Please consider adding support for the VIA padlock hardware AES acceleration in loop-aes. From the loop-aes README: "If you want enable VIA padlock hardware AES xcrypt instructions, specify PADLOCK=y on make command line. If enabled, presence of VIA padlock instructions is detected at run time, but code still works on non-padlock processors too." I see two options here, either make PADLOCK=y the default (should still run on other processors) or (probably more preferable) add a padlock USE flag and do something like: use padlock && BUILD_PARAMS="${BUILD_PARAMS} PADLOCK=y"
*** This bug has been marked as a duplicate of 145537 *** *** This bug has been marked as a duplicate of 145537 ***
Bug 145537 does not appear to be related? Padlock support in loop-aes would neither use nor require support in openssl. Upstream loop-aes already contains it's own padlock implementation as a compile time option but the ebuild does not enable it. All that's needed is for the ebuild to allow that compile time option to be set?
CCing main contributor for this :) can you please contribute a patch?
Created attachment 97891 [details, diff] loop-aes-3.1d.ebuild.diff Sure! Although I don't have the hardware to test it, it seems to work without the hardware, but I notice a performance impact. Add new use flag, padlock: Use VIA padlock instructions, detected at run time, code still works on non-padlock processors too.
Works great, thanks! Nearly triples the IO speed on a C7. If you need any testing on this hardware in the future please feel free to CC me.
thanks, fixed