Message ID | 1374659615-29447-1-git-send-email-amit.daniel@samsung.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Amit Daniel Kachhap wrote: > > This patch skips the deep C1(AFTR -Arm off top running) state for > exynos5440 > soc as this soc does not support this state. All the cpu's only allows the > basic > C0 state. > > Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > --- > arch/arm/mach-exynos/cpuidle.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- > exynos/cpuidle.c > index 17a18ff..9a776a1 100644 > --- a/arch/arm/mach-exynos/cpuidle.c > +++ b/arch/arm/mach-exynos/cpuidle.c > @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > device->cpu = cpu_id; > > /* Support IDLE only */ > - if (cpu_id != 0) > + if (soc_is_exynos5440() || cpu_id != 0) > device->state_count = 1; > > ret = cpuidle_register_device(device); > -- > 1.7.1 Applied, thanks. - Kukjin
[ added Daniel and linux-pm to cc: ] Hi, On Wednesday, July 24, 2013 03:23:35 PM Amit Daniel Kachhap wrote: > This patch skips the deep C1(AFTR -Arm off top running) state for exynos5440 > soc as this soc does not support this state. All the cpu's only allows the basic > C0 state. > > Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > --- > arch/arm/mach-exynos/cpuidle.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach-exynos/cpuidle.c > index 17a18ff..9a776a1 100644 > --- a/arch/arm/mach-exynos/cpuidle.c > +++ b/arch/arm/mach-exynos/cpuidle.c > @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > device->cpu = cpu_id; > > /* Support IDLE only */ > - if (cpu_id != 0) > + if (soc_is_exynos5440() || cpu_id != 0) > device->state_count = 1; > > ret = cpuidle_register_device(device); This may not achieve what you are intending to do and conflicts with Daniel's patch (which removes code that you're modifying). Please see: http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 How's about changing exynos4_idle_driver.state_count for EXYNOS5540 at the beginning of exynos4_init_cpuidle() instead? Best regards, -- Bartlomiej Zolnierkiewicz Samsung R&D Institute Poland Samsung Electronics
On 07/24/2013 01:47 PM, Kukjin Kim wrote: > Amit Daniel Kachhap wrote: >> >> This patch skips the deep C1(AFTR -Arm off top running) state for >> exynos5440 >> soc as this soc does not support this state. All the cpu's only allows the >> basic >> C0 state. >> >> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> >> --- >> arch/arm/mach-exynos/cpuidle.c | 2 +- >> 1 files changed, 1 insertions(+), 1 deletions(-) >> >> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- >> exynos/cpuidle.c >> index 17a18ff..9a776a1 100644 >> --- a/arch/arm/mach-exynos/cpuidle.c >> +++ b/arch/arm/mach-exynos/cpuidle.c >> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) >> device->cpu = cpu_id; >> >> /* Support IDLE only */ >> - if (cpu_id != 0) >> + if (soc_is_exynos5440() || cpu_id != 0) >> device->state_count = 1; >> >> ret = cpuidle_register_device(device); >> -- >> 1.7.1 > > Applied, thanks. You shouldn't have. This patch means exynos5540 has no cpuidle driver at all. It should be fixed in the Kconfig to unselect CONFIG_CPU_IDLE for an exynos5540.
On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: > On 07/24/2013 01:47 PM, Kukjin Kim wrote: > > Amit Daniel Kachhap wrote: > >> This patch skips the deep C1(AFTR -Arm off top running) state for > >> exynos5440 > >> soc as this soc does not support this state. All the cpu's only > >> allows the basic > >> C0 state. > >> > >> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > >> --- > >> > >> arch/arm/mach-exynos/cpuidle.c | 2 +- > >> 1 files changed, 1 insertions(+), 1 deletions(-) > >> > >> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- > >> exynos/cpuidle.c > >> index 17a18ff..9a776a1 100644 > >> --- a/arch/arm/mach-exynos/cpuidle.c > >> +++ b/arch/arm/mach-exynos/cpuidle.c > >> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > >> > >> device->cpu = cpu_id; > >> > >> /* Support IDLE only */ > >> > >> - if (cpu_id != 0) > >> + if (soc_is_exynos5440() || cpu_id != 0) > >> > >> device->state_count = 1; > >> > >> ret = cpuidle_register_device(device); > >> > >> -- > >> 1.7.1 > > > > Applied, thanks. > > You shouldn't have. This patch means exynos5540 has no cpuidle driver at > all. It should be fixed in the Kconfig to unselect CONFIG_CPU_IDLE for > an exynos5540. To shed more light on this, let me add that you need to register a cpuidle driver only if you have more states than a simple WFI or you need some crazy steps to enter WFI. Default setup falls back to generic ARM WFI. (Daniel, do we get the nice idle stats as provided by cpuidle core then?) Anyway, Exynos cpuidle is using an initcall to initialize and we support multiple Exynos SoCs in single