[PATCH 00/11] sched/wait: Replace wait_on_atomic_t() with wait_var_event()

[PATCH 00/11] sched/wait: Replace wait_on_atomic_t() with wait_var_event()

[Peter Zijlstra]

I never really liked the whole wait_on_atomic_t() because of the hard-coded
type and condition. And when Dan proposed to add 50+ lines of duplication to
implement wait_on_atomic_one() I figured that was it.

Instead implement wait_var_event(), which at the fundamental level is much the
same hashed-waitqueue as wait_on_atomic_t() is, but it doesn't care about the
type or condition.

It is further modelled on the existing wait_event() API and is otherwise
consistent with that.

The existing:

  wait_for_atomic_t(&my_atomic, wait_atomic_t, TASK_UNINTERRUPTIBLE);
  wake_up_atomic_t(&my_atomic);

can be replaced with:

  wait_var_event(&my_atomic, !atomic_read(&my_atomic));
  wake_up_var(&my_atomic);

And Dan can do what he needs done using:

  wait_var_event(&my_atomic, atomic_read(&my_atomic) == 1);


---
 arch/mips/kernel/process.c                         |   2 +
 arch/mips/kernel/traps.c                           |   4 +-
 drivers/gpu/drm/drm_dp_aux_dev.c                   |  13 +--
 drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c |  14 +--
 drivers/media/platform/qcom/venus/hfi.c            |   8 +-
 fs/afs/cell.c                                      |   6 +-
 fs/afs/rxrpc.c                                     |   6 +-
 fs/afs/server.c                                    |   6 +-
 fs/btrfs/extent-tree.c                             |  14 ++-
 fs/btrfs/ioctl.c                                   |   2 +-
 fs/fscache/cookie.c                                |   7 +-
 fs/nfs/inode.c                                     |   5 -
 fs/nfs/pagelist.c                                  |   6 +-
 fs/nfs/pnfs_nfs.c                                  |   2 +-
 fs/nfs/write.c                                     |   6 +-
 fs/ocfs2/filecheck.c                               |   9 +-
 include/linux/fscache-cache.h                      |   2 +-
 include/linux/wait_bit.h                           |  95 +++++++++++++-----
 kernel/sched/wait_bit.c                            | 110 +++++----------------
 19 files changed, 147 insertions(+), 170 deletions(-)

[PATCH 01/11] sched/wait: Introduce wait_var_event()

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-1.patch --]
[-- Type: text/plain, Size: 5461 bytes --]

As a replacement for the wait_on_atomic_t() API provide the
wait_var_event() API.

The wait_var_event() API is based on the very same hashed-waitqueue
idea, but doesn't care about the type (atomic_t) or the specific
condition (atomic_read() == 0). IOW. it's much more widely
applicable/flexible.

It shares all the benefits/disadvantages of a hashed-waitqueue
approach with the existing wait_on_atomic_t/wait_on_bit() APIs.

The API is modeled after the existing wait_event() API, but instead of
taking a wait_queue_head, it takes an address. This addresses is
hashed to obtain a wait_queue_head from the bit_wait_table.

Similar to the wait_event() API, it takes a condition expression as
second argument and will wait until this expression becomes true.

The following are (mostly) identical replacements:

  wait_on_atomic_t(&my_atomic, atomic_t_wait, TASK_UNINTERRUPTIBLE);
  wake_up_atomic_t(&my_atomic);

  wait_var_event(&my_atomic, !atomic_read(&my_atomic));
  wake_up_var(&my_atomic);

The only difference is that wake_up_var() is an unconditional wakeup
and doesn't check the previously hard-coded (atomic_read() == 0)
condition here. This is of little concequence, since most callers are
already conditional on atomic_dec_and_test() and the ones that are
not, are trivial to make so.

Cc: David Howells <dhowells@redhat.com>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 include/linux/wait_bit.h |   70 +++++++++++++++++++++++++++++++++++++++++++++++
 kernel/sched/wait_bit.c  |   48 ++++++++++++++++++++++++++++++++
 2 files changed, 118 insertions(+)

--- a/include/linux/wait_bit.h
+++ b/include/linux/wait_bit.h
@@ -262,4 +262,74 @@ int wait_on_atomic_t(atomic_t *val, wait
 	return out_of_line_wait_on_atomic_t(val, action, mode);
 }
 
