Skip to contentSkip to conversation

pcre2: add patch fixing a build issue on m68k

PATCH New Thomas Petazzoni 2 participants 4 messages Feb 26, 2018 – Feb 27, 2018

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pcre2: add patch fixing a build issue on m68k

A pretty nasty difference in alignment rules between most architectures and m68k lead pcre2 to fail building its 16-bit and 32-bit variants on m68k. This commit adds a patch that fixes that.

Fixes:

http://autobuild.buildroot.net/results/f6a45df5cb80e3fd94d57163bd28a0014a02bf4/

---
 package/pcre2/0001-fix-heapframe-alignment.patch | 47 ++++++++++++++++++++++++
 1 file changed, 47 insertions(+)
 create mode 100644 package/pcre2/0001-fix-heapframe-alignment.patch
package/pcre2/0001-fix-heapframe-alignment.patch new file +47 -0
new file mode 100644
index 0000000000..3a9bd8ec47
@@ -0,0 +1,47 @@
+src/pcre2_intmodedep.h: fix alignment of fields in struct heapframe
+
+pcre2_intmodedep.h has a check to verify that the size of the struct
+heapframe is a multiple of 4 bytes. On most architectures this works
+fine, but not on m68k. Indeed, when building the 16-bit variant of
+pcre2, the heapframe structure contains:
+
+ PCRE2_UCHAR occu[2];
+ PCRE2_SPTR  eptr;
+
+Where PCRE2_UCHAR is a 16-bit data type, and PCRE2_SPTR is a
+pointer. The occu[] array starts at byte 0x32, so not aligned on a
+32-bit boundary. With 2 x 16-bit, the occur[] array ends at byte 0x36.
+
+Now, on most architectures, the alignment required for a pointer will
+make the eptr field start at 0x38 (on 32 bit architectures). However,
+on m68k, it is fine to have a pointer aligned only on a 16-bit
+boundary, and the eptr pointer will be at offset 0x36.
+
+This doesn't cause a problem per-se, but breaks the check that
+heapframe should be a multiple of 4 bytes.
+
+To fix this, we make sure eptr is aligned by introducing an unused
+field of 16 bits before the occu[] array
+
+Fixes the following build failure on m68k:
+
+src/pcre2_intmodedep.h:818:14: error: size of array 'check_heapframe_size' is negative
+ typedef char check_heapframe_size[
+              ^~~~~~~~~~~~~~~~~~~~
+
+Signed-off-by: Thomas Petazzoni <[email protected]>
+Upstream: https://bugs.exim.org/show_bug.cgi?id=2247
+
+Index: src/pcre2_intmodedep.h
+===================================================================
+--- a/src/pcre2_intmodedep.h	(revision 923)
++++ b/src/pcre2_intmodedep.h	(working copy)
+@@ -797,8 +797,10 @@
+   PCRE2_UCHAR occu[6];       /* Used for other case code units */
+ #elif PCRE2_CODE_UNIT_WIDTH == 16
+   PCRE2_UCHAR occu[2];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */
+ #else
+   PCRE2_UCHAR occu[1];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */
+ #endif
Signature · 1 line
2.14.3
#1 lore raw

PATCH 1/1 pcre2: add patch fixing a build issue on m68k

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Thomas" == Thomas Petazzoni <[email protected]> writes:

Hi,

Quoted 6 lines

A pretty nasty difference in alignment rules between most architectures and m68k lead pcre2 to fail building its 16-bit and 32-bit variants on m68k. This commit adds a patch that fixes that.

package/pcre2/0001-fix-heapframe-alignment.patch new file +2 -0
+Index: src/pcre2_intmodedep.h
+===================================================================

package/pcre2/0001-fix-heapframe-alignment.patch line 46

Quoted 10 lines
package/pcre2/0001-fix-heapframe-alignment.patch new file +10 -0
+--- a/src/pcre2_intmodedep.h	(revision 923)
++++ b/src/pcre2_intmodedep.h	(working copy)
+@@ -797,8 +797,10 @@
+   PCRE2_UCHAR occu[6];       /* Used for other case code units */
+ #elif PCRE2_CODE_UNIT_WIDTH == 16
+   PCRE2_UCHAR occu[2];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */
+ #else
+   PCRE2_UCHAR occu[1];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */

I haven't looked at the code, but doesn't this #else refer to the 32bit variant, which may require 4 byte alignment depending on architecture, so the 2 bytes padding you are adding would cause eptr to move to offset 0x3c instead of 0x38?

