This documentation change was being written at the same time as Richard's

changes. So I'm committing this version to overwrite his changes for now,
and he can always take his turn to overwrite my words if he wants :-)

PR: 86

doc/crypto/RSA_check_key.pod

@@ -18,7 +18,9 @@ in fact prime, and that B<n = p*q>.
 It also checks that B<d*e = 1 mod (p-1*q-1)>,
 and that B<dmp1>, B<dmq1> and B<iqmp> are set correctly or are B<NULL>.
 
-The key's public components may not be B<NULL>.
+As such, this function can not be used with any arbitrary RSA key object,
+even if it is otherwise fit for regular RSA operation. See B<NOTES> for more
+information.
 
 =head1 RETURN VALUE
 
@@ -30,12 +32,29 @@ obtained using L<ERR_get_error(3)|ERR_get_error(3)>.
 
 =head1 NOTES
 
-RSA_check_key() can only check soft keys.  If given hard keys, i.e. keys
-that were retreived from an ENGINE (with ENGINE_load_private_key()),
-RSA_check_key() will always return 0, as if the key was invalid.  The
-reason is that the private components B<p>, B<q> and B<d> are normally
-not available to OpenSSL for hard keys, and the checks can therefore not
-be done.
+This function does not work on RSA public keys that have only the modulus
+and public exponent elements populated. It performs integrity checks on all
+the RSA key material, so the RSA key structure must contain all the private
+key data too.
+
+Unlike most other RSA functions, this function does B<not> work
+transparently with any underlying ENGINE implementation because it uses the
+key data in the RSA structure directly. An ENGINE implementation can
+override the way key data is stored and handled, and can even provide
+support for HSM keys - in which case the RSA structure may contain B<no>
+key data at all! If the ENGINE in question is only being used for
+acceleration or analysis purposes, then in all likelihood the RSA key data
+is complete and untouched, but this can't be assumed in the general case.
+
+=head1 BUGS
+
+A method of verifying the RSA key using opaque RSA API functions might need
+to be considered. Right now RSA_check_key() simply uses the RSA structure
+elements directly, bypassing the RSA_METHOD table altogether (and
+completely violating encapsulation and object-orientation in the process).
+The best fix will probably be to introduce a "check_key()" handler to the
+RSA_METHOD function table so that alternative implementations can also
+provide their own verifiers.
 
 =head1 SEE ALSO