Re: [PATCH] mips: Realtek RTL: select NO_EXCEPT_FILL

[Olliver Schinagl]

Hey Thomas,

I'm digging deeper into this and understand less of it :)

For one, we currently use both ioread32 (our timer driver for example) 
and ioread32be (our GPIO driver) and thus both variants of register 
access I would think without problems.

I also see that a few devices override? (is that possible at all) 
mangle-port.h, though I didn't dig into this too much, just noticed some 
other boards with those headers. Is this an option at all? just supply a 
SoC specific mangle-port.h?

Loonking at mangle-port.h (and only at the 32 bit case), basically, 
there's ioswabl and __mem_ioswabl that I should be concerned about. 
Grepping for those functions however, only yields me one result, where  
__relaxed_ioswabl gets defined to  __ioswabl, __mem_iosawbl is unused. 
Even more frustrating, is that git grep also doesn't find a user for 
__relaxed_ioswabl. I'm sure there's some non-greppable magic missing :) 
as the failures of setting this define of course are real.

So to further my search, how does this work then?

Olliver

On 19-02-2023 17:07, Olliver Schinagl wrote:
> Hey Thomas,
>
> On 19-02-2023 11:31, Thomas Bogendoerfer wrote:
> > On Sun, Feb 19, 2023 at 10:27:17AM +0100, Olliver Schinagl wrote:
> > > It's still odd though; as we do not have _anything_ PCI, but it
> > > SWAP_IO_SPACE causes the crash.
> > but something uses readX/write() calls. If you aren't using any driver
> > existing driver but only newly written dedicated for that SOC
> > you could use raw_read/raw_writeX() instead. These type of functions
> > are always using native endianess.
> Ok that is valueable information. I think currently the only driver
> (that I can think of right now) that we use 'off the shelf' is the uart
> driver, so I hope that's written cleanly :)
>
> As for readX/writeX, for sure that is being used, and I intended to
> refactor that while going to proper and clean drivers (the realtek
> support in openwrt is a big mess right now).
>
> So you say raw_readX, but what about ioread32 which I thought was preferred?
>
> I'll meanwhile read into what readX vs raw_readX to learn more with
> regards to endianess.
>
> Thank you so mcuh so far!
>
> > > What makes SWAP_IO_SPACE generic then? :)
> > als long as hardware presents memory used with readX/writeX is
> > seen as little endian independant from CPU endianess it's generic.
> 'little endian independent'? What does that mean, that the register maps
> follow the CPU endianess?
>
> > Thomas.