zImage, so deselecting CONFIG_CPU_IDLE is not an option. Considering multiplatform requirements, the driver has to be modified to initialize only on supported platforms, either by: a) dropping the initcall and calling the init function directly from arch/arm/mach-exynos or b) checking if machine we are running on is supported, which would mean a long list of all Exynos SoCs that needs to be checked. An evolution of option a) is registering a platform device somewhere in arch/arm/mach-exynos and making exynos-cpuidle a platform driver. The problem is that you must register a static platform device from arch code, because cpuidle is not a real hardware block that can be put into device tree. Best regards, Tomasz
On 07/28/2013 11:22 AM, Tomasz Figa wrote: > On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: >> On 07/24/2013 01:47 PM, Kukjin Kim wrote: >>> Amit Daniel Kachhap wrote: >>>> This patch skips the deep C1(AFTR -Arm off top running) state for >>>> exynos5440 >>>> soc as this soc does not support this state. All the cpu's only >>>> allows the basic >>>> C0 state. >>>> >>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> >>>> --- >>>> >>>> arch/arm/mach-exynos/cpuidle.c | 2 +- >>>> 1 files changed, 1 insertions(+), 1 deletions(-) >>>> >>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- >>>> exynos/cpuidle.c >>>> index 17a18ff..9a776a1 100644 >>>> --- a/arch/arm/mach-exynos/cpuidle.c >>>> +++ b/arch/arm/mach-exynos/cpuidle.c >>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) >>>> >>>> device->cpu = cpu_id; >>>> >>>> /* Support IDLE only */ >>>> >>>> - if (cpu_id != 0) >>>> + if (soc_is_exynos5440() || cpu_id != 0) >>>> >>>> device->state_count = 1; >>>> >>>> ret = cpuidle_register_device(device); >>>> >>>> -- >>>> 1.7.1 >>> >>> Applied, thanks. >> >> You shouldn't have. This patch means exynos5540 has no cpuidle driver at >> all. It should be fixed in the Kconfig to unselect CONFIG_CPU_IDLE for >> an exynos5540. > > To shed more light on this, let me add that you need to register a cpuidle > driver only if you have more states than a simple WFI or you need some > crazy steps to enter WFI. Default setup falls back to generic ARM WFI. > (Daniel, do we get the nice idle stats as provided by cpuidle core then?) Nope, but with one state, idle vs busy stats do the trick. BTW, I am writing a tool to do some stats based on the idle events [1]. It is still at a very early development stage but we can get some interesting informations. > Anyway, Exynos cpuidle is using an initcall to initialize and we support > multiple Exynos SoCs in single zImage, so deselecting CONFIG_CPU_IDLE is > not an option. Good point. > Considering multiplatform requirements, the driver has to > be modified to initialize only on supported platforms, either by: > > a) dropping the initcall and calling the init function directly from > arch/arm/mach-exynos > > or > > b) checking if machine we are running on is supported, which would mean a > long list of all Exynos SoCs that needs to be checked. > > An evolution of option a) is registering a platform device somewhere in > arch/arm/mach-exynos and making exynos-cpuidle a platform driver. Yes, I am favorable to this solution [2]. > The > problem is that you must register a static platform device from arch code, > because cpuidle is not a real hardware block that can be put into device > tree. Thanks -- Daniel [1] https://git.linaro.org/gitweb?p=people/dlezcano/idlestat.git;a=summary [2] http://patches.linaro.org/18368/
On Sunday 28 of July 2013 12:31:51 Daniel Lezcano wrote: > On 07/28/2013 11:22 AM, Tomasz Figa wrote: > > On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: > >> On 07/24/2013 01:47 PM, Kukjin Kim wrote: > >>> Amit Daniel Kachhap wrote: > >>>> This patch skips the deep C1(AFTR -Arm off top running) state for > >>>> exynos5440 > >>>> soc as this soc does not support this state. All the cpu's only > >>>> allows the basic > >>>> C0 state. > >>>> > >>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > >>>> --- > >>>> > >>>> arch/arm/mach-exynos/cpuidle.c | 2 +- > >>>> 1 files changed, 1 insertions(+), 1 deletions(-) > >>>> > >>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- > >>>> exynos/cpuidle.c > >>>> index 17a18ff..9a776a1 100644 > >>>> --- a/arch/arm/mach-exynos/cpuidle.c > >>>> +++ b/arch/arm/mach-exynos/cpuidle.c > >>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > >>>> > >>>> device->cpu = cpu_id; > >>>> > >>>> /* Support IDLE only */ > >>>> > >>>> - if (cpu_id != 0) > >>>> + if (soc_is_exynos5440() || cpu_id != 0) > >>>> > >>>> device->state_count = 1; > >>>> > >>>> ret = cpuidle_register_device(device); > >>>> > >>>> -- > >>>> 1.7.1 > >>> > >>> Applied, thanks. > >> > >> You shouldn't