[U-Boot] [PREVIEW] LL TEMAC v10 refactored -- for evaluation only

[Stephan Linz]

Am Donnerstag, den 23.02.2012, 13:22 +0100 schrieb Michal Simek: 
> Hi Stephan,

Hi Michal,

I will take your comments seriously and ar working on a solution without
a complex conversion matrix and a ll_temac standard initialization. Give
me some time to do so ...

Stephan

> 
> Stephan Linz wrote:
> > Hi Michal,
> > 
> > back to your requests and objections I've built up a new patch set for
> > evaluation. Some of the changes relate to the Microblaze platform rather
> > than the network driver.
> > 
> > 1) U-Boot configuration variables vs. xparameters.h
> > ----------------------------------------------------------------------
> > I've tried to find a compromise between the U-Boot coding and
> > configuration style and your preference. But at least in the point of
> > the many many CPP conditional directives I can not reduce the code. I
> > can move the code only to another place. I need a converstion from the
> > automaticaly generated content of xparameters.h to the U-Boot
> > configuration space. Now you can find this conversion matrix in a new
> > platform header, xconversions.h -- I've placed the header in
> > board/xilinx/microblaze-generic/ I think, we can find a better place.
> 
> Sorry I don't like that idea in general. I think that make no sense to
> create infrastructure for several ll_temac driver in bootloader.
> 
> The same code in different location is the same solution which I won't like.
> 
> > 
> > The conversion matrix avoids all the wrong configuration variables
> > inside U-Boot code beginning with XILINX_ (and not CONFIG_) and we can
> > expand the configuration to new software related options we never get
> > from hardware (example special interface naming). This mechanism can
> > also be used by other driver and platform code, for example
> > arch/microblaze/cpu/cache.c and arch/microblaze/lib/bootm.c (use
> > XILINX_USE_DCACHE instead of ex. CONFIG_XILINX_DCACHE),
> > common/cmd_mfsl.c (use XILINX_FSL_NUMBER instead of ex.
> > CONFIG_XILINX_FSL), serial_xuartlite.c (use XILINX_UARTLITE_BASEADDR,
> > XILINX_UARTLITE_BASEADDR1, ...), and probably also in xilinx_emaclite.c
> > and xilinx_axi_emac.c when we extand a multi-IP support as we have done
> > for the LL TEMAC now. Sure we have to work on, but it's another case (I
> > can make some patches if you want)
> 
> I agree that we should use the same variable name if has to start with CONFIG_
> prefix.
> Based on my experience make more sense to move all functionality and configuration
> thing to BSP and keep them out of mainline code because none cares in community
> that we can change almost everything.
> That's why u-boot uses xparamters.h, linux dts and 5 kernel parameters.
> 
> Creating any conversions out of BSPs is just wrong way and I won't agree with that.
> If you want to talk about renaming some variables I am ok to do so and I can also
> change my bsp to cover backward compatibility.
> 
> Pretty soon the same situation will be with u-boot for Xilinx Zynq arm platform
> where we will have to use the same style to be acceptable.
> 
> > Another but not currently existing feasibility would be to generate a
> > U-Boot conform xparameters.h, for example
> > CONFIG_XILINX_LL_TEMAC0_BASE_ADDR instead of XILINX_LLTEMAC_BASEADDR.
> > You and I, we have it in hand. We can change the Xilinx BSP in the near
> > future. You in PataLinux and I in my own home-brewed BSP generator TPOS.
> 
> As I wrote above. Suggest what we should change and we can do that.
> 
> > 
> > 2) Standard initialization vs. multiple direct calls from platform
> > ----------------------------------------------------------------------
> > Returning to the topic "standard" initialization. The last patch will
> > show the differences bwetween your prefered way to do all in board init
> > and my applicable standard initializition (a single call to the driver).
> > Please look at it in the code and decide for yourself.
> 
> IRC. I describe you how I would like to see that configuration. It should be
> the same as I use for axiemac and emaclite.
> 
> > 
> > We need a forward-looking decision, what kind of initialization, we want
> > to use. Do we want to copy and paste the standard initialization steps
> > from ip core to ip core (board init) and from platform to platform
> > (Microplace, PPC405, PPC440) or make a simple call to the driver?
> > 
> > My preferences is to call the driver from platform and let him run all
> > the necessary steps, that I do not want knowing on platform side.
> 
> You have to know platform all the time.
> I want to keep microblaze-generic platform as simple as possible.
> 
> If you are not OK with my preferred way you can of course create your new
> platform with conversion tables as you like.
> In PetaLinux we uses separate platform because of configuration things which lie
> above of it.
> 
> Honestly if I look at patch v10 1/8 my eyes hurting me because the origin patch (v3 or so)
> had ~670 line of code and yours 1943 and do the same things.
>  From my point of view I am OK to keep this driver completely out of mainline because
> it is based on plb which will be deprecated in near future.
> I know that people will still use it for a long time but the question is if make sense
> to push to mainline.
> I will definitely use v3 in PetaLinux and keep things simpler.
> Future is axi driver is in the mainline.
> 
> Regards,
> Michal
> 
> 

-- 
Viele Grüße,
Stephan Linz
______________________________________________________________________________
MB-Ref: http://www.li-pro.de/xilinx_mb:mbref:start
OpenDCC: http://www.li-pro.net/opendcc.phtml
PC/M: http://www.li-pro.net/pcm.phtml
Sourceforge: http://sourceforge.net/users/slz
Gitorious: https://gitorious.org/~slz