[parisc-linux] Re: CCIO dma io_command and related io_tlb format questions.

[Grant Grundler]

On Mon, Oct 16, 2006 at 01:51:27PM +0200, Joel Soete wrote:
...
>     but it seems that at in the history num_physpages (now ram size in bytes)
>     was well the physical number of page:
...
>     nowaday num_physpages is bytes so the comparision is eronous?
>     (compare bytes against page number)

Where did you get the idea num_physpages is in bytes?

These two examples suggest it's counting PAGE_SIZE pages :

mm/slab.c:      if (num_physpages > (32 << 20) >> PAGE_SHIFT)
mm/swap.c:      unsigned long megs = num_physpages >> (20 - PAGE_SHIFT);


>     the same way in:
>     iov_order = get_order(iova_space_size << PAGE_SHIFT);
> 
>     iova_space_size << PAGE_SHIFT == iova_space_size * 2^PAGE_SHIFT
>                                   == iova_space_size * PAGE_SIZE
> 
>     made sense when iova_space_size was a number of pages,

Look at get_order() then you'll understand why the "<< PAGE_SHIFT".
iov_order is being set to log2(iova_space_size)+1.

grant