have. This patch means exynos5540 has no cpuidle driver > >> at all. It should be fixed in the Kconfig to unselect > >> CONFIG_CPU_IDLE for an exynos5540. > > > > To shed more light on this, let me add that you need to register a > > cpuidle driver only if you have more states than a simple WFI or you > > need some crazy steps to enter WFI. Default setup falls back to > > generic ARM WFI. (Daniel, do we get the nice idle stats as provided > > by cpuidle core then?) > Nope, but with one state, idle vs busy stats do the trick. > > BTW, I am writing a tool to do some stats based on the idle events [1]. > It is still at a very early development stage but we can get some > interesting informations. Sounds good. By the way, a random idea that just came to my mind is that since we already support multiple cpuidle drivers, what about providing a default cpuidle-wfi driver that gets used in case no other driver is available? This would give us cpuidle stats at the same place as with other cpuidle drivers and might also unify some code paths. As I said, it's just a random idea, so feel free to redirect it to /dev/null. > > Anyway, Exynos cpuidle is using an initcall to initialize and we > > support multiple Exynos SoCs in single zImage, so deselecting > > CONFIG_CPU_IDLE is not an option. > > Good point. > > > Considering multiplatform requirements, the driver has to > > > > be modified to initialize only on supported platforms, either by: > > a) dropping the initcall and calling the init function directly from > > > > arch/arm/mach-exynos > > > > or > > > > b) checking if machine we are running on is supported, which would > > mean a> > > long list of all Exynos SoCs that needs to be checked. > > > > An evolution of option a) is registering a platform device somewhere > > in > > arch/arm/mach-exynos and making exynos-cpuidle a platform driver. > > Yes, I am favorable to this solution [2]. OK, looks good. Best regards, Tomasz
Hi Daniel/Tomasz, From the discussion I can conclude that SOC check is needed in the cpuidle driver for deeper C states. Only the question is where to insert this. Also to perform the SOC there can be 2 ways like 1) each SOC check 4120, 4412, 5250 etc (long list) 2) negate the nonsupporting SOC's like 5440 (small list like current patch) Any opinion? As Bartlomiej suggested that this patch conflicts with Daniel's earlier patch http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 So I can re-base my patch on top of this one if needed. Thanks, Amit Daniel On Sun, Jul 28, 2013 at 4:01 PM, Daniel Lezcano <daniel.lezcano@linaro.org> wrote: > On 07/28/2013 11:22 AM, Tomasz Figa wrote: >> On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: >>> On 07/24/2013 01:47 PM, Kukjin Kim wrote: >>>> Amit Daniel Kachhap wrote: >>>>> This patch skips the deep C1(AFTR -Arm off top running) state for >>>>> exynos5440 >>>>> soc as this soc does not support this state. All the cpu's only >>>>> allows the basic >>>>> C0 state. >>>>> >>>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> >>>>> --- >>>>> >>>>> arch/arm/mach-exynos/cpuidle.c | 2 +- >>>>> 1 files changed, 1 insertions(+), 1 deletions(-) >>>>> >>>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- >>>>> exynos/cpuidle.c >>>>> index 17a18ff..9a776a1 100644 >>>>> --- a/arch/arm/mach-exynos/cpuidle.c >>>>> +++ b/arch/arm/mach-exynos/cpuidle.c >>>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) >>>>> >>>>> device->cpu = cpu_id; >>>>> >>>>> /* Support IDLE only */ >>>>> >>>>> - if (cpu_id != 0) >>>>> + if (soc_is_exynos5440() || cpu_id != 0) >>>>> >>>>> device->state_count = 1; >>>>> >>>>> ret = cpuidle_register_device(device); >>>>> >>>>> -- >>>>> 1.7.1 >>>> >>>> Applied, thanks. >>> >>> You shouldn't have. This patch means exynos5540 has no cpuidle driver at >>> all. It should be fixed in the Kconfig to unselect CONFIG_CPU_IDLE for >>> an exynos5540. >> >> To shed more light on this, let me add that you need to register a cpuidle >> driver only if you have more states than a simple WFI or you need some >> crazy steps to enter WFI. Default setup falls back to generic ARM WFI. >> (Daniel, do we get the nice idle stats as provided by cpuidle core then?) > > Nope, but with one state, idle vs busy stats do the trick. > > BTW, I am writing a tool to do some stats based on the idle events [1]. > It is still at a very early development stage but we can get some > interesting informations. > > >> Anyway, Exynos cpuidle is using an initcall to initialize and we support >> multiple Exynos SoCs in single zImage, so deselecting CONFIG_CPU_IDLE is >> not an option. > > Good point. > >> Considering multiplatform requirements, the driver has to >> be modified to initialize only on supported platforms, either