Skip to contentSkip to conversation

fat: fix unaligned errors

PATCH New Mike Frysinger 3 participants 9 messages Jan 2, 2009 – Jan 24, 2009 VFAT/FAT/msdos filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fat: fix unaligned errors

From: Bryan Wu <[email protected]>

A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

---
 fs/fat/fat.c |   19 +++++++++++++++----
 1 files changed, 15 insertions(+), 4 deletions(-)
fs/fat/fat.c +15 -4
index a9dde7d..06736d9 100644
@@ -429,12 +429,16 @@ slot2str(dir_slot *slotptr, char *l_name, int *idx)
 
 
 /*
+ * We need align this buffer to 16bit, cause it will be cast up to
+ * a dir_slot structure
+ */
+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));
+/*
  * Extract the full long filename starting at 'retdent' (which is really
  * a slot) into 'l_name'. If successful also copy the real directory entry
  * into 'retdent'
  * Return 0 on success, -1 otherwise.
  */
-__u8	 get_vfatname_block[MAX_CLUSTSIZE];
 static int
 get_vfatname(fsdata *mydata, int curclust, __u8 *cluster,
 	     dir_entry *retdent, char *l_name)
@@ -517,10 +521,14 @@ mkcksum(const char *str)
 
 
 /*
+ * We need align this buffer to 32bit, cause it will be cast up to
+ * a dir_entry structure
+ */
+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));
+/*
  * Get the directory entry associated with 'filename' from the directory
  * starting at 'startsect'
  */
-__u8 get_dentfromdir_block[MAX_CLUSTSIZE];
 static dir_entry *get_dentfromdir (fsdata * mydata, int startsect,
 				   char *filename, dir_entry * retdent,
 				   int dols)
@@ -725,8 +733,11 @@ read_bootsectandvi(boot_sector *bs, volume_info *volinfo, int *fatsize)
 	return -1;
 }
 
-
-__u8 do_fat_read_block[MAX_CLUSTSIZE];  /* Block buffer */
+/*
+ * We need align this buffer to 32bit, cause it will be cast up to
+ * a dir_entry structure
+ */
+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));
 long
 do_fat_read (const char *filename, void *buffer, unsigned long maxsize,
 	     int dols)
Signature · 1 line
1.6.0.6
#1 lore raw

PATCH 1/1 fat: fix unaligned errors

Hello Mike,

2009/1/2 Mike Frysinger <[email protected]>:

Quoted 5 lines

From: Bryan Wu <[email protected]>
A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

Just curious (because before the Christmas holidays I was also debugging some FAT problems, and maybe this is related)
What problem/symptoms does this patch fix? How did the bug show itself?

Kind Regards,

Remy

Quoted 67 lines
---
 fs/fat/fat.c |   19 +++++++++++++++----
 1 files changed, 15 insertions(+), 4 deletions(-)
fs/fat/fat.c +15 -5
index a9dde7d..06736d9 100644
@@ -429,12 +429,16 @@ slot2str(dir_slot *slotptr, char *l_name, int *idx)
 
 
 /*
+ * We need align this buffer to 16bit, cause it will be cast up to
+ * a dir_slot structure
+ */
+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));
+/*
 * Extract the full long filename starting at 'retdent' (which is really
 * a slot) into 'l_name'. If successful also copy the real directory entry
 * into 'retdent'
 * Return 0 on success, -1 otherwise.
 */
-__u8    get_vfatname_block[MAX_CLUSTSIZE];
 static int
 get_vfatname(fsdata *mydata, int curclust, __u8 *cluster,
            dir_entry *retdent, char *l_name)
@@ -517,10 +521,14 @@ mkcksum(const char *str)
 
 
 /*
+ * We need align this buffer to 32bit, cause it will be cast up to
+ * a dir_entry structure
+ */
+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));
+/*
 * Get the directory entry associated with 'filename' from the directory
 * starting at 'startsect'
 */
-__u8 get_dentfromdir_block[MAX_CLUSTSIZE];
 static dir_entry *get_dentfromdir (fsdata * mydata, int startsect,
                                  char *filename, dir_entry * retdent,
                                  int dols)
@@ -725,8 +733,11 @@ read_bootsectandvi(boot_sector *bs, volume_info *volinfo, int *fatsize)
       return -1;
 }
 
