You just need to show that you can evaluate $\phi$ in polynomial-time.
This is clearly true for any reasonable encoding of $\phi$ and $c$, but the details depend on the specific encoding you choose.
One of the many possible "natural" encodings could be as follows: $\phi$ is encoded by specifying the number of variables $n$, the number of clauses $m$, and list of clauses, where each clause is a list of literals.
Each literal is just a pair whose first element is the index (between $1$ and $n$) of a variable and the second elements is a boolean that encodes whether the variable is negated.
The assignment $c$ is an array of $n$ boolean values, the $i$-th value is the truth value of the $i$-th variable.
Assume that you already checked that the input is a valid encoding.
Then you can just scan the list of clauses. For each clause, scan the list of literals. For each literal check whether the literal is positive and the corresponding variable is set to true in $c$, or the literal is negated and the variable is set to false. If no literal in the clause satisfies the above conditions, reject. If you run of clauses without rejecting, accept.
This takes time proportional to $m+n$ on the RAM model. If you want to use a Turing machine, describing the algorithm is more annoying and depends on even more fine-grained details of the encoding, but it can be done.