by: >> >> a) dropping the initcall and calling the init function directly from >> arch/arm/mach-exynos >> >> or >> >> b) checking if machine we are running on is supported, which would mean a >> long list of all Exynos SoCs that needs to be checked. >> >> An evolution of option a) is registering a platform device somewhere in >> arch/arm/mach-exynos and making exynos-cpuidle a platform driver. > > Yes, I am favorable to this solution [2]. > >> The >> problem is that you must register a static platform device from arch code, >> because cpuidle is not a real hardware block that can be put into device >> tree. > > Thanks > -- Daniel > > [1] https://git.linaro.org/gitweb?p=people/dlezcano/idlestat.git;a=summary > [2] http://patches.linaro.org/18368/ > > -- > <http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs > > Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | > <http://twitter.com/#!/linaroorg> Twitter | > <http://www.linaro.org/linaro-blog/> Blog > > -- > To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
amit daniel kachhap wrote: > > Hi Daniel/Tomasz, > > From the discussion I can conclude that SOC check is needed in the > cpuidle driver for deeper C states. Only the question is where to > insert this. > Also to perform the SOC there can be 2 ways like > 1) each SOC check 4120, 4412, 5250 etc (long list) > 2) negate the nonsupporting SOC's like 5440 (small list like current patch) > Any opinion? I’d preferred to use 2nd :) > As Bartlomiej suggested that this patch conflicts with Daniel's > earlier patch http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 > So I can re-base my patch on top of this one if needed. > Sounds good to me. Thanks, Kukjin > Thanks, > Amit Daniel > > On Sun, Jul 28, 2013 at 4:01 PM, Daniel Lezcano > <daniel.lezcano@linaro.org> wrote: > > On 07/28/2013 11:22 AM, Tomasz Figa wrote: > >> On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: > >>> On 07/24/2013 01:47 PM, Kukjin Kim wrote: > >>>> Amit Daniel Kachhap wrote: > >>>>> This patch skips the deep C1(AFTR -Arm off top running) state for > >>>>> exynos5440 > >>>>> soc as this soc does not support this state. All the cpu's only > >>>>> allows the basic > >>>>> C0 state. > >>>>> > >>>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > >>>>> --- > >>>>> > >>>>> arch/arm/mach-exynos/cpuidle.c | 2 +- > >>>>> 1 files changed, 1 insertions(+), 1 deletions(-) > >>>>> > >>>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- > >>>>> exynos/cpuidle.c > >>>>> index 17a18ff..9a776a1 100644 > >>>>> --- a/arch/arm/mach-exynos/cpuidle.c > >>>>> +++ b/arch/arm/mach-exynos/cpuidle.c > >>>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > >>>>> > >>>>> device->cpu = cpu_id; > >>>>> > >>>>> /* Support IDLE only */ > >>>>> > >>>>> - if (cpu_id != 0) > >>>>> + if (soc_is_exynos5440() || cpu_id != 0) > >>>>> > >>>>> device->state_count = 1; > >>>>> > >>>>> ret = cpuidle_register_device(device); > >>>>> > >>>>> -- > >>>>> 1.7.1 > >>>> > >>>> Applied, thanks. > >>> > >>> You shouldn't have. This patch means exynos5540 has no cpuidle driver > at > >>> all. It should be fixed in the Kconfig to unselect CONFIG_CPU_IDLE for > >>> an exynos5540. > >> > >> To shed more light on this, let me add that you need to register a > cpuidle > >> driver only if you have more states than a simple WFI or you need some > >> crazy steps to enter WFI. Default setup falls back to generic ARM WFI. > >> (Daniel, do we get the nice idle stats as provided by cpuidle core > then?) > > > > Nope, but with one state, idle vs busy stats do the trick. > > > > BTW, I am writing a tool to do some stats based on the idle events [1]. > > It is still at a very early development stage but we can get some > > interesting informations. > > > > > >> Anyway, Exynos cpuidle is using an initcall to initialize and we > support > >> multiple Exynos SoCs in single zImage, so deselecting CONFIG_CPU_IDLE > is > >> not an option. > > > > Good point. > > > >> Considering multiplatform requirements, the driver has to > >> be modified to initialize only on supported platforms, either by: > >> > >> a) dropping the initcall and calling the init function directly from > >> arch/arm/mach-exynos > >> > >> or > >> > >> b) checking if machine we are running on is supported, which would > mean a > >> long list of all Exynos SoCs that needs to be checked. > >> > >> An evolution of option a) is registering a platform device somewhere in > >> arch/arm/mach-exynos and making exynos-cpuidle a platform driver. > > > > Yes, I am favorable to this solution [2]. > > > >> The > >> problem is that you must register a static platform device from arch > code, > >> because cpuidle is not a real hardware block that can be put into > device > >> tree.
