Moreover, the error register shoudn't be affeceted by the HOB bit, so this flag doesn't make sense.I meant IDE_TFLAG_IN_HOB_FEATURE, of course.
Once again blame the certain ioctl (you know which one)... :)
Oh, horror... and I know that it wasn't completely ungrounded since both ATA/PI-6 adn -7 have words about reading the features register (depending on HOB). At least ATA/PI-8 got rid of this. Anyway, I think we can safely get rid of this flag and just return the same value in 'features' and 'hob_features'.