diff --git a/compiler/rustc_middle/src/ty/query/plumbing.rs b/compiler/rustc_middle/src/ty/query/plumbing.rs
index 27bf22dac75..1a8aacc4869 100644
--- a/compiler/rustc_middle/src/ty/query/plumbing.rs
+++ b/compiler/rustc_middle/src/ty/query/plumbing.rs
@@ -40,7 +40,8 @@ impl QueryContext for TyCtxt<'tcx> {
 
     fn try_collect_active_jobs(
         &self,
-    ) -> Option<FxHashMap<QueryJobId<Self::DepKind>, QueryJobInfo<Self>>> {
+    ) -> Option<FxHashMap<QueryJobId<Self::DepKind>, QueryJobInfo<Self::DepKind, Self::Query>>>
+    {
         self.queries.try_collect_active_jobs()
     }
 
@@ -353,7 +354,7 @@ macro_rules! define_queries_inner {
             $(pub type $name<$tcx> = $V;)*
         }
 
-        $(impl<$tcx> QueryConfig<TyCtxt<$tcx>> for queries::$name<$tcx> {
+        $(impl<$tcx> QueryConfig for queries::$name<$tcx> {
             type Key = $($K)*;
             type Value = $V;
             type Stored = <
@@ -372,7 +373,7 @@ macro_rules! define_queries_inner {
             type Cache = query_storage!([$($modifiers)*][$($K)*, $V]);
 
             #[inline(always)]
-            fn query_state<'a>(tcx: TyCtxt<$tcx>) -> &'a QueryState<TyCtxt<$tcx>, Self::Cache> {
+            fn query_state<'a>(tcx: TyCtxt<$tcx>) -> &'a QueryState<crate::dep_graph::DepKind, <TyCtxt<$tcx> as QueryContext>::Query, Self::Cache> {
                 &tcx.queries.$name
             }
 
@@ -454,7 +455,7 @@ macro_rules! define_queries_inner {
             #[inline(always)]
             #[must_use]
             pub fn $name(self, key: query_helper_param_ty!($($K)*))
-                -> <queries::$name<$tcx> as QueryConfig<TyCtxt<$tcx>>>::Stored
+                -> <queries::$name<$tcx> as QueryConfig>::Stored
             {
                 self.at(DUMMY_SP).$name(key.into_query_param())
             })*
@@ -493,7 +494,7 @@ macro_rules! define_queries_inner {
             $($(#[$attr])*
             #[inline(always)]
             pub fn $name(self, key: query_helper_param_ty!($($K)*))
-                -> <queries::$name<$tcx> as QueryConfig<TyCtxt<$tcx>>>::Stored
+                -> <queries::$name<$tcx> as QueryConfig>::Stored
             {
                 get_query::<queries::$name<'_>, _>(self.tcx, self.span, key.into_query_param())
             })*
@@ -527,7 +528,8 @@ macro_rules! define_queries_struct {
             fallback_extern_providers: Box<Providers>,
 
             $($(#[$attr])*  $name: QueryState<
-                TyCtxt<$tcx>,
+                crate::dep_graph::DepKind,
+                <TyCtxt<$tcx> as QueryContext>::Query,
                 <queries::$name<$tcx> as QueryAccessors<TyCtxt<'tcx>>>::Cache,
             >,)*
         }
@@ -548,7 +550,7 @@ macro_rules! define_queries_struct {
 
             pub(crate) fn try_collect_active_jobs(
                 &self
-            ) -> Option<FxHashMap<QueryJobId<crate::dep_graph::DepKind>, QueryJobInfo<TyCtxt<'tcx>>>> {
+            ) -> Option<FxHashMap<QueryJobId<crate::dep_graph::DepKind>, QueryJobInfo<crate::dep_graph::DepKind, <TyCtxt<$tcx> as QueryContext>::Query>>> {
                 let mut jobs = FxHashMap::default();
 
                 $(
diff --git a/compiler/rustc_middle/src/ty/query/profiling_support.rs b/compiler/rustc_middle/src/ty/query/profiling_support.rs
index 4e8db3194bd..cbcecb88491 100644
--- a/compiler/rustc_middle/src/ty/query/profiling_support.rs
+++ b/compiler/rustc_middle/src/ty/query/profiling_support.rs
@@ -5,8 +5,7 @@ use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::profiling::SelfProfiler;
 use rustc_hir::def_id::{CrateNum, DefId, DefIndex, LocalDefId, CRATE_DEF_INDEX, LOCAL_CRATE};
 use rustc_hir::definitions::DefPathData;
-use rustc_query_system::query::QueryCache;
-use rustc_query_system::query::QueryState;
+use rustc_query_system::query::{QueryCache, QueryContext, QueryState};
 use std::fmt::Debug;
 use std::io::Write;
 
@@ -231,7 +230,7 @@ where
 pub(super) fn alloc_self_profile_query_strings_for_query_cache<'tcx, C>(
     tcx: TyCtxt<'tcx>,
     query_name: &'static str,
-    query_state: &QueryState<TyCtxt<'tcx>, C>,
+    query_state: &QueryState<crate::dep_graph::DepKind, <TyCtxt<'tcx> as QueryContext>::Query, C>,
     string_cache: &mut QueryKeyStringCache,
 ) where
     C: QueryCache,
diff --git a/compiler/rustc_middle/src/ty/query/stats.rs b/compiler/rustc_middle/src/ty/query/stats.rs
index b496bf839ab..877f88d380a 100644
--- a/compiler/rustc_middle/src/ty/query/stats.rs
+++ b/compiler/rustc_middle/src/ty/query/stats.rs
@@ -1,11 +1,10 @@
 use crate::ty::query::queries;
 use crate::ty::TyCtxt;
 use rustc_hir::def_id::{DefId, LOCAL_CRATE};
-use rustc_query_system::query::QueryCache;
-use rustc_query_system::query::QueryState;
-use rustc_query_system::query::{QueryAccessors, QueryContext};
+use rustc_query_system::query::{QueryAccessors, QueryCache, QueryContext, QueryState};
 
 use std::any::type_name;
+use std::hash::Hash;
 use std::mem;
 #[cfg(debug_assertions)]
 use std::sync::atomic::Ordering;
@@ -38,10 +37,12 @@ struct QueryStats {
     local_def_id_keys: Option<usize>,
 }
 
