[PATCH 08/18] serial: add uart driver for MediaTek MT7620 SoC

[Simon Glass]

Hi Weijie,

On Thu, 29 Oct 2020 at 03:47, Weijie Gao <weijie.gao at mediatek.com> wrote:
>
> On Mon, 2020-10-26 at 22:51 -0600, Simon Glass wrote:
> > Hi Weijie,
> >
> > On Fri, 16 Oct 2020 at 01:36, Weijie Gao <weijie.gao at mediatek.com> wrote:
> > >
> > > This patch adds uart support for MediaTek MT7620 and earlier SoCs.
> > >
> > > The UART used by MT7620 is incompatible with the ns16550a driver.
> > > All registers of this UART have different addresses. A special 16-bit
> > > register for Divisor Latch is used to set the baudrate instead of the
> > > original two 8-bit registers (DLL and DLM).
> > >
> > > The driver can be built without DM which is useful for tiny SPL.
> > >
> > > Signed-off-by: Weijie Gao <weijie.gao at mediatek.com>
> > > ---
> > >  drivers/serial/Kconfig         |  20 ++
> > >  drivers/serial/Makefile        |   1 +
> > >  drivers/serial/serial.c        |   2 +
> > >  drivers/serial/serial_mt7620.c | 350 +++++++++++++++++++++++++++++++++
> > >  4 files changed, 373 insertions(+)
> > >  create mode 100644 drivers/serial/serial_mt7620.c
> >
> > Why do you need to build without DM? We have of-platdata which
> > provides for smaller images. What is the SRAM size available?
> >
>
> Actually it's the total size (spl+u-boot-lzma.img) that matters.
> mt7620_rfb_defconfig is configured to be compatible with the original
> u-boot from MTK SDK, and the maximum size is 192KiB.

What is lzma.img ?

>
> Currently without SPL_DM, the final image u-boot-with-spl.bin is about
> 170KiB, the u-boot-spl.bin is only 20KiB.

That is definitely huge.

> If SPL_DM is enabled, the total size is 189KiB, which is very close to
> the limit 192KiB, and the size of spl is almost doubled, 39KiB.
> (Without DM, only serial_mt7620 is needed. However with DM, both
> serial_mt7620, clk-mt7620 and misc are needed)
>
> Considering that we may still add some feature to mt7620_rfb_defconfig
> in the future, only 3KiB left is not enough. Building without DM is
> apparently better.
>
> I believe using of-platdata can only reduce few KiBs of SPL, because the
> DM itself is the largest overhead.

That's not my experience. The core DM overhead is fairly small,
perhaps 3-4KB. The device tree itself is typically at least another
3KB, depending on how many nodes you mark for inclusion in SPL.

With of-platdata the overhead of DM is not that terrible. You can see
some stats here:

http://u-boot.10912.n7.nabble.com/RFC-PATCH-v2-0-3-RFC-tiny-dm-Proposal-for-using-driver-model-in-SPL-td418552.html

Regards,
SImon