Re: [RFC][PATCH] OMAP3: PM: Fix workaround implementation for OMAP3 errata (1.142)

[Kalle Jokiniemi <kalle.jokiniemi@digia.com>]

On Fri, 2009-08-07 at 11:48 +0300, Roger Quadros wrote:
> ext Kalle Jokiniemi wrote:
> > On Fri, 2009-08-07 at 11:03 +0300, Roger Quadros wrote:
> >> ext Kalle Jokiniemi wrote:
> >>> On Fri, 2009-08-07 at 01:14 +0300, Kevin Hilman wrote:
> >>>> Roger Quadros <ext-roger.quadros@nokia.com> writes:
> >>>>
> >>>>> As per errata 1.142, on EMU/HS devices, SDRC_POWER should be programmed
> >>>>> for automatic self-refresh before transition to OFF mode.
> >>>>> In the current implementation this is done in omap3_scratchpad_contents()
> >>>>> which is wrong, since this is the value that will be restored while
> >>>>> resuming from OFF mode and not while transitioning to it.
> >>>>>
> >>>>> This patch implements the workaround in the correct way as per errata.
> >>>>>
> >>>>> Signed-off-by: Roger Quadros <ext-roger.quadros@nokia.com>
> >>>> This looks right to me.
> >>>>
> >>>> Rajendra, Kalle, any comments? since you were the originators of
> >>>> the original patch.
> >>>>
> >>>> Kevin
> >>>>
> >>>>> ---
> >>>>>  arch/arm/mach-omap2/control.c |   16 ++--------------
> >>>>>  arch/arm/mach-omap2/pm34xx.c  |   20 ++++++++++++++------
> >>>>>  2 files changed, 16 insertions(+), 20 deletions(-)
> >>>>>
> >>>>> diff --git a/arch/arm/mach-omap2/control.c b/arch/arm/mach-omap2/control.c
> >>>>> index a7159a9..0a563c8 100644
> >>>>> --- a/arch/arm/mach-omap2/control.c
> >>>>> +++ b/arch/arm/mach-omap2/control.c
> >>>>> @@ -272,20 +272,8 @@ void omap3_save_scratchpad_contents(void)
> >>>>>  			(sdrc_read_reg(SDRC_ERR_TYPE) & 0xFFFF);
> >>>>>  	sdrc_block_contents.dll_a_ctrl = sdrc_read_reg(SDRC_DLLA_CTRL);
> >>>>>  	sdrc_block_contents.dll_b_ctrl = 0x0;
> >>>>> -	/*
> >>>>> -	 * Due to a OMAP3 errata (1.142), on EMU/HS devices SRDC should
> >>>>> -	 * be programed to issue automatic self refresh on timeout
> >>>>> -	 * of AUTO_CNT = 1 prior to any transition to OFF mode.
> >>>>> -	 */
> >>>>> -	if ((omap_type() != OMAP2_DEVICE_TYPE_GP)
> >>>>> -			&& (omap_rev() >= OMAP3430_REV_ES3_0))
> >>>>> -		sdrc_block_contents.power = (sdrc_read_reg(SDRC_POWER) &
> >>>>> -				~(SDRC_POWER_AUTOCOUNT_MASK|
> >>>>> -				SDRC_POWER_CLKCTRL_MASK)) |
> >>>>> -				(1 << SDRC_POWER_AUTOCOUNT_SHIFT) |
> >>>>> -				SDRC_SELF_REFRESH_ON_AUTOCOUNT;
> >>>>> -	else
> >>>>> -		sdrc_block_contents.power = sdrc_read_reg(SDRC_POWER);
> >>>>> +
> >>>>> +	sdrc_block_contents.power = sdrc_read_reg(SDRC_POWER);
> >>> Why do you want to remove the workaround from scratchpad?
> >>>
> >>> When we wake up from off mode, the boot ROM code writes these settings
> >>> as the first sdrc_power settings after wakeup. If that setting does not
> >>> include the workaround, we'll have a period of time where sdram is not
> >>> being refreshed. This hits especially HS devices that do long secure
> >>> context restore in ROM code as well.
> >>>
> >>>
> >> I suppose SDRAM is always refreshed (i.e. auto refresh) when ON. However when we 
> >> enter OFF mode it should be _self_ refreshed. That is done by the below code 
> >> before we switch to OFF mode.
> > 
> > The erratum speaks of auto-refresh failing after waking up from CORE off
> > mode. I think there is the described workaround in place in sleep34xx.S
> > code to kick up the auto refresh, but this still leaves a short gap
> > between boot ROM and the sleep code.
> > 
> > Unless the self-refresh mode does not get changed by the scratchpad
> > settings?
> > 
> > - Kalle
> 
> There are 2 parts in the errata 1.142.
> 
> The 1st part is what you are talking about. And it needs to be applied only on 
> ES3.0 devices.
> 
> The 2nd part is valid only for HS devices.
> "..it is mandatory on HS device to have the SDRC issuing automatic self-refresh
> entries on inactivity periods.."
> 
> My patch is only fixing the implementation of the 2nd part without changing the 
> work around for the 1st part. no?

