Small signal linearity is often an -assumption- used to
simplify (or make at all tractable) circuit analysis. You
will find very few dead-linear active devices; linearity
is an ideal. Many devices have regions of "good enough".
If a device is large signal linear then it will usually be
small signal linear. However a small signal linear (as
tested) device may become large signal nonlinear due
to things like clipping, or being moved out of the small
signal bias conditions, by larger signals. I have run
into some instances of a device that looks large signal
linear, being in fact small signal nonlinear - for example,
MOSFETs in development with "channel hookup" problems
that appear high impedance with Vds~0V, Vgs>>VT and
need significant Vds to connect the channel to the drain.
In "RF land" the nonlinearity of interest may be the Cdb,
Csb, Cgb capacitances' C-V swing more than ohmic
nonlinearity - especially for well-driven CMOS switches.
This is why FDSOI wins - bottom plate and sidewall Cxx
are now non-semiconductor, so linear (constant) terms.