Re: [v2 PATCH] crypto: sun4i-ss - Fix sparse endianness markers

[Corentin Labbe]

On Fri, Sep 11, 2020 at 02:13:55PM +1000, Herbert Xu wrote:
> On Thu, Sep 10, 2020 at 02:22:48PM +0200, Corentin Labbe wrote:
> >
> > I get some md5 error on both A20+BE:
> > alg: ahash: md5 test failed (wrong result) on test vector \"random: psize=129 ksize=0\", cfg=\"random: inplace use_finup nosimd src_divs=[<reimport,nosimd>85.99%@+3999, 5.85%@+30, <reimport>0.96%@+25, <reimport,nosimd>5.9%@+2263, <flush,nosimd>2.11%@+1950] iv_offset=2 key_offset=43\"
> > and A33+BE:
> > [   84.469045] alg: ahash: md5 test failed (wrong result) on test vector \"random: psize=322 ksize=0\", cfg=\"random: inplace may_sleep use_finup src_divs=[<reimport>99.1%@+2668, <reimport>0.88%@alignmask+3630, 0.11%@+3403] iv_offset=33\"
> > +[   84.469074] need:35966fc8 b31ea266 2bf064e9 f20f40ad
> > +[   84.469084] have:e29e4491 f3b6effc fa366691 00e04bd9
> > 
> > Thoses errors are random. (1 boot out of 2)
> 
> Do these really go away without this patch applied? AFAICS the
> generated code should be identical.
> 

I got this on next-20200910/multi_v7_defconfig BigEndian
[   12.137856] alg: hash: skipping comparison tests for md5-sun4i-ss because md5-generic is unavailable
md5-sun4i-ss md5 reqs=763
[   98.286632] alg: ahash: md5 test failed (wrong result) on test vector \"random: psize=65 ksize=0\", cfg=\"random: use_finup src_divs=[95.28%@+1052, <reimport>0.61%@+4046, 0.87%@+24, <reimport,nosimd>3.24%@+542] key_offset=54\"

So sun4i-ss is not involved.
Strangely /proc/crypto show:
name         : md5                                                                                  
driver       : md5-generic                                                                          
module       : md5                                                                                  
priority     : 0                                                                                    
refcnt       : 1                                                                                    
selftest     : passed                                                                               
internal     : no                                                                                   
type         : shash                                                                                
blocksize    : 64                                                                                   
digestsize   : 16

and I didnt see anything failed/unknow in /proc/crypto

Why the failed algorithm is not visible ?