[U-Boot] [PATCH 10/10] ARM: socfpga: arria10: add support for building Arria10

[Chin Liang See]

Hi Marek,

On Tue, 2015-11-24 at 10:31 +0100, Marek Vasut wrote:
> On Tuesday, November 24, 2015 at 04:17:34 AM, Chin Liang See wrote:
> > On Mon, 2015-11-23 at 17:25 -0600, Dinh Nguyen wrote:
> > > On 11/23/2015 05:20 PM, Marek Vasut wrote:
> > > > On Tuesday, November 24, 2015 at 12:04:10 AM, Dinh Nguyen
> > > > wrote:
> > > > > On 11/23/2015 05:03 PM, Marek Vasut wrote:
> > > > > > On Monday, November 23, 2015 at 11:50:15 PM, Dinh Nguyen
> > > > > > wrote:
> > > > > > > On 11/23/2015 04:46 PM, Marek Vasut wrote:
> > > > > > > > On Monday, November 23, 2015 at 11:32:27 PM, Dinh
> > > > > > > > Nguyen
> > > > > > > > wrote:
> > > > > > > > [...]
> > > > > > > > 
> > > > > > > > > > > > > The main point is that we need to program the
> > > > > > > > > > > > > FPGA
> > > > > > > > > > > > > during U-Boot booting up with a ~>10 MB rbf
> > > > > > > > > > > > > file
> > > > > > > > > > > > > while being
> > > > > > > > > > > > > limited to the OCRAM's size. I would like to
> > > > > > > > > > > > > contain this
> > > > > > > > > > > > > ugliness in it's own directory.
> > > > > > > > > > > > 
> > > > > > > > > > > > What's the problem with this ? We already
> > > > > > > > > > > > support
> > > > > > > > > > > > loading files
> > > > > > > > > > > > from storage in SPL, so just compile the FPGA
> > > > > > > > > > > > manager into SPL as
> > > > > > > > > > > > well and use it.
> > > > > > > > > > > 
> > > > > > > > > > > Ok, let me re-work it all under the c5/a5
> > > > > > > > > > > directory.
> > > > > > > > > > > Thanks for
> > > > > > > > > > > reviewing.
> > > > > > > > > > 
> > > > > > > > > > But you didn't really answer my question -- what is
> > > > > > > > > > the
> > > > > > > > > > problem with
> > > > > > > > > > the FPGA loader in SPL ?
> > > > > > > > > 
> > > > > > > > > I thought you've already answered your own question.
> > > > > > > > > For
> > > > > > > > > whatever
> > > > > > > > > reason, the downstream A10 is re-doing the FPGA
> > > > > > > > > manager
> > > > > > > > > just for this
> > > > > > > > > purpose.
> > > > > > > > 
> > > > > > > > Could the reason be that the FPGA manager in it's
> > > > > > > > current
> > > > > > > > state expects
> > > > > > > > one big buffer with the entire FPGA bitstream ? When
> > > > > > > > you're
> > > > > > > > in SPL and
> > > > > > > > you still don't have DRAM running, you cannot create
> > > > > > > > such
> > > > > > > > buffer
> > > > > > > > anywhere. Thus, what you need to do is to have some
> > > > > > > > sort of
> > > > > > > > code which
> > > > > > > > loads a bit of the bitstream file at time and feeds it
> > > > > > > > into
> > > > > > > > the FPGA
> > > > > > > > manager, piece by piece. This should be doable pretty
> > > > > > > > easily, what do
> > > > > > > > you think ?
> > > > > > > 
> > > > > > > That's exactly what is being in the mach-socfpga
> > > > > > > directory.
> > > > > > 
> > > > > > Um, am I missing it in this patchset ?
> > > > > 
> > > > > No, you're not missing it, I have not sent it up yet. That
> > > > > support is in
> > > > > the downstream, but also with the fpga-manager driver re
> > > > > -written.
> > > > > I need
> > > > > to clean this up before I can send it.
> > > > 
> > > > OK, thanks :)
> > > > 
> > > > > > > Yes, but should that code go into mach-socfpga or
> > > > > > > drivers?
> > > > > > 
> > > > > > The FPGA manager bits are already in drivers/fpga/ , so
> > > > > > that's
> > > > > > where the
> > > > > > improvements should go. If you need some special handling
> > > > > > in
> > > > > > the SPL,
> > > > > > that should be in mach-socfpga . In case it's too much
> > > > > > change
> > > > > > to the
> > > > > > current SPL, moving the spl.c to spl-gen5.c and creating
> > > > > > new
> > > > > > spl-gen10.c
> > > > > > might make sense ... or something like that, possibly even
> > > > > > with
> > > > > > some
> > > > > > spl-common.c .
> > > > > 
> > > > > Ok. BTW, Arria10 is not using SPL. It has 256K of OCRAM that
> > > > > U
> > > > > -Boot can
> > > > > use to setup the SDRAM.
> > > > 
> > > > So what do I do if I want to boot arria10 from NAND ? UBI and
> > > > UBIFS
> > > > won't
> > > > fit into 256kiB, so I think using SPL might be the sensible
> > > > thing
> > > > afterall,
> > > > since you would be able to use arbitrarily-sized U-Boot.
> > > 
> > > I hope Chin Liang can chime here, I know that we have support for
> > > NAND,
> > > but I haven't been part of that task, so I don't know how it's
> > > being
> > > done.
> > 
> > Finally my email is back online :)
> > 
> > We do have NAND support but not with UBI and UBIFS.
> 
> I just hope you're not using raw NAND and just hoping it will work.
> 
> > For this support, user can use U-Boot to load arbitrarily-sized U
> > -Boot
> > that run on SDRAM.
> > 
> > One of the nice thing of U-Boot over SPL is the console support and
> > ability to troubleshoot.
> > This is possible with Arria 10 SoC as we have larger OCRAM (256kB
> > vs CV
> > SoC 64kB).
> 
> OK, that's not really the point here -- the point is, if you compile
> enough
> features into U-Boot, it will be bigger than those 256k. What will
> you do
> then ?

For this case, we will have 4 boot stages where a minimal U-Boot that
run on OCRAM to load a larger U-Boot which will run on SDRAM. For
better explanation, here are the events for this 4 boot stages.

1. BootROM load minimal U-Boot to OCRAM
2. Minimal U-Boot initialize all the critical HW such as clocks, DDR
3. Minimal U-Boot load larger U-Boot to SDRAM
4. Larger U-Boot will then load Linux
5. Linux boot

Thanks
Chin Liang


> 
> Best regards,
> Marek Vasut