Re: [PATCH 3/5] powerpc/lib/sstep: Add bpermd instruction emulation

From: Segher Boessenkool <segher@kernel.crashing.org>
To: Matt Brown <matthew.brown.dev@gmail.com>
Cc: linuxppc-dev@lists.ozlabs.org, Anton Blanchard <anton@samba.org>
Subject: Re: [PATCH 3/5] powerpc/lib/sstep: Add bpermd instruction emulation
Date: Fri, 14 Jul 2017 11:17:01 -0500	[thread overview]
Message-ID: <20170714161701.GV13471@gate.crashing.org> (raw)
In-Reply-To: <CAPoR37Ymuy95zzshsgwQwsCyMJwdc3MOzneL4=dq_xABY7o_dA@mail.gmail.com>

On Fri, Jul 14, 2017 at 02:19:34PM +1000, Matt Brown wrote:
> >> +static nokprobe_inline void do_bpermd(struct pt_regs *regs, unsigned long v1,
> >> +                             unsigned long v2, int ra)
> >> +{
> >> +     unsigned int idx, i;
> >> +     unsigned char perm;
> >> +
> >> +     perm = 0x0;
> >> +     for (i = 0; i < 8; i++) {
> >> +             idx = (v1 >> (i * 8)) & 0xff;
> >> +             if (idx < 64)
> >> +                     perm |= (v2 & (1 << idx)) >> (idx - i);
> >
> > That doesn't work I think, the bit numbers ("idx") are big-endian?
> 
> Why would it be big-endian? Wouldn't it be in the same endian form as the arch?

Because that is what the ISA says.  Bit ordering is always BE.  If any
instruction behaves differently in LE mode that is explicitly described.

Please somehow test that the emulation works correctly, and describe
how you tested it, to give people the warm fuzzies.

Segher