Re: [PATCH] ns16550: Fix DM serial operation with non-DM SPL

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 3363 bytes --]

On Tue, Jan 17, 2023 at 04:09:14PM +0000, Andre Przywara wrote:
> On Tue, 17 Jan 2023 08:15:18 -0500
> Tom Rini <trini@konsulko.com> wrote:
> 
> Hi Tom,
> 
> > On Tue, Jan 17, 2023 at 12:09:38PM +0000, Andre Przywara wrote:
> > 
> > > Commit 9591b63531fa ("Convert CONFIG_SYS_NS16550_MEM32 et al to Kconfig")
> > > moved some NS16550 configuration variables into Kconfig.
> > > Among those there is CONFIG_SYS_NS16550_REG_SIZE, which used to be only
> > > defined for SPL build runs, but now is always set (thanks for Kconfig).
> > > However this breaks the gating logic in ns16550.h, where we *override*
> > > this variable for DM build, as we learn this setting from the DT instead.
> > > 
> > > As a consequence, we did the register shift twice: once when building
> > > the register struct (as required for non-DM SPL builds), but then also
> > > again in the driver after we parsed the reg-shift DT property.
> > > 
> > > Change the logic to match what the comment says: only observe
> > > CONFIG_SYS_NS16550_REG_SIZE when not using DM, and ignore it otherwise.
> > > 
> > > This fixes U-Boot proper for all sunxi boards, since they are relying
> > > on this driver being build non-DM for the SPL, but DM for U-Boot proper.
> > > 
> > > Fixes: 9591b63531fa ("Convert CONFIG_SYS_NS16550_MEM32 et al to Kconfig")
> > > Signed-off-by: Andre Przywara <andre.przywara@arm.com>
> > > ---
> > > Hi,
> > > 
> > > this is admittedly a quick fix, to get sunxi booting again. This whole
> > > code around the register access looks somewhat bonkers, to be honest,
> > > but at the moment I don't have time to rework this.
> > > Another possible "quicker fix" would be use a separate variable for the
> > > register shift, so that we don't have to redefine a Kconfig variable.
> > > Another idea would be to get rid of the struct for the registers
> > > altogether, to remove the hacks about when to shift.
> > > 
> > > Cheers,
> > > Andre
> > > 
> > >  include/ns16550.h | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/include/ns16550.h b/include/ns16550.h
> > > index 243226fc3d9..e9e2aeedd16 100644
> > > --- a/include/ns16550.h
> > > +++ b/include/ns16550.h
> > > @@ -26,11 +26,12 @@
> > >  
> > >  #include <linux/types.h>
> > >  
> > > -#if CONFIG_IS_ENABLED(DM_SERIAL) && !defined(CONFIG_SYS_NS16550_REG_SIZE)
> > > +#if CONFIG_IS_ENABLED(DM_SERIAL)
> > >  /*
> > >   * For driver model we always use one byte per register, and sort out the
> > >   * differences in the driver
> > >   */
> > > +#undef CONFIG_SYS_NS16550_REG_SIZE
> > >  #define CONFIG_SYS_NS16550_REG_SIZE (-1)
> > >  #endif  
> > 
> > Does:
> > https://patchwork.ozlabs.org/project/uboot/patch/20230110161946.3816866-5-trini@konsulko.com/
> > work for you as well?
> 
> Unfortunately not. This changed condition there does not trigger for
> sunxi. If I hack it that it applies, but only to the proper build, it
> works, though. But I don't feel like adding to this #if even more ;-)

I've got a version that does work, now, on my pine64 plus, and I've
asked Quentin to test his Rockchip platform that was also broken, before
I post a new version there.

A better long term fix would likely start by seeing which non-DM_SERIAL
cases we have, and what they need, still.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]