Re: [PATCH] mtd: rawnand: omap2: Actually prevent invalid configuration and build error

[Miquel Raynal]

Hi Roger,

rogerq@xxxxxxxxxx wrote on Mon, 7 Mar 2022 14:25:48 +0200:

> Hi Miquel,
> 
> On 07/03/2022 12:03, Miquel Raynal wrote:
> > Hi Roger,
> > 
> > rogerq@xxxxxxxxxx wrote on Sat, 5 Mar 2022 00:50:14 +0200:
> >   
> >> Hi Miquel,
> >>
> >> On 04/03/2022 17:54, Miquel Raynal wrote:  
> >>> Hi Guenter, Roger,
> >>>
> >>> rdunlap@xxxxxxxxxxxxx wrote on Sat, 26 Feb 2022 22:55:28 -0800:
> >>>     
> >>>> On 2/19/22 16:44, Guenter Roeck wrote:    
> >>>>> On Sat, Feb 19, 2022 at 09:36:00PM +0200, Roger Quadros wrote:      
> >>>>>> The root of the problem is that we are selecting symbols that have
> >>>>>> dependencies. This can cause random configurations that can fail.
> >>>>>> The cleanest solution is to avoid using select.
> >>>>>>
> >>>>>> This driver uses interfaces from the OMAP_GPMC driver so we have to
> >>>>>> depend on it instead.
> >>>>>>
> >>>>>> Fixes: 4cd335dae3cf ("mtd: rawnand: omap2: Prevent invalid configuration and build error")
> >>>>>> Signed-off-by: Roger Quadros <rogerq@xxxxxxxxxx>      
> >>>>>
> >>>>> Tested-by: Guenter Roeck <linux@xxxxxxxxxxxx>      
> >>>>
> >>>> Tested-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>    
> >>>
> >>> Sorry for noticing that just now, but there is still a problem with
> >>> this patch: we now always compile-in the OMAP_GPMC driver whenever we
> >>> need the NAND controller, even though it is not needed. This grows the
> >>> kernel for no reason.    
> >>
> >> Sorry, I did not understand what you meant.
> >>
> >> We no longer explicitly enable OMAP_GPMC since we dropped the "select".
> >> This fixes all build issues that were reported recently.
> >>
> >> MTD_NAND_OMAP2 will not be enabled if OMAP_GPMC is not since we added
> >> the "depends on". This fixes the original build issue that we started to
> >> fix with select initially.  
> > 
> > Yes, this side is fine.
> > 
> > In the initial commit, you proposed:
> > 
> > --- a/drivers/mtd/nand/raw/Kconfig
> > +++ b/drivers/mtd/nand/raw/Kconfig
> > @@ -42,7 +42,8 @@ config MTD_NAND_OMAP2
> >         tristate "OMAP2, OMAP3, OMAP4 and Keystone NAND controller"
> >         depends on ARCH_OMAP2PLUS || ARCH_KEYSTONE || ARCH_K3 || COMPILE_TEST
> >         depends on HAS_IOMEM
> > +       select OMAP_GPMC if ARCH_K3
> > 
> > Which creates a dependency over OMAP_GPMC only for a single
> > architecture. Which means that other OMAP platforms do not necessarily
> > need OMAP_GPMC for the NAND controller to work. Now, you propose:  
> 
> No that is not true. Other platforms that need MTD_NAND_OMAP2 are
> explicitly selecting OMAP_GPMC
> i.e. in arch/arm/mach-omap2/Kconfig

Ok, in this case the fix is legit, but as you said there is certainly
some clean up to do on this side in a second time.

> > --- a/drivers/mtd/nand/raw/Kconfig
> > +++ b/drivers/mtd/nand/raw/Kconfig
> > @@ -42,8 +42,7 @@ config MTD_NAND_OMAP2
> >  	tristate "OMAP2, OMAP3, OMAP4 and Keystone NAND controller"
> >  	depends on ARCH_OMAP2PLUS || ARCH_KEYSTONE || ARCH_K3 || COMPILE_TEST
> >  	depends on HAS_IOMEM
> > 	depends on OMAP_GPMC
> > 
> > This means any of the other OMAP architectures will compile the GPMC
> > driver even though they might not need it, which would unnecessarily
> > increase the kernel size.
> > 
> > Am I missing something?  
> 
> MTD_NAND_OMAP2 NAND controller is a submodule of the OMAP GPMC IP. So it
> cannot work without OMAP_GPMC driver.
> 
> Hope this clarifies the doubts.

I was not sure anymore if there was a proper hardware dependency here
because of the "select GPMC if ARCH_K3" addition. But your
clarification make this more understandable now.

I will send the fix to Linus.

Cheers,
Miquèl

> >>> In fact, Roger once said:
> >>>
> >>> 	"We will figure out how to enable OMAP_GPMC for K3 architecture
> >>> 	some other way."
> >>>
> >>> It turns out this is not what was finally proposed. Could we try yet
> >>> another solution?    
> >>
> >> This issue is still present i.e. we cannot enable MTD_NAND_OMAP2 driver on
> >> K3 platform since OMAP_GPMC config is hidden and not select-able
> >> by user or defconfig file.
> >>
> >> But it is not yet a deal breaker since NAND on K3 is not yet enabled upstream.
> >>
> >> For this I think OMAP_GPMC has to be a visible config entry and select-able
> >> from a defconfig file as I had done initially [1].
> >>
> >> Now we have a lot of explanation to write as to why we need to do it ;)  
> > 
> > We certainly do :)
> >   
> >> [1] - https://lore.kernel.org/lkml/20211123102607.13002-3-rogerq@xxxxxxxxxx/
> >>  
> > 
> > Thanks,
> > Miquèl  
> 
> cheers,
> -roger