It sounds like your only IV(s) is (are) categorical, and you will be examining the differences across categories using dummy variables (scored 0,1; or perhaps 1,2). Linearity is not an issue in that situation. If every predictor has only two values, then every predictor has a perfectly linear effect on the DV because of a basic axiom of Euclidean geometry: two points uniquely determine a straight line.

On the other hand, if one or more of your IVs is/are being treated in your model as continuous rather than categorical, then you should examine their linearity in the usual ways, primarily through bivariate scatterplots and multivariate residual plots. You could do that by running your model first in the Regression procedure, with regression diagnostics chosen from the Statistics options. If linearity diagnostics look good, you could then do final estimation with Process if you think that's best for addressing other potential violations of assumptions.

Dear Marcus Hohlfeld

These two websites will answer your question

http://people.duke.edu/~rnau/testing.htm

https://www.statisticssolutions.com/assumptions-of-linear-regression/

Also, please look at this article

https://www.open.ac.uk/socialsciences/spsstutorial/files/tutorials/assumptions.pdf

Thank you Burke D. Grandjean

So I assume it is enough to simply state that the Assumption of Linearity is not given for a regression of multicategorical IV? Everybody will understand that?

Because I simply cannot find any reference for that...

I don't think you need to even mention it. Or maybe just a footnote or a parenthetical to say it's "not relevant for a categorical IV" or words to that effect. That should be clear enough to anyone who understands why the regression assumptions otherwise matter.