Jeremy Rosen wrote:
> for the TX and RX buffer, they use memory they get for
> m8xx_cpm_hostalloc... why ?
It's convenient for small, uncached buffer space. Don't be allocating
large (more than 32-64 bytes) out of this space, as there is only a
page or two shared among all CPM devices. This is normally used for
things like SCC/SMC UART "FIFOs".
> Does it have something to do with the MMU ?
No, but with caches.
> I would like to avoid doing that, is their a way to use normal ( kernel
> space) memory ?
An alternate and proper approach is to use consistent_alloc(). This will
allocate page size memory, aligned on page boundaries, with cache disabled.
You must use the virtual/physical addresses returned from this function
and keep track of them in the driver. The virt_to_* or __va/__pa() macros
won't work.
A third approach is to allocate kernel memory either with kmalloc() or
getpages(). This is not cache coherent with CPM DMA, so you need to
call the cache flush/invalidate functions to manage this space.
All of these have advantages and disadvantages you need to trade off in
your design.
-- Dan
** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/