-
-__u8 do_fat_read_block[MAX_CLUSTSIZE];  /* Block buffer */
+/*
+ * We need align this buffer to 32bit, cause it will be cast up to
+ * a dir_entry structure
+ */
+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));
 long
 do_fat_read (const char *filename, void *buffer, unsigned long maxsize,
            int dols)
--

1.6.0.6

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

#2 lore raw

PATCH 1/1 fat: fix unaligned errors

Quoted 10 lines · On Friday 02 January 2009 04:32:39 Remy Bohmer wrote:

2009/1/2 Mike Frysinger <[email protected]>:

Quoted 5 lines

From: Bryan Wu <[email protected]>
A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

Just curious (because before the Christmas holidays I was also debugging some FAT problems, and maybe this is related)
What problem/symptoms does this patch fix? How did the bug show itself?

i think should cover all your questions ...
http://blackfin.uclinux.org/gf/tracker/4765
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20090102/19a9146d/attachment.pgp

#3 lore raw

PATCH 1/1 fat: fix unaligned errors

In reply to Mike Frysinger

Dear Mike Frysinger,

Quoted 6 lines · In message <[email protected]> you wrote:

From: Bryan Wu <[email protected]>

A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

...

Quoted 1 line

+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));

...

Quoted 1 line

+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

...

Quoted 1 line

+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

What makes you sure that a 16 resp. 32 bit alignment is sufficient, and that gcc does not decide to align such structures even stricter?

Wouldn't it make more sense to use "__alignof__ ()" here to be on the safe side?

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
One friend in a lifetime is much; two  are  many;  three  are  hardly
possible. Friendship needs a certain parallelism of life, a community
of thought, a rivalry of aim.                     - Henry Brook Adams
#4 lore raw

PATCH 1/1 fat: fix unaligned errors

Quoted 18 lines · On Friday 02 January 2009 17:20:47 Wolfgang Denk wrote:
Quoted 6 lines · In message <[email protected]> you wrote:

From: Bryan Wu <[email protected]>

A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

...

Quoted 2 lines

+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));

...

Quoted 2 lines

+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

Quoted 7 lines

...

Quoted 2 lines

+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

What makes you sure that a 16 resp. 32 bit alignment is sufficient, and that gcc does not decide to align such structures even stricter?

read the structures in question. the largest member is 16bit/32bit. i really dont know what other alignment gcc could randomly generate.

Quoted 2 lines

Wouldn't it make more sense to use "__alignof__ ()" here to be on the safe side?

i found no such alignment attribute in the gcc docs which is why i suggested Bryan use aligned(sizeof(...)).
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Function-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Variable-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Type-Attributes.html
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20090102/1ce92c0d/attachment.pgp

#5 lore raw

PATCH 1/1 fat: fix unaligned errors

Dear Mike,

Quoted 7 lines · In message <[email protected]> you wrote:
Quoted 7 lines
Quoted 3 lines

+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));

...

Quoted 2 lines

+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

...

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

What makes you sure that a 16 resp. 32 bit alignment is sufficient, and that gcc does not decide to align such structures even stricter?

read the structures in question. the largest member is 16bit/32bit. i really dont know what other alignment gcc could randomly generate.

It is not (only) the size of the "largest" member in a struct which determins global alignment, but also the size of the struct itself.

See for example
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Type-Attributes.html

Read especially this part:

In the example above, if the size of each short is 2 bytes, then the size of the entire struct S type is 6 bytes. The smallest power of two which is greater than or equal to that is 8, so the compiler sets the alignment for the entire struct S type to 8 bytes.

Quoted 8 lines
Quoted 2 lines

Wouldn't it make more sense to use "__alignof__ ()" here to be on the safe side?

i found no such alignment attribute in the gcc docs which is why i suggested Bryan use aligned(sizeof(...)).
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Function-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Variable-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Type-Attributes.html

Maybe you should have searched for the (IMHO pretty obvious) term "Alignment" as well, i.e. look at
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Alignment.html#Alignment

Quote:

The keyword __alignof__ allows you to inquire about how an object is aligned, or the minimum alignment usually required by a type. Its syntax is just like sizeof.

Best regards,

Wolfgang Denk

Signature · 4 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The more sins you confess, the more books you will sell.
#6 lore raw

PATCH 1/1 fat: fix unaligned errors

Quoted 26 lines · On Friday 02 January 2009 18:48:50 Wolfgang Denk wrote:
Quoted 10 lines · In message <[email protected]> you wrote:
Quoted 10 lines
Quoted 2 lines

+__u8 get_vfatname_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u16))));

...

