Skip to contentSkip to conversation

btrfs: simplify how first hit is passed to __btrfs_abort_transaction()

PATCH In review David Sterba 3 participants 3 messages Apr 30 – May 6 Btrfs file system

lore
Share & tools
lore

David Sterba

author maintainer
#0 lore raw

PATCH 1/1 btrfs: simplify how first hit is passed to __btrfs_abort_transaction()

Optimize the btrfs_abort_transaction() for size as it (by our convention) must be put right after the error condition is detected. The exact file:line is reported so there's a portion that must be inlined. As this is cold code it bloats functions. In previous patch "btrfs: move transaction abort message to __btrfs_abort_transaction()" the error message was moved to the common helper, saving like 20KiB of btrfs.ko and several instructions per call site and some stack space.

There's little left to be optimized, we need to keep the atomic test_and_set_bit() and to convey that as 'first hit' to __btrfs_abort_transaction().

Right now it's a bool, which takes 8 bytes on stack for each call but it's 1 bit of information. We can encode that to some of the other parameters.

For that let's use the 'error' parameter, by convention it's negative errno so we can reliably detect if it's the first hit or a later error. Also the negation is usually implemented by a single instruction (NEG on x86_64) so the resulting object code is kept short.

This reduces btrfs.ko by 8K and stack in several functions by 8 bytes.

Cumulative effect with the other commit is -30K of btrfs.ko. While the encoding is an implementation detail, it's contained within the API. Making the transaction abort calls very light is desired.

---
 fs/btrfs/transaction.c | 13 ++++++++++++-
 fs/btrfs/transaction.h | 15 +++++++++------
 2 files changed, 21 insertions(+), 7 deletions(-)
fs/btrfs/transaction.c +12 -1
index 01f66328966f..bbef3ae86ac4 100644
@@ -2723,12 +2723,23 @@ int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
  *
  * We'll complete the cleanup in btrfs_end_transaction and
  * btrfs_commit_transaction.
+ *
+ * Note: the parameter @error encodes whether the transactin abort was first hit
+ *       (setting the FS_ERROR state bit in btrfs_abort_transaction())
+ *       - positive number - first hit
+ *       - negative number - abort after it was already done
  */
 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 				      const char *function,
-				      unsigned int line, int error, bool first_hit)
+				      unsigned int line, int error)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
+	bool first_hit = false;
+
+	if (error > 0) {
+		error = -error;
+		first_hit = true;
+	}
 
 	WRITE_ONCE(trans->aborted, error);
 	WRITE_ONCE(trans->transaction->aborted, error);
fs/btrfs/transaction.h +9 -6
index f1cb05460cec..e1d4e7330198 100644
@@ -243,20 +243,23 @@ static inline bool btrfs_abort_should_print_stack(int error)
 }
 
 /*
- * Call btrfs_abort_transaction as early as possible when an error condition is
- * detected, that way the exact stack trace is reported for some errors.
+ * Call btrfs_abort_transaction() as early as possible when an error condition
+ * is detected, that way the exact stack trace is reported for some errors.
+ *
+ * Error number must be negative as it encodes wheather it's the first abort.
  */
 #define btrfs_abort_transaction(trans, error)		\
 do {								\
-	bool __first = false;					\
+	int __error = (error);					\
+								\
 	/* Report first abort since mount */			\
 	if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,	\
 			&((trans)->fs_info->fs_state))) {	\
-		__first = true;					\
+		__error = -__error;				\
 		WARN_ON(btrfs_abort_should_print_stack(error));	\
 	}							\
 	__btrfs_abort_transaction((trans), __func__,		\
-				  __LINE__, (error), __first);	\
+				  __LINE__, __error);		\
 } while (0)
 
 int btrfs_end_transaction(struct btrfs_trans_handle *trans);
@@ -294,7 +297,7 @@ void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 				      const char *function,
-				      unsigned int line, int error, bool first_hit);
+				      unsigned int line, int error);
 
 int __init btrfs_transaction_init(void);
 void __cold btrfs_transaction_exit(void);
Signature · 1 line
2.51.0
#1 lore raw
Quoted 67 lines · David Sterba

Optimize the btrfs_abort_transaction() for size as it (by our convention) must be put right after the error condition is detected. The exact file:line is reported so there's a portion that must be inlined. As this is cold code it bloats functions. In previous patch "btrfs: move transaction abort message to __btrfs_abort_transaction()" the error message was moved to the common helper, saving like 20KiB of btrfs.ko and several instructions per call site and some stack space.