On Monday 29 of July 2013 10:16:14 amit daniel kachhap wrote: > Hi Daniel/Tomasz, > > From the discussion I can conclude that SOC check is needed in the > cpuidle driver for deeper C states. A check is needed for the whole cpuidle driver, so it registers only on Exynos SoCs which support deeper C states. > Only the question is where to > insert this. Exynos doesn't support multiplatform yet, but we must make sure that any code being added is multiplatform-aware. So initcall is not a good idea. I would put this somewhere on Exynos-specific initialization path, i.e. something that would not called for all platforms compiled in (in case of multiplatform). As I discussed with Daniel, this should be using a traditional platform_driver model, with the difference that it can't be registered from device tree, but rather statically in mach code. For example, you can add an exynos_register_cpuidle() function in arch/arm/mach-exynos/common.c, which registers such platform device and always call it from exynos4_dt_machine_init() in mach-exynos4-dt.c. In mach-exynos5-dt.c you could make this conditional and check if !soc_is_exynos5440(). Best regards, Tomasz > Also to perform the SOC there can be 2 ways like > 1) each SOC check 4120, 4412, 5250 etc (long list) > 2) negate the nonsupporting SOC's like 5440 (small list like current > patch) Any opinion? > As Bartlomiej suggested that this patch conflicts with Daniel's > earlier patch http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 > So I can re-base my patch on top of this one if needed. > > Thanks, > Amit Daniel > > On Sun, Jul 28, 2013 at 4:01 PM, Daniel Lezcano > > <daniel.lezcano@linaro.org> wrote: > > On 07/28/2013 11:22 AM, Tomasz Figa wrote: > >> On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: > >>> On 07/24/2013 01:47 PM, Kukjin Kim wrote: > >>>> Amit Daniel Kachhap wrote: > >>>>> This patch skips the deep C1(AFTR -Arm off top running) state for > >>>>> exynos5440 > >>>>> soc as this soc does not support this state. All the cpu's only > >>>>> allows the basic > >>>>> C0 state. > >>>>> > >>>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> > >>>>> --- > >>>>> > >>>>> arch/arm/mach-exynos/cpuidle.c | 2 +- > >>>>> 1 files changed, 1 insertions(+), 1 deletions(-) > >>>>> > >>>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- > >>>>> exynos/cpuidle.c > >>>>> index 17a18ff..9a776a1 100644 > >>>>> --- a/arch/arm/mach-exynos/cpuidle.c > >>>>> +++ b/arch/arm/mach-exynos/cpuidle.c > >>>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) > >>>>> > >>>>> device->cpu = cpu_id; > >>>>> > >>>>> /* Support IDLE only */ > >>>>> > >>>>> - if (cpu_id != 0) > >>>>> + if (soc_is_exynos5440() || cpu_id != 0) > >>>>> > >>>>> device->state_count = 1; > >>>>> > >>>>> ret = cpuidle_register_device(device); > >>>>> > >>>>> -- > >>>>> 1.7.1 > >>>> > >>>> Applied, thanks. > >>> > >>> You shouldn't have. This patch means exynos5540 has no cpuidle > >>> driver at all. It should be fixed in the Kconfig to unselect > >>> CONFIG_CPU_IDLE for an exynos5540. > >> > >> To shed more light on this, let me add that you need to register a > >> cpuidle driver only if you have more states than a simple WFI or you > >> need some crazy steps to enter WFI. Default setup falls back to > >> generic ARM WFI. (Daniel, do we get the nice idle stats as provided > >> by cpuidle core then?)> > > Nope, but with one state, idle vs busy stats do the trick. > > > > BTW, I am writing a tool to do some stats based on the idle events > > [1]. > > It is still at a very early development stage but we can get some > > interesting informations. > > > >> Anyway, Exynos cpuidle is using an initcall to initialize and we > >> support multiple Exynos SoCs in single zImage, so deselecting > >> CONFIG_CPU_IDLE is not an option. > > > > Good point. > > > >> Considering multiplatform requirements, the driver has to > >> > >> be modified to initialize only on supported platforms, either by: > >> a) dropping the initcall and calling the init function directly from > >> > >> arch/arm/mach-exynos > >> > >> or > >> > >> b) checking if machine we are running on is supported, which would > >> mean a>> > >> long list of all Exynos SoCs that needs to be checked. > >> > >> An evolution of option a) is registering a platform device somewhere > >> in > >> arch/arm/mach-exynos and making exynos-cpuidle a platform driver. > > > > Yes, I am favorable to this solution [2]. > > > >> The > >> problem is that you must register a static platform device from arch > >> code, because cpuidle is not a real hardware block that can be put > >> into device tree. > > > > Thanks > > > > -- Daniel > > > > [1] > > https://git.linaro.org/gitweb?p=people/dlezcano/idlestat.git;a=summar > > y [2] http://patches.linaro.org/18368/ > > > > -- > > > > <http://www.linaro.org/> Linaro.org ? Open source software for ARM > > SoCs > > > > Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | > > <http://twitter.com/#!