I'll just add a bit of fuel to the above fire. As Sumelic notes, Zulu (and other Nguni languages) have /ɮ, ɬ, l/. The fact that /ɮ, l/ contrast suggests that /ɬ/ which is a voiceless version of /ɮ/ is not "voiceless l", it is a voiceless lateral fricative (as he notes), and not a voiceless /l/. Similarly, the existence of /ɬ/ in Lushootseed and numerous other PNW languages, plus the fact that there aren't other voiceless versions of sonorants, indicates that Lushootseed /ɬ/ is a voiceless lateral fricative, not a voiceless /l/. Same goes for lateral fricatives in Chadic.
Contrarily, evidence from Klamath indicates that would-be voiceless sonorants are phonologically aspirated. The language has /l l' ɬ/, that is plain, glottalized and "voiceless" /l/ (there is also a plain, glottalized and voiceless contrast in obstruents and sonorants). When /l/ or /n/ stand before /ɬ l'/, the result is [lʔ, lh]. This makes sense if /ɬ/ is really an aspirated sonorant, and the rule splits a geminated aspirated sonorant into a plain sonorant plus the corresponding laryngeal glide. Ito & Mester 1989 in their Rendaku paper mention a number of reasons for treating the "voiceless sonorants" of Burmese as being aspirated. (A similar argument from Lhasa Tibetan could probably be made, based on aspiration throwback in negative verb, but I don't have a datum on any /ɬ/-initial H toned verb).
It has been observed for English that r,l "devoice" in "pray, clay", but not "spray, splay", but this is exactly where you get aspiration differences on stops in English. Although English does not have "voiceless" liquids, it does (in some dialects) have voiceless glides, e.g. "which; human". There is evidence that these should be treated as "aspirated". The argument centers on the general distribution of aspiration in English: it has to be foot-initial (at the beginning of a syllable, and not immediately preceded by a stressed syllable unless the syllable with aspiration is also stressed). This not only determines aspiration of voiceless stops, it also explains the distribution of h and its deletion in prò(h)ibítion vs. prohíbit. This also affects the realization of wh which is "voiceless" in where, what, but voiced in anywhere, somewhat.
So it has been conjectured that there are no "voiceless sonorants", instead, there are aspirated sonorants (and glottalized sonorants), and plain sonorants. I have not seen a convincing example of [-voice] sonorants in any languages, though that is the usual description of how aspirated sonorants are produced.