+extern void init_wait_var_entry(struct wait_bit_queue_entry *wbq_entry, void *var, int flags);
+extern void wake_up_var(void *var);
+extern wait_queue_head_t *__var_waitqueue(void *p);
+
+#define ___wait_var_event(var, condition, state, exclusive, ret, cmd)	\
+({									\
+	__label__ __out;						\
+	struct wait_queue_head *__wq_head = __var_waitqueue(var);	\
+	struct wait_bit_queue_entry __wbq_entry;			\
+	long __ret = ret; /* explicit shadow */				\
+									\
+	init_wait_var_entry(&__wbq_entry, var,				\
+			    exclusive ? WQ_FLAG_EXCLUSIVE : 0);		\
+	for (;;) {							\
+		long __int = prepare_to_wait_event(__wq_head,		\
+						   &__wbq_entry.wq_entry, \
+						   state);		\
+		if (condition)						\
+			break;						\
+									\
+		if (___wait_is_interruptible(state) && __int) {		\
+			__ret = __int;					\
+			goto __out;					\
+		}							\
+									\
+		cmd;							\
+	}								\
+	finish_wait(__wq_head, &__wbq_entry.wq_entry);			\
+__out:	__ret;								\
+})
+
+#define __wait_var_event(var, condition)				\
+	___wait_var_event(var, condition, TASK_UNINTERRUPTIBLE, 0, 0,	\
+			  schedule())
+
+#define wait_var_event(var, condition)					\
+do {									\
+	might_sleep();							\
+	if (condition)							\
+		break;							\
+	__wait_var_event(var, condition);				\
+} while (0)
+
+#define __wait_var_event_killable(var, condition)			\
+	___wait_var_event(var, condition, TASK_KILLABLE, 0, 0,		\
+			  schedule())
+
+#define wait_var_event_killable(var, condition)				\
+({									\
+	int __ret = 0;							\
+	might_sleep();							\
+	if (!(condition))						\
+		__ret = __wait_var_event_killable(var, condition);	\
+	__ret;								\
+})
+
+#define __wait_var_event_timeout(var, condition, timeout)		\
+	___wait_var_event(var, ___wait_cond_timeout(condition),		\
+			  TASK_UNINTERRUPTIBLE, 0, timeout,		\
+			  __ret = schedule_timeout(__ret))
+
+#define wait_var_event_timeout(var, condition, timeout)			\
+({									\
+	long __ret = timeout;						\
+	might_sleep();							\
+	if (!___wait_cond_timeout(condition))				\
+		__ret = __wait_var_event_timeout(var, condition, timeout); \
+	__ret;								\
+})
+
 #endif /* _LINUX_WAIT_BIT_H */
--- a/kernel/sched/wait_bit.c
+++ b/kernel/sched/wait_bit.c
@@ -149,6 +149,54 @@ void wake_up_bit(void *word, int bit)
 }
 EXPORT_SYMBOL(wake_up_bit);
 
+wait_queue_head_t *__var_waitqueue(void *p)
+{
+	if (BITS_PER_LONG == 64) {
+		unsigned long q = (unsigned long)p;
+
+		return bit_waitqueue((void *)(q & ~1), q & 1);
+	}
+	return bit_waitqueue(p, 0);
+}
+EXPORT_SYMBOL(__var_waitqueue);
+
+static int
+var_wake_function(struct wait_queue_entry *wq_entry, unsigned int mode,
+		  int sync, void *arg)
+{
+	struct wait_bit_key *key = arg;
+	struct wait_bit_queue_entry *wbq_entry =
+		container_of(wq_entry, struct wait_bit_queue_entry, wq_entry);
+
+	if (wbq_entry->key.flags != key->flags ||
+	    wbq_entry->key.bit_nr != key->bit_nr)
+		return 0;
+
+	return autoremove_wake_function(wq_entry, mode, sync, key);
+}
+
+void init_wait_var_entry(struct wait_bit_queue_entry *wbq_entry, void *var, int flags)
+{
+	*wbq_entry = (struct wait_bit_queue_entry){
+		.key = {
+			.flags	= (var),
+			.bit_nr = -1,
+		},
+		.wq_entry = {
+			.private = current,
+			.func	 = var_wake_function,
+			.entry	 = LIST_HEAD_INIT(wbq_entry->wq_entry.entry),
+		},
+	};
+}
+EXPORT_SYMBOL(init_wait_var_entry);
+
+void wake_up_var(void *var)
+{
+	__wake_up_bit(__var_waitqueue(var), var, -1);
+}
+EXPORT_SYMBOL(wake_up_var);
+
 /*
  * Manipulate the atomic_t address to produce a better bit waitqueue table hash
  * index (we're keying off bit -1, but that would produce a horrible hash

Re: [PATCH 01/11] sched/wait: Introduce wait_var_event()

[Dan Williams]

