Skip to contentSkip to conversation

btrfs: zstd ensure reclaim timer is properly cleaned up

PATCH In review Dennis Zhou 2 participants 5 messages Feb 22, 2019 – Feb 27, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 btrfs: zstd ensure reclaim timer is properly cleaned up

The timer function, zstd_reclaim_timer_fn(), reschedules itself under certain conditions. When cleaning up, take the lock and remove all workspaces. This prevents the timer from rearming itself. Lastly, switch to del_timer_sync() to ensure that the timer function can't trigger as we're unloading.

---
v2:
- cleanup workspaces and then disable the timer

 fs/btrfs/zstd.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
fs/btrfs/zstd.c +4 -2
index 3e418a3aeb11..6b9e29d050f3 100644
@@ -195,8 +195,7 @@ static void zstd_cleanup_workspace_manager(void)
 	struct workspace *workspace;
 	int i;
 
-	del_timer(&wsm.timer);
-
+	spin_lock(&wsm.lock);
 	for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
 		while (!list_empty(&wsm.idle_ws[i])) {
 			workspace = container_of(wsm.idle_ws[i].next,
@@ -206,6 +205,9 @@ static void zstd_cleanup_workspace_manager(void)
 			wsm.ops->free_workspace(&workspace->list);
 		}
 	}
+	spin_unlock(&wsm.lock);
+
+	del_timer_sync(&wsm.timer);
 }
 
 /*
Signature · 1 line
2.17.1
Reviewed-by on v2 1/1
#1 lore raw
Quoted 7 lines · Dennis Zhou

The timer function, zstd_reclaim_timer_fn(), reschedules itself under certain conditions. When cleaning up, take the lock and remove all workspaces. This prevents the timer from rearming itself. Lastly, switch to del_timer_sync() to ensure that the timer function can't trigger as we're unloading.

Quoted 22 lines

---
v2:
- cleanup workspaces and then disable the timer

 fs/btrfs/zstd.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
fs/btrfs/zstd.c +1 -2
index 3e418a3aeb11..6b9e29d050f3 100644
@@ -195,8 +195,7 @@ static void zstd_cleanup_workspace_manager(void)
 	struct workspace *workspace;
 	int i;
 
-	del_timer(&wsm.timer);
-
+	spin_lock(&wsm.lock);
 	for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
 		while (!list_empty(&wsm.idle_ws[i])) {
 			workspace = container_of(wsm.idle_ws[i].next,

fs/btrfs/zstd.c static void zstd_cleanup_workspace_manager(void) · line 205

Quoted 1 line
fs/btrfs/zstd.c +0 -0
@@ -206,6 +205,9 @@ static void zstd_cleanup_workspace_manager(void)
 			wsm.ops->free_workspace(&workspace->list);

I've noticed while reading the code, why do you use the indirect call here? The wsm.ops points to btrfs_zstd_compress so free_workspace is always zstd_free_workspace.

The compiler is usually smart to replace such things by direct call if the type has not escaped, but this is not true for btrfs_compress_op so the indirect function call must be preserved.

#2 lore raw
Quoted 11 lines · David Sterba
Quoted 7 lines · Dennis Zhou

The timer function, zstd_reclaim_timer_fn(), reschedules itself under certain conditions. When cleaning up, take the lock and remove all workspaces. This prevents the timer from rearming itself. Lastly, switch to del_timer_sync() to ensure that the timer function can't trigger as we're unloading.

Thanks!

Quoted 28 lines
Quoted 22 lines

---
v2:
- cleanup workspaces and then disable the timer

 fs/btrfs/zstd.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
fs/btrfs/zstd.c +1 -2
index 3e418a3aeb11..6b9e29d050f3 100644
@@ -195,8 +195,7 @@ static void zstd_cleanup_workspace_manager(void)
 	struct workspace *workspace;
 	int i;
 
-	del_timer(&wsm.timer);
-
+	spin_lock(&wsm.lock);
 	for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
 		while (!list_empty(&wsm.idle_ws[i])) {
 			workspace = container_of(wsm.idle_ws[i].next,

fs/btrfs/zstd.c static void zstd_cleanup_workspace_manager(void) · line 205

Quoted 1 line
fs/btrfs/zstd.c +0 -0
@@ -206,6 +205,9 @@ static void zstd_cleanup_workspace_manager(void)
 			wsm.ops->free_workspace(&workspace->list);

I've noticed while reading the code, why do you use the indirect call here? The wsm.ops points to btrfs_zstd_compress so free_workspace is always zstd_free_workspace.

Quoted 3 lines

The compiler is usually smart to replace such things by direct call if the type has not escaped, but this is not true for btrfs_compress_op so the indirect function call must be preserved.

I don't have a strong reason to use the indirect call here. It was just to make it consistent for everyone to use the indirection. This at least is in the cleanup path, so I don't think performance is that important? But I don't feel strongly for or against calling zstd_free_workspace() directly.

Thanks,
Dennis

#3 lore raw
Quoted 11 lines · Dennis Zhou
Quoted 7 lines

I've noticed while reading the code, why do you use the indirect call here? The wsm.ops points to btrfs_zstd_compress so free_workspace is always zstd_free_workspace.

The compiler is usually smart to replace such things by direct call if the type has not escaped, but this is not true for btrfs_compress_op so the indirect function call must be preserved.

I don't have a strong reason to use the indirect call here. It was just to make it consistent for everyone to use the indirection. This at least is in the cleanup path, so I don't think performance is that important?

It's not just that, the timer uses it too and there are indirect calls of the alloc_workspace callback. The indirection is not used by lzo nor zlib code, so I don't see what 'everyone' you mean. In the generic compression code it makes sense, I see that.

Quoted 2 lines

But I don't feel strongly for or against calling zstd_free_workspace() directly.

I feel strongly about not using the indirection when not necessary :)

#4 lore raw
Quoted 18 lines · David Sterba
Quoted 11 lines · Dennis Zhou
Quoted 7 lines

I've noticed while reading the code, why do you use the indirect call here? The wsm.ops points to btrfs_zstd_compress so free_workspace is always zstd_free_workspace.

The compiler is usually smart to replace such things by direct call if the type has not escaped, but this is not true for btrfs_compress_op so the indirect function call must be preserved.

I don't have a strong reason to use the indirect call here. It was just to make it consistent for everyone to use the indirection. This at least is in the cleanup path, so I don't think performance is that important?

It's not just that, the timer uses it too and there are indirect calls of the alloc_workspace callback. The indirection is not used by lzo nor zlib code, so I don't see what 'everyone' you mean. In the generic compression code it makes sense, I see that.

Quoted 4 lines
Quoted 2 lines

But I don't feel strongly for or against calling zstd_free_workspace() directly.

I feel strongly about not using the indirection when not necessary :)

Great :). I sent you a patch just now to remove the indirection [1].

[1] https://lore.kernel.org/linux-btrfs/[email protected]

Thanks,
Dennis