[okl4-developer] memsection_create() and eas_map()

[Geoffrey Lee]

On Tue, Nov 11, 2008 at 09:35:14AM +0100, Robert Wetzel wrote:
> Hi,
> 
> I've hacked together a small test application, which creates an EAS
> and tries to map memory from it's own memory space into the EAS. One
> of the things it tries to map is the framebuffer; it gets mapped, but
> the graphics output is kinda broken (starts at the wrong address).

Robert - 

Are you using memory section created from memsection_create() as
the store for your framebuffer memory?  We don't typically use
this to allocate framebuffer memory as we have no control
over the physical backing.  Note that sometimes this works
anyway if the physical chunk allocated is contigous and enough
to fill the memory section.

> 
> Digging a bit deeper it turns out memsection_create() and eas_map()
> don't really work together, because they approach memory allocation /
> mapping differently. Even more so, eas_map() seems to map only one 4k
> page and then stops there, regardless how big the fpage is given to it
> (seems to be a bug; the return value of l4e_map() differs from what is
> assumed in eas_map(), at least in 2.1).

OK Linux relies on eas_map() to map to external address spaces
from a memory section backing store (the heap), so that should work.

Yes, what you point out appears to be a problem, I think the intension
is to loop over the whole fpage rather than doing the 4K mapping,
as it would appear in the code.

> 
> What I wanted to ask is how is eas_map() _supposed_ to work, and was
> there ever the intention to get it working together with
> memsection_create()?

Would it be possible to provide more detail on how the memory section
is created and how you are doing the mapping?  A simple testcase
to show the situation would be really good.

> 
> 
> Robert

	-gl

> 
> _______________________________________________
> Developer mailing list
> Developer at okl4.org
> https://lists.okl4.org/mailman/listinfo/developer
> 

--