-fn stats<CTX: QueryContext, C: QueryCache>(
-    name: &'static str,
-    map: &QueryState<CTX, C>,
-) -> QueryStats {
+fn stats<D, Q, C>(name: &'static str, map: &QueryState<D, Q, C>) -> QueryStats
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+    C: QueryCache,
+{
     let mut stats = QueryStats {
         name,
         #[cfg(debug_assertions)]
@@ -127,7 +128,8 @@ macro_rules! print_stats {
 
             $($(
                 queries.push(stats::<
-                    TyCtxt<'_>,
+                    crate::dep_graph::DepKind,
+                    <TyCtxt<'_> as QueryContext>::Query,
                     <queries::$name<'_> as QueryAccessors<TyCtxt<'_>>>::Cache,
                 >(
                     stringify!($name),
diff --git a/compiler/rustc_query_system/src/query/caches.rs b/compiler/rustc_query_system/src/query/caches.rs
index 1839e1af45e..7bc6ae1d1c6 100644
--- a/compiler/rustc_query_system/src/query/caches.rs
+++ b/compiler/rustc_query_system/src/query/caches.rs
@@ -1,12 +1,12 @@
 use crate::dep_graph::DepNodeIndex;
 use crate::query::plumbing::{QueryLookup, QueryState};
-use crate::query::QueryContext;
 
 use rustc_arena::TypedArena;
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::sharded::Sharded;
 use rustc_data_structures::sync::WorkerLocal;
 use std::default::Default;
+use std::fmt::Debug;
 use std::hash::Hash;
 use std::marker::PhantomData;
 
@@ -24,16 +24,16 @@ pub trait QueryStorage: Default {
 }
 
 pub trait QueryCache: QueryStorage {
-    type Key: Hash;
+    type Key: Hash + Eq + Clone + Debug;
     type Sharded: Default;
 
     /// Checks if the query is already computed and in the cache.
     /// It returns the shard index and a lock guard to the shard,
     /// which will be used if the query is not in the cache and we need
     /// to compute it.
-    fn lookup<CTX: QueryContext, R, OnHit, OnMiss>(
+    fn lookup<D, Q, R, OnHit, OnMiss>(
         &self,
-        state: &QueryState<CTX, Self>,
+        state: &QueryState<D, Q, Self>,
         key: Self::Key,
         // `on_hit` can be called while holding a lock to the query state shard.
         on_hit: OnHit,
@@ -41,7 +41,7 @@ pub trait QueryCache: QueryStorage {
     ) -> R
     where
         OnHit: FnOnce(&Self::Stored, DepNodeIndex) -> R,
-        OnMiss: FnOnce(Self::Key, QueryLookup<'_, CTX, Self::Key, Self::Sharded>) -> R;
+        OnMiss: FnOnce(Self::Key, QueryLookup<'_, D, Q, Self::Key, Self::Sharded>) -> R;
 
     fn complete(
         &self,
@@ -86,21 +86,25 @@ impl<K: Eq + Hash, V: Clone> QueryStorage for DefaultCache<K, V> {
     }
 }
 
-impl<K: Eq + Hash, V: Clone> QueryCache for DefaultCache<K, V> {
+impl<K, V> QueryCache for DefaultCache<K, V>
+where
+    K: Eq + Hash + Clone + Debug,
+    V: Clone,
+{
     type Key = K;
     type Sharded = FxHashMap<K, (V, DepNodeIndex)>;
 
     #[inline(always)]
-    fn lookup<CTX: QueryContext, R, OnHit, OnMiss>(
+    fn lookup<D, Q, R, OnHit, OnMiss>(
         &self,
-        state: &QueryState<CTX, Self>,
+        state: &QueryState<D, Q, Self>,
         key: K,
         on_hit: OnHit,
         on_miss: OnMiss,
     ) -> R
     where
         OnHit: FnOnce(&V, DepNodeIndex) -> R,
-        OnMiss: FnOnce(K, QueryLookup<'_, CTX, K, Self::Sharded>) -> R,
+        OnMiss: FnOnce(K, QueryLookup<'_, D, Q, K, Self::Sharded>) -> R,
     {
         let mut lookup = state.get_lookup(&key);
         let lock = &mut *lookup.lock;
@@ -164,21 +168,24 @@ impl<'tcx, K: Eq + Hash, V: 'tcx> QueryStorage for ArenaCache<'tcx, K, V> {
     }
 }
 
-impl<'tcx, K: Eq + Hash, V: 'tcx> QueryCache for ArenaCache<'tcx, K, V> {
+impl<'tcx, K, V: 'tcx> QueryCache for ArenaCache<'tcx, K, V>
+where
+    K: Eq + Hash + Clone + Debug,
+{
     type Key = K;
     type Sharded = FxHashMap<K, &'tcx (V, DepNodeIndex)>;
 
     #[inline(always)]
-    fn lookup<CTX: QueryContext, R, OnHit, OnMiss>(
+    fn lookup<D, Q, R, OnHit, OnMiss>(
         &self,
-        state: &QueryState<CTX, Self>,
+        state: &QueryState<D, Q, Self>,
         key: K,
         on_hit: OnHit,
         on_miss: OnMiss,
     ) -> R
     where
         OnHit: FnOnce(&&'tcx V, DepNodeIndex) -> R,
-        OnMiss: FnOnce(K, QueryLookup<'_, CTX, K, Self::Sharded>) -> R,
+        OnMiss: FnOnce(K, QueryLookup<'_, D, Q, K, Self::Sharded>) -> R,
     {
         let mut lookup = state.get_lookup(&key);
         let lock = &mut *lookup.lock;
diff --git a/compiler/rustc_query_system/src/query/config.rs b/compiler/rustc_query_system/src/query/config.rs
index 549056570f9..6c9849e8708 100644
--- a/compiler/rustc_query_system/src/query/config.rs
+++ b/compiler/rustc_query_system/src/query/config.rs
@@ -12,9 +12,7 @@ use std::borrow::Cow;
 use std::fmt::Debug;
 use std::hash::Hash;
 
