| Message ID | 1405928755-19413-3-git-send-email-shawn.guo@freescale.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Jul 21, 2014 at 08:45:54AM +0100, Shawn Guo wrote: > With SCU standby enabled, SCU CLK will be turned off when all processors > are in WFI mode. And the clock will be turned on when any processor > leaves WFI mode. > > This behavior should be preferable in terms of power efficiency of > system idle. So let's set the SCU standby bit to enable the support in > function scu_enable(). > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > --- > arch/arm/kernel/smp_scu.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/arch/arm/kernel/smp_scu.c b/arch/arm/kernel/smp_scu.c > index c947508f84e6..9f29d167d02c 100644 > --- a/arch/arm/kernel/smp_scu.c > +++ b/arch/arm/kernel/smp_scu.c > @@ -18,6 +18,7 @@ > > #define SCU_CTRL 0x00 > #define SCU_ENABLE (1 << 0) > +#define SCU_STANDBY_ENABLE (1 << 5) > #define SCU_CONFIG 0x04 > #define SCU_CPU_STATUS 0x08 > #define SCU_INVALIDATE 0x0c > @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) > if (scu_ctrl & SCU_ENABLE) > return; > > - scu_ctrl |= SCU_ENABLE; > + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; I don't think this bit exists on all revisions of the A9. Will
On Mon, Jul 21, 2014 at 09:51:50AM +0100, Will Deacon wrote: > On Mon, Jul 21, 2014 at 08:45:54AM +0100, Shawn Guo wrote: > > With SCU standby enabled, SCU CLK will be turned off when all processors > > are in WFI mode. And the clock will be turned on when any processor > > leaves WFI mode. > > > > This behavior should be preferable in terms of power efficiency of > > system idle. So let's set the SCU standby bit to enable the support in > > function scu_enable(). > > > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > > --- > > arch/arm/kernel/smp_scu.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/arch/arm/kernel/smp_scu.c b/arch/arm/kernel/smp_scu.c > > index c947508f84e6..9f29d167d02c 100644 > > --- a/arch/arm/kernel/smp_scu.c > > +++ b/arch/arm/kernel/smp_scu.c > > @@ -18,6 +18,7 @@ > > > > #define SCU_CTRL 0x00 > > #define SCU_ENABLE (1 << 0) > > +#define SCU_STANDBY_ENABLE (1 << 5) > > #define SCU_CONFIG 0x04 > > #define SCU_CPU_STATUS 0x08 > > #define SCU_INVALIDATE 0x0c > > @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) > > if (scu_ctrl & SCU_ENABLE) > > return; > > > > - scu_ctrl |= SCU_ENABLE; > > + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; > > I don't think this bit exists on all revisions of the A9. Thanks for the info, Will. Is there any side-effect to write the standby bit on those revisions which do not define the bit? Shawn
On Mon, Jul 21, 2014 at 10:27:04AM +0100, Shawn Guo wrote: > On Mon, Jul 21, 2014 at 09:51:50AM +0100, Will Deacon wrote: > > On Mon, Jul 21, 2014 at 08:45:54AM +0100, Shawn Guo wrote: > > > With SCU standby enabled, SCU CLK will be turned off when all processors > > > are in WFI mode. And the clock will be turned on when any processor > > > leaves WFI mode. > > > > > > This behavior should be preferable in terms of power efficiency of > > > system idle. So let's set the SCU standby bit to enable the support in > > > function scu_enable(). > > > > > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > > > --- > > > arch/arm/kernel/smp_scu.c | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/arch/arm/kernel/smp_scu.c b/arch/arm/kernel/smp_scu.c > > > index c947508f84e6..9f29d167d02c 100644 > > > --- a/arch/arm/kernel/smp_scu.c > > > +++ b/arch/arm/kernel/smp_scu.c > > > @@ -18,6 +18,7 @@ > > > > > > #define SCU_CTRL 0x00 > > > #define SCU_ENABLE (1 << 0) > > > +#define SCU_STANDBY_ENABLE (1 << 5) > > > #define SCU_CONFIG 0x04 > > > #define SCU_CPU_STATUS 0x08 > > > #define SCU_INVALIDATE 0x0c > > > @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) > > > if (scu_ctrl & SCU_ENABLE) > > > return; > > > > > > - scu_ctrl |= SCU_ENABLE; > > > + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; > > > > I don't think this bit exists on all revisions of the A9. > > Thanks for the info, Will. Is there any side-effect to write the > standby bit on those revisions which do not define the bit? I'm not actually sure what happens at the hardware level, but the TRM is pretty clear in its definition of `RESERVED': `All reserved bits not used by the implementation must be written as 0 and read as 0.' Will
