On 2019/10/22 14:15, Herbert Xu wrote: > On Mon, Oct 21, 2019 at 04:45:14PM +0800, Zhou Wang wrote: >> >> I made CRYPTO_MANAGER_DISABLE_TESTS=n and CRYPTO_TEST=m. After loading >> hisi_qm and hisi_zip modules, I got: >> >> [ 138.232605] hisi_zip 0000:75:00.0: Adding to iommu group 40 >> [ 138.239325] hisi_zip 0000:75:00.0: enabling device (0000 -> 0002) >> [ 138.245896] hisi_zip 0000:b5:00.0: Adding to iommu group 41 >> [ 138.252435] hisi_zip 0000:b5:00.0: enabling device (0000 -> 0002) >> [ 138.260393] alg: No test for gzip (hisi-gzip-acomp) >> >> This is OK: as the test of zlib-deflate of hisi_zip was successful, so >> it was quiet, as there is no test case for gzip, so it printed above message. >> >> cat /proc/crypto, I got: >> name : gzip >> driver : hisi-gzip-acomp >> module : hisi_zip >> priority : 300 >> refcnt : 1 >> selftest : passed >> internal : no >> type : acomp >> >> name : zlib-deflate >> driver : hisi-zlib-acomp >> module : hisi_zip >> priority : 300 >> refcnt : 1 >> selftest : passed >> internal : no >> type : acomp >> [...] >> >> However, seems we can not trigger a test by loading tcrypto. > > The test has already been carried out when the algorithm is > registered. Testing twice doesn't change anything. To trigger > a new test, unload the algorithm and then run tcrypt again. > >> Do you mean as crypto_has_alg can detect if an alg has already >> been tested, so it directly breaks in the case 0 in do test in tcrypto? > > As I said, tests are always carried out at registration time so > by triggering the registration tcrypt knows that the test would > have already occured. > > In fact this tcrypt code exists only for legacy reasons. You can > also trigger the registration directly by loading your modules or > for templates use crconf. I got it, thanks for your explanation. Will stop to do this. Best, Zhou > > Cheers, >