| Message ID | 1431115050-23693-1-git-send-email-balbi@ti.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On 05/08/2015 02:57 PM, Felipe Balbi wrote: > By adding operating points, cpufreq-dt has a > chance of running and doing something useful. > > Signed-off-by: Felipe Balbi <balbi@ti.com> > --- > arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ > 1 file changed, 9 insertions(+) > > diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi > index c80a3e233792..ea1db20f64fc 100644 > --- a/arch/arm/boot/dts/am4372.dtsi > +++ b/arch/arm/boot/dts/am4372.dtsi > @@ -38,6 +38,15 @@ > clocks = <&dpll_mpu_ck>; > clock-names = "cpu"; > > + operating-points = < > + /* kHz uV */ > + 1000000 1325000 /* OPP_NITRO */ > + 800000 1260000 /* OPP_TURBO */ > + 720000 1200000 /* OPP_120 */ > + 600000 1100000 /* OPP_100 */ > + 300000 950000 /* OPP_50 */ > + >; > + > clock-latency = <300000>; /* From omap-cpufreq driver */ > }; > }; > which of these OPPs need AVS? which of these are dependent on Efuse bit dependent?
On 05/08/2015 03:09 PM, Nishanth Menon wrote: > On 05/08/2015 02:57 PM, Felipe Balbi wrote: >> By adding operating points, cpufreq-dt has a >> chance of running and doing something useful. >> >> Signed-off-by: Felipe Balbi <balbi@ti.com> >> --- >> arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ >> 1 file changed, 9 insertions(+) >> >> diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi >> index c80a3e233792..ea1db20f64fc 100644 >> --- a/arch/arm/boot/dts/am4372.dtsi >> +++ b/arch/arm/boot/dts/am4372.dtsi >> @@ -38,6 +38,15 @@ >> clocks = <&dpll_mpu_ck>; >> clock-names = "cpu"; >> >> + operating-points = < >> + /* kHz uV */ >> + 1000000 1325000 /* OPP_NITRO */ >> + 800000 1260000 /* OPP_TURBO */ >> + 720000 1200000 /* OPP_120 */ >> + 600000 1100000 /* OPP_100 */ >> + 300000 950000 /* OPP_50 */ >> + >; >> + >> clock-latency = <300000>; /* From omap-cpufreq driver */ >> }; >> }; >> > which of these OPPs need AVS? which of these are dependent on Efuse bit > dependent? > You can use http://git.ti.com/ti-linux-kernel/ti-linux-kernel/blobs/ti-linux-3.14.y/arch/arm/mach-omap2/opp43xx_data.c for reference.
On Fri, May 08, 2015 at 03:12:27PM -0500, Nishanth Menon wrote: > On 05/08/2015 03:09 PM, Nishanth Menon wrote: > > On 05/08/2015 02:57 PM, Felipe Balbi wrote: > >> By adding operating points, cpufreq-dt has a > >> chance of running and doing something useful. > >> > >> Signed-off-by: Felipe Balbi <balbi@ti.com> > >> --- > >> arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ > >> 1 file changed, 9 insertions(+) > >> > >> diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi > >> index c80a3e233792..ea1db20f64fc 100644 > >> --- a/arch/arm/boot/dts/am4372.dtsi > >> +++ b/arch/arm/boot/dts/am4372.dtsi > >> @@ -38,6 +38,15 @@ > >> clocks = <&dpll_mpu_ck>; > >> clock-names = "cpu"; > >> > >> + operating-points = < > >> + /* kHz uV */ > >> + 1000000 1325000 /* OPP_NITRO */ > >> + 800000 1260000 /* OPP_TURBO */ > >> + 720000 1200000 /* OPP_120 */ > >> + 600000 1100000 /* OPP_100 */ > >> + 300000 950000 /* OPP_50 */ > >> + >; > >> + > >> clock-latency = <300000>; /* From omap-cpufreq driver */ > >> }; > >> }; > >> > > which of these OPPs need AVS? which of these are dependent on Efuse bit > > dependent? > > > > > You can use > http://git.ti.com/ti-linux-kernel/ti-linux-kernel/blobs/ti-linux-3.14.y/arch/arm/mach-omap2/opp43xx_data.c > for reference. heh, why isn't that upstream yet ? Seems to be ready already. The point is that as of now, u-boot will set maximum OPP it can find and, for AM437x, that will be 800MHz or 1GHz depending on your board. 1GHz might not be supported in all SoCs and letting that be used all the time is likely going to reduce silicon lifetime. At least allowing ondemand governor run, we will be mostly running at 300MHz and only jump to "invalid" OPPs under load which, granted, is still not perfect, but better than running at 1GHz all the time, don't you agree ?