On Thu, Mar 15, 2018 at 5:19 AM, Peter Zijlstra <peterz@infradead.org> wrote:
> As a replacement for the wait_on_atomic_t() API provide the
> wait_var_event() API.
>
> The wait_var_event() API is based on the very same hashed-waitqueue
> idea, but doesn't care about the type (atomic_t) or the specific
> condition (atomic_read() == 0). IOW. it's much more widely
> applicable/flexible.
>
> It shares all the benefits/disadvantages of a hashed-waitqueue
> approach with the existing wait_on_atomic_t/wait_on_bit() APIs.
>
> The API is modeled after the existing wait_event() API, but instead of
> taking a wait_queue_head, it takes an address. This addresses is
> hashed to obtain a wait_queue_head from the bit_wait_table.
>
> Similar to the wait_event() API, it takes a condition expression as
> second argument and will wait until this expression becomes true.
>
> The following are (mostly) identical replacements:
>
>   wait_on_atomic_t(&my_atomic, atomic_t_wait, TASK_UNINTERRUPTIBLE);
>   wake_up_atomic_t(&my_atomic);
>
>   wait_var_event(&my_atomic, !atomic_read(&my_atomic));
>   wake_up_var(&my_atomic);
>
> The only difference is that wake_up_var() is an unconditional wakeup
> and doesn't check the previously hard-coded (atomic_read() == 0)
> condition here. This is of little concequence, since most callers are
> already conditional on atomic_dec_and_test() and the ones that are
> not, are trivial to make so.
>
> Cc: David Howells <dhowells@redhat.com>
> Tested-by: Dan Williams <dan.j.williams@intel.com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  include/linux/wait_bit.h |   70 +++++++++++++++++++++++++++++++++++++++++++++++
>  kernel/sched/wait_bit.c  |   48 ++++++++++++++++++++++++++++++++
>  2 files changed, 118 insertions(+)

Hi Peter,

Is this commit ready to hit tip/sched/core? I'm looking for an
immutable branch that I can use as a basis for the "dax vs dma vs
truncate" fix series.

Thanks in advance.

[PATCH 02/11] sched/wait,drm: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-2.patch --]
[-- Type: text/plain, Size: 3245 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Unlike the wake_up_atomic_t(), wake_up_var() will issue the wakeup
even if the variable is not 0.

Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: David Airlie <airlied@linux.ie>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 drivers/gpu/drm/drm_dp_aux_dev.c                   |   13 +++++++------
 drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c |   14 ++++----------
 2 files changed, 11 insertions(+), 16 deletions(-)

--- a/drivers/gpu/drm/drm_dp_aux_dev.c
+++ b/drivers/gpu/drm/drm_dp_aux_dev.c
@@ -177,8 +177,9 @@ static ssize_t auxdev_read_iter(struct k
 		res = pos - iocb->ki_pos;
 	iocb->ki_pos = pos;
 
-	atomic_dec(&aux_dev->usecount);
-	wake_up_atomic_t(&aux_dev->usecount);
+	if (atomic_dec_and_test(&aux_dev->usecount))
+		wake_up_var(&aux_dev->usecount);
+
 	return res;
 }
 
@@ -218,8 +219,9 @@ static ssize_t auxdev_write_iter(struct
 		res = pos - iocb->ki_pos;
 	iocb->ki_pos = pos;
 
-	atomic_dec(&aux_dev->usecount);
-	wake_up_atomic_t(&aux_dev->usecount);
+	if (atomic_dec_and_test(&aux_dev->usecount))
+		wake_up_var(&aux_dev->usecount);
+
 	return res;
 }
 
@@ -277,8 +279,7 @@ void drm_dp_aux_unregister_devnode(struc
 	mutex_unlock(&aux_idr_mutex);
 
 	atomic_dec(&aux_dev->usecount);
-	wait_on_atomic_t(&aux_dev->usecount, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&aux_dev->usecount, !atomic_read(&aux_dev->usecount));
 
 	minor = aux_dev->index;
 	if (aux_dev->dev)
--- a/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/selftests/intel_breadcrumbs.c
@@ -271,18 +271,13 @@ struct igt_wakeup {
 	u32 seqno;
 };
 
-static int wait_atomic_timeout(atomic_t *p, unsigned int mode)
-{
-	return schedule_timeout(10 * HZ) ? 0 : -ETIMEDOUT;
-}
-
 static bool wait_for_ready(struct igt_wakeup *w)
 {
 	DEFINE_WAIT(ready);
 
 	set_bit(IDLE, &w->flags);
 	if (atomic_dec_and_test(w->done))
-		wake_up_atomic_t(w->done);
+		wake_up_var(w->done);
 
 	if (test_bit(STOP, &w->flags))
 		goto out;
@@ -299,7 +294,7 @@ static bool wait_for_ready(struct igt_wa
 out:
 	clear_bit(IDLE, &w->flags);
 	if (atomic_dec_and_test(w->set))
-		wake_up_atomic_t(w->set);
+		wake_up_var(w->set);
 
 	return !test_bit(STOP, &w->flags);
 }
@@ -342,7 +337,7 @@ static void igt_wake_all_sync(atomic_t *
 	atomic_set(ready, 0);
 	wake_up_all(wq);
 
