Getting the selected text of a page is fairly easy, you can do something like
var text = window.getSelection().toString();
and you'll get a text representation of the currently selected text that you can pass from a content script to a background page or a popup.
Getting HTML content is a lot more difficult, mostly because the selection isn't always at a clean HTML boundary in the document (what if you only select a small part of a long link, or a few cells of a table for example). The most direct way to get all of the html associated with a selection is to reference commonAncestorContainer, which is a property on a selection range that corresponds with the deepest node which contains both the start and end of the selection. To get this, you'd do something like:
var selection = window.getSelection();
// Only works with a single range - add extra logic to
// iterate over more ranges if needed
var range = selection.getRangeAt(0);
var container = range.commonAncestorContainer;
var html = container.innerHTML
Of course, this will likely contain a lot of HTML that wasn't actually selected. It's possible that you could iterate through the children of the common ancestor and prune out anything that wasn't in the selection, but that's going to be a bit more involved and may not be necessary depending on what you're trying to do.
To show how to wrap this all up into an extension, I've written a short sample which you can reference:
http://github.com/kurrik/chrome-extensions/tree/master/contentscript-selection/
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…