Re: [PATCH RFC 0/2] KVM: s390: avoid having to enable vm.alloc_pgste

[Martin Schwidefsky]

On Fri, 2 Jun 2017 12:28:48 +0200
Heiko Carstens <heiko.carstens@xxxxxxxxxx> wrote:

> On Fri, Jun 02, 2017 at 11:46:47AM +0200, Martin Schwidefsky wrote:
> > On Fri, 2 Jun 2017 09:02:10 +0200
> > Heiko Carstens <heiko.carstens@xxxxxxxxxx> wrote:  
> > > Maybe this is a bit over-simplified, but might work.  
> > This is not over-simplified at all, that does work:  
> 
> Good!
> 
> > +struct arch_elf_state {
> > +};
> > +
> > +#define INIT_ARCH_ELF_STATE { }
> > +
> > +#define arch_elf_pt_proc(ehdr, phdr, elf, interp, state) (0)
> > +#define arch_check_elf(ehdr, interp, interp_ehdr, state)	\
> > +({								\
> > +	struct elf64_hdr *hdr = (void*) ehdr;			\
> > +	int _rc = 0;						\
> > +	if (hdr->e_ident[EI_CLASS] == ELFCLASS64 &&		\
> > +	    (hdr->e_flags & 0x00000002) &&			\
> > +	    !page_table_allocate_pgste &&			\
> > +	    !current->mm->context.alloc_pgste) {		\
> > +		current->mm->context.alloc_pgste = 1;		\  
> 
> However, I think this is over-simplified, unless I'm mistaken.
> 
> If you set current->mm->context.alloc_pgste here, then that means that 4k
> page tables will be freed when the original mm will be released, instead of
> the correct 2k ones.
> 
> I think you need an additional intermediate context flag here. Something
> like current->mm->context.request_pgste or whatever, no?

Yes, the flags for the current mm and the next mm have to be different.
request_pgste is a nice name for the new flag, I'll use it.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.