[PATCH] ARM: OMAP3: Fix iva2_pwrdm settings for 3703

[Mark A. Greer]

On Thu, May 16, 2013 at 11:35:16AM -0700, Tony Lindgren wrote:
> * Yegor Yefremov <yegor_sub1 at visionsystems.de> [130516 05:13]:
> > On 15.05.2013 23:50, Mark A. Greer wrote:
> > > On Wed, May 15, 2013 at 10:07:35AM -0700, Tony Lindgren wrote:
> > >> * Mark A. Greer <mgreer at animalcreek.com> [130514 18:57]:
> > >>> On Tue, May 14, 2013 at 05:25:37PM -0700, Tony Lindgren wrote:
> > >>>> Commit a819c4f1 (ARM: OMAP3: PM: Only access IVA if one exists)
> > >>>> changed PM to not access IVA registers on omaps that don't have
> > >>>> them. Turns out we still need to idle iva2 as otherwise iva2_pwrdm
> > >>>> will stay on and block deeper idle states.
> > >>>>
> > >>>> Signed-off-by: Tony Lindgren <tony at atomide.com>
> > >>>>
> > >>>> ---
> > >>>>
> > >>>> Paul & Kevin, do you have better ideas for fixing this?
> > >>>>
> > >>>> --- a/arch/arm/mach-omap2/pm34xx.c
> > >>>> +++ b/arch/arm/mach-omap2/pm34xx.c
> > >>>> @@ -546,8 +546,10 @@ static void __init prcm_setup_regs(void)
> > >>>>  	/* Clear any pending PRCM interrupts */
> > >>>>  	omap2_prm_write_mod_reg(0, OCP_MOD, OMAP3_PRM_IRQSTATUS_MPU_OFFSET);
> > >>>>  
> > >>>> -	if (omap3_has_iva())
> > >>>> -		omap3_iva_idle();
> > >>>> +	/*
> > >>>> +	 * We need to idle iva2_pwrdm even on am3703 with no iva2.
> > >>>> +	 */
> > >>>> +	omap3_iva_idle();
> > >>>>  
> > >>>>  	omap3_d2d_idle();
> > >>>>  }
> > >>>
> > >>> [Kevin, Paul, some background: Tony discovered that the am3703 needs
> > >>> to have omap3_iva_idle() called even though its not supposed to have
> > >>> an IVA.]
> > >>>
> > >>> This is potentially bad for other SoC's that don't have an IVA so I don't
> > >>> think its the way to go.  My preference would be to set the OMAP3_HAS_IVA
> > >>> flag for the am3703 only since its the one with the bug.  Maybe something
> > >>> in id.c:omap3xxx_check_features() but I don't see any existing way to check
> > >>> for an am3703 vs. other am37xx SoCs.  Any ideas?
> > >>>
> > >>> Another option, I suppose, is a dts entry but I don't like that at all.
> > >>
> > >> It seems that iva2_pwrdm is there for all omap3 even if OMAP3_HAS_IVA
> > >> flag is unset. And if that's the case, iva2 clocks still need to be idled
> > >> in all cases.
> > > 
> > > Ahh, this takes us to the "Great TI Docs Mystery" :) -- its basically
> > > impossible to tell because despite what their docs may say, the hardware
> > > can be quite different.  I'm not sure how to proceed other than trial &
> > > error with as many different SoCs as we can find.
> > > 
> > >> It's possible that not all steps in omap3_iva_idle() are needed though.
> > >> I can debug further to see which part of the omap3_iva_idle() are needed.
> > >>
> > >> Mark, do you have some omap3 with no iva (other than am3703) to test the
> > >> idle states with?
> > > 
> > > I have an am35xx which isn't supposed to have an IVA so I can test with it
> > > (although, I'm not sure how well the kernel works on the am35xx these days).
> > > 
> > > I'm a bit busy today but I'll try booting the am35xx tomorrow and if it
> > > comes up, see what I can figure out.
> > 
> > I think this issue is relevant to am3517 as you can see from this thread: http://thread.gmane.org/gmane.linux.ports.arm.omap/97903
> > I could boot only with your patch http://www.spinics.net/lists/arm-kernel/msg168865.html
> 
> OK sounds like Mark's patch as http://www.spinics.net/lists/arm-kernel/msg168865.html
> is needed as a fix.

FYI, that patch works (until recently, anyway) but is an unacceptable approach.

> > I have such a system running, so if you have any other patches/ideas to test, I would do it.
> 
> Well I think my patch does not matter for am3517 as that one has iva2
> while am3703 does not.

To make sure there is no confusion, the am3517 does NOT have an iva2.

Tony was misled by a boot log from an older kernel that didn't have the
patch that unsets the IVA feature flag.

Aside: I have so far been unable to get my am3517evm to come up.

Mark
--