-// The parameter `CTX` is required in librustc_middle:
-// implementations may need to access the `'tcx` lifetime in `CTX = TyCtxt<'tcx>`.
-pub trait QueryConfig<CTX> {
+pub trait QueryConfig {
     const NAME: &'static str;
     const CATEGORY: ProfileCategory;
 
@@ -70,7 +68,7 @@ impl<CTX: QueryContext, K, V> QueryVtable<CTX, K, V> {
     }
 }
 
-pub trait QueryAccessors<CTX: QueryContext>: QueryConfig<CTX> {
+pub trait QueryAccessors<CTX: QueryContext>: QueryConfig {
     const ANON: bool;
     const EVAL_ALWAYS: bool;
     const DEP_KIND: CTX::DepKind;
@@ -78,7 +76,7 @@ pub trait QueryAccessors<CTX: QueryContext>: QueryConfig<CTX> {
     type Cache: QueryCache<Key = Self::Key, Stored = Self::Stored, Value = Self::Value>;
 
     // Don't use this method to access query results, instead use the methods on TyCtxt
-    fn query_state<'a>(tcx: CTX) -> &'a QueryState<CTX, Self::Cache>;
+    fn query_state<'a>(tcx: CTX) -> &'a QueryState<CTX::DepKind, CTX::Query, Self::Cache>;
 
     fn to_dep_node(tcx: CTX, key: &Self::Key) -> DepNode<CTX::DepKind>
     where
diff --git a/compiler/rustc_query_system/src/query/job.rs b/compiler/rustc_query_system/src/query/job.rs
index 190312bb330..c1d3210b617 100644
--- a/compiler/rustc_query_system/src/query/job.rs
+++ b/compiler/rustc_query_system/src/query/job.rs
@@ -1,16 +1,16 @@
-use crate::dep_graph::{DepContext, DepKind};
 use crate::query::plumbing::CycleError;
-use crate::query::QueryContext;
 
 use rustc_data_structures::fx::FxHashMap;
 use rustc_span::Span;
 
 use std::convert::TryFrom;
+use std::hash::Hash;
 use std::marker::PhantomData;
 use std::num::NonZeroU32;
 
 #[cfg(parallel_compiler)]
 use {
+    super::QueryContext,
     parking_lot::{Condvar, Mutex},
     rustc_data_structures::fx::FxHashSet,
     rustc_data_structures::stable_hasher::{HashStable, StableHasher},
@@ -31,7 +31,7 @@ pub struct QueryInfo<Q> {
     pub query: Q,
 }
 
-type QueryMap<CTX> = FxHashMap<QueryJobId<<CTX as DepContext>::DepKind>, QueryJobInfo<CTX>>;
+pub(crate) type QueryMap<D, Q> = FxHashMap<QueryJobId<D>, QueryJobInfo<D, Q>>;
 
 /// A value uniquely identifiying an active query job within a shard in the query cache.
 #[derive(Copy, Clone, Eq, PartialEq, Hash)]
@@ -39,71 +39,75 @@ pub struct QueryShardJobId(pub NonZeroU32);
 
 /// A value uniquely identifiying an active query job.
 #[derive(Copy, Clone, Eq, PartialEq, Hash)]
-pub struct QueryJobId<K> {
+pub struct QueryJobId<D> {
     /// Which job within a shard is this
     pub job: QueryShardJobId,
 
     /// In which shard is this job
     pub shard: u16,
 
-    /// What kind of query this job is
-    pub kind: K,
+    /// What kind of query this job is.
+    pub kind: D,
 }
 
-impl<K: DepKind> QueryJobId<K> {
-    pub fn new(job: QueryShardJobId, shard: usize, kind: K) -> Self {
+impl<D> QueryJobId<D>
+where
+    D: Copy + Clone + Eq + Hash,
+{
+    pub fn new(job: QueryShardJobId, shard: usize, kind: D) -> Self {
         QueryJobId { job, shard: u16::try_from(shard).unwrap(), kind }
     }
 
-    fn query<CTX: QueryContext<DepKind = K>>(self, map: &QueryMap<CTX>) -> CTX::Query {
+    fn query<Q: Clone>(self, map: &QueryMap<D, Q>) -> Q {
         map.get(&self).unwrap().info.query.clone()
     }
 
     #[cfg(parallel_compiler)]
-    fn span<CTX: QueryContext<DepKind = K>>(self, map: &QueryMap<CTX>) -> Span {
+    fn span<Q: Clone>(self, map: &QueryMap<D, Q>) -> Span {
         map.get(&self).unwrap().job.span
     }
 
     #[cfg(parallel_compiler)]
-    fn parent<CTX: QueryContext<DepKind = K>>(self, map: &QueryMap<CTX>) -> Option<QueryJobId<K>> {
+    fn parent<Q: Clone>(self, map: &QueryMap<D, Q>) -> Option<QueryJobId<D>> {
         map.get(&self).unwrap().job.parent
     }
 
     #[cfg(parallel_compiler)]
-    fn latch<'a, CTX: QueryContext<DepKind = K>>(
-        self,
-        map: &'a QueryMap<CTX>,
-    ) -> Option<&'a QueryLatch<CTX>> {
+    fn latch<'a, Q: Clone>(self, map: &'a QueryMap<D, Q>) -> Option<&'a QueryLatch<D, Q>> {
         map.get(&self).unwrap().job.latch.as_ref()
     }
 }
 
-pub struct QueryJobInfo<CTX: QueryContext> {
-    pub info: QueryInfo<CTX::Query>,
-    pub job: QueryJob<CTX>,
+pub struct QueryJobInfo<D, Q> {
+    pub info: QueryInfo<Q>,
+    pub job: QueryJob<D, Q>,
 }
 
