) +static char cpufreq_param_governor[CPUFREQ_NAME_LEN ]; +static struct cpufreq_governor *default_governor; +/** * The "cpufreq driver" - the arch- or hardware-dependent low * level driver of CPUFreq support, and its spinlock. This lock @@ -1055,7 +1058,6 @@ __weak struct cpufreq_governor *cpufreq_default_governor(void)static int cpufreq_init_policy (struct cpufreq_policy *policy) { -struct cpufreq_governor *def_gov =cpufreq_default_governor ();struct cpufreq_governor *gov =NULL ;unsigned int pol =CPUFREQ_POLICY_UNKNOWN ;@@ -1065,8 +1067,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)if (gov) {pr_debug ("Restoring governor %s for cpu %d\n" , policy->governor->name, policy->cpu); - }else if (def_gov) { - gov = def_gov; + }else if (default_governor) { + gov = default_governor; }else {return -ENODATA ; }
Quoted 15 lines
drivers/cpufreq/cpufreq.c static int cpufreq_init_policy(struct cpufreq_policy *policy) · line 1080
Quoted 8 lines
drivers/cpufreq/cpufreq.c +2
-2@@ -1074,8 +1076,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (def_gov) { - pol =cpufreq_parse_policy (def_gov->name); + }else if (default_governor) { + pol =cpufreq_parse_policy (default_governor->name);This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Sounds right.
Quoted 4 lines
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Well, AFAICS the idea was to use the default governor as a kind of default policy proxy, but I agree that strings should be sufficient for that.
I'll have a look at what to do with that code.
Re: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 10 lines · Rafael J. Wysocki
Quoted 6 lines · Viresh Kumar
This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Sounds right.
Quoted 8 lines
Quoted 4 lines
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Well, AFAICS the idea was to use the default governor as a kind of default policy proxy, but I agree that strings should be sufficient for that.
I agree with all the above. I'd much rather not rely on the default governor name to populate the default policy, too, so +1 from me.
Thanks,
Quentin
Rafael J. Wysocki
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 17 lines · Quentin Perret
Quoted 15 lines · Rafael J. Wysocki
Quoted 6 lines · Viresh Kumar
This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Sounds right.
Quoted 4 lines
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Quoted 6 lines
Quoted 3 lines
Well, AFAICS the idea was to use the default governor as a kind of default policy proxy, but I agree that strings should be sufficient for that.
I agree with all the above. I'd much rather not rely on the default governor name to populate the default policy, too, so +1 from me.
So before this series the default governor was selected at the kernel configuration time (pre-build) and was always built-in. Because it could not go away, its name could be used to indicate the default policy for the "setpolicy" drivers.
After this series, however, it cannot be used this way reliably, but you can still pass cpufreq_param_governor to cpufreq_parse_policy() instead of def_gov->name in cpufreq_init_policy(), can't you?
Re: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 25 lines · Rafael J. Wysocki
Quoted 17 lines · Quentin Perret
Quoted 15 lines · Rafael J. Wysocki
Quoted 6 lines · Viresh Kumar
This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Sounds right.
Quoted 4 lines
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Quoted 6 lines
Quoted 3 lines
Well, AFAICS the idea was to use the default governor as a kind of default policy proxy, but I agree that strings should be sufficient for that.
I agree with all the above. I'd much rather not rely on the default governor name to populate the default policy, too, so +1 from me.
Quoted 8 lines
So before this series the default governor was selected at the kernel configuration time (pre-build) and was always built-in. Because it could not go away, its name could be used to indicate the default policy for the "setpolicy" drivers.
After this series, however, it cannot be used this way reliably, but you can still pass
cpufreq_param_governortocpufreq_parse_policy()instead ofdef_gov->nameincpufreq_init_policy(), can't you?
