diff --git a/src/librustc_expand/mbe/macro_parser.rs b/src/librustc_expand/mbe/macro_parser.rs
index 5bf7602ea6e8f..6599e92222c75 100644
--- a/src/librustc_expand/mbe/macro_parser.rs
+++ b/src/librustc_expand/mbe/macro_parser.rs
@@ -856,8 +856,6 @@ fn parse_nt(p: &mut Parser<'_>, sp: Span, name: Symbol) -> Nonterminal {
     if name == sym::tt {
         return token::NtTT(p.parse_token_tree());
     }
-    // check at the beginning and the parser checks after each bump
-    p.process_potential_macro_variable();
     match parse_nt_inner(p, sp, name) {
         Ok(nt) => nt,
         Err(mut err) => {
diff --git a/src/librustc_expand/mbe/macro_rules.rs b/src/librustc_expand/mbe/macro_rules.rs
index f3c827b1816a8..52e581e30f537 100644
--- a/src/librustc_expand/mbe/macro_rules.rs
+++ b/src/librustc_expand/mbe/macro_rules.rs
@@ -267,7 +267,6 @@ fn generic_extension<'cx>(
                     cx.current_expansion.module.mod_path.last().map(|id| id.to_string());
                 p.last_type_ascription = cx.current_expansion.prior_type_ascription;
 
-                p.process_potential_macro_variable();
                 // Let the context choose how to interpret the result.
                 // Weird, but useful for X-macros.
                 return Box::new(ParserAnyMacro {
diff --git a/src/librustc_parse/lib.rs b/src/librustc_parse/lib.rs
index 4aad2c0f68a29..a0b8415b3e17e 100644
--- a/src/librustc_parse/lib.rs
+++ b/src/librustc_parse/lib.rs
@@ -9,7 +9,7 @@ use rustc_errors::{Diagnostic, FatalError, Level, PResult};
 use rustc_session::parse::ParseSess;
 use rustc_span::{FileName, SourceFile, Span};
 use syntax::ast;
-use syntax::token::{self, Nonterminal};
+use syntax::token::{self, Nonterminal, Token};
 use syntax::tokenstream::{self, TokenStream, TokenTree};
 
 use std::path::{Path, PathBuf};
@@ -170,8 +170,9 @@ fn maybe_source_file_to_parser(
     let (stream, unclosed_delims) = maybe_file_to_stream(sess, source_file, None)?;
     let mut parser = stream_to_parser(sess, stream, None);
     parser.unclosed_delims = unclosed_delims;
-    if parser.token == token::Eof && parser.token.span.is_dummy() {
-        parser.token.span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
+    if parser.token == token::Eof {
+        let span = Span::new(end_pos, end_pos, parser.token.span.ctxt());
+        parser.set_token(Token::new(token::Eof, span));
     }
 
     Ok(parser)
diff --git a/src/librustc_parse/parser/expr.rs b/src/librustc_parse/parser/expr.rs
index 51822ab2ea5a1..97daa91eed196 100644
--- a/src/librustc_parse/parser/expr.rs
+++ b/src/librustc_parse/parser/expr.rs
@@ -166,7 +166,7 @@ impl<'a> Parser<'a> {
         while let Some(op) = self.check_assoc_op() {
             // Adjust the span for interpolated LHS to point to the `$lhs` token
             // and not to what it refers to.
-            let lhs_span = match self.unnormalized_prev_token().kind {
+            let lhs_span = match self.unnormalized_prev_token.kind {
                 TokenKind::Interpolated(..) => self.prev_span,
                 _ => lhs.span,
             };
@@ -527,7 +527,7 @@ impl<'a> Parser<'a> {
     ) -> PResult<'a, (Span, P<Expr>)> {
         expr.map(|e| {
             (
-                match self.unnormalized_prev_token().kind {
+                match self.unnormalized_prev_token.kind {
                     TokenKind::Interpolated(..) => self.prev_span,
                     _ => e.span,
                 },
diff --git a/src/librustc_parse/parser/item.rs b/src/librustc_parse/parser/item.rs
index d7b8d9778f0d2..5dc50a0cf2fde 100644
--- a/src/librustc_parse/parser/item.rs
+++ b/src/librustc_parse/parser/item.rs
@@ -1400,8 +1400,9 @@ impl<'a> Parser<'a> {
     }
 