 /// Represents an active query job.
 #[derive(Clone)]
-pub struct QueryJob<CTX: QueryContext> {
+pub struct QueryJob<D, Q> {
     pub id: QueryShardJobId,
 
     /// The span corresponding to the reason for which this query was required.
     pub span: Span,
 
     /// The parent query job which created this job and is implicitly waiting on it.
-    pub parent: Option<QueryJobId<CTX::DepKind>>,
+    pub parent: Option<QueryJobId<D>>,
 
     /// The latch that is used to wait on this job.
     #[cfg(parallel_compiler)]
-    latch: Option<QueryLatch<CTX>>,
+    latch: Option<QueryLatch<D, Q>>,
 
-    dummy: PhantomData<QueryLatch<CTX>>,
+    dummy: PhantomData<QueryLatch<D, Q>>,
 }
 
-impl<CTX: QueryContext> QueryJob<CTX> {
+impl<D, Q> QueryJob<D, Q>
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+{
     /// Creates a new query job.
-    pub fn new(id: QueryShardJobId, span: Span, parent: Option<QueryJobId<CTX::DepKind>>) -> Self {
+    pub fn new(id: QueryShardJobId, span: Span, parent: Option<QueryJobId<D>>) -> Self {
         QueryJob {
             id,
             span,
@@ -115,7 +119,7 @@ impl<CTX: QueryContext> QueryJob<CTX> {
     }
 
     #[cfg(parallel_compiler)]
-    pub(super) fn latch(&mut self, _id: QueryJobId<CTX::DepKind>) -> QueryLatch<CTX> {
+    pub(super) fn latch(&mut self, _id: QueryJobId<D>) -> QueryLatch<D, Q> {
         if self.latch.is_none() {
             self.latch = Some(QueryLatch::new());
         }
@@ -123,7 +127,7 @@ impl<CTX: QueryContext> QueryJob<CTX> {
     }
 
     #[cfg(not(parallel_compiler))]
-    pub(super) fn latch(&mut self, id: QueryJobId<CTX::DepKind>) -> QueryLatch<CTX> {
+    pub(super) fn latch(&mut self, id: QueryJobId<D>) -> QueryLatch<D, Q> {
         QueryLatch { id, dummy: PhantomData }
     }
 
@@ -143,19 +147,26 @@ impl<CTX: QueryContext> QueryJob<CTX> {
 
 #[cfg(not(parallel_compiler))]
 #[derive(Clone)]
-pub(super) struct QueryLatch<CTX: QueryContext> {
-    id: QueryJobId<CTX::DepKind>,
-    dummy: PhantomData<CTX>,
+pub(super) struct QueryLatch<D, Q> {
+    id: QueryJobId<D>,
+    dummy: PhantomData<Q>,
 }
 
 #[cfg(not(parallel_compiler))]
-impl<CTX: QueryContext> QueryLatch<CTX> {
-    pub(super) fn find_cycle_in_stack(&self, tcx: CTX, span: Span) -> CycleError<CTX::Query> {
-        let query_map = tcx.try_collect_active_jobs().unwrap();
-
-        // Get the current executing query (waiter) and find the waitee amongst its parents
-        let mut current_job = tcx.current_query_job();
+impl<D, Q> QueryLatch<D, Q>
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+{
+    pub(super) fn find_cycle_in_stack(
+        &self,
+        query_map: QueryMap<D, Q>,
+        current_job: &Option<QueryJobId<D>>,
+        span: Span,
+    ) -> CycleError<Q> {
+        // Find the waitee amongst `current_job` parents
         let mut cycle = Vec::new();
+        let mut current_job = Option::clone(current_job);
 
         while let Some(job) = current_job {
             let info = query_map.get(&job).unwrap();
@@ -186,15 +197,15 @@ impl<CTX: QueryContext> QueryLatch<CTX> {
 }
 
 #[cfg(parallel_compiler)]
-struct QueryWaiter<CTX: QueryContext> {
-    query: Option<QueryJobId<CTX::DepKind>>,
+struct QueryWaiter<D, Q> {
+    query: Option<QueryJobId<D>>,
     condvar: Condvar,
     span: Span,
-    cycle: Lock<Option<CycleError<CTX::Query>>>,
+    cycle: Lock<Option<CycleError<Q>>>,
 }
 
 #[cfg(parallel_compiler)]
-impl<CTX: QueryContext> QueryWaiter<CTX> {
+impl<D, Q> QueryWaiter<D, Q> {
     fn notify(&self, registry: &rayon_core::Registry) {
         rayon_core::mark_unblocked(registry);
         self.condvar.notify_one();
@@ -202,19 +213,19 @@ impl<CTX: QueryContext> QueryWaiter<CTX> {
 }
 
 #[cfg(parallel_compiler)]
-struct QueryLatchInfo<CTX: QueryContext> {
+struct QueryLatchInfo<D, Q> {
     complete: bool,
-    waiters: Vec<Lrc<QueryWaiter<CTX>>>,
+    waiters: Vec<Lrc<QueryWaiter<D, Q>>>,
 }
 
 #[cfg(parallel_compiler)]
 #[derive(Clone)]
-pub(super) struct QueryLatch<CTX: QueryContext> {
-    info: Lrc<Mutex<QueryLatchInfo<CTX>>>,
+pub(super) struct QueryLatch<D, Q> {
+    info: Lrc<Mutex<QueryLatchInfo<D, Q>>>,
 }
 
 #[cfg(parallel_compiler)]
-impl<CTX: QueryContext> QueryLatch<CTX> {
+impl<D: Eq + Hash, Q: Clone> QueryLatch<D, Q> {
     fn new() -> Self {
         QueryLatch {
             info: Lrc::new(Mutex::new(QueryLatchInfo { complete: false, waiters: Vec::new() })),
@@ -223,10 +234,13 @@ impl<CTX: QueryContext> QueryLatch<CTX> {
 }
 
