Re: [PATCH 0/6] freepgt: free_pgtables shakeup

[Hugh Dickins <hugh@veritas.com>]

On Sat, 26 Mar 2005, Russell King wrote:
> On Sun, Mar 27, 2005 at 12:51:14AM +1100, Nick Piggin wrote:
> 
> > But I don't quite understand (should really look at the code more),
> > how come you aren't leaking memory?
> 
> The ARM free_pgd_slow() knows about this special first L1 page table, and
> knows to free it if it exists when its called.
> 
> > Do _all_ processes share this same first L1 page table?
> 
> No.  It has to be specific to each process.  Each L1 page table entry
> covers 2MB, but executables start at virtual 0x8000.

I'm not satisfied with Nick's patch because it defines FIRST_USER_ADDRESS
as FIRST_USER_PGD_NR * PGDIR_SIZE i.e. 2MB.  And if that is the first
user address, then there is no need for his patch, because the new
free_pgtables will never touch that pgd because there's no vma in it.

That's why I thought arm needed no special code for this
(overlooking the nr_ptes bug issue, sorry).

But above you say FIRST_USER_ADDRESS should actually be 0x8000?
In that case, I think we should define FIRST_USER_ADDRESS to 0,
either in every asm-arch or in asm-generic, with arm overriding
it to 0x8000 - or better, to whatever #define you already have
for that 0x8000, but I didn't find it.

If vmas can occur in between 0x8000 and 2MB, then with Nick's patch
we'd be liable to pass free_pgtables a vma with vm_start lower than
floor, which is not a case I've thought through, nor wish to.

And, whether FIRST_USER_ADDRESS is 0x8000 or 2MB,
shouldn't your arch_get_unmapped_area be enforcing it?

Hugh
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/