On 05/08/2015 03:24 PM, Felipe Balbi wrote: > On Fri, May 08, 2015 at 03:12:27PM -0500, Nishanth Menon wrote: >> On 05/08/2015 03:09 PM, Nishanth Menon wrote: >>> On 05/08/2015 02:57 PM, Felipe Balbi wrote: >>>> By adding operating points, cpufreq-dt has a >>>> chance of running and doing something useful. >>>> >>>> Signed-off-by: Felipe Balbi <balbi@ti.com> >>>> --- >>>> arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ >>>> 1 file changed, 9 insertions(+) >>>> >>>> diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi >>>> index c80a3e233792..ea1db20f64fc 100644 >>>> --- a/arch/arm/boot/dts/am4372.dtsi >>>> +++ b/arch/arm/boot/dts/am4372.dtsi >>>> @@ -38,6 +38,15 @@ >>>> clocks = <&dpll_mpu_ck>; >>>> clock-names = "cpu"; >>>> >>>> + operating-points = < >>>> + /* kHz uV */ >>>> + 1000000 1325000 /* OPP_NITRO */ >>>> + 800000 1260000 /* OPP_TURBO */ >>>> + 720000 1200000 /* OPP_120 */ >>>> + 600000 1100000 /* OPP_100 */ >>>> + 300000 950000 /* OPP_50 */ >>>> + >; >>>> + >>>> clock-latency = <300000>; /* From omap-cpufreq driver */ >>>> }; >>>> }; >>>> >>> which of these OPPs need AVS? which of these are dependent on Efuse bit >>> dependent? >>> >> >> >> You can use >> http://git.ti.com/ti-linux-kernel/ti-linux-kernel/blobs/ti-linux-3.14.y/arch/arm/mach-omap2/opp43xx_data.c >> for reference. > > heh, why isn't that upstream yet ? Seems to be ready already. The point we have tried in the past[1] - unfortunately that was just bandaid since the existing OPP device tree bindings have very limiting behavior across multiple SoCs.This was one of the reasons why we stopped adding more OPPs, since we are hoping to see the new bindings[2] which is under discussion settle down and help enable support for cases like that we have on TI SoCs, iMX SoCs etc. > is that as of now, u-boot will set maximum OPP it can find and, for > AM437x, that will be 800MHz or 1GHz depending on your board. 1GHz might > not be supported in all SoCs and letting that be used all the time is > likely going to reduce silicon lifetime. > At least allowing ondemand governor run, we will be mostly running at > 300MHz and only jump to "invalid" OPPs under load which, granted, is > still not perfect, but better than running at 1GHz all the time, don't > you agree ? The fix for this should ideally be in u-boot - we are trying not to introduce dts changes in the hopes that the new proposed bindings that Viresh has on discussion comes to conclusion and help introduce new dtb support with proper SoC description. allowing an SoC to enter an invalid OPP is broken by design. we must attempt to curb it. unfortunately, we already do have a broken implementation for AM335x, DRA7 in place which went with the assumption that we will be able to do modifiers on top of existing dt description and the hopes that [1] will eventually get upstream. But, as is clear now, [1] has no future path in upstream kernel. following the same broken path for newer SoC definitions will probably very short sighted by us. in my opinion, doing a temporary hack in upstream kernel is not an elegant approach. I suggest helping review and approving Viresh's new bindings so that we have a longer term solution is more the way to do this. Just my 2 cents. Thanks once again for identifying the urgent need for helping Viresh's series along - will be great if folks could help review and approve his series to get them upstream and help all of us ARM SoC variations along. [1] http://comments.gmane.org/gmane.linux.ports.arm.kernel/309466 [2] http://marc.info/?l=devicetree&m=143039569104090&w=2
