for some reasons, i was looking into libgcrypt source code and found this :
if i understand this correctly, while asking to use /dev/random for some bytes; it actually check first if rdrand is available (without being very explicit about it) and use it up to 25%, which is fine i guess if it was used standalone.
Where i find it a little tricky is that it is activated by default. Also isnt /dev/random rely on rdrand as well? thus increasing the real % of entropy coming out of the intel/amd rdrand?
Isnt this a bit too much relying on rdrand closed implementation?