mm.c: function virt_to_xen_l2e

[PUCCETTI Armand <armand.puccetti@cea.fr>]

I am trying to understand the function virt_to_xen_l2e.
The actual code in xen V3.0.1 begins with:

l2_pgentry_t *virt_to_xen_l2e(unsigned long v)
{
    l4_pgentry_t *pl4e;
    l3_pgentry_t *pl3e;
    l2_pgentry_t *pl2e;

    pl4e = &idle_pg_table[l4_table_offset(v)];
    if ( !(l4e_get_flags(*pl4e) & _PAGE_PRESENT) )
    {
        pl3e = page_to_virt(alloc_xen_pagetable());
        clear_page(pl3e);
        *pl4e = l4e_from_paddr(__pa(pl3e), __PAGE_HYPERVISOR);
    }
   
    pl3e = l4e_to_l3e(*pl4e) + l3_table_offset(v);
...

What puzzles me is pl3e. If I decompose the first expression
of the rhs of the last statement above, introducing some temporary vars, 
I get:

    intpte_t a, b;
    physaddr_t c;

    a=(*pl4e).l4;
   
    b = (a & (PADDR_MASK & PAGE_MASK));
   
    c = ((physaddr_t) b);   
   
    pl3e= ((l3_pgentry_t *)__va(c)) + ...; // same line as original

pl3e is apparently made by some arithmetic transformations and then a 
cast to a pointer.
How do we know if the final address pl3e is defined, and the address it 
refers to is correct??

Same question about pl2e, as similar transformations are made to 
calculate it some lines
later.

thanks for any insight.

Armand