[U-Boot] [PATCH] ARM SOCFPGA: add resetmgr command so reset can be deasserted in bootcmd (for example on peripheral dma interfaces after fpga has been programmed).

[Marek Vasut]

On 12/16/2017 06:03 PM, Frank Mori Hess wrote:
> Marek Vasut wrote:
>> Please always CC the list. Do NOT top-post.
> 
> You do realize I was replying to an email you sent to my personal address and 
> you didn't even send to the list?

The reply was To: to the u-boot ML, CC others, check the headers.

>> What is your goal here ?
> 
> To put things in context, my larger goal is to update u-boot from the old 
> version altera integrates into Quartus to a reasonably recent version of 
> mainline u-boot being provided by the distro we are using.  My naive hope was 
> the new version of u-boot would work at least as well as the old altera one.  
> Experience so far:  infinite reboot loop due to broken cadence driver:
> 
> https://lists.denx.de/pipermail/u-boot/2017-December/313470.html
> 
> No response except from author responsible for breaking the driver insisting 
> his changes be kept.

You need to figure out how to make it work for both platforms. Propose
some sort of patch to Vignesh and ask him to test it, I think he'd be
more than willing to ... CC me if you get stuck.

> And now: DMA peripheral requests for FPGA are non-functional due to mainline 
> u-boot ignoring the reset_config.h in the handoff files generated by Quartus.  
> Apparently, the mainline uboot position is that it is inappropriate to provide 
> any more support for initializing the resets than providing the ability to 
> write to memory addresses with "mw".

So far, it seems to be that mw is really all you need.
Do you have an argument for why mw is not sufficient ?

>> Going back to my initial question -- what is your usecase and your aim
>> here ? Usually you use FPGA manager in Linux to load the FPGA.
>>>>
>>>> But you can really just do mw to the correct address or create a U-Boot
>>>> script , so this command is not really needed, is it ?
>>
> 
> Ok, I am running Linux on the board.  I don't see how it would help to load 
> the FPGA from Linux rather than u-boot.  The dma peripheral requests would 
> still be just as disabled.  The only difference would be that I would be 
> forced to deassert the resets from Linux rather than u-boot, which I guess 
> would make it not your problem?

You might want to tone down this hostile attitude, it could be that this
is the same thing which triggered Vignesh in the QSPI NOR thread ...

> In principle, the fpga manager provides a 
> write_complete hook that the socfpga fpga manager could use to deassert resets 
> (perhaps based on device tree settings), but looking at my 4.1.33 fpga/
> socfpga.c it doesn't seem to.  

Oh, so you're using some ancient hacked-up vendor kernel, not mainline
Linux. If you model your hardware correctly in DT or whatever you use
and your drivers (Linux or U-Boot) are correct, then the DMA resets are
also toggled correctly and you won't need to start hacking in all these
special commands. That's how things should be done ...

-- 
Best regards,
Marek Vasut