Re: Discussion about iopa()

On Feb 9, 2007, at 5:33 PM, Timur Tabi wrote:

Kumar Gala wrote:

quoted

We should fix it so the driver just has to do dma_alloc_coherent,  
dma_free_coherent, and dma_map_*/dma_unmap_*

Ok, I'm getting confused now.  The driver currently uses ioremap 
().  You're saying that it should use the dma_xxx functions  
instead, but not until those functions support MURAM?  When that  
happens, won't that make ioremap() obsolete?

ioremap() is for IO space not normal memory or things that behave  
like normal memory (MURAM, on chip SRAM, etc).  ioremap will always  
be needed to ensure proper non-cachable/guarded access to registers  
on devices, etc...  The problem is that we don't have a mechanism  
between kmalloc/vmalloc (for system memory) and ioremap (for devices)  
to handle other memories in the system.  So we end up with a  
specialized allocator (rheap + muram wrappers) to deal with it.

So, I'd rather mapping also be specialized for MURAM.  I don't see  
any benefit in making mapping behave the same while not addressing  
the other functionality (allocation, deallocation, etc.)

- k