     fn report_invalid_macro_expansion_item(&self, args: &MacArgs) {
+        let span = args.span().expect("undelimited macro call");
         let mut err = self.struct_span_err(
-            self.prev_span,
+            span,
             "macros that expand to items must be delimited with braces or followed by a semicolon",
         );
         if self.unclosed_delims.is_empty() {
@@ -1416,14 +1417,14 @@ impl<'a> Parser<'a> {
             );
         } else {
             err.span_suggestion(
-                self.prev_span,
+                span,
                 "change the delimiters to curly braces",
                 " { /* items */ }".to_string(),
                 Applicability::HasPlaceholders,
             );
         }
         err.span_suggestion(
-            self.prev_span.shrink_to_hi(),
+            span.shrink_to_hi(),
             "add a semicolon",
             ';'.to_string(),
             Applicability::MaybeIncorrect,
diff --git a/src/librustc_parse/parser/mod.rs b/src/librustc_parse/parser/mod.rs
index 79944dc35e523..937e5e3cd695b 100644
--- a/src/librustc_parse/parser/mod.rs
+++ b/src/librustc_parse/parser/mod.rs
@@ -93,18 +93,16 @@ pub struct Parser<'a> {
     /// Use span from this token if you need an isolated span.
     pub token: Token,
     /// The current non-normalized token if it's different from `token`.
-    /// Preferable use is through the `unnormalized_token()` getter.
     /// Use span from this token if you need to concatenate it with some neighbouring spans.
-    unnormalized_token: Option<Token>,
+    unnormalized_token: Token,
     /// The previous normalized token.
     /// Use span from this token if you need an isolated span.
     prev_token: Token,
     /// The previous non-normalized token if it's different from `prev_token`.
-    /// Preferable use is through the `unnormalized_prev_token()` getter.
     /// Use span from this token if you need to concatenate it with some neighbouring spans.
-    unnormalized_prev_token: Option<Token>,
-    /// Equivalent to `unnormalized_prev_token().span`.
-    /// FIXME: Remove in favor of `(unnormalized_)prev_token().span`.
+    unnormalized_prev_token: Token,
+    /// Equivalent to `unnormalized_prev_token.span`.
+    /// FIXME: Remove in favor of `(unnormalized_)prev_token.span`.
     pub prev_span: Span,
     restrictions: Restrictions,
     /// Used to determine the path to externally loaded source files.
@@ -378,9 +376,9 @@ impl<'a> Parser<'a> {
         let mut parser = Parser {
             sess,
             token: Token::dummy(),
-            unnormalized_token: None,
+            unnormalized_token: Token::dummy(),
             prev_token: Token::dummy(),
-            unnormalized_prev_token: None,
+            unnormalized_prev_token: Token::dummy(),
             prev_span: DUMMY_SP,
             restrictions: Restrictions::empty(),
             recurse_into_file_modules,
@@ -404,7 +402,8 @@ impl<'a> Parser<'a> {
             subparser_name,
         };
 
-        parser.token = parser.next_tok();
+        // Make parser point to the first token.
+        parser.bump();
 
         if let Some(directory) = directory {
             parser.directory = directory;
@@ -418,19 +417,10 @@ impl<'a> Parser<'a> {
             }
         }
 
-        parser.process_potential_macro_variable();
         parser
     }
 
-    fn unnormalized_token(&self) -> &Token {
-        self.unnormalized_token.as_ref().unwrap_or(&self.token)
-    }
-
-    fn unnormalized_prev_token(&self) -> &Token {
-        self.unnormalized_prev_token.as_ref().unwrap_or(&self.prev_token)
-    }
-
-    fn next_tok(&mut self) -> Token {
+    fn next_tok(&mut self, fallback_span: Span) -> Token {
         let mut next = if self.desugar_doc_comments {
             self.token_cursor.next_desugared()
         } else {
@@ -438,7 +428,7 @@ impl<'a> Parser<'a> {
         };
         if next.span.is_dummy() {
             // Tweak the location for better diagnostics, but keep syntactic context intact.
-            next.span = self.unnormalized_token().span.with_ctxt(next.span.ctxt());
+            next.span = fallback_span.with_ctxt(next.span.ctxt());
         }
         next
     }
@@ -896,6 +886,23 @@ impl<'a> Parser<'a> {
         self.parse_delim_comma_seq(token::Paren, f)
     }
 
+    // Interpolated identifier (`$i: ident`) and lifetime (`$l: lifetime`)
+    // tokens are replaced with usual identifier and lifetime tokens,
+    // so the former are never encountered during normal parsing.
+    crate fn set_token(&mut self, token: Token) {
+        self.unnormalized_token = token;
+        self.token = match &self.unnormalized_token.kind {
+            token::Interpolated(nt) => match **nt {
+                token::NtIdent(ident, is_raw) => {
+                    Token::new(token::Ident(ident.name, is_raw), ident.span)
+                }
+                token::NtLifetime(ident) => Token::new(token::Lifetime(ident.name), ident.span),
+                _ => self.unnormalized_token.clone(),
+            },
+            _ => self.unnormalized_token.clone(),
+        }
+    }
+
     /// Advance the parser by one token.
     pub fn bump(&mut self) {
         if self.prev_token.kind == TokenKind::Eof {
@@ -905,16 +912,15 @@ impl<'a> Parser<'a> {
         }
 