Hi, On Sun, May 10, 2015 at 07:23:34PM -0500, Nishanth Menon wrote: > On 05/08/2015 03:24 PM, Felipe Balbi wrote: > > On Fri, May 08, 2015 at 03:12:27PM -0500, Nishanth Menon wrote: > >> On 05/08/2015 03:09 PM, Nishanth Menon wrote: > >>> On 05/08/2015 02:57 PM, Felipe Balbi wrote: > >>>> By adding operating points, cpufreq-dt has a > >>>> chance of running and doing something useful. > >>>> > >>>> Signed-off-by: Felipe Balbi <balbi@ti.com> > >>>> --- > >>>> arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ > >>>> 1 file changed, 9 insertions(+) > >>>> > >>>> diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi > >>>> index c80a3e233792..ea1db20f64fc 100644 > >>>> --- a/arch/arm/boot/dts/am4372.dtsi > >>>> +++ b/arch/arm/boot/dts/am4372.dtsi > >>>> @@ -38,6 +38,15 @@ > >>>> clocks = <&dpll_mpu_ck>; > >>>> clock-names = "cpu"; > >>>> > >>>> + operating-points = < > >>>> + /* kHz uV */ > >>>> + 1000000 1325000 /* OPP_NITRO */ > >>>> + 800000 1260000 /* OPP_TURBO */ > >>>> + 720000 1200000 /* OPP_120 */ > >>>> + 600000 1100000 /* OPP_100 */ > >>>> + 300000 950000 /* OPP_50 */ > >>>> + >; > >>>> + > >>>> clock-latency = <300000>; /* From omap-cpufreq driver */ > >>>> }; > >>>> }; > >>>> > >>> which of these OPPs need AVS? which of these are dependent on Efuse bit > >>> dependent? > >>> > >> > >> > >> You can use > >> http://git.ti.com/ti-linux-kernel/ti-linux-kernel/blobs/ti-linux-3.14.y/arch/arm/mach-omap2/opp43xx_data.c > >> for reference. > > > > heh, why isn't that upstream yet ? Seems to be ready already. The point > > we have tried in the past[1] - unfortunately that was just bandaid since > the existing OPP device tree bindings have very limiting behavior across > multiple SoCs.This was one of the reasons why we stopped adding more > OPPs, since we are hoping to see the new bindings[2] which is under > discussion settle down and help enable support for cases like that we > have on TI SoCs, iMX SoCs etc. fair enough. > > is that as of now, u-boot will set maximum OPP it can find and, for > > AM437x, that will be 800MHz or 1GHz depending on your board. 1GHz might > > not be supported in all SoCs and letting that be used all the time is > > likely going to reduce silicon lifetime. > > > At least allowing ondemand governor run, we will be mostly running at > > 300MHz and only jump to "invalid" OPPs under load which, granted, is > > still not perfect, but better than running at 1GHz all the time, don't > > you agree ? > The fix for this should ideally be in u-boot - we are trying not to right, ideally, yeah; but then we go back to the discussion regarding kernel vs bootloader dependencies :-) > introduce dts changes in the hopes that the new proposed bindings that > Viresh has on discussion comes to conclusion and help introduce new dtb > support with proper SoC description. allowing an SoC to enter an invalid > OPP is broken by design. we must attempt to curb it. unfortunately, we > already do have a broken implementation for AM335x, DRA7 in place which > went with the assumption that we will be able to do modifiers on top of > existing dt description and the hopes that [1] will eventually get > upstream. But, as is clear now, [1] has no future path in upstream > kernel. following the same broken path for newer SoC definitions will > probably very short sighted by us. > > in my opinion, doing a temporary hack in upstream kernel is not an > elegant approach. I suggest helping review and approving Viresh's new however this is not a hack, right ? If we get rid of OPP_NITRO and OPP_TURBO, then we will more than likely always be dealing with safe OPPs (yeah, I need to confirm this since it's not on public TRM, so as of now, take this statement with a grain of salt :-), moreover, even though we're trying to change opp bindings, the current situation is still very much accepted and will remain valid even after changing binding :-) Not to mention that people using AM43xx today might be using it under invalid OPPs and decreasing silicon life; I'd assume that's a very urgent detail to sort out. > bindings so that we have a longer term solution is more the way to do this. > > Just my 2 cents. Thanks once again for identifying the urgent need for > helping Viresh's series along - will be great if folks could help review > and approve his series to get them upstream and help all of us ARM SoC > variations along. np.