-	wait_on_atomic_t(set, atomic_t_wait, TASK_UNINTERRUPTIBLE);
+	wait_var_event(set, !atomic_read(set));
 	atomic_set(ready, count);
 	atomic_set(done, count);
 }
@@ -350,7 +345,6 @@ static void igt_wake_all_sync(atomic_t *
 static int igt_wakeup(void *arg)
 {
 	I915_RND_STATE(prng);
-	const int state = TASK_UNINTERRUPTIBLE;
 	struct intel_engine_cs *engine = arg;
 	struct igt_wakeup *waiters;
 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
@@ -418,7 +412,7 @@ static int igt_wakeup(void *arg)
 		 * that they are ready for the next test. We wait until all
 		 * threads are complete and waiting for us (i.e. not a seqno).
 		 */
-		err = wait_on_atomic_t(&done, wait_atomic_timeout, state);
+		err = wait_var_event_timeout(&done, !atomic_read(&done), 10 * HZ);
 		if (err) {
 			pr_err("Timed out waiting for %d remaining waiters\n",
 			       atomic_read(&done));

Re: [PATCH 02/11] sched/wait,drm: Replace wait_on_atomic_t usage

[Chris Wilson]

Quoting Peter Zijlstra (2018-03-15 12:19:04)
> Convert the wait_on_atomic_t() usage to the new wait_var_event() API.
> 
> Unlike the wake_up_atomic_t(), wake_up_var() will issue the wakeup
> even if the variable is not 0.
> 
> Cc: Daniel Vetter <daniel.vetter@intel.com>
> Cc: David Airlie <airlied@linux.ie>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris

[PATCH 03/11] sched/wait,media: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-3.patch --]
[-- Type: text/plain, Size: 1148 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Unlike the wake_up_atomic_t(), wake_up_var() will issue the wakeup
even if the variable is not 0.

Cc: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 drivers/media/platform/qcom/venus/hfi.c |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

--- a/drivers/media/platform/qcom/venus/hfi.c
+++ b/drivers/media/platform/qcom/venus/hfi.c
@@ -106,8 +106,8 @@ int hfi_core_deinit(struct venus_core *c
 
 	if (!empty) {
 		mutex_unlock(&core->lock);
-		wait_on_atomic_t(&core->insts_count, atomic_t_wait,
-				 TASK_UNINTERRUPTIBLE);
+		wait_var_event(&core->insts_count,
+			       !atomic_read(&core->insts_count));
 		mutex_lock(&core->lock);
 	}
 
@@ -229,8 +229,8 @@ void hfi_session_destroy(struct venus_in
 
 	mutex_lock(&core->lock);
 	list_del_init(&inst->list);
-	atomic_dec(&core->insts_count);
-	wake_up_atomic_t(&core->insts_count);
+	if (atomic_dec_and_test(&core->insts_count))
+		wake_up_var(&core->insts_count);
 	mutex_unlock(&core->lock);
 }
 EXPORT_SYMBOL_GPL(hfi_session_destroy);

[PATCH 04/11] sched/wait,afs: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-4.patch --]
[-- Type: text/plain, Size: 2300 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 fs/afs/cell.c   |    6 +++---
 fs/afs/rxrpc.c  |    6 +++---
 fs/afs/server.c |    6 +++---
 3 files changed, 9 insertions(+), 9 deletions(-)

--- a/fs/afs/cell.c
+++ b/fs/afs/cell.c
@@ -25,7 +25,7 @@ static void afs_manage_cell(struct work_
 static void afs_dec_cells_outstanding(struct afs_net *net)
 {
 	if (atomic_dec_and_test(&net->cells_outstanding))
-		wake_up_atomic_t(&net->cells_outstanding);
+		wake_up_var(&net->cells_outstanding);
 }
 
 /*
@@ -764,7 +764,7 @@ void afs_cell_purge(struct afs_net *net)
 	afs_queue_cell_manager(net);
 
 	_debug("wait");
-	wait_on_atomic_t(&net->cells_outstanding, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&net->cells_outstanding,
+		       !atomic_read(&net->cells_outstanding));
 	_leave("");
 }
--- a/fs/afs/rxrpc.c
+++ b/fs/afs/rxrpc.c
@@ -103,8 +103,8 @@ void afs_close_socket(struct afs_net *ne
 	}
 
 	_debug("outstanding %u", atomic_read(&net->nr_outstanding_calls));
-	wait_on_atomic_t(&net->nr_outstanding_calls, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&net->nr_outstanding_calls,
+		       !atomic_read(&net->nr_outstanding_calls));
 	_debug("no outstanding calls");
 
 	kernel_sock_shutdown(net->socket, SHUT_RDWR);
@@ -175,7 +175,7 @@ void afs_put_call(struct afs_call *call)
 		trace_afs_call(call, afs_call_trace_free, 0, o,
 			       __builtin_return_address(0));
 		if (o == 0)
-			wake_up_atomic_t(&net->nr_outstanding_calls);
+			wake_up_var(&net->nr_outstanding_calls);
 	}
 }
 
--- a/fs/afs/server.c
+++ b/fs/afs/server.c
@@ -25,7 +25,7 @@ static void afs_inc_servers_outstanding(
 static void afs_dec_servers_outstanding(struct afs_net *net)
 {
 	if (atomic_dec_and_test(&net->servers_outstanding))
-		wake_up_atomic_t(&net->servers_outstanding);
+		wake_up_var(&net->servers_outstanding);
 }
 