There's little left to be optimized, we need to keep the atomic test_and_set_bit() and to convey that as 'first hit' to __btrfs_abort_transaction().

Right now it's a bool, which takes 8 bytes on stack for each call but it's 1 bit of information. We can encode that to some of the other parameters.

For that let's use the 'error' parameter, by convention it's negative errno so we can reliably detect if it's the first hit or a later error. Also the negation is usually implemented by a single instruction (NEG on x86_64) so the resulting object code is kept short.

This reduces btrfs.ko by 8K and stack in several functions by 8 bytes.

Cumulative effect with the other commit is -30K of btrfs.ko. While the encoding is an implementation detail, it's contained within the API. Making the transaction abort calls very light is desired.

---
  fs/btrfs/transaction.c | 13 ++++++++++++-
  fs/btrfs/transaction.h | 15 +++++++++------
  2 files changed, 21 insertions(+), 7 deletions(-)
fs/btrfs/transaction.c +12 -1
index 01f66328966f..bbef3ae86ac4 100644
@@ -2723,12 +2723,23 @@ int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
   *
   * We'll complete the cleanup in btrfs_end_transaction and
   * btrfs_commit_transaction.
+ *
+ * Note: the parameter @error encodes whether the transactin abort was first hit
+ *       (setting the FS_ERROR state bit in btrfs_abort_transaction())
+ *       - positive number - first hit
+ *       - negative number - abort after it was already done
   */
  void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
  				      const char *function,
-				      unsigned int line, int error, bool first_hit)
+				      unsigned int line, int error)
  {
  	struct btrfs_fs_info *fs_info = trans->fs_info;
+	bool first_hit = false;
+
+	if (error > 0) {
+		error = -error;
+		first_hit = true;
+	}
 
  	WRITE_ONCE(trans->aborted, error);
  	WRITE_ONCE(trans->transaction->aborted, error);
fs/btrfs/transaction.h +0 -0
index f1cb05460cec..e1d4e7330198 100644

fs/btrfs/transaction.h static inline bool btrfs_abort_should_print_stack(int error) · line 249

Quoted 9 lines
fs/btrfs/transaction.h +4 -2
@@ -243,20 +243,23 @@ static inline bool btrfs_abort_should_print_stack(int error)
  }
 
  /*
- * Call btrfs_abort_transaction as early as possible when an error condition is
- * detected, that way the exact stack trace is reported for some errors.
+ * Call btrfs_abort_transaction() as early as possible when an error condition
+ * is detected, that way the exact stack trace is reported for some errors.
+ *
+ * Error number must be negative as it encodes wheather it's the first abort.

What about an ASSERT() for it?

It's not that rare to forgot the minus sign for immediately error code like "-EIO".

fs/btrfs/transaction.h static inline bool btrfs_abort_should_print_stack(int error) · line 251

Quoted 2 lines
fs/btrfs/transaction.h +0 -0
   */
  #define btrfs_abort_transaction(trans, error)		\

And I do not like to use macros when inlined function can be used.
There is no special requirement for things like #name, thus this can be converted to a regular static inline function.

Not sure how an static inlined version would affect the text section size though.

Thanks,
Qu

Quoted 26 lines
fs/btrfs/transaction.h +5 -4
  do {								\
-	bool __first = false;					\
+	int __error = (error);					\
+								\
  	/* Report first abort since mount */			\
  	if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,	\
  			&((trans)->fs_info->fs_state))) {	\
-		__first = true;					\
+		__error = -__error;				\
  		WARN_ON(btrfs_abort_should_print_stack(error));	\
  	}							\
  	__btrfs_abort_transaction((trans), __func__,		\
-				  __LINE__, (error), __first);	\
+				  __LINE__, __error);		\
  } while (0)
 
  int btrfs_end_transaction(struct btrfs_trans_handle *trans);
@@ -294,7 +297,7 @@ void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
  void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
  void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
  				      const char *function,
-				      unsigned int line, int error, bool first_hit);
+				      unsigned int line, int error);
 
  int __init btrfs_transaction_init(void);
  void __cold btrfs_transaction_exit(void);
#2 lore raw
Quoted 82 lines · Qu Wenruo
Quoted 67 lines · David Sterba

