In a compliant DOM API there is nothing you need to do. Simply save actual newline characters to the attribute, the API will encode them correctly on its own (see Canonical XML spec, section 5.2).
If you do your own encoding (i.e. replacing
with
before saving the attribute value), the API will encode your input again, resulting in 

in the XML file.
Bottom line is, the string value is saved verbatim. You get out what you put in, no need to interfere.
However… some implementations are not compliant. For example, they will encode &
characters in attribute values, but forget about newline characters or tabs. This puts you in a losing position since you can't simply replace newlines with
beforehand.
These implementations will save newline characters unencoded, like this:
<xml attribute="line 1
line 2" />
Upon parsing such a document, literal newlines in attributes are normalized into a single space (again, in accordance to the spec) - and thus they are lost.
Saving (and retaining!) newlines in attributes is impossible in these implementations.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…