/linaroorg> Twitter | > > <http://www.linaro.org/linaro-blog/> Blog > > > > -- > > To unsubscribe from this list: send the line "unsubscribe > > linux-samsung-soc" in the body of a message to > > majordomo@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Tomasz, On Mon, Jul 29, 2013 at 2:03 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: > On Monday 29 of July 2013 10:16:14 amit daniel kachhap wrote: >> Hi Daniel/Tomasz, >> >> From the discussion I can conclude that SOC check is needed in the >> cpuidle driver for deeper C states. > > A check is needed for the whole cpuidle driver, so it registers only on > Exynos SoCs which support deeper C states. > >> Only the question is where to >> insert this. > > Exynos doesn't support multiplatform yet, but we must make sure that any > code being added is multiplatform-aware. So initcall is not a good idea. I > would put this somewhere on Exynos-specific initialization path, i.e. > something that would not called for all platforms compiled in (in case of > multiplatform). > > As I discussed with Daniel, this should be using a traditional > platform_driver model, with the difference that it can't be registered > from device tree, but rather statically in mach code. > > For example, you can add an exynos_register_cpuidle() function in > arch/arm/mach-exynos/common.c, which registers such platform device and > always call it from exynos4_dt_machine_init() in mach-exynos4-dt.c. In > mach-exynos5-dt.c you could make this conditional and check if > !soc_is_exynos5440(). thanks for the detailed explanation. Certainly initcall is not good for multikernel. Still I have few concerns. 1) Each C state might be SOC specific. Say there are 3 C states C0(basic wfi), c1(single CPU power down) and C2 (System wide power down like AFTR). In this case some exynos soc may support few of these states and cpuidle driver should bind C-state to correct SOC. 2) Even for default C0 ( simple wfi) currently there is no sysfs information if the cpuidle driver is not registered(Not quite sure). I will try to post the new driver by dynamically creating the cpuidle_state table based on SOC capability and removing the initcall. Regards, Amit D > > Best regards, > Tomasz > >> Also to perform the SOC there can be 2 ways like >> 1) each SOC check 4120, 4412, 5250 etc (long list) >> 2) negate the nonsupporting SOC's like 5440 (small list like current >> patch) Any opinion? >> As Bartlomiej suggested that this patch conflicts with Daniel's >> earlier patch http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 >> So I can re-base my patch on top of this one if needed. >> >> Thanks, >> Amit Daniel >> >> On Sun, Jul 28, 2013 at 4:01 PM, Daniel Lezcano >> >> <daniel.lezcano@linaro.org> wrote: >> > On 07/28/2013 11:22 AM, Tomasz Figa wrote: >> >> On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: >> >>> On 07/24/2013 01:47 PM, Kukjin Kim wrote: >> >>>> Amit Daniel Kachhap wrote: >> >>>>> This patch skips the deep C1(AFTR -Arm off top running) state for >> >>>>> exynos5440 >> >>>>> soc as this soc does not support this state. All the cpu's only >> >>>>> allows the basic >> >>>>> C0 state. >> >>>>> >> >>>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> >> >>>>> --- >> >>>>> >> >>>>> arch/arm/mach-exynos/cpuidle.c | 2 +- >> >>>>> 1 files changed, 1 insertions(+), 1 deletions(-) >> >>>>> >> >>>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- >> >>>>> exynos/cpuidle.c >> >>>>> index 17a18ff..9a776a1 100644 >> >>>>> --- a/arch/arm/mach-exynos/cpuidle.c >> >>>>> +++ b/arch/arm/mach-exynos/cpuidle.c >> >>>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) >> >>>>> >> >>>>> device->cpu = cpu_id; >> >>>>> >> >>>>> /* Support IDLE only */ >> >>>>> >> >>>>> - if (cpu_id != 0) >> >>>>> + if (soc_is_exynos5440() || cpu_id != 0) >> >>>>> >> >>>>> device->state_count = 1; >> >>>>> >> >>>>> ret = cpuidle_register_device(device); >> >>>>> >> >>>>> -- >> >>>>> 1.7.1 >> >>>> >> >>>> Applied, thanks. >> >>> >> >>> You shouldn't have. This patch means exynos5540 has no cpuidle >> >>> driver at all. It should be fixed in the Kconfig to unselect >> >>> CONFIG_CPU_IDLE for an exynos5540. >> >> >> >> To shed more light on this, let me add that you need to register a >> >> cpuidle driver only if you have more states than a simple WFI or you >> >> need some crazy steps to enter WFI. Default setup falls back to >> >> generic ARM WFI. (Daniel, do we get the nice idle stats as provided >> >> by cpuidle core then?)