[U-Boot] [PATCH v5 4/4] riscv: Remove redundant a2 store on DRAM base in start.S

[Anup Patel]

On Tue, Nov 27, 2018 at 11:14 AM Anup Patel <anup at brainfault.org> wrote:
>
> On Tue, Nov 27, 2018 at 10:50 AM Rick Chen <rickchen36 at gmail.com> wrote:
> >
> > Anup Patel <anup at brainfault.org> 於 2018年11月27日 週二 上午11:28寫道：
> > >
> > > On Tue, Nov 27, 2018 at 8:50 AM Rick Chen <rickchen36 at gmail.com> wrote:
> > > >
> > > > > > > > Currently, the RISC-V U-Boot is saving a2 register at
> > > > > > > > CONFIG_SYS_DRAM_BASE in start.S which does not make sense because
> > > > > > > > there is no information passed by previous booting stage in a2
> > > > > > > > register.
> > > > > > > >
> > > > > > > > This patch removes redundant a2 store on DRAM base.
> > > > > > > >
> > > > > > > > Signed-off-by: Anup Patel <anup at brainfault.org>
> > > > > > > > ---
> > > > > > > >  arch/riscv/cpu/start.S | 2 --
> > > > > > > >  1 file changed, 2 deletions(-)
> > > > > > > >
> > > > > > > > diff --git a/arch/riscv/cpu/start.S b/arch/riscv/cpu/start.S index
> > > > > > > > 704190f946..e4276e8e19 100644
> > > > > > > > --- a/arch/riscv/cpu/start.S
> > > > > > > > +++ b/arch/riscv/cpu/start.S
> > > > > > > > @@ -38,8 +38,6 @@ _start:
> > > > > > > >         mv      s0, a0
> > > > > > > >         mv      s1, a1
> > > > > > > >
> > > > > > > > -       li      t0, CONFIG_SYS_SDRAM_BASE
> > > > > > > > -       SREG    a2, 0(t0)
> > > > > > > >         la      t0, trap_entry
> > > > > > > >  #ifdef CONFIG_RISCV_SMODE
> > > > > > > >         csrw    stvec, t0
> > > > > > > > --
> > > > > > >
> > > > > > > This is weird. I remember these two lines were already removed by
> > > > > > > Lukas's patch series before? Did not have time to dig out the history
> > > > > > > though.
> > > > > > >
> > > >
> > > > > > > Regards,
> > > > > > > Bin
> > > > > >
> > > > > > You are correct, however I removed it again, because I did not want to break
> > > > > > Rick's board. He did add a commit to the last pull request that removes these
> > > > > > two lines and adjusts his board accordingly, but it is not in the current one.
> > > > > >
> > > >
> > > > Hi Likas
> > > >
> > > > Thanks for your explanation.
> > > >
> > > > RIck's commit as below
> > > > https://www.mail-archive.com/u-boot@lists.denx.de/msg305880.html
> > >
> > > When we run U-Boot in S-mode the BBL runs from 0x80000000 so this
> > > two lines corrupts BBL instructions.
> > >
> > > If this is important for some board then please have it around #ifdef.
> > >
> >
> > Hi Anup
> >
> > In the discussion as below :
> > https://www.mail-archive.com/u-boot@lists.denx.de/msg305880.html
> >
> > I try to solve this issue with the aptch
> > [PATCH] riscv: ax25-ae350: Pass dtb address to u-boot with a1 register
> > diff --git a/arch/riscv/cpu/start.S b/arch/riscv/cpu/start.S
> > -       li      t0, CONFIG_SYS_SDRAM_BASE
> > -       SREG    a2, 0(t0)
> >
> > diff --git a/board/AndesTech/ax25-ae350/ax25-ae350.c
> > b/board/AndesTech/ax25-ae350/ax25-ae350.c
> >  void *board_fdt_blob_setup(void)
> >  {
> > -       void **ptr = (void *)CONFIG_SYS_SDRAM_BASE;
> > +       void **ptr = (void *)&prior_stage_fdt_address;
> >
> > in the previous pull request.
> >
> > But Bin do not agree with that I use prior_stage_fdt_address in
> > board_fdt_blob_setup( )
> > I try to explain why I use it like that way.
> > Then Bin have not any reply in the following mail.
> > Finally I decide to drop this patch in the next pull request.
> >
> > Hi Bin
> >
> > How do you think about I recovery this patch to fix this issue ?
>
> Actually, previous booting stage can pass location of FDT stored in flash
> to U-Boot. U-Boot requires FDT at a DRAM location which it can modify
> in-place before passing on to Linux kernel so we should relocate the FDT
> passed by previous booting stage to some board specific DRAM location.
>
> My suggestion is as follows:
>
> Instead of SDRAM_BASE, we can have new board specific config
> CONFIG_RISCV_PRIOR_FDT_BASE
>
> If CONFIG_RISCV_PRIOR_FDT_BASE is defined/selected by
> config then in start.S copy-over the FDT from location pointed by
> "a1" register to location pointed by CONFIG_RISCV_PRIOR_FDT_BASE.

I think you can copy-over FDT in C code too. You don't need to do
in start.S.

>
> In your board_fdt_blob_setup(), you can safely do:
> void **ptr = (void *)CONFIG_RISCV_PRIOR_FDT_BASE;
>
> Regards,
> Anup