On Mon, May 11, 2015 at 10:19 AM, Felipe Balbi <balbi@ti.com> wrote: > >> in my opinion, doing a temporary hack in upstream kernel is not an >> elegant approach. I suggest helping review and approving Viresh's new > > however this is not a hack, right ? If we get rid of OPP_NITRO and > OPP_TURBO, then we will more than likely always be dealing with safe > OPPs (yeah, I need to confirm this since it's not on public TRM, so as > of now, take this statement with a grain of salt :-), moreover, even > though we're trying to change opp bindings, the current situation is > still very much accepted and will remain valid even after changing > binding :-) yes - if we do have a documented subset of OPPs that are valid for all "variants" of AM437x, we could add that in using the legacy bindings, but, we will have to do a transition over to the new bindings when they are finalized to support all OPPs appropriately. > Not to mention that people using AM43xx today might be using it under > invalid OPPs and decreasing silicon life; I'd assume that's a very > urgent detail to sort out. While I do agree that there is always a debate between fixing things in kernel for bootloader issues, but it does not mean that we should just postpone fixing the bootloader in this case - since, at this very moment, we are already in broken configuration - example sitting on a bootloader shell does have the same impact we have at this time. Lets try to help Viresh in getting his series sorted out meanwhile - maybe others can help as well :( --- Regards, Nishanth Menon -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
HI, On Mon, May 11, 2015 at 11:46:13AM -0500, Nishanth Menon wrote: > On Mon, May 11, 2015 at 10:19 AM, Felipe Balbi <balbi@ti.com> wrote: > > > >> in my opinion, doing a temporary hack in upstream kernel is not an > >> elegant approach. I suggest helping review and approving Viresh's new > > > > however this is not a hack, right ? If we get rid of OPP_NITRO and > > OPP_TURBO, then we will more than likely always be dealing with safe > > OPPs (yeah, I need to confirm this since it's not on public TRM, so as > > of now, take this statement with a grain of salt :-), moreover, even > > though we're trying to change opp bindings, the current situation is > > still very much accepted and will remain valid even after changing > > binding :-) > > yes - if we do have a documented subset of OPPs that are valid for all > "variants" of AM437x, we could add that in using the legacy bindings, > but, we will have to do a transition over to the new bindings when > they are finalized to support all OPPs appropriately. sounds fair to me. > > Not to mention that people using AM43xx today might be using it under > > invalid OPPs and decreasing silicon life; I'd assume that's a very > > urgent detail to sort out. > > While I do agree that there is always a debate between fixing things > in kernel for bootloader issues, but it does not mean that we should > just postpone fixing the bootloader in this case - since, at this very > moment, we are already in broken configuration - example sitting on a > bootloader shell does have the same impact we have at this time. agreed, I'll cook up a patch for bootloader too. > Lets try to help Viresh in getting his series sorted out meanwhile - > maybe others can help as well :( sure.
diff --git a/arch/arm/boot/dts/am4372.dtsi b/arch/arm/boot/dts/am4372.dtsi index c80a3e233792..ea1db20f64fc 100644 --- a/arch/arm/boot/dts/am4372.dtsi +++ b/arch/arm/boot/dts/am4372.dtsi @@ -38,6 +38,15 @@ clocks = <&dpll_mpu_ck>; clock-names = "cpu"; + operating-points = < + /* kHz uV */ + 1000000 1325000 /* OPP_NITRO */ + 800000 1260000 /* OPP_TURBO */ + 720000 1200000 /* OPP_120 */ + 600000 1100000 /* OPP_100 */ + 300000 950000 /* OPP_50 */ + >; + clock-latency = <300000>; /* From omap-cpufreq driver */ }; };
By adding operating points, cpufreq-dt has a chance of running and doing something useful. Signed-off-by: Felipe Balbi <balbi@ti.com> --- arch/arm/boot/dts/am4372.dtsi | 9 +++++++++ 1 file changed, 9 insertions(+)