         // Update the current and previous tokens.
-        let next_token = self.next_tok();
-        self.prev_token = mem::replace(&mut self.token, next_token);
+        self.prev_token = self.token.take();
         self.unnormalized_prev_token = self.unnormalized_token.take();
+        let next_token = self.next_tok(self.unnormalized_prev_token.span);
+        self.set_token(next_token);
 
         // Update fields derived from the previous token.
-        self.prev_span = self.unnormalized_prev_token().span;
+        self.prev_span = self.unnormalized_prev_token.span;
 
         self.expected_tokens.clear();
-        // Check after each token.
-        self.process_potential_macro_variable();
     }
 
     /// Advances the parser using provided token as a next one. Use this when
@@ -924,12 +930,12 @@ impl<'a> Parser<'a> {
     /// Correct token kinds and spans need to be calculated instead.
     fn bump_with(&mut self, next: TokenKind, span: Span) {
         // Update the current and previous tokens.
-        let next_token = Token::new(next, span);
-        self.prev_token = mem::replace(&mut self.token, next_token);
+        self.prev_token = self.token.take();
         self.unnormalized_prev_token = self.unnormalized_token.take();
+        self.set_token(Token::new(next, span));
 
         // Update fields derived from the previous token.
-        self.prev_span = self.unnormalized_prev_token().span.with_hi(span.lo());
+        self.prev_span = self.unnormalized_prev_token.span.with_hi(span.lo());
 
         self.expected_tokens.clear();
     }
@@ -1066,39 +1072,6 @@ impl<'a> Parser<'a> {
         }
     }
 
-    pub fn process_potential_macro_variable(&mut self) {
-        let normalized_token = match self.token.kind {
-            token::Dollar
-                if self.token.span.from_expansion() && self.look_ahead(1, |t| t.is_ident()) =>
-            {
-                self.bump();
-                let name = match self.token.kind {
-                    token::Ident(name, _) => name,
-                    _ => unreachable!(),
-                };
-                let span = self.prev_span.to(self.token.span);
-                self.struct_span_err(span, &format!("unknown macro variable `{}`", name))
-                    .span_label(span, "unknown macro variable")
-                    .emit();
-                self.bump();
-                return;
-            }
-            token::Interpolated(ref nt) => {
-                // Interpolated identifier and lifetime tokens are replaced with usual identifier
-                // and lifetime tokens, so the former are never encountered during normal parsing.
-                match **nt {
-                    token::NtIdent(ident, is_raw) => {
-                        Token::new(token::Ident(ident.name, is_raw), ident.span)
-                    }
-                    token::NtLifetime(ident) => Token::new(token::Lifetime(ident.name), ident.span),
-                    _ => return,
-                }
-            }
-            _ => return,
-        };
-        self.unnormalized_token = Some(mem::replace(&mut self.token, normalized_token));
-    }
-
     /// Parses a single token tree from the input.
     pub fn parse_token_tree(&mut self) -> TokenTree {
         match self.token.kind {
@@ -1107,15 +1080,14 @@ impl<'a> Parser<'a> {
                     &mut self.token_cursor.frame,
                     self.token_cursor.stack.pop().unwrap(),
                 );
-                self.token.span = frame.span.entire();
+                self.set_token(Token::new(TokenKind::CloseDelim(frame.delim), frame.span.close));
                 self.bump();
                 TokenTree::Delimited(frame.span, frame.delim, frame.tree_cursor.stream.into())
             }
             token::CloseDelim(_) | token::Eof => unreachable!(),
             _ => {
-                let token = self.token.clone();
                 self.bump();
-                TokenTree::Token(token)
+                TokenTree::Token(self.prev_token.clone())
             }
         }
     }
diff --git a/src/librustc_parse/parser/path.rs b/src/librustc_parse/parser/path.rs
index 761c06b70ee8b..18e57c6a5d49f 100644
--- a/src/librustc_parse/parser/path.rs
+++ b/src/librustc_parse/parser/path.rs
@@ -134,7 +134,7 @@ impl<'a> Parser<'a> {
             path
         });
 
