Does XIP_MTD really exist?

[Nicolas Pitre]

On Thu, 15 Jun 2017, Chris Brandt wrote:

> In the MTD directory, there are hints that an XIP from MTD exits, but 
> mostly only to make code run from RAM when modifying a parallel NOR flash.
> 
> Is this code all abandoned?

It used to work on some SA1100 and PXA targets... more than 10 years 
ago. No idea if it is still 
in use. Last time I suggested ripping it out entirely, someone raised a 
hand to say they're still relying on it. That too was a few years back.

> Also, I don't really see of any mainline filesystem driver that would 
> make use of it.

Upon a time, the cramfs filesystem had the ability to flag some files 
that would not be compressed, aligned to a page boundary, and be mapped 
directly out of flash. I might have to revive that functionality in the 
near future if no longer available.

> Since as of v4.0, filemap_xip.c was removed (from the guys that added 
> DAX) which broke AXFS which I had been using.
> 
> So, now I'm trying to decide if I need to rework AXFS to work with a 
> mainline kernel (basically by adding filemap_xip.c back in under an fs/axfs
> directory), or if there is some other obvious XIP file system method 
> that I'm missing.
> 
> The NO_MMU + XIP_KERNEL people must be using something for a file 
> system....I'm just not sure what that is.

So far I've seen initramfs packaged with the kernel. The kernel is XIP 
but user space ends up living in RAM forever.


Nicolas