The fix for second part is ok. But I have understood that the scratchpad
change was an additional safeguard against the 1st part. Rajendra, any
comments? 

I don't know how fast SDRAM corrupts when not refreshed, but on low OPP
it takes several msecs to reach public code from boot ROM. So keeping
SDRAM in self-refresh also in this period, seems sensible to me (even
though the erratum does not mention a requirement for this).

- Kalle

> 
> -roger
> 
> > 
> >> When we wake up from OFF mode (the scratchpad contents are used), we should 
> >> restore the original register setting that was there before we went to OFF mode 
> >> and not the self refresh mode setting. Hence the removal of above code.
> >>
> >> -roger
> >>>>>  
> >>>>>  	sdrc_block_contents.cs_0 = 0x0;
> >>>>>  	sdrc_block_contents.mcfg_0 = sdrc_read_reg(SDRC_MCFG_0);
> >>>>> diff --git a/arch/arm/mach-omap2/pm34xx.c b/arch/arm/mach-omap2/pm34xx.c
> >>>>> index 14f10bc..eb3c9e5 100644
> >>>>> --- a/arch/arm/mach-omap2/pm34xx.c
> >>>>> +++ b/arch/arm/mach-omap2/pm34xx.c
> >>>>> @@ -405,15 +405,23 @@ void omap_sram_idle(void)
> >>>>>  	}
> >>>>>  
> >>>>>  	/*
> >>>>> -	* On EMU/HS devices ROM code restores a SRDC value
> >>>>> -	* from scratchpad which has automatic self refresh on timeout
> >>>>> -	* of AUTO_CNT = 1 enabled. This takes care of errata 1.142.
> >>>>> -	* Hence store/restore the SDRC_POWER register here.
> >>>>> -	*/
> >>>>> +	 * Due to a OMAP3 errata (1.142), on EMU/HS devices SRDC should
> >>>>> +	 * be programed to issue automatic self refresh on timeout
> >>>>> +	 * of AUTO_CNT = 1 prior to any transition to OFF mode.
> >>>>> +	 */
> >>>>> +
> >>>>>  	if (omap_rev() >= OMAP3430_REV_ES3_0 &&
> >>>>>  	    omap_type() != OMAP2_DEVICE_TYPE_GP &&
> >>>>> -	    core_next_state == PWRDM_POWER_OFF)
> >>>>> +	    core_next_state == PWRDM_POWER_OFF) {
> >>>>> +
> >>>>>  		sdrc_pwr = sdrc_read_reg(SDRC_POWER);
> >>>>> +		sdrc_write_reg((sdrc_pwr &
> >>>>> +				~(SDRC_POWER_AUTOCOUNT_MASK|
> >>>>> +				SDRC_POWER_CLKCTRL_MASK)) |
> >>>>> +				(1 << SDRC_POWER_AUTOCOUNT_SHIFT) |
> >>>>> +				SDRC_SELF_REFRESH_ON_AUTOCOUNT,
> >>>>> +					SDRC_POWER);
> >>>>> +	}
> >>> This part seems ok to me.
> >>>
> >>> - Kalle
> >>>
> >>>
> >>>>>  
> >>>>>  	if (regset_save_on_suspend)
> >>>>>  		pm_dbg_regset_save(1);
> >>>>> -- 
> >>>>> 1.6.0.4
> > 
>