On Thu, May 14, 2026 at 10:22:41PM +0500, Stepan Ionichev wrote: > uuid_kunit currently exercises only guid_parse() and uuid_parse() (plus > their invalid-input paths). The four random generators exported from > lib/uuid.c -- generate_random_uuid(), generate_random_guid(), > uuid_gen() and guid_gen() -- have no direct kunit coverage. > > Random output cannot be compared against a fixed expected value, but > RFC 4122 section 4.4 specifies two invariants that any version-4 > random UUID/GUID must satisfy: > > - version 4 in the high nibble of the version byte > (byte 6 in the wire uuid_t layout, byte 7 in the byte-swapped > guid_t layout); > - variant DCE 1.1 (binary 10x) in the high bits of byte 8. > > Add four test cases that invoke each generator several times and > verify these bit patterns hold. The same checks catch a regression > in either the mask/OR sequence in the generators or the layout > constants. Run the loop a handful of times to cover the small but > non-zero chance that an unmasked random byte happens to satisfy > the version/variant pattern by accident on a single call.
Test cases are always good, although here I'm not sure we are so picky about following this RFC. Whatever, feel free to add Acked-by: Andy Shevchenko <[email protected]> ... > +static void uuid_test_uuid_gen(struct kunit *test) > +{ > + unsigned int i; > + uuid_t u; > + for (i = 0; i < 8; i++) { Can be for (unsigned int i = 0; i < 8; i++) { > + uuid_gen(&u); > + KUNIT_EXPECT_EQ(test, u.b[6] & 0xf0, 0x40); > + KUNIT_EXPECT_EQ(test, u.b[8] & 0xc0, 0x80); > + } > +} Same for the rest. -- With Best Regards, Andy Shevchenko
