The length function $\ell_q(r,R)$ is the smallest length of a $ q $-ary linear code with codimension (redundancy) $r$ and covering radius $R$. In this work, new upper bounds on $\ell_q(tR+1,R)$ are obtained in the following forms: \begin{equation*} \begin{split} &(a)~\ell_q(r,R)\le cq^{(r-R)/R}\cdot\sqrt[R]{\ln q},~ R\ge3,~r=tR+1,~t\ge1, &\phantom{(a)~} q\text{ is an arbitrary prime power},~c\text{ is independent of }q. \end{split} \end{equation*} \begin{equation*} \begin{split} &(b)~\ell_q(r,R)< 3.43Rq^{(r-R)/R}\cdot\sqrt[R]{\ln q},~ R\ge3,~r=tR+1,~t\ge1, &\phantom{(b)~} q\text{ is an arbitrary prime power},~q\text{ is large enough}. \end{split} \end{equation*} In the literature, for $q=(q')^R$ with $q'$ a prime power, smaller upper bounds are known; however, when $q$ is an arbitrary prime power, the bounds of this paper are better than the known ones. For $t=1$, we use a one-to-one correspondence between $[n,n-(R+1)]_qR$ codes and $(R-1)$-saturating $n$-sets in the projective space $\mathrm{PG}(R,q)$. A new construction of such saturating sets providing sets of small size is proposed. Then the $[n,n-(R+1)]_qR$ codes, obtained by geometrical methods, are taken as the starting ones in the lift-constructions (so-called "$q^m$-concatenating constructions") for covering codes to obtain infinite families of codes with growing codimension $r=tR+1$, $t\ge1$.