One possible solution is based on knapsack.
Consider the list elements $a_1$, $a_2$, $\ldots$, $a_n$ in any fixed order.
Calculate the following boolean function: $f (k, t)$ is true if it is possible to have the sum of exactly $t$ from some subset of $a_1$, $a_2$, $\ldots$, $a_k$, and false otherwise.
Here, $k$ spans from $0$ to $n$, inclusive, and $t$ goes from $0$ to $s$: we don't need larger sums.
The calculation can be performed in $O (n \cdot s)$ using dynamic programming.
Just observe that $f (k, t) = f (k - 1, t)\mathrm{~or~} f (k, t - a_k)$: for the element $a_k$ and for every possible target sum $t$, we either drop the element from consideration, or take it into our sum.
In the latter case, we must have the chance to take it again, hence the transition to $(k, t - a_k)$ instead of $(k - 1, t - a_k)$.
Now, knowing which $(k, t)$ pairs are possible and which are not, we can recursively construct all possible answers in $O((n + s) \cdot R)$, where $R$ is the number of answers.
Indeed, descend from $(n, s)$ down to $(0, 0)$.
When we are at $(k, t)$, we can see whether $(k - 1, t)$ and $(k, t - a_k)$ are possible, and descend only into the possible branches.
As every descent takes at most $n + s$ steps, finding $R$ possible answers will happen in $(n + s) \cdot R$ time.
Actually, it will be faster since some descent paths share a common start, and some $a_k$ are greater than $1$.