 #[cfg(parallel_compiler)]
-impl<CTX: QueryContext> QueryLatch<CTX> {
+impl<D, Q> QueryLatch<D, Q> {
     /// Awaits for the query job to complete.
-    pub(super) fn wait_on(&self, tcx: CTX, span: Span) -> Result<(), CycleError<CTX::Query>> {
-        let query = tcx.current_query_job();
+    pub(super) fn wait_on(
+        &self,
+        query: Option<QueryJobId<D>>,
+        span: Span,
+    ) -> Result<(), CycleError<Q>> {
         let waiter =
             Lrc::new(QueryWaiter { query, span, cycle: Lock::new(None), condvar: Condvar::new() });
         self.wait_on_inner(&waiter);
@@ -239,12 +253,9 @@ impl<CTX: QueryContext> QueryLatch<CTX> {
             Some(cycle) => Err(cycle),
         }
     }
-}
 
-#[cfg(parallel_compiler)]
-impl<CTX: QueryContext> QueryLatch<CTX> {
     /// Awaits the caller on this latch by blocking the current thread.
-    fn wait_on_inner(&self, waiter: &Lrc<QueryWaiter<CTX>>) {
+    fn wait_on_inner(&self, waiter: &Lrc<QueryWaiter<D, Q>>) {
         let mut info = self.info.lock();
         if !info.complete {
             // We push the waiter on to the `waiters` list. It can be accessed inside
@@ -278,7 +289,7 @@ impl<CTX: QueryContext> QueryLatch<CTX> {
 
     /// Removes a single waiter from the list of waiters.
     /// This is used to break query cycles.
-    fn extract_waiter(&self, waiter: usize) -> Lrc<QueryWaiter<CTX>> {
+    fn extract_waiter(&self, waiter: usize) -> Lrc<QueryWaiter<D, Q>> {
         let mut info = self.info.lock();
         debug_assert!(!info.complete);
         // Remove the waiter from the list of waiters
@@ -288,7 +299,7 @@ impl<CTX: QueryContext> QueryLatch<CTX> {
 
 /// A resumable waiter of a query. The usize is the index into waiters in the query's latch
 #[cfg(parallel_compiler)]
-type Waiter<K> = (QueryJobId<K>, usize);
+type Waiter<D> = (QueryJobId<D>, usize);
 
 /// Visits all the non-resumable and resumable waiters of a query.
 /// Only waiters in a query are visited.
@@ -300,13 +311,15 @@ type Waiter<K> = (QueryJobId<K>, usize);
 /// required information to resume the waiter.
 /// If all `visit` calls returns None, this function also returns None.
 #[cfg(parallel_compiler)]
-fn visit_waiters<CTX: QueryContext, F>(
-    query_map: &QueryMap<CTX>,
-    query: QueryJobId<CTX::DepKind>,
+fn visit_waiters<D, Q, F>(
+    query_map: &QueryMap<D, Q>,
+    query: QueryJobId<D>,
     mut visit: F,
-) -> Option<Option<Waiter<CTX::DepKind>>>
+) -> Option<Option<Waiter<D>>>
 where
-    F: FnMut(Span, QueryJobId<CTX::DepKind>) -> Option<Option<Waiter<CTX::DepKind>>>,
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+    F: FnMut(Span, QueryJobId<D>) -> Option<Option<Waiter<D>>>,
 {
     // Visit the parent query which is a non-resumable waiter since it's on the same stack
     if let Some(parent) = query.parent(query_map) {
@@ -335,13 +348,17 @@ where
 /// If a cycle is detected, this initial value is replaced with the span causing
 /// the cycle.
 #[cfg(parallel_compiler)]
-fn cycle_check<CTX: QueryContext>(
-    query_map: &QueryMap<CTX>,
-    query: QueryJobId<CTX::DepKind>,
+fn cycle_check<D, Q>(
+    query_map: &QueryMap<D, Q>,
+    query: QueryJobId<D>,
     span: Span,
-    stack: &mut Vec<(Span, QueryJobId<CTX::DepKind>)>,
-    visited: &mut FxHashSet<QueryJobId<CTX::DepKind>>,
-) -> Option<Option<Waiter<CTX::DepKind>>> {
+    stack: &mut Vec<(Span, QueryJobId<D>)>,
+    visited: &mut FxHashSet<QueryJobId<D>>,
+) -> Option<Option<Waiter<D>>>
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+{
     if !visited.insert(query) {
         return if let Some(p) = stack.iter().position(|q| q.1 == query) {
             // We detected a query cycle, fix up the initial span and return Some
@@ -376,11 +393,15 @@ fn cycle_check<CTX: QueryContext>(
 /// from `query` without going through any of the queries in `visited`.
 /// This is achieved with a depth first search.
 #[cfg(parallel_compiler)]
-fn connected_to_root<CTX: QueryContext>(
-    query_map: &QueryMap<CTX>,
-    query: QueryJobId<CTX::DepKind>,
-    visited: &mut FxHashSet<QueryJobId<CTX::DepKind>>,
-) -> bool {
+fn connected_to_root<D, Q>(
+    query_map: &QueryMap<D, Q>,
+    query: QueryJobId<D>,
+    visited: &mut FxHashSet<QueryJobId<D>>,
+) -> bool
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+{
     // We already visited this or we're deliberately ignoring it
     if !visited.insert(query) {
         return false;
@@ -399,7 +420,12 @@ fn connected_to_root<CTX: QueryContext>(
 