 /*
@@ -521,8 +521,8 @@ void afs_purge_servers(struct afs_net *n
 	afs_queue_server_manager(net);
 
 	_debug("wait");
-	wait_on_atomic_t(&net->servers_outstanding, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&net->servers_outstanding,
+		       !atomic_read(&net->servers_outstanding));
 	_leave("");
 }
 

[PATCH 05/11] sched/wait,btrfs: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-5.patch --]
[-- Type: text/plain, Size: 2414 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Cc: Chris Mason <clm@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 fs/btrfs/extent-tree.c |   14 ++++++--------
 fs/btrfs/ioctl.c       |    2 +-
 2 files changed, 7 insertions(+), 9 deletions(-)

--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3990,7 +3990,7 @@ void btrfs_dec_nocow_writers(struct btrf
 	bg = btrfs_lookup_block_group(fs_info, bytenr);
 	ASSERT(bg);
 	if (atomic_dec_and_test(&bg->nocow_writers))
-		wake_up_atomic_t(&bg->nocow_writers);
+		wake_up_var(&bg->nocow_writers);
 	/*
 	 * Once for our lookup and once for the lookup done by a previous call
 	 * to btrfs_inc_nocow_writers()
@@ -4001,8 +4001,7 @@ void btrfs_dec_nocow_writers(struct btrf
 
 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
 {
-	wait_on_atomic_t(&bg->nocow_writers, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
 }
 
 static const char *alloc_name(u64 flags)
@@ -6526,7 +6525,7 @@ void btrfs_dec_block_group_reservations(
 	bg = btrfs_lookup_block_group(fs_info, start);
 	ASSERT(bg);
 	if (atomic_dec_and_test(&bg->reservations))
-		wake_up_atomic_t(&bg->reservations);
+		wake_up_var(&bg->reservations);
 	btrfs_put_block_group(bg);
 }
 
@@ -6552,8 +6551,7 @@ void btrfs_wait_block_group_reservations
 	down_write(&space_info->groups_sem);
 	up_write(&space_info->groups_sem);
 
-	wait_on_atomic_t(&bg->reservations, atomic_t_wait,
-			 TASK_UNINTERRUPTIBLE);
+	wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
 }
 
 /**
@@ -11061,7 +11059,7 @@ void btrfs_wait_for_snapshot_creation(st
 		ret = btrfs_start_write_no_snapshotting(root);
 		if (ret)
 			break;
-		wait_on_atomic_t(&root->will_be_snapshotted, atomic_t_wait,
-				 TASK_UNINTERRUPTIBLE);
+		wait_var_event(&root->will_be_snapshotted,
+			       !atomic_read(&root->will_be_snapshotted));
 	}
 }
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -723,7 +723,7 @@ static int create_snapshot(struct btrfs_
 	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 dec_and_free:
 	if (atomic_dec_and_test(&root->will_be_snapshotted))
-		wake_up_atomic_t(&root->will_be_snapshotted);
+		wake_up_var(&root->will_be_snapshotted);
 free_pending:
 	kfree(pending_snapshot->root_item);
 	btrfs_free_path(pending_snapshot->path);

Re: [PATCH 05/11] sched/wait,btrfs: Replace wait_on_atomic_t usage

[David Sterba]

On Thu, Mar 15, 2018 at 01:19:07PM +0100, Peter Zijlstra wrote:
> Convert the wait_on_atomic_t() usage to the new wait_var_event() API.
> 
> Cc: Chris Mason <clm@fb.com>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

Acked-by: David Sterba <dsterba@suse.com>

[PATCH 06/11] sched/wait,fscache: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-6.patch --]
[-- Type: text/plain, Size: 1201 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 fs/fscache/cookie.c           |    7 ++++---
 include/linux/fscache-cache.h |    2 +-
 2 files changed, 5 insertions(+), 4 deletions(-)

--- a/fs/fscache/cookie.c
+++ b/fs/fscache/cookie.c
@@ -557,9 +557,10 @@ void __fscache_disable_cookie(struct fsc
 	 * n_active reaches 0).  This makes sure outstanding reads and writes
 	 * have completed.
 	 */
-	if (!atomic_dec_and_test(&cookie->n_active))
-		wait_on_atomic_t(&cookie->n_active, atomic_t_wait,
-				 TASK_UNINTERRUPTIBLE);
+	if (!atomic_dec_and_test(&cookie->n_active)) {
+		wait_var_event(&cookie->n_active,
+			       !atomic_read(&cookie->n_active));
+	}
 
 	/* Make sure any pending writes are cancelled. */
 	if (cookie->def->type != FSCACHE_COOKIE_TYPE_INDEX)
--- a/include/linux/fscache-cache.h
+++ b/include/linux/fscache-cache.h
@@ -496,7 +496,7 @@ static inline bool __fscache_unuse_cooki
 
 static inline void __fscache_wake_unused_cookie(struct fscache_cookie *cookie)
 {
-	wake_up_atomic_t(&cookie->n_active);
+	wake_up_var(&cookie->n_active);
 }
 
