On Fri, Sep 07, 2018 at 02:57:42AM +0100, Ben Hutchings wrote:
> The current constraints for inline "rep xcrypt*" instructions mark ecx
> as an input only. The compiler can therefore assume wrongly that ecx
> holds the same value afterward, but in reality it will contain 0.
>
> This previously led to data corruption, which was fixed around by
> commit 46d8c4b28652 ("crypto: padlock-aes - Fix Nano workaround data
> corruption"). But a future compiler or different optimisation
> configuration could reintroduce the problem. Update the constraints
> to avoid this.
>
> Fixes: a76c1c23d0c3 ("crypto: padlock-aes - work around Nano CPU ...")
> Fixes: 46d8c4b28652 ("crypto: padlock-aes - Fix Nano workaround data ...")
> Signed-off-by: Ben Hutchings <[email protected]>
> ---
> This is totally untested, so don't assume I know what I'm talking
> about. :-)
Could you please test it by combining this patch with a revert
of my fix to confirm that it actually works?
Thanks!
--
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