Good point. I also need to fallback to the default builtin governor if the command line parameter isn't valid (or non-existent), so perhaps something like so?
iff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index dad6b85f4c89..20a2020abf88 100644
drivers/cpufreq/cpufreq.c +19 -2
@@ -653,6 +653,23 @@ static unsigned int cpufreq_parse_policy(char *str_governor)return CPUFREQ_POLICY_UNKNOWN ; } +static unsigned int cpufreq_default_policy (void ) +{ +unsigned int pol; + + pol =cpufreq_parse_policy (cpufreq_param_governor); +if (pol !=CPUFREQ_POLICY_UNKNOWN ) +return pol; + +if (IS_BUILTIN (CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE )) +return CPUFREQ_POLICY_PERFORMANCE ; + +if (IS_BUILTIN (CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE )) +return CPUFREQ_POLICY_POWERSAVE ; + +return CPUFREQ_POLICY_UNKNOWN ; +} +/** * cpufreq_parse_governor - parse a governor string only for has_target() * @str_governor: Governor name.
@@ -1085,8 +1102,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (default_governor) { - pol =cpufreq_parse_policy (default_governor->name); + }else { + pol =cpufreq_default_policy ();/* * In case the default governor is neiter "performance" * nor "powersave", fall back to the initial policy
Rafael J. Wysocki
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 32 lines · Quentin Perret
Quoted 30 lines · Rafael J. Wysocki
Quoted 17 lines · Quentin Perret
Quoted 15 lines · Rafael J. Wysocki
Quoted 6 lines · Viresh Kumar
This change is not right IMO. This part handles the set-policy case, where there are no governors. Right now this code, for some reasons unknown to me, forcefully uses the default governor set to indicate the policy, which is not a great idea in my opinion TBH. This doesn't and shouldn't care about governor modules and should only be looking at strings instead of governor pointer.
Sounds right.
Quoted 4 lines
Rafael, I even think we should remove this code completely and just rely on what the driver has sent to us. Using the selected governor for set policy drivers is very confusing and also we shouldn't be forced to compiling any governor for the set-policy case.
Quoted 6 lines
Quoted 3 lines
Well, AFAICS the idea was to use the default governor as a kind of default policy proxy, but I agree that strings should be sufficient for that.
I agree with all the above. I'd much rather not rely on the default governor name to populate the default policy, too, so +1 from me.
So before this series the default governor was selected at the kernel configuration time (pre-build) and was always built-in. Because it could not go away, its name could be used to indicate the default policy for the "setpolicy" drivers.
Quoted 7 lines
Quoted 3 lines
After this series, however, it cannot be used this way reliably, but you can still pass
cpufreq_param_governortocpufreq_parse_policy()instead ofdef_gov->nameincpufreq_init_policy(), can't you?Good point. I also need to fallback to the default builtin governor if the command line parameter isn't valid (or non-existent), so perhaps something like so?
Yes, that should work if I haven't missed anything.
Quoted 39 lines
iff --git
a/drivers/cpufreq/cpufreq.cb/drivers/cpufreq/cpufreq.cindexdad6b85f4c89..20a2020abf88100644drivers/cpufreq/cpufreq.c +19
-2@@ -653,6 +653,23 @@ static unsigned int cpufreq_parse_policy(char *str_governor)return CPUFREQ_POLICY_UNKNOWN ; } +static unsigned int cpufreq_default_policy (void ) +{ +unsigned int pol; + + pol =cpufreq_parse_policy (cpufreq_param_governor); +if (pol !=CPUFREQ_POLICY_UNKNOWN ) +return pol; + +if (IS_BUILTIN (CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE )) +return CPUFREQ_POLICY_PERFORMANCE ; + +if (IS_BUILTIN (CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE )) +return CPUFREQ_POLICY_POWERSAVE ; + +return CPUFREQ_POLICY_UNKNOWN ; +} +/** * cpufreq_parse_governor - parse a governor string only for has_target() * @str_governor: Governor name. @@ -1085,8 +1102,8 @@ static int cpufreq_init_policy(struct cpufreq_policy *policy)/* Use the default policy if there is no last_policy. */ if (policy->last_policy) { pol = policy->last_policy; - }else if (default_governor) { - pol =cpufreq_parse_policy (default_governor->name); + }else { + pol =cpufreq_default_policy ();/* * In case the default governor is neiter "performance" * nor "powersave", fall back to the initial policy
Viresh Kumar
maintainerRe: [PATCH v2 2/2] cpufreq: Specify default governor on command line
In reply to Quentin Perretdrivers/cpufreq/cpufreq.c static int __init cpufreq_core_init(void) · line 2800
Quoted 6 lines · Quentin Perret
drivers/cpufreq/cpufreq.c +2
-0@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject); +mutex_lock (&cpufreq_governor_mutex); +if (!default_governor)
Also is this check really required ? The pointer will always be NULL at this point, isn't it ?