 /**

[PATCH 07/11] sched/wait,nfs: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-7.patch --]
[-- Type: text/plain, Size: 2586 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Cc: Anna Schumaker <anna.schumaker@netapp.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 fs/nfs/inode.c    |    5 -----
 fs/nfs/pagelist.c |    6 +++---
 fs/nfs/pnfs_nfs.c |    2 +-
 fs/nfs/write.c    |    6 +++---
 4 files changed, 7 insertions(+), 12 deletions(-)

--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -85,11 +85,6 @@ int nfs_wait_bit_killable(struct wait_bi
 }
 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
 
-int nfs_wait_atomic_killable(atomic_t *p, unsigned int mode)
-{
-	return nfs_wait_killable(mode);
-}
-
 /**
  * nfs_compat_user_ino64 - returns the user-visible inode number
  * @fileid: 64-bit fileid
--- a/fs/nfs/pagelist.c
+++ b/fs/nfs/pagelist.c
@@ -98,8 +98,8 @@ nfs_page_free(struct nfs_page *p)
 int
 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
 {
-	return wait_on_atomic_t(&l_ctx->io_count, nfs_wait_atomic_killable,
-			TASK_KILLABLE);
+	return wait_var_event_killable(&l_ctx->io_count,
+				       !atomic_read(&l_ctx->io_count));
 }
 
 /**
@@ -395,7 +395,7 @@ static void nfs_clear_request(struct nfs
 	}
 	if (l_ctx != NULL) {
 		if (atomic_dec_and_test(&l_ctx->io_count)) {
-			wake_up_atomic_t(&l_ctx->io_count);
+			wake_up_var(&l_ctx->io_count);
 			if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
 				rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
 		}
--- a/fs/nfs/pnfs_nfs.c
+++ b/fs/nfs/pnfs_nfs.c
@@ -245,7 +245,7 @@ pnfs_generic_commit_cancel_empty_pagelis
 {
 	if (list_empty(pages)) {
 		if (atomic_dec_and_test(&cinfo->mds->rpcs_out))
-			wake_up_atomic_t(&cinfo->mds->rpcs_out);
+			wake_up_var(&cinfo->mds->rpcs_out);
 		/* don't call nfs_commitdata_release - it tries to put
 		 * the open_context which is not acquired until nfs_init_commit
 		 * which has not been called on @data */
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1620,8 +1620,8 @@ static void nfs_writeback_result(struct
 
 static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
 {
-	return wait_on_atomic_t(&cinfo->rpcs_out,
-			nfs_wait_atomic_killable, TASK_KILLABLE);
+	return wait_var_event_killable(&cinfo->rpcs_out,
+				       !atomic_read(&cinfo->rpcs_out));
 }
 
 static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
@@ -1632,7 +1632,7 @@ static void nfs_commit_begin(struct nfs_
 static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
 {
 	if (atomic_dec_and_test(&cinfo->rpcs_out))
-		wake_up_atomic_t(&cinfo->rpcs_out);
+		wake_up_var(&cinfo->rpcs_out);
 }
 
 void nfs_commitdata_release(struct nfs_commit_data *data)

[PATCH 08/11] sched/wait,ocfs2: Replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-8.patch --]
[-- Type: text/plain, Size: 1046 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 fs/ocfs2/filecheck.c |    9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

--- a/fs/ocfs2/filecheck.c
+++ b/fs/ocfs2/filecheck.c
@@ -134,9 +134,10 @@ ocfs2_filecheck_sysfs_free(struct ocfs2_
 {
 	struct ocfs2_filecheck_entry *p;
 
-	if (!atomic_dec_and_test(&entry->fs_count))
-		wait_on_atomic_t(&entry->fs_count, atomic_t_wait,
-				 TASK_UNINTERRUPTIBLE);
+	if (!atomic_dec_and_test(&entry->fs_count)) {
+		wait_var_event(&entry->fs_count,
+			       !atomic_read(&entry->fs_count));
+	}
 
 	spin_lock(&entry->fs_fcheck->fc_lock);
 	while (!list_empty(&entry->fs_fcheck->fc_head)) {
@@ -183,7 +184,7 @@ static void
 ocfs2_filecheck_sysfs_put(struct ocfs2_filecheck_sysfs_entry *entry)
 {
 	if (atomic_dec_and_test(&entry->fs_count))
-		wake_up_atomic_t(&entry->fs_count);
+		wake_up_var(&entry->fs_count);
 }
 
 static struct ocfs2_filecheck_sysfs_entry *

[PATCH 09/11] sched/wait,mips: Fix and replace wait_on_atomic_t usage

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-9.patch --]
[-- Type: text/plain, Size: 1169 bytes --]

Convert the wait_on_atomic_t() usage to the new wait_var_event() API.

And while there, add the missing wakeup...