 // Deterministically pick an query from a list
 #[cfg(parallel_compiler)]
-fn pick_query<'a, CTX, T, F>(query_map: &QueryMap<CTX>, tcx: CTX, queries: &'a [T], f: F) -> &'a T
+fn pick_query<'a, CTX, T, F>(
+    query_map: &QueryMap<CTX::DepKind, CTX::Query>,
+    tcx: CTX,
+    queries: &'a [T],
+    f: F,
+) -> &'a T
 where
     CTX: QueryContext,
     F: Fn(&T) -> (Span, QueryJobId<CTX::DepKind>),
@@ -429,9 +455,9 @@ where
 /// the function returns false.
 #[cfg(parallel_compiler)]
 fn remove_cycle<CTX: QueryContext>(
-    query_map: &QueryMap<CTX>,
+    query_map: &QueryMap<CTX::DepKind, CTX::Query>,
     jobs: &mut Vec<QueryJobId<CTX::DepKind>>,
-    wakelist: &mut Vec<Lrc<QueryWaiter<CTX>>>,
+    wakelist: &mut Vec<Lrc<QueryWaiter<CTX::DepKind, CTX::Query>>>,
     tcx: CTX,
 ) -> bool {
     let mut visited = FxHashSet::default();
diff --git a/compiler/rustc_query_system/src/query/mod.rs b/compiler/rustc_query_system/src/query/mod.rs
index 49097725bc9..da45565dbe6 100644
--- a/compiler/rustc_query_system/src/query/mod.rs
+++ b/compiler/rustc_query_system/src/query/mod.rs
@@ -15,8 +15,8 @@ mod config;
 pub use self::config::{QueryAccessors, QueryConfig, QueryDescription};
 
 use crate::dep_graph::{DepContext, DepGraph};
+use crate::query::job::QueryMap;
 
-use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::stable_hasher::HashStable;
 use rustc_data_structures::sync::Lock;
 use rustc_data_structures::thin_vec::ThinVec;
@@ -38,9 +38,7 @@ pub trait QueryContext: DepContext {
     /// Get the query information from the TLS context.
     fn current_query_job(&self) -> Option<QueryJobId<Self::DepKind>>;
 
-    fn try_collect_active_jobs(
-        &self,
-    ) -> Option<FxHashMap<QueryJobId<Self::DepKind>, QueryJobInfo<Self>>>;
+    fn try_collect_active_jobs(&self) -> Option<QueryMap<Self::DepKind, Self::Query>>;
 
     /// Executes a job by changing the `ImplicitCtxt` to point to the
     /// new query job while it executes. It returns the diagnostics
diff --git a/compiler/rustc_query_system/src/query/plumbing.rs b/compiler/rustc_query_system/src/query/plumbing.rs
index ae042cc8081..50f443716f4 100644
--- a/compiler/rustc_query_system/src/query/plumbing.rs
+++ b/compiler/rustc_query_system/src/query/plumbing.rs
@@ -7,7 +7,7 @@ use crate::dep_graph::{DepNodeIndex, SerializedDepNodeIndex};
 use crate::query::caches::QueryCache;
 use crate::query::config::{QueryDescription, QueryVtable, QueryVtableExt};
 use crate::query::job::{QueryInfo, QueryJob, QueryJobId, QueryJobInfo, QueryShardJobId};
-use crate::query::QueryContext;
+use crate::query::{QueryContext, QueryMap};
 
 #[cfg(not(parallel_compiler))]
 use rustc_data_structures::cold_path;
@@ -20,8 +20,6 @@ use rustc_errors::{Diagnostic, FatalError};
 use rustc_span::source_map::DUMMY_SP;
 use rustc_span::Span;
 use std::collections::hash_map::Entry;
-use std::convert::TryFrom;
-use std::fmt::Debug;
 use std::hash::{Hash, Hasher};
 use std::mem;
 use std::num::NonZeroU32;
@@ -29,33 +27,33 @@ use std::ptr;
 #[cfg(debug_assertions)]
 use std::sync::atomic::{AtomicUsize, Ordering};
 
-pub(super) struct QueryStateShard<CTX: QueryContext, K, C> {
+pub(super) struct QueryStateShard<D, Q, K, C> {
     pub(super) cache: C,
-    active: FxHashMap<K, QueryResult<CTX>>,
+    active: FxHashMap<K, QueryResult<D, Q>>,
 
     /// Used to generate unique ids for active jobs.
     jobs: u32,
 }
 
-impl<CTX: QueryContext, K, C: Default> Default for QueryStateShard<CTX, K, C> {
-    fn default() -> QueryStateShard<CTX, K, C> {
+impl<D, Q, K, C: Default> Default for QueryStateShard<D, Q, K, C> {
+    fn default() -> QueryStateShard<D, Q, K, C> {
         QueryStateShard { cache: Default::default(), active: Default::default(), jobs: 0 }
     }
 }
 
-pub struct QueryState<CTX: QueryContext, C: QueryCache> {
+pub struct QueryState<D, Q, C: QueryCache> {
     cache: C,
-    shards: Sharded<QueryStateShard<CTX, C::Key, C::Sharded>>,
+    shards: Sharded<QueryStateShard<D, Q, C::Key, C::Sharded>>,
     #[cfg(debug_assertions)]
     pub cache_hits: AtomicUsize,
 }
 
-impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
+impl<D, Q, C: QueryCache> QueryState<D, Q, C> {
     #[inline]
     pub(super) fn get_lookup<'tcx>(
         &'tcx self,
         key: &C::Key,
-    ) -> QueryLookup<'tcx, CTX, C::Key, C::Sharded> {
+    ) -> QueryLookup<'tcx, D, Q, C::Key, C::Sharded> {
         // We compute the key's hash once and then use it for both the
         // shard lookup and the hashmap lookup. This relies on the fact
         // that both of them use `FxHasher`.
@@ -70,16 +68,21 @@ impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
 }
 
 /// Indicates the state of a query for a given key in a query map.
-enum QueryResult<CTX: QueryContext> {
+enum QueryResult<D, Q> {
     /// An already executing query. The query job can be used to await for its completion.
-    Started(QueryJob<CTX>),
+    Started(QueryJob<D, Q>),
 
     /// The query panicked. Queries trying to wait on this will raise a fatal error which will
     /// silently panic.
     Poisoned,
 }
 
-impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
+impl<D, Q, C> QueryState<D, Q, C>
+where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+    C: QueryCache,
+{
     #[inline(always)]
     pub fn iter_results<R>(
         &self,
@@ -98,13 +101,10 @@ impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
 
     pub fn try_collect_active_jobs(
         &self,
-        kind: CTX::DepKind,
-        make_query: fn(C::Key) -> CTX::Query,
-        jobs: &mut FxHashMap<QueryJobId<CTX::DepKind>, QueryJobInfo<CTX>>,
-    ) -> Option<()>
-    where
-        C::Key: Clone,
-    {
+        kind: D,
+        make_query: fn(C::Key) -> Q,
+        jobs: &mut QueryMap<D, Q>,
+    ) -> Option<()> {
         // We use try_lock_shards here since we are called from the
         // deadlock handler, and this shouldn't be locked.
         let shards = self.shards.try_lock_shards()?;
@@ -112,8 +112,7 @@ impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
         jobs.extend(shards.flat_map(|(shard_id, shard)| {
             shard.active.iter().filter_map(move |(k, v)| {
                 if let QueryResult::Started(ref job) = *v {
-                    let id =
-                        QueryJobId { job: job.id, shard: u16::try_from(shard_id).unwrap(), kind };
+                    let id = QueryJobId::new(job.id, shard_id, kind);
                     let info = QueryInfo { span: job.span, query: make_query(k.clone()) };
                     Some((id, QueryJobInfo { info, job: job.clone() }))
                 } else {
@@ -126,8 +125,8 @@ impl<CTX: QueryContext, C: QueryCache> QueryState<CTX, C> {
     }
 }
 
-impl<CTX: QueryContext, C: QueryCache> Default for QueryState<CTX, C> {
-    fn default() -> QueryState<CTX, C> {
+impl<D, Q, C: QueryCache> Default for QueryState<D, Q, C> {
+    fn default() -> QueryState<D, Q, C> {
         QueryState {
             cache: C::default(),
             shards: Default::default(),
@@ -138,28 +137,30 @@ impl<CTX: QueryContext, C: QueryCache> Default for QueryState<CTX, C> {
 }
 
 /// Values used when checking a query cache which can be reused on a cache-miss to execute the query.
-pub struct QueryLookup<'tcx, CTX: QueryContext, K, C> {
+pub struct QueryLookup<'tcx, D, Q, K, C> {
     pub(super) key_hash: u64,
     shard: usize,
-    pub(super) lock: LockGuard<'tcx, QueryStateShard<CTX, K, C>>,
+    pub(super) lock: LockGuard<'tcx, QueryStateShard<D, Q, K, C>>,
 }
 
 /// A type representing the responsibility to execute the job in the `job` field.
 /// This will poison the relevant query if dropped.
-struct JobOwner<'tcx, CTX: QueryContext, C>
+struct JobOwner<'tcx, D, Q, C>
 where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
     C: QueryCache,
-    C::Key: Eq + Hash + Clone + Debug,
 {
-    state: &'tcx QueryState<CTX, C>,
+    state: &'tcx QueryState<D, Q, C>,
     key: C::Key,
-    id: QueryJobId<CTX::DepKind>,
+    id: QueryJobId<D>,
 }
 
-impl<'tcx, CTX: QueryContext, C> JobOwner<'tcx, CTX, C>
+impl<'tcx, D, Q, C> JobOwner<'tcx, D, Q, C>
 where
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
     C: QueryCache,
-    C::Key: Eq + Hash + Clone + Debug,
 {
     /// Either gets a `JobOwner` corresponding the query, allowing us to
     /// start executing the query, or returns with the result of the query.
@@ -170,14 +171,14 @@ where
     /// This function is inlined because that results in a noticeable speed-up
     /// for some compile-time benchmarks.
     #[inline(always)]
-    fn try_start<'a, 'b>(
+    fn try_start<'a, 'b, CTX>(
         tcx: CTX,
-        state: &'b QueryState<CTX, C>,
+        state: &'b QueryState<CTX::DepKind, CTX::Query, C>,
         span: Span,
         key: &C::Key,
-        mut lookup: QueryLookup<'a, CTX, C::Key, C::Sharded>,
+        mut lookup: QueryLookup<'a, CTX::DepKind, CTX::Query, C::Key, C::Sharded>,
         query: &QueryVtable<CTX, C::Key, C::Value>,
-    ) -> TryGetJob<'b, CTX, C>
+    ) -> TryGetJob<'b, CTX::DepKind, CTX::Query, C>
     where
         CTX: QueryContext,
     {
@@ -229,7 +230,12 @@ where
         // so we just return the error.
         #[cfg(not(parallel_compiler))]
         return TryGetJob::Cycle(cold_path(|| {
-            let value = query.handle_cycle_error(tcx, latch.find_cycle_in_stack(tcx, span));
+            let error: CycleError<CTX::Query> = latch.find_cycle_in_stack(
+                tcx.try_collect_active_jobs().unwrap(),
+                &tcx.current_query_job(),
+                span,
+            );
+            let value = query.handle_cycle_error(tcx, error);
             state.cache.store_nocache(value)
         }));
 
@@ -237,7 +243,7 @@ where
         // thread.
         #[cfg(parallel_compiler)]
         {
-            let result = latch.wait_on(tcx, span);
+            let result = latch.wait_on(tcx.current_query_job(), span);
 
             if let Err(cycle) = result {
                 let value = query.handle_cycle_error(tcx, cycle);
@@ -297,9 +303,11 @@ where
     (result, diagnostics.into_inner())
 }
 
-impl<'tcx, CTX: QueryContext, C: QueryCache> Drop for JobOwner<'tcx, CTX, C>
+impl<'tcx, D, Q, C> Drop for JobOwner<'tcx, D, Q, C>
 where
-    C::Key: Eq + Hash + Clone + Debug,
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+    C: QueryCache,
 {
     #[inline(never)]
     #[cold]
@@ -330,12 +338,14 @@ pub struct CycleError<Q> {
 }
 