On Mon, Jul 21, 2014 at 05:27:04PM +0800, Shawn Guo wrote: > On Mon, Jul 21, 2014 at 09:51:50AM +0100, Will Deacon wrote: > > On Mon, Jul 21, 2014 at 08:45:54AM +0100, Shawn Guo wrote: > > > With SCU standby enabled, SCU CLK will be turned off when all processors > > > are in WFI mode. And the clock will be turned on when any processor > > > leaves WFI mode. > > > > > > This behavior should be preferable in terms of power efficiency of > > > system idle. So let's set the SCU standby bit to enable the support in > > > function scu_enable(). > > > > > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > > > --- > > > arch/arm/kernel/smp_scu.c | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/arch/arm/kernel/smp_scu.c b/arch/arm/kernel/smp_scu.c > > > index c947508f84e6..9f29d167d02c 100644 > > > --- a/arch/arm/kernel/smp_scu.c > > > +++ b/arch/arm/kernel/smp_scu.c > > > @@ -18,6 +18,7 @@ > > > > > > #define SCU_CTRL 0x00 > > > #define SCU_ENABLE (1 << 0) > > > +#define SCU_STANDBY_ENABLE (1 << 5) > > > #define SCU_CONFIG 0x04 > > > #define SCU_CPU_STATUS 0x08 > > > #define SCU_INVALIDATE 0x0c > > > @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) > > > if (scu_ctrl & SCU_ENABLE) > > > return; > > > > > > - scu_ctrl |= SCU_ENABLE; > > > + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; > > > > I don't think this bit exists on all revisions of the A9. > > Thanks for the info, Will. Is there any side-effect to write the > standby bit on those revisions which do not define the bit? The usual question - could the firmware enable this bit before Linux starts? We already do a read/modify/write sequence here and are only supposed to write the enable bit as the rest are implementation defined.
On Mon, Jul 21, 2014 at 10:44:52AM +0100, Will Deacon wrote: > > > > @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) > > > > if (scu_ctrl & SCU_ENABLE) > > > > return; > > > > > > > > - scu_ctrl |= SCU_ENABLE; > > > > + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; > > > > > > I don't think this bit exists on all revisions of the A9. > > > > Thanks for the info, Will. Is there any side-effect to write the > > standby bit on those revisions which do not define the bit? > > I'm not actually sure what happens at the hardware level, but the TRM is > pretty clear in its definition of `RESERVED': > > `All reserved bits not used by the implementation must be written as 0 and > read as 0.' Okay. Can you please tell which revisions of A9 do not implement this standby bit, so that I can check and skip them? The oldest revision I can find on http://infocenter.arm.com/ is r2p0, which already implements the bit. Shawn
On Mon, Jul 21, 2014 at 11:26:38AM +0100, Catalin Marinas wrote: > The usual question - could the firmware enable this bit before Linux > starts? It could, I guess. Actually, on i.MX we're setting this bit in platform code right now. But I think setting this bit makes sense for most of the platforms, so it can reasonably be done in SCU core function. Isn't it the point of having core/common function after all? > We already do a read/modify/write sequence here and are only > supposed to write the enable bit as the rest are implementation defined. Isn't standby bit implemented by all A9 SCU except a couple of very early revisions (per Will)? Shawn
On Tue, Jul 22, 2014 at 03:09:12AM +0100, Shawn Guo wrote: > On Mon, Jul 21, 2014 at 11:26:38AM +0100, Catalin Marinas wrote: > > The usual question - could the firmware enable this bit before Linux > > starts? > > It could, I guess. Actually, on i.MX we're setting this bit in platform > code right now. But I think setting this bit makes sense for most of > the platforms, so it can reasonably be done in SCU core function. Isn't > it the point of having core/common function after all? Only that it wouldn't be consistent. If a platform (like some OMAPs) boots in non-secure mode, the SCU would be already enabled the firmware and your patch would not have any effect (that's one reason on arm64 I try to get some consistency between various platforms and not rely on the kernel which may or may not be able to enable certain features; but it's late to enforced this on arm32). > > We already do a read/modify/write sequence here and are only > > supposed to write the enable bit as the rest are implementation defined. > > Isn't standby bit implemented by all A9 SCU except a couple of very > early revisions (per Will)? And we don't know the behaviour of setting this bit on such A9 early revisions. So we can try to (1) find out if there are any in the field, (2) read the RTL to see if anything happens or (3) add a check in Linux for such revisions. I think (3) should be the case but you need to figure out which revisions these are.