That in itself may not be an issue, but it is a change of behaviour.

Signature · 1 line
Bye, Peter Korsgaard
#2 lore raw

PATCH 1/1 pcre2: add patch fixing a build issue on m68k

Hello,

Quoted 17 lines · Peter Korsgaard
Quoted 6 lines

A pretty nasty difference in alignment rules between most architectures and m68k lead pcre2 to fail building its 16-bit and 32-bit variants on m68k. This commit adds a patch that fixes that.

package/pcre2/0001-fix-heapframe-alignment.patch new file +2 -0
+Index: src/pcre2_intmodedep.h
+===================================================================

package/pcre2/0001-fix-heapframe-alignment.patch line 46

Quoted 10 lines
package/pcre2/0001-fix-heapframe-alignment.patch new file +10 -0
+--- a/src/pcre2_intmodedep.h	(revision 923)
++++ b/src/pcre2_intmodedep.h	(working copy)
+@@ -797,8 +797,10 @@
+   PCRE2_UCHAR occu[6];       /* Used for other case code units */
+ #elif PCRE2_CODE_UNIT_WIDTH == 16
+   PCRE2_UCHAR occu[2];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */
+ #else
+   PCRE2_UCHAR occu[1];       /* Used for other case code units */
++  uint8_t     unused[2];     /* Ensure 32 bit alignment */  
Quoted 4 lines

I haven't looked at the code, but doesn't this #else refer to the 32bit variant, which may require 4 byte alignment depending on architecture, so the 2 bytes padding you are adding would cause eptr to move to offset 0x3c instead of 0x38?

Without this additional unused[2] for the 32-bit variant, pcre2 32-bit variant would not build on m68k.

Indeed, the occu[] array starts at offset 0x32, which is a multiple of 2, but not a multiple of 4. Then you add a single PCRE2_UCHAR, which in the 32-bit variant is 32-bit long, you end up with the eptr pointer at offset 0x32 + 4, which is 0x36, which again is a multiple of 2, but not a multiple of 4.

By adding these additional two bytes, I make sure that the eptr field is at offset 0x38. This is why the unused[2] in the #else case is needed for the m68k to build.

However, it is true that on non-m68k platform, the single PCRE2_UCHAR occu[1] will probably be aligned on a 32-bit boundary, which indeed would shift the rest of the structure.

So perhaps I should instead do something like this:

uint8_t return_id;
uint8_t op;

#if PCRE2_CODE_UNIT_WIDTH == 8
PCRE2_UCHAR occu[6]; /* Used for other case code units */
#elif PCRE2_CODE_UNIT_WIDTH == 16
PCRE2_UCHAR occu[2]; /* Used for other case code units */
uint8_t unused[2]; /* Ensure 32 bit alignment */
#else
uint8_t unused[2]; /* Ensure 32 bit alignment */
PCRE2_UCHAR occu[1]; /* Used for other case code units */
#endif

PCRE2_SPTR eptr;

This should give the same offset for the different fields, regardless of the architecture, correct ?

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com
#3 lore raw

PATCH 1/1 pcre2: add patch fixing a build issue on m68k

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Thomas" == Thomas Petazzoni <[email protected]> writes:

Hi,

Quoted 19 lines

However, it is true that on non-m68k platform, the single PCRE2_UCHAR occu[1] will probably be aligned on a 32-bit boundary, which indeed would shift the rest of the structure.

So perhaps I should instead do something like this:

uint8_t return_id;
uint8_t op;

#if PCRE2_CODE_UNIT_WIDTH == 8
PCRE2_UCHAR occu[6]; /* Used for other case code units */
#elif PCRE2_CODE_UNIT_WIDTH == 16
PCRE2_UCHAR occu[2]; /* Used for other case code units */
uint8_t unused[2]; /* Ensure 32 bit alignment */
#else
uint8_t unused[2]; /* Ensure 32 bit alignment */
PCRE2_UCHAR occu[1]; /* Used for other case code units */
#endif

Quoted 4 lines

PCRE2_SPTR eptr;

This should give the same offset for the different fields, regardless of the architecture, correct ?

Yes, indeed!

Signature · 1 line
Bye, Peter Korsgaard