Quoted 10 lines
drivers/cpufreq/cpufreq.c +4
-0+ default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex); +return 0 ; }module_param (off, int,0444 ); +module_param_string (default_governor, cpufreq_param_governor,CPUFREQ_NAME_LEN ,0444 );core_initcall (cpufreq_core_init);Signature · 1 line
2.27.0.111.gc72c7da667-goog
Signature · 1 line
viresh
Re: [PATCH v2 2/2] cpufreq: Specify default governor on command line
Quoted 10 lines · Viresh Kumar
drivers/cpufreq/cpufreq.c static int __init cpufreq_core_init(void) · line 2800
Quoted 6 lines · Quentin Perret
drivers/cpufreq/cpufreq.c +2
-0@@ -2789,7 +2796,13 @@ static int __init cpufreq_core_init(void) cpufreq_global_kobject =kobject_create_and_add ("cpufreq" , &cpu_subsys.dev_root->kobj);BUG_ON (!cpufreq_global_kobject); +mutex_lock (&cpufreq_governor_mutex); +if (!default_governor)Also is this check really required ? The pointer will always be NULL at this point, isn't it ?
Not necessarily in this implementation -- the governors are registered at core_initcall time too, so I don't think we can assume any ordering there.
But it looks like your new version has fixed that by design, so I'll go look at it some more, and try it out.
Thanks for the help!
Quentin
Quoted 14 lines
Quoted 11 lines
drivers/cpufreq/cpufreq.c +4
-0+ default_governor =cpufreq_default_governor (); +mutex_unlock (&cpufreq_governor_mutex); +return 0 ; }module_param (off, int,0444 ); +module_param_string (default_governor, cpufreq_param_governor,CPUFREQ_NAME_LEN ,0444 );core_initcall (cpufreq_core_init);Signature · 1 line
2.27.0.111.gc72c7da667-googSignature · 1 line
viresh
Quoted 46 lines · On 2020.06.23 07:22 Quentin Perret wrote:
This series enables users of prebuilt kernels (e.g. distro kernels) to specify their CPUfreq governor of choice using the kernel command line, instead of having to wait for the system to fully boot to userspace to switch using the sysfs interface. This is helpful for 2 reasons:
1. users get to choose the governor that runs during the actual boot;
2. it simplifies the userspace boot procedure a bit (one less thing to worry about).To enable this, the first patch moves all governor init calls to
core_initcall, to make sure they are registered by the time the drivers probe. This should be relatively low impact as registering a governor is a simple procedure (it gets added to a llist), and all governors already load atcore_initcallanyway when they're set as the default inKconfig. This also allows to clean-up the governors' init/exit code, and reduces boilerplate.The second patch introduces the new command line parameter, inspired by its cpuidle counterpart. More details can be found in the respective patch headers.
Changes in v2:
- added Viresh's ack to patch 01
- moved the assignment of 'default_governor' in patch 02 to the governor registration path instead of the driver registration (Viresh)Thanks,
QuentinQuentin Perret (2):
cpufreq: Register governors atcore_initcall
cpufreq: Specify default governor on command line.../admin-guide/kernel-parameters.txt | 5 ++++ Documentation/admin-guide/pm/cpufreq.rst | 6 ++--- .../platforms/cell/cpufreq_spudemand.c | 26 ++----------------- drivers/cpufreq/cpufreq.c | 23 ++++++++++++---- drivers/cpufreq/cpufreq_conservative.c | 22 ++++------------ drivers/cpufreq/cpufreq_ondemand.c | 24 +++++------------ drivers/cpufreq/cpufreq_performance.c | 14 ++-------- drivers/cpufreq/cpufreq_powersave.c | 18 +++---------- drivers/cpufreq/cpufreq_userspace.c | 18 +++---------- include/linux/cpufreq.h | 14 ++++++++++ kernel/sched/cpufreq_schedutil.c | 6 +---- 11 files changed, 62 insertions(+), 114 deletions(-)
Quoted 2 lines
--
2.27.0.111.gc72c7da667-goog
Hi Quentin,
Because I am lazy and sometimes do not want to recompile the distro source, I have a need/desire for this.