Quoted 2 lines

+__u8 get_dentfromdir_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

...

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

+__u8 do_fat_read_block[MAX_CLUSTSIZE] __attribute__ ((aligned(sizeof(__u32))));

What makes you sure that a 16 resp. 32 bit alignment is sufficient, and that gcc does not decide to align such structures even stricter?

read the structures in question. the largest member is 16bit/32bit. i really dont know what other alignment gcc could randomly generate.

It is not (only) the size of the "largest" member in a struct which determins global alignment, but also the size of the struct itself.

See for example
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Type-Attributes.html

Quoted 7 lines

Read especially this part:

In the example above, if the size of each short is 2 bytes, then the size of the entire struct S type is 6 bytes. The smallest power of two which is greater than or equal to that is 8, so the compiler sets the alignment for the entire struct S type to 8 bytes.

i was not aware of this, thanks for the highlight

Quoted 13 lines
Quoted 8 lines
Quoted 2 lines

Wouldn't it make more sense to use "__alignof__ ()" here to be on the safe side?

i found no such alignment attribute in the gcc docs which is why i suggested Bryan use aligned(sizeof(...)).
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Function-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Variable-Attributes.html
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Type-Attributes.html

Maybe you should have searched for the (IMHO pretty obvious) term "Alignment" as well, i.e. look at
http://gcc.gnu.org/onlinedocs/gcc-4.3.2/gcc/Alignment.html#Alignment

Quoted 5 lines

Quote:

The keyword __alignof__ allows you to inquire about how an object is aligned, or the minimum alignment usually required by a type. Its syntax is just like sizeof.

well i guess we cant all be "Wolfgang Denk"s
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20090102/b1503012/attachment-0001.pgp

#7 lore raw

PATCH v2 1/1 fat: fix unaligned errors

In reply to Wolfgang Denk

From: Bryan Wu <[email protected]>

A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

---
v1
- use alignof() rather than sizeof()

 fs/fat/fat.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)
fs/fat/fat.c +5 -3
index a9dde7d..28c7805 100644
@@ -434,7 +434,8 @@ slot2str(dir_slot *slotptr, char *l_name, int *idx)
  * into 'retdent'
  * Return 0 on success, -1 otherwise.
  */
-__u8	 get_vfatname_block[MAX_CLUSTSIZE];
+__attribute__ ((__aligned__(__alignof__(dir_entry))))
+__u8 get_vfatname_block[MAX_CLUSTSIZE];
 static int
 get_vfatname(fsdata *mydata, int curclust, __u8 *cluster,
 	     dir_entry *retdent, char *l_name)
@@ -520,6 +521,7 @@ mkcksum(const char *str)
  * Get the directory entry associated with 'filename' from the directory
  * starting at 'startsect'
  */
+__attribute__ ((__aligned__(__alignof__(dir_entry))))
 __u8 get_dentfromdir_block[MAX_CLUSTSIZE];
 static dir_entry *get_dentfromdir (fsdata * mydata, int startsect,
 				   char *filename, dir_entry * retdent,
@@ -725,8 +727,8 @@ read_bootsectandvi(boot_sector *bs, volume_info *volinfo, int *fatsize)
 	return -1;
 }
 
-
-__u8 do_fat_read_block[MAX_CLUSTSIZE];  /* Block buffer */
+__attribute__ ((__aligned__(__alignof__(dir_entry))))
+__u8 do_fat_read_block[MAX_CLUSTSIZE];
 long
 do_fat_read (const char *filename, void *buffer, unsigned long maxsize,
 	     int dols)
Signature · 1 line
1.6.0.6
#8 lore raw

PATCH v2 1/1 fat: fix unaligned errors

Dear Mike Frysinger,

Quoted 7 lines · In message <[email protected]> you wrote:

From: Bryan Wu <[email protected]>

A couple of buffers in the fat code are declared as an array of bytes. But it is then cast up to a structure with 16bit and 32bit members. Since GCC assumes structure alignment here, we have to force the buffers to be aligned according to the structure usage.

Quoted 8 lines

---
v1
- use alignof() rather than sizeof()

 fs/fat/fat.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)

Applied, thanks.

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
"There is such a fine line between genius and stupidity."
- David St. Hubbins, "Spinal Tap"