Cc: James Hogan <jhogan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/mips/kernel/process.c |    2 ++
 arch/mips/kernel/traps.c   |    4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)

--- a/arch/mips/kernel/process.c
+++ b/arch/mips/kernel/process.c
@@ -781,6 +781,8 @@ int mips_set_process_fp_mode(struct task
 	atomic_set(&task->mm->context.fp_mode_switching, 0);
 	preempt_enable();
 
+	wake_up_var(&task->mm->context.fp_mode_switching);
+
 	return 0;
 }
 
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -1248,8 +1248,8 @@ static int enable_restore_fp_context(int
 	 * If an FP mode switch is currently underway, wait for it to
 	 * complete before proceeding.
 	 */
-	wait_on_atomic_t(&current->mm->context.fp_mode_switching,
-			 atomic_t_wait, TASK_KILLABLE);
+	wait_var_event(&current->mm->context.fp_mode_switching,
+		       !atomic_read(&current->mm->context.fp_mode_switching));
 
 	if (!used_math()) {
 		/* First time FP context user. */

[PATCH 10/11] sched/wait: Remove wait_on_atomic_t()

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-10.patch --]
[-- Type: text/plain, Size: 6301 bytes --]

There are no users left, remove it.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 include/linux/wait_bit.h |   27 ------------
 kernel/sched/wait_bit.c  |  101 -----------------------------------------------
 2 files changed, 128 deletions(-)

--- a/include/linux/wait_bit.h
+++ b/include/linux/wait_bit.h
@@ -10,7 +10,6 @@
 struct wait_bit_key {
 	void			*flags;
 	int			bit_nr;
-#define WAIT_ATOMIC_T_BIT_NR	-1
 	unsigned long		timeout;
 };
 
@@ -22,21 +21,15 @@ struct wait_bit_queue_entry {
 #define __WAIT_BIT_KEY_INITIALIZER(word, bit)					\
 	{ .flags = word, .bit_nr = bit, }
 
-#define __WAIT_ATOMIC_T_KEY_INITIALIZER(p)					\
-	{ .flags = p, .bit_nr = WAIT_ATOMIC_T_BIT_NR, }
-
 typedef int wait_bit_action_f(struct wait_bit_key *key, int mode);
-typedef int wait_atomic_t_action_f(atomic_t *counter, unsigned int mode);
 
 void __wake_up_bit(struct wait_queue_head *wq_head, void *word, int bit);
 int __wait_on_bit(struct wait_queue_head *wq_head, struct wait_bit_queue_entry *wbq_entry, wait_bit_action_f *action, unsigned int mode);
 int __wait_on_bit_lock(struct wait_queue_head *wq_head, struct wait_bit_queue_entry *wbq_entry, wait_bit_action_f *action, unsigned int mode);
 void wake_up_bit(void *word, int bit);
-void wake_up_atomic_t(atomic_t *p);
 int out_of_line_wait_on_bit(void *word, int, wait_bit_action_f *action, unsigned int mode);
 int out_of_line_wait_on_bit_timeout(void *word, int, wait_bit_action_f *action, unsigned int mode, unsigned long timeout);
 int out_of_line_wait_on_bit_lock(void *word, int, wait_bit_action_f *action, unsigned int mode);
-int out_of_line_wait_on_atomic_t(atomic_t *p, wait_atomic_t_action_f action, unsigned int mode);
 struct wait_queue_head *bit_waitqueue(void *word, int bit);
 extern void __init wait_bit_init(void);
 
@@ -57,7 +50,6 @@ extern int bit_wait(struct wait_bit_key
 extern int bit_wait_io(struct wait_bit_key *key, int mode);
 extern int bit_wait_timeout(struct wait_bit_key *key, int mode);
 extern int bit_wait_io_timeout(struct wait_bit_key *key, int mode);
-extern int atomic_t_wait(atomic_t *counter, unsigned int mode);
 
 /**
  * wait_on_bit - wait for a bit to be cleared
@@ -243,25 +235,6 @@ wait_on_bit_lock_action(unsigned long *w
 	return out_of_line_wait_on_bit_lock(word, bit, action, mode);
 }
 
-/**
- * wait_on_atomic_t - Wait for an atomic_t to become 0
- * @val: The atomic value being waited on, a kernel virtual address
- * @action: the function used to sleep, which may take special actions
- * @mode: the task state to sleep in
- *
- * Wait for an atomic_t to become 0.  We abuse the bit-wait waitqueue table for
- * the purpose of getting a waitqueue, but we set the key to a bit number
- * outside of the target 'word'.
- */
-static inline
-int wait_on_atomic_t(atomic_t *val, wait_atomic_t_action_f action, unsigned mode)
-{
-	might_sleep();
-	if (atomic_read(val) == 0)
-		return 0;
-	return out_of_line_wait_on_atomic_t(val, action, mode);
-}
-
 extern void init_wait_var_entry(struct wait_bit_queue_entry *wbq_entry, void *var, int flags);
 extern void wake_up_var(void *var);
 extern wait_queue_head_t *__var_waitqueue(void *p);
--- a/kernel/sched/wait_bit.c
+++ b/kernel/sched/wait_bit.c
@@ -197,107 +197,6 @@ void wake_up_var(void *var)
 }
 EXPORT_SYMBOL(wake_up_var);
 
