Re: [PATCH 2/3] arm64: use PAGE_KERNEL_ROX directly in alloc_insn_page

[Christoph Hellwig]

On Thu, Jun 18, 2020 at 12:35:06PM +0200, Peter Zijlstra wrote:
> On Thu, Jun 18, 2020 at 10:55:58AM +0200, David Hildenbrand wrote:
> > On 18.06.20 08:43, Christoph Hellwig wrote:
> > > Use PAGE_KERNEL_ROX directly instead of allocating RWX and setting the
> > > page read-only just after the allocation.
> > > 
> > > Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> > > ---
> > >  arch/arm64/kernel/probes/kprobes.c | 12 +++---------
> > >  1 file changed, 3 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/arch/arm64/kernel/probes/kprobes.c b/arch/arm64/kernel/probes/kprobes.c
> > > index d1c95dcf1d7833..cbe49cd117cfec 100644
> > > --- a/arch/arm64/kernel/probes/kprobes.c
> > > +++ b/arch/arm64/kernel/probes/kprobes.c
> > > @@ -120,15 +120,9 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
> > >  
> > >  void *alloc_insn_page(void)
> > >  {
> > > -	void *page;
> > > -
> > > -	page = vmalloc_exec(PAGE_SIZE);
> > > -	if (page) {
> > > -		set_memory_ro((unsigned long)page, 1);
> > > -		set_vm_flush_reset_perms(page);
> > > -	}
> > > -
> > > -	return page;
> > > +	return __vmalloc_node_range(PAGE_SIZE, 1, VMALLOC_START, VMALLOC_END,
> > > +			GFP_KERNEL, PAGE_KERNEL_ROX, VM_FLUSH_RESET_PERMS,
> > > +			NUMA_NO_NODE, __func__);
> > 
> > I do wonder if something like vmalloc_prot(size, prot) would make this
> > (and the other two users) easier to read.
> > 
> > So instead of ripping out vmalloc_exec(), converting it into
> > vmalloc_prot() instead.
> > 
> > Did you consider that?
> 
> For x86 Christoph did module_alloc_prot(), which is in his more
> extensive set of patches addressing this. I suspect that would be the
> right thing for ARM64 as well.

Yes.  The somewhat hacky way I added it cause problems for UML, so I
instead plan to do a series converting all architectures over to
module_alloc_prot, plus lots of other cleanups in the area that I
noticed.

I don't think vmalloc_prot is a good idea per se, as there only few
potential users, and I don't want too many vmalloc APIs.