Re: [PATCH 3/3] riscv: fix __user annotation for __copy_user()

[Palmer Dabbelt]

On Tue, 12 Jun 2018 11:19:57 PDT (-0700), luc.vanoostenryck@xxxxxxxxx wrote:
> On Tue, Jun 12, 2018 at 7:12 PM, Palmer Dabbelt <palmer@xxxxxxxxxx> wrote:
> > On Mon, 11 Jun 2018 20:00:08 PDT (-0700), luc.vanoostenryck@xxxxxxxxx wrote:
> > > On Mon, Jun 11, 2018 at 12:01:37PM -0700, Palmer Dabbelt wrote:
> > > > OK, I'll try and figure out what's going on.  We've had a handful of
> > > > headaches trying to get things like '.align 2; .align 2' to actually
> > > > produce
> > > > no NOPs for the second alignment directive, which is surprisingly
> > > > complicated due to the aggressive linker relaxation we do.
> > >
> > >
> > > OK. I imagine indeed but note that no linker is involved here so,
> > > if the problem is still present, it must already be in the assembler.
> >
> >
> > Ah, OK -- in that case then it's just not a bug.  In RISC-V land we handle
> > alignment as part of relaxation in the linker, so if you're looking at the
> > output of the assembler then you'll always see a bunch of NOPs for every
> > alignment directive.  If you 'objdump -dr' you should be able to see the
> > relocations that get emitted, there should be a R_RISCV_ALIGN that points to
> > the run of NOPs.
>
> Ah OK. Indeed I see the R_RISCV_ALIGN.
> Thanks for the explanation.

No problem.  There's heaps of info about this in some blog posts I made a while 
ago if you're interested.