 /// The result of `try_start`.
-enum TryGetJob<'tcx, CTX: QueryContext, C: QueryCache>
+enum TryGetJob<'tcx, D, Q, C>
 where
-    C::Key: Eq + Hash + Clone + Debug,
+    D: Copy + Clone + Eq + Hash,
+    Q: Clone,
+    C: QueryCache,
 {
     /// The query is not yet started. Contains a guard to the cache eventually used to start it.
-    NotYetStarted(JobOwner<'tcx, CTX, C>),
+    NotYetStarted(JobOwner<'tcx, D, Q, C>),
 
     /// The query was already completed.
     /// Returns the result of the query and its dep-node index
@@ -354,7 +364,7 @@ where
 #[inline(always)]
 fn try_get_cached<CTX, C, R, OnHit, OnMiss>(
     tcx: CTX,
-    state: &QueryState<CTX, C>,
+    state: &QueryState<CTX::DepKind, CTX::Query, C>,
     key: C::Key,
     // `on_hit` can be called while holding a lock to the query cache
     on_hit: OnHit,
@@ -364,7 +374,7 @@ where
     C: QueryCache,
     CTX: QueryContext,
     OnHit: FnOnce(&C::Stored, DepNodeIndex) -> R,
-    OnMiss: FnOnce(C::Key, QueryLookup<'_, CTX, C::Key, C::Sharded>) -> R,
+    OnMiss: FnOnce(C::Key, QueryLookup<'_, CTX::DepKind, CTX::Query, C::Key, C::Sharded>) -> R,
 {
     state.cache.lookup(
         state,
@@ -386,19 +396,20 @@ where
 #[inline(always)]
 fn try_execute_query<CTX, C>(
     tcx: CTX,
-    state: &QueryState<CTX, C>,
+    state: &QueryState<CTX::DepKind, CTX::Query, C>,
     span: Span,
     key: C::Key,
-    lookup: QueryLookup<'_, CTX, C::Key, C::Sharded>,
+    lookup: QueryLookup<'_, CTX::DepKind, CTX::Query, C::Key, C::Sharded>,
     query: &QueryVtable<CTX, C::Key, C::Value>,
 ) -> C::Stored
 where
     C: QueryCache,
-    C::Key: Eq + Clone + Debug + crate::dep_graph::DepNodeParams<CTX>,
-    C::Stored: Clone,
+    C::Key: crate::dep_graph::DepNodeParams<CTX>,
     CTX: QueryContext,
 {
-    let job = match JobOwner::try_start(tcx, state, span, &key, lookup, query) {
+    let job = match JobOwner::<'_, CTX::DepKind, CTX::Query, C>::try_start(
+        tcx, state, span, &key, lookup, query,
+    ) {
         TryGetJob::NotYetStarted(job) => job,
         TryGetJob::Cycle(result) => return result,
         #[cfg(parallel_compiler)]
@@ -559,14 +570,12 @@ fn incremental_verify_ich<CTX, K, V>(
 fn force_query_with_job<C, CTX>(
     tcx: CTX,
     key: C::Key,
-    job: JobOwner<'_, CTX, C>,
+    job: JobOwner<'_, CTX::DepKind, CTX::Query, C>,
     dep_node: DepNode<CTX::DepKind>,
     query: &QueryVtable<CTX, C::Key, C::Value>,
 ) -> (C::Stored, DepNodeIndex)
 where
     C: QueryCache,
-    C::Key: Eq + Clone + Debug,
-    C::Stored: Clone,
     CTX: QueryContext,
 {
     // If the following assertion triggers, it can have two reasons:
@@ -617,7 +626,7 @@ where
 #[inline(never)]
 fn get_query_impl<CTX, C>(
     tcx: CTX,
-    state: &QueryState<CTX, C>,
+    state: &QueryState<CTX::DepKind, CTX::Query, C>,
     span: Span,
     key: C::Key,
     query: &QueryVtable<CTX, C::Key, C::Value>,
@@ -625,8 +634,7 @@ fn get_query_impl<CTX, C>(
 where
     CTX: QueryContext,
     C: QueryCache,
-    C::Key: Eq + Clone + crate::dep_graph::DepNodeParams<CTX>,
-    C::Stored: Clone,
+    C::Key: crate::dep_graph::DepNodeParams<CTX>,
 {
     try_get_cached(
         tcx,
@@ -650,12 +658,12 @@ where
 #[inline(never)]
 fn ensure_query_impl<CTX, C>(
     tcx: CTX,
-    state: &QueryState<CTX, C>,
+    state: &QueryState<CTX::DepKind, CTX::Query, C>,
     key: C::Key,
     query: &QueryVtable<CTX, C::Key, C::Value>,
 ) where
     C: QueryCache,
-    C::Key: Eq + Clone + crate::dep_graph::DepNodeParams<CTX>,
+    C::Key: crate::dep_graph::DepNodeParams<CTX>,
     CTX: QueryContext,
 {
     if query.eval_always {
@@ -687,14 +695,14 @@ fn ensure_query_impl<CTX, C>(
 #[inline(never)]
 fn force_query_impl<CTX, C>(
     tcx: CTX,
-    state: &QueryState<CTX, C>,
+    state: &QueryState<CTX::DepKind, CTX::Query, C>,
     key: C::Key,
     span: Span,
     dep_node: DepNode<CTX::DepKind>,
     query: &QueryVtable<CTX, C::Key, C::Value>,
 ) where
     C: QueryCache,
-    C::Key: Eq + Clone + crate::dep_graph::DepNodeParams<CTX>,
+    C::Key: crate::dep_graph::DepNodeParams<CTX>,
     CTX: QueryContext,
 {
     // We may be concurrently trying both execute and force a query.
@@ -708,7 +716,9 @@ fn force_query_impl<CTX, C>(
             // Cache hit, do nothing
         },
         |key, lookup| {
-            let job = match JobOwner::try_start(tcx, state, span, &key, lookup, query) {
+            let job = match JobOwner::<'_, CTX::DepKind, CTX::Query, C>::try_start(
+                tcx, state, span, &key, lookup, query,
+            ) {
                 TryGetJob::NotYetStarted(job) => job,
                 TryGetJob::Cycle(_) => return,
                 #[cfg(parallel_compiler)]