> >> > Nope, but with one state, idle vs busy stats do the trick. >> > >> > BTW, I am writing a tool to do some stats based on the idle events >> > [1]. >> > It is still at a very early development stage but we can get some >> > interesting informations. >> > >> >> Anyway, Exynos cpuidle is using an initcall to initialize and we >> >> support multiple Exynos SoCs in single zImage, so deselecting >> >> CONFIG_CPU_IDLE is not an option. >> > >> > Good point. >> > >> >> Considering multiplatform requirements, the driver has to >> >> >> >> be modified to initialize only on supported platforms, either by: >> >> a) dropping the initcall and calling the init function directly from >> >> >> >> arch/arm/mach-exynos >> >> >> >> or >> >> >> >> b) checking if machine we are running on is supported, which would >> >> mean a>> >> >> long list of all Exynos SoCs that needs to be checked. >> >> >> >> An evolution of option a) is registering a platform device somewhere >> >> in >> >> arch/arm/mach-exynos and making exynos-cpuidle a platform driver. >> > >> > Yes, I am favorable to this solution [2]. >> > >> >> The >> >> problem is that you must register a static platform device from arch >> >> code, because cpuidle is not a real hardware block that can be put >> >> into device tree. >> > >> > Thanks >> > >> > -- Daniel >> > >> > [1] >> > https://git.linaro.org/gitweb?p=people/dlezcano/idlestat.git;a=summar >> > y [2] http://patches.linaro.org/18368/ >> > >> > -- >> > >> > <http://www.linaro.org/> Linaro.org ? Open source software for ARM >> > SoCs >> > >> > Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | >> > <http://twitter.com/#!/linaroorg> Twitter | >> > <http://www.linaro.org/linaro-blog/> Blog >> > >> > -- >> > To unsubscribe from this list: send the line "unsubscribe >> > linux-samsung-soc" in the body of a message to >> > majordomo@vger.kernel.org >> > More majordomo info at http://vger.kernel.org/majordomo-info.html > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On 07/29/2013 11:33 AM, amit daniel kachhap wrote: > Hi Tomasz, > > On Mon, Jul 29, 2013 at 2:03 PM, Tomasz Figa <tomasz.figa@gmail.com> wrote: >> On Monday 29 of July 2013 10:16:14 amit daniel kachhap wrote: >>> Hi Daniel/Tomasz, >>> >>> From the discussion I can conclude that SOC check is needed in the >>> cpuidle driver for deeper C states. >> >> A check is needed for the whole cpuidle driver, so it registers only on >> Exynos SoCs which support deeper C states. > >> >>> Only the question is where to >>> insert this. >> >> Exynos doesn't support multiplatform yet, but we must make sure that any >> code being added is multiplatform-aware. So initcall is not a good idea. I >> would put this somewhere on Exynos-specific initialization path, i.e. >> something that would not called for all platforms compiled in (in case of >> multiplatform). >> >> As I discussed with Daniel, this should be using a traditional >> platform_driver model, with the difference that it can't be registered >> from device tree, but rather statically in mach code. >> >> For example, you can add an exynos_register_cpuidle() function in >> arch/arm/mach-exynos/common.c, which registers such platform device and >> always call it from exynos4_dt_machine_init() in mach-exynos4-dt.c. In >> mach-exynos5-dt.c you could make this conditional and check if >> !soc_is_exynos5440(). > thanks for the detailed explanation. Certainly initcall is not good > for multikernel. Still I have few concerns. > 1) Each C state might be SOC specific. Say there are 3 C states > C0(basic wfi), c1(single CPU power down) and C2 (System wide power > down like AFTR). In this case some exynos soc may support few of these > states and cpuidle driver should bind C-state to correct SOC. > 2) Even for default C0 ( simple wfi) currently there is no sysfs > information if the cpuidle driver is not registered(Not quite sure). > I will try to post the new driver by dynamically creating the > cpuidle_state table based on SOC capability and removing the initcall. Why don't you create a cpuidle driver for each SoC family ? I am cleaning up all the drivers to unify the code, hoping we can factor all the drivers through a single ARM driver and assign the ops for the different states. As Thomasz said, the platform driver model should be used, no need of initcall, the code path will initialize the driver or not depending of the SoC. You should not have any "soc_is_*" checks. This is the direction all the ARM drivers are going to. Look at the example I sent with the ux500, it illustrates the idea. There are other examples also: cpuidle-kirkwood.c, cpuidle for davinci. Concerning the "single cpu power down" state, is the exynos able to handle that ? I mean, is it possible to power down a core and have it woken up by an interrupt ? Concerning the AFTR state : Is there a way to use this state with *all* cpus online ? If not, shall we use the same approach than omap4, that is CPU0 power down CPU1 and goes to AFTR (all interrupts migrated to CPU0) and then wake up CPU1 when CPU0 is woken up ? Concerning WFI and sysfs, if there are no cpuidle driver registered, no sysfs. I would not focus on and spent too much effort for the statistics because a lot of changes will certainly happen with the power aware scheduler and the statistics will likely appear at this level. Regards -- Daniel >> Best regards, >> Tomasz >> >>> Also to perform the SOC there can be 2 ways like >>> 1) each SOC check 4120, 4412, 5250 etc (long list) >>> 2) negate the nonsupporting SOC's