-/*
- * Manipulate the atomic_t address to produce a better bit waitqueue table hash
- * index (we're keying off bit -1, but that would produce a horrible hash
- * value).
- */
-static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p)
-{
-	if (BITS_PER_LONG == 64) {
-		unsigned long q = (unsigned long)p;
-
-		return bit_waitqueue((void *)(q & ~1), q & 1);
-	}
-	return bit_waitqueue(p, 0);
-}
-
-static int wake_atomic_t_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync,
-				  void *arg)
-{
-	struct wait_bit_key *key = arg;
-	struct wait_bit_queue_entry *wait_bit = container_of(wq_entry, struct wait_bit_queue_entry, wq_entry);
-	atomic_t *val = key->flags;
-
-	if (wait_bit->key.flags != key->flags ||
-	    wait_bit->key.bit_nr != key->bit_nr ||
-	    atomic_read(val) != 0)
-		return 0;
-
-	return autoremove_wake_function(wq_entry, mode, sync, key);
-}
-
-/*
- * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
- * the actions of __wait_on_atomic_t() are permitted return codes.  Nonzero
- * return codes halt waiting and return.
- */
-static __sched
-int __wait_on_atomic_t(struct wait_queue_head *wq_head, struct wait_bit_queue_entry *wbq_entry,
-		       wait_atomic_t_action_f action, unsigned int mode)
-{
-	atomic_t *val;
-	int ret = 0;
-
-	do {
-		prepare_to_wait(wq_head, &wbq_entry->wq_entry, mode);
-		val = wbq_entry->key.flags;
-		if (atomic_read(val) == 0)
-			break;
-		ret = (*action)(val, mode);
-	} while (!ret && atomic_read(val) != 0);
-	finish_wait(wq_head, &wbq_entry->wq_entry);
-
-	return ret;
-}
-
-#define DEFINE_WAIT_ATOMIC_T(name, p)					\
-	struct wait_bit_queue_entry name = {				\
-		.key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p),		\
-		.wq_entry = {						\
-			.private	= current,			\
-			.func		= wake_atomic_t_function,	\
-			.entry		=				\
-				LIST_HEAD_INIT((name).wq_entry.entry),	\
-		},							\
-	}
-
-__sched int out_of_line_wait_on_atomic_t(atomic_t *p,
-					 wait_atomic_t_action_f action,
-					 unsigned int mode)
-{
-	struct wait_queue_head *wq_head = atomic_t_waitqueue(p);
-	DEFINE_WAIT_ATOMIC_T(wq_entry, p);
-
-	return __wait_on_atomic_t(wq_head, &wq_entry, action, mode);
-}
-EXPORT_SYMBOL(out_of_line_wait_on_atomic_t);
-
-__sched int atomic_t_wait(atomic_t *counter, unsigned int mode)
-{
-	schedule();
-	if (signal_pending_state(mode, current))
-		return -EINTR;
-
-	return 0;
-}
-EXPORT_SYMBOL(atomic_t_wait);
-
-/**
- * wake_up_atomic_t - Wake up a waiter on a atomic_t
- * @p: The atomic_t being waited on, a kernel virtual address
- *
- * Wake up anyone waiting for the atomic_t to go to zero.
- *
- * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
- * check is done by the waiter's wake function, not the by the waker itself).
- */
-void wake_up_atomic_t(atomic_t *p)
-{
-	__wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR);
-}
-EXPORT_SYMBOL(wake_up_atomic_t);
-
 __sched int bit_wait(struct wait_bit_key *word, int mode)
 {
 	schedule();

[PATCH 11/11] sched/wait: Improve __var_waitqueue()

[Peter Zijlstra]

[-- Attachment #1: peterz-wait_var-11.patch --]
[-- Type: text/plain, Size: 857 bytes --]

Since we fixed hash_64 to not suck, there is no need to play games to
attempt to improve the hash value on 64bit. Also, since we don't use
the bit value for the variables, use hash_ptr() directly.

Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: George Spelvin <linux@sciencehorizons.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 kernel/sched/wait_bit.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

--- a/kernel/sched/wait_bit.c
+++ b/kernel/sched/wait_bit.c
@@ -151,12 +151,7 @@ EXPORT_SYMBOL(wake_up_bit);
 
 wait_queue_head_t *__var_waitqueue(void *p)
 {
-	if (BITS_PER_LONG == 64) {
-		unsigned long q = (unsigned long)p;
-
-		return bit_waitqueue((void *)(q & ~1), q & 1);
-	}
-	return bit_waitqueue(p, 0);
+	return bit_wait_table + hash_ptr(p, WAIT_TABLE_BITS);
 }
 EXPORT_SYMBOL(__var_waitqueue);