Optimize the btrfs_abort_transaction() for size as it (by our convention) must be put right after the error condition is detected. The exact file:line is reported so there's a portion that must be inlined. As this is cold code it bloats functions. In previous patch "btrfs: move transaction abort message to __btrfs_abort_transaction()" the error message was moved to the common helper, saving like 20KiB of btrfs.ko and several instructions per call site and some stack space.

There's little left to be optimized, we need to keep the atomic test_and_set_bit() and to convey that as 'first hit' to __btrfs_abort_transaction().

Right now it's a bool, which takes 8 bytes on stack for each call but it's 1 bit of information. We can encode that to some of the other parameters.

For that let's use the 'error' parameter, by convention it's negative errno so we can reliably detect if it's the first hit or a later error. Also the negation is usually implemented by a single instruction (NEG on x86_64) so the resulting object code is kept short.

This reduces btrfs.ko by 8K and stack in several functions by 8 bytes.

Cumulative effect with the other commit is -30K of btrfs.ko. While the encoding is an implementation detail, it's contained within the API. Making the transaction abort calls very light is desired.

---
  fs/btrfs/transaction.c | 13 ++++++++++++-
  fs/btrfs/transaction.h | 15 +++++++++------
  2 files changed, 21 insertions(+), 7 deletions(-)
fs/btrfs/transaction.c +12 -1
index 01f66328966f..bbef3ae86ac4 100644
@@ -2723,12 +2723,23 @@ int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info)
   *
   * We'll complete the cleanup in btrfs_end_transaction and
   * btrfs_commit_transaction.
+ *
+ * Note: the parameter @error encodes whether the transactin abort was first hit
+ *       (setting the FS_ERROR state bit in btrfs_abort_transaction())
+ *       - positive number - first hit
+ *       - negative number - abort after it was already done
   */
  void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
  				      const char *function,
-				      unsigned int line, int error, bool first_hit)
+				      unsigned int line, int error)
  {
  	struct btrfs_fs_info *fs_info = trans->fs_info;
+	bool first_hit = false;
+
+	if (error > 0) {
+		error = -error;
+		first_hit = true;
+	}
 
  	WRITE_ONCE(trans->aborted, error);
  	WRITE_ONCE(trans->transaction->aborted, error);
fs/btrfs/transaction.h +0 -0
index f1cb05460cec..e1d4e7330198 100644

fs/btrfs/transaction.h static inline bool btrfs_abort_should_print_stack(int error) · line 249

Quoted 9 lines
fs/btrfs/transaction.h +4 -2
@@ -243,20 +243,23 @@ static inline bool btrfs_abort_should_print_stack(int error)
  }
 
  /*
- * Call btrfs_abort_transaction as early as possible when an error condition is
- * detected, that way the exact stack trace is reported for some errors.
+ * Call btrfs_abort_transaction() as early as possible when an error condition
+ * is detected, that way the exact stack trace is reported for some errors.
+ *
+ * Error number must be negative as it encodes wheather it's the first abort.

What about an ASSERT() for it?

Quoted 2 lines

It's not that rare to forgot the minus sign for immediately error code like "-EIO".

Yes I can add it, but ASSERT is evaluated at run time so this would not catch "EIO". I have figured out how to do it at compile time, will be in in v2.

Quoted 6 lines

fs/btrfs/transaction.h static inline bool btrfs_abort_should_print_stack(int error) · line 251

Quoted 2 lines
fs/btrfs/transaction.h +0 -0
   */
  #define btrfs_abort_transaction(trans, error)		\

And I do not like to use macros when inlined function can be used. There is no special requirement for things like #name, thus this can be converted to a regular static inline function.

No it can't, static inline changes how __LINE__ and __func__ (and __FILE__ for that matter) work so we'd get reports from the same function where the function is defined. While with macro we have the exact location, this is a standard trick. It's been like that since forever, 49b25e0540904b ("btrfs: enhance transaction abort infrastructure").