like 5440 (small list like current >>> patch) Any opinion? >>> As Bartlomiej suggested that this patch conflicts with Daniel's >>> earlier patch http://marc.info/?l=linux-arm-kernel&m=137467935712513&w=2 >>> So I can re-base my patch on top of this one if needed. >>> >>> Thanks, >>> Amit Daniel >>> >>> On Sun, Jul 28, 2013 at 4:01 PM, Daniel Lezcano >>> >>> <daniel.lezcano@linaro.org> wrote: >>>> On 07/28/2013 11:22 AM, Tomasz Figa wrote: >>>>> On Sunday 28 of July 2013 09:10:09 Daniel Lezcano wrote: >>>>>> On 07/24/2013 01:47 PM, Kukjin Kim wrote: >>>>>>> Amit Daniel Kachhap wrote: >>>>>>>> This patch skips the deep C1(AFTR -Arm off top running) state for >>>>>>>> exynos5440 >>>>>>>> soc as this soc does not support this state. All the cpu's only >>>>>>>> allows the basic >>>>>>>> C0 state. >>>>>>>> >>>>>>>> Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> >>>>>>>> --- >>>>>>>> >>>>>>>> arch/arm/mach-exynos/cpuidle.c | 2 +- >>>>>>>> 1 files changed, 1 insertions(+), 1 deletions(-) >>>>>>>> >>>>>>>> diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach- >>>>>>>> exynos/cpuidle.c >>>>>>>> index 17a18ff..9a776a1 100644 >>>>>>>> --- a/arch/arm/mach-exynos/cpuidle.c >>>>>>>> +++ b/arch/arm/mach-exynos/cpuidle.c >>>>>>>> @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) >>>>>>>> >>>>>>>> device->cpu = cpu_id; >>>>>>>> >>>>>>>> /* Support IDLE only */ >>>>>>>> >>>>>>>> - if (cpu_id != 0) >>>>>>>> + if (soc_is_exynos5440() || cpu_id != 0) >>>>>>>> >>>>>>>> device->state_count = 1; >>>>>>>> >>>>>>>> ret = cpuidle_register_device(device); >>>>>>>> >>>>>>>> -- >>>>>>>> 1.7.1 >>>>>>> >>>>>>> Applied, thanks. >>>>>> >>>>>> You shouldn't have. This patch means exynos5540 has no cpuidle >>>>>> driver at all. It should be fixed in the Kconfig to unselect >>>>>> CONFIG_CPU_IDLE for an exynos5540. >>>>> >>>>> To shed more light on this, let me add that you need to register a >>>>> cpuidle driver only if you have more states than a simple WFI or you >>>>> need some crazy steps to enter WFI. Default setup falls back to >>>>> generic ARM WFI. (Daniel, do we get the nice idle stats as provided >>>>> by cpuidle core then?)> >>>> Nope, but with one state, idle vs busy stats do the trick. >>>> >>>> BTW, I am writing a tool to do some stats based on the idle events >>>> [1]. >>>> It is still at a very early development stage but we can get some >>>> interesting informations. >>>> >>>>> Anyway, Exynos cpuidle is using an initcall to initialize and we >>>>> support multiple Exynos SoCs in single zImage, so deselecting >>>>> CONFIG_CPU_IDLE is not an option. >>>> >>>> Good point. >>>> >>>>> Considering multiplatform requirements, the driver has to >>>>> >>>>> be modified to initialize only on supported platforms, either by: >>>>> a) dropping the initcall and calling the init function directly from >>>>> >>>>> arch/arm/mach-exynos >>>>> >>>>> or >>>>> >>>>> b) checking if machine we are running on is supported, which would >>>>> mean a>> >>>>> long list of all Exynos SoCs that needs to be checked. >>>>> >>>>> An evolution of option a) is registering a platform device somewhere >>>>> in >>>>> arch/arm/mach-exynos and making exynos-cpuidle a platform driver. >>>> >>>> Yes, I am favorable to this solution [2]. >>>> >>>>> The >>>>> problem is that you must register a static platform device from arch >>>>> code, because cpuidle is not a real hardware block that can be put >>>>> into device tree. >>>> >>>> Thanks >>>> >>>> -- Daniel >>>> >>>> [1] >>>> https://git.linaro.org/gitweb?p=people/dlezcano/idlestat.git;a=summar >>>> y [2] http://patches.linaro.org/18368/ >>>> >>>> -- >>>> >>>> <http://www.linaro.org/> Linaro.org ? Open source software for ARM >>>> SoCs >>>> >>>> Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | >>>> <http://twitter.com/#!/linaroorg> Twitter | >>>> <http://www.linaro.org/linaro-blog/> Blog >>>> >>>> -- >>>> To unsubscribe from this list: send the line "unsubscribe >>>> linux-samsung-soc" in the body of a message to >>>> majordomo@vger.kernel.org >>>> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> _______________________________________________ >> linux-arm-kernel mailing list >> linux-arm-kernel@lists.infradead.org >> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
diff --git a/arch/arm/mach-exynos/cpuidle.c b/arch/arm/mach-exynos/cpuidle.c index 17a18ff..9a776a1 100644 --- a/arch/arm/mach-exynos/cpuidle.c +++ b/arch/arm/mach-exynos/cpuidle.c @@ -210,7 +210,7 @@ static int __init exynos4_init_cpuidle(void) device->cpu = cpu_id; /* Support IDLE only */ - if (cpu_id != 0) + if (soc_is_exynos5440() || cpu_id != 0) device->state_count = 1; ret = cpuidle_register_device(device);
This patch skips the deep C1(AFTR -Arm off top running) state for exynos5440 soc as this soc does not support this state. All the cpu's only allows the basic C0 state. Signed-off-by: Amit Daniel Kachhap <amit.daniel@samsung.com> --- arch/arm/mach-exynos/cpuidle.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)