On Tue, Jul 22, 2014 at 05:13:31PM +0100, Catalin Marinas wrote: > And we don't know the behaviour of setting this bit on such A9 early > revisions. So we can try to (1) find out if there are any in the field, > (2) read the RTL to see if anything happens or (3) add a check in Linux > for such revisions. I think (3) should be the case but you need to figure > out which revisions these are. Can I ask the favor from you ARM folks on that, since I cannot figure it out from any public information? Shawn
On Wed, Jul 23, 2014 at 05:50:57AM +0100, Shawn Guo wrote: > On Tue, Jul 22, 2014 at 05:13:31PM +0100, Catalin Marinas wrote: > > And we don't know the behaviour of setting this bit on such A9 early > > revisions. So we can try to (1) find out if there are any in the field, > > (2) read the RTL to see if anything happens or (3) add a check in Linux > > for such revisions. I think (3) should be the case but you need to figure > > out which revisions these are. > > Can I ask the favor from you ARM folks on that, since I cannot figure > it out from any public information? Looking through some of the older A9 TRMs, it seems that this bit was added in r2p0. Whether it works as advertised on the relevant SoCs I can't tell.
On Wed, Jul 23, 2014 at 05:45:21PM +0100, Catalin Marinas wrote: > On Wed, Jul 23, 2014 at 05:50:57AM +0100, Shawn Guo wrote: > > On Tue, Jul 22, 2014 at 05:13:31PM +0100, Catalin Marinas wrote: > > > And we don't know the behaviour of setting this bit on such A9 early > > > revisions. So we can try to (1) find out if there are any in the field, > > > (2) read the RTL to see if anything happens or (3) add a check in Linux > > > for such revisions. I think (3) should be the case but you need to figure > > > out which revisions these are. > > > > Can I ask the favor from you ARM folks on that, since I cannot figure > > it out from any public information? > > Looking through some of the older A9 TRMs, it seems that this bit was > added in r2p0. Whether it works as advertised on the relevant SoCs I > can't tell. Thanks for checking, Catalin. Shawn
diff --git a/arch/arm/kernel/smp_scu.c b/arch/arm/kernel/smp_scu.c index c947508f84e6..9f29d167d02c 100644 --- a/arch/arm/kernel/smp_scu.c +++ b/arch/arm/kernel/smp_scu.c @@ -18,6 +18,7 @@ #define SCU_CTRL 0x00 #define SCU_ENABLE (1 << 0) +#define SCU_STANDBY_ENABLE (1 << 5) #define SCU_CONFIG 0x04 #define SCU_CPU_STATUS 0x08 #define SCU_INVALIDATE 0x0c @@ -54,7 +55,7 @@ void scu_enable(void __iomem *scu_base) if (scu_ctrl & SCU_ENABLE) return; - scu_ctrl |= SCU_ENABLE; + scu_ctrl |= SCU_ENABLE | SCU_STANDBY_ENABLE; writel_relaxed(scu_ctrl, scu_base + SCU_CTRL); /*
With SCU standby enabled, SCU CLK will be turned off when all processors are in WFI mode. And the clock will be turned on when any processor leaves WFI mode. This behavior should be preferable in terms of power efficiency of system idle. So let's set the SCU standby bit to enable the support in function scu_enable(). Signed-off-by: Shawn Guo <shawn.guo@freescale.com> --- arch/arm/kernel/smp_scu.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)