-        let lo = self.unnormalized_token().span;
+        let lo = self.unnormalized_token.span;
         let mut segments = Vec::new();
         let mod_sep_ctxt = self.token.span.ctxt();
         if self.eat(&token::ModSep) {
diff --git a/src/test/ui/issues/issue-6596-1.rs b/src/test/ui/issues/issue-6596-1.rs
index 5da54451346a5..25f1d6500726e 100644
--- a/src/test/ui/issues/issue-6596-1.rs
+++ b/src/test/ui/issues/issue-6596-1.rs
@@ -1,7 +1,7 @@
 macro_rules! e {
     ($inp:ident) => (
         $nonexistent
-        //~^ ERROR unknown macro variable `nonexistent`
+        //~^ ERROR expected expression, found `$`
     );
 }
 
diff --git a/src/test/ui/issues/issue-6596-1.stderr b/src/test/ui/issues/issue-6596-1.stderr
index 4f29d8a927456..216fe6472a503 100644
--- a/src/test/ui/issues/issue-6596-1.stderr
+++ b/src/test/ui/issues/issue-6596-1.stderr
@@ -1,8 +1,8 @@
-error: unknown macro variable `nonexistent`
+error: expected expression, found `$`
   --> $DIR/issue-6596-1.rs:3:9
    |
 LL |         $nonexistent
-   |         ^^^^^^^^^^^^ unknown macro variable
+   |         ^^^^^^^^^^^^ expected expression
 ...
 LL |     e!(foo);
    |     -------- in this macro invocation
diff --git a/src/test/ui/issues/issue-6596-2.rs b/src/test/ui/issues/issue-6596-2.rs
index b19700efe5ad3..8f7c98d9a67a7 100644
--- a/src/test/ui/issues/issue-6596-2.rs
+++ b/src/test/ui/issues/issue-6596-2.rs
@@ -3,7 +3,7 @@
 macro_rules! g {
     ($inp:ident) => (
         { $inp $nonexistent }
-        //~^ ERROR unknown macro variable `nonexistent`
+        //~^ ERROR expected one of `!`, `.`, `::`, `;`, `?`, `{`, `}`, or an operator, found `$`
     );
 }
 
diff --git a/src/test/ui/issues/issue-6596-2.stderr b/src/test/ui/issues/issue-6596-2.stderr
index 4fcb0176faafd..3d13c64f762ea 100644
--- a/src/test/ui/issues/issue-6596-2.stderr
+++ b/src/test/ui/issues/issue-6596-2.stderr
@@ -1,8 +1,8 @@
-error: unknown macro variable `nonexistent`
+error: expected one of `!`, `.`, `::`, `;`, `?`, `{`, `}`, or an operator, found `$`
   --> $DIR/issue-6596-2.rs:5:16
    |
 LL |         { $inp $nonexistent }
-   |                ^^^^^^^^^^^^ unknown macro variable
+   |                ^^^^^^^^^^^^ expected one of 8 possible tokens
 ...
 LL |     g!(foo);
    |     -------- in this macro invocation
diff --git a/src/test/ui/proc-macro/auxiliary/generate-dollar-ident.rs b/src/test/ui/proc-macro/auxiliary/generate-dollar-ident.rs
new file mode 100644
index 0000000000000..c9f0664c3a3ac
--- /dev/null
+++ b/src/test/ui/proc-macro/auxiliary/generate-dollar-ident.rs
@@ -0,0 +1,17 @@
+// force-host
+// no-prefer-dynamic
+
+#![feature(proc_macro_hygiene)]
+#![feature(proc_macro_quote)]
+#![crate_type = "proc-macro"]
+
+extern crate proc_macro;
+use proc_macro::*;
+
+#[proc_macro]
+pub fn dollar_ident(input: TokenStream) -> TokenStream {
+    let black_hole = input.into_iter().next().unwrap();
+    quote! {
+        $black_hole!($$var);
+    }
+}
diff --git a/src/test/ui/proc-macro/generate-dollar-ident.rs b/src/test/ui/proc-macro/generate-dollar-ident.rs
new file mode 100644
index 0000000000000..b838be9fb9f2c
--- /dev/null
+++ b/src/test/ui/proc-macro/generate-dollar-ident.rs
@@ -0,0 +1,18 @@
+// Proc macros can generate token sequence `$ IDENT`
+// without it being recognized as an unknown macro variable.
+
+// check-pass
+// aux-build:generate-dollar-ident.rs
+
+extern crate generate_dollar_ident;
+use generate_dollar_ident::*;
+
+macro_rules! black_hole {
+    ($($tt:tt)*) => {};
+}
+
+black_hole!($var);
+
+dollar_ident!(black_hole);
+
+fn main() {}