Tested these two grub command lines:
GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=300 intel_pstate=disable cpufreq.default_governor=schedutil cpuidle_sysfs_switch cpuidle.governor=teo"
And
#GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=450 intel_pstate=passive cpufreq.default_governor=schedutil cpuidle_sysfs_switch cpuidle.governor=teo"
And all worked as expected. I use Ubuntu as my distro, and also had to disable a startup script that switches to "ondemand", or similar, after 1 minute.
As a side note (separate subject, but is one reason I tried it):
My i5-9600K based computer seems to hit a power limit during boot approximately 3 seconds after kernel selection on grub. This had no effect on that issue (even when selecting powersave governor).
... Doug
Hi Doug,
Quoted 4 lines · Doug Smythies
Hi Quentin,
Because I am lazy and sometimes do not want to recompile the distro source, I have a need/desire for this.
Good to know I'm not the only one ;-)
Quoted 9 lines
Tested these two grub command lines:
GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=300 intel_pstate=disable cpufreq.default_governor=schedutil
cpuidle_sysfs_switchcpuidle.governor=teo"And
#GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=450 intel_pstate=passive cpufreq.default_governor=schedutil
cpuidle_sysfs_switchcpuidle.governor=teo"And all worked as expected. I use Ubuntu as my distro, and also had to disable a startup script that switches to "ondemand", or similar, after 1 minute.
Good, thanks for giving it a try.
Quoted 3 lines
As a side note (separate subject, but is one reason I tried it):
My i5-9600K based computer seems to hit a power limit during boot approximately 3 seconds after kernel selection on grub. This had no effect on that issue (even when selecting powersave governor).
Interesting ... Could you confirm that compiling with powersave as default doesn't fix the issue either?
Other question, when does the intel_pstate driver start on your device? Before or after that 3 seconds boot time?
Thanks,
Quentin
Hi Quentin,
Thanks for your quick reply.
Quoted 25 lines · On 2020.06.23 11:05 Quentin Perret wrote:
Hi Doug,
Quoted 4 lines · Doug Smythies
Hi Quentin,
Because I am lazy and sometimes do not want to recompile the distro source, I have a need/desire for this.
Good to know I'm not the only one ;-)
Quoted 3 lines
Tested these two grub command lines:
GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=300 intel_pstate=disable
cpufreq.default_governor=schedutil
cpuidle_sysfs_switchcpuidle.governor=teo"Quoted 4 lines
And
#GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 consoleblank=450 intel_pstate=passive
cpufreq.default_governor=schedutil
cpuidle_sysfs_switchcpuidle.governor=teo"Quoted 2 lines
And all worked as expected. I use Ubuntu as my distro, and also had to disable a startup script that
switches to "ondemand", or similar, after 1 minute.
Good, thanks for giving it a try.
Quoted 7 lines
Quoted 2 lines
As a side note (separate subject, but is one reason I tried it):
My i5-9600K based computer seems to hit a power limit during boot approximately 3 seconds afterkernel selection on grub.
Quoted 1 line
This had no effect on that issue (even when selecting powersave governor).
Interesting ... Could you confirm that compiling with powersave as default doesn't fix the issue either?
No, it doesn't (good idea for a test though).
However, the big mains spike is also gone. So, I no longer know why those power limit log bits are always set after boot.
Quoted 3 lines
Other question, when does the
intel_pstatedriver start on your device? Before or after that 3 seconds boot time?
Before, if I understand correctly (from dmesg):
[ 0.468969] intel_pstate: Intel P-state driver initializing
I'll attach a couple of annotated mains power graphs.
(which will likely get stripped from the on-list version of this e-mail).
Currently, I am drowning in stuff that doesn't work, and will put this aside for now. I'll revive this as a new thread or a bugzilla eventually.
I also tried booting with turbo disabled, no difference.
Thanks for this patch set.
... Doug