Patchwork MIPS Prelinker problem -- possible fix

login
register
mail settings
Submitter Mark Hatle
Date Jan. 5, 2012, 4:07 p.m.
Message ID <4F05CAB9.4020001@windriver.com>
Download mbox | patch
Permalink /patch/18445/
State New
Headers show

Comments

Mark Hatle - Jan. 5, 2012, 4:07 p.m.
I have created a patch (with help from Khem Raj) to resolve a long standing 
issue w/ the MIPS prelinker.  The prelinker was detecting a error condition 
based on what appears to be some incomplete math calculations.

Attached is a patch that I believe will fix the problem, but I would like folks 
on both MIPS and non MIPS system to try it out.

If I don't hear about any issues (and my testing continues to progress 
successfully), I intend to commit it to the upstream cross-prelink repository 
early next week.

--Mark
commit 65b8dfda2bb1d23a194daafec8fdd23e8997e528
Author: Mark Hatle <mark.hatle@windriver.com>
Date:   Wed Jan 4 14:59:00 2012 -0600

    MIPS Prelink fix: Testing do not merge!
    
    This is a possible fix for MIPS prelinking problems as described in the
    yoctoproject bugzilla: http://bugzilla.yoctoproject.org/show_bug.cgi?id=1463
    
    The purpose of this patch is to allow folks to test it before it's
    applied to the upstream sources.
    
    Signed-off-by: Mark Hatle <mark.hatle@windriver.com>
Phil Blundell - Jan. 6, 2012, 1:25 p.m.
On Thu, 2012-01-05 at 10:07 -0600, Mark Hatle wrote:
> I have created a patch (with help from Khem Raj) to resolve a long standing 
> issue w/ the MIPS prelinker.  The prelinker was detecting a error condition 
> based on what appears to be some incomplete math calculations.
> 
> Attached is a patch that I believe will fix the problem, but I would like folks 
> on both MIPS and non MIPS system to try it out.

Thanks.  I tried this out last night and it does seem to solve the
problem for me.  (I was previously using a hack in image-prelink.bbclass
which just deleted all the .mdebug.abi32 sections before prelinking.)

p.

Patch

diff --git a/meta/recipes-devtools/prelink/prelink/prelink-mips.patch b/meta/recipes-devtools/prelink/prelink/prelink-mips.patch
new file mode 100644
index 0000000..d7d4a80
--- /dev/null
+++ b/meta/recipes-devtools/prelink/prelink/prelink-mips.patch
@@ -0,0 +1,81 @@ 
+commit 3233963574d9e6a13b7d6977879c0dfd7caeb310
+Author: Mark Hatle <mark.hatle@windriver.com>
+Date:   Wed Jan 4 14:30:38 2012 -0600
+
+    Fix MIPS non-NOBITS after NOBITS in segment error
+    
+    With recent compilers and linkers it is typical for a MIPS binary to
+    be built with the sections such as:
+    
+      [23] .got              PROGBITS        004aafb0 09afb0 00055c 04 WAp  0   0 16
+      [24] .sdata            PROGBITS        004ab50c 09b50c 000004 00 WAp  0   0  4
+      [25] .sbss             NOBITS          004ab510 09b510 00003d 00 WAp  0   0  4
+      [26] .bss              NOBITS          004ab550 09b510 0021cc 00  WA  0   0 16
+      [27] .dynstr           STRTAB          004bb510 09b510 000b99 00   A  0   0  1
+      [28] .gnu.attributes   LOOS+ffffff5    00000000 09c0a9 000010 00      0   0  1
+      [29] .mdebug.abi32     PROGBITS        004ad71c 09c0b9 000000 00      0   0  1
+      [30] .gnu_debuglink    PROGBITS        00000000 09c0b9 00000c 00      0   0  1
+      [31] .gnu.prelink_undo PROGBITS        00000000 09c0c8 0005bc 01      0   0  4
+    
+    When determining if a section is within a segment, simply checking
+    that it fits within the boundries of that segment is not enough.  In
+    the above example, the .mdebug.abi32 fits within the boundries of the
+    third segment.  However with a size of 0 there is nothing for the
+    prelinker to do.  The validation that non-NOBITS sections can't come
+    after a PROGBITS fails, because it incorectly assumes that it is
+    part of the same section as .sbss, and .bss.
+    
+    Using elfutils readelf as a guide, a check was missing in the mapping
+    function that the section had a size > 0.  Binutils has a similar
+    check, but it is a bit more specific that the address of a section
+    can not start at the end of a segment.  The size > 0 was an easier
+    and more obvious check, which is why it was used.
+    
+    2012-01-04  Mark Hatle <mark.hatle@windriver.com>
+            * exec.c: Check that a section is larger then 0 bytes when
+              determining the section to segment mapping.  This matches
+              the behavior of elfutils - readelf.  Otherwise an empty
+              PROGBITS section at the end of a segment will cause a
+              failure.
+    
+    Signed-off-by: Mark Hatle <mark.hatle@windriver.com>
+
+diff --git a/trunk/ChangeLog.cross b/trunk/ChangeLog.cross
+index ccb974c..6e350f9 100644
+--- trunk/ChangeLog.cross
++++ trunk/ChangeLog.cross
+@@ -1,3 +1,10 @@
++2012-01-04  Mark Hatle <mark.hatle@windriver.com>
++	* exec.c: Check that a section is larger then 0 bytes when
++	  determining the section to segment mapping.  This matches
++	  the behavior of elfutils - readelf.  Otherwise an empty
++	  PROGBITS section at the end of a segment will cause a
++	  failure.
++
+ 2011-12-08  Mark Hatle <mark.hatle@windriver.com>
+ 	* rtld/rtld.c: Fix an issue where missing objects would trigger
+ 	  an assert in dl-version.c
+diff --git a/trunk/src/exec.c b/trunk/src/exec.c
+index 7c8f38f..cf38b72 100644
+--- trunk/src/exec.c
++++ trunk/src/exec.c
+@@ -500,7 +500,8 @@ error_out:
+ 	int sfirst = 0, slast = 0, last = 0;
+ 
+ 	for (j = 1; j < dso->ehdr.e_shnum; ++j)
+-	  if (dso->shdr[j].sh_addr >= dso->phdr[i].p_vaddr
++	  if (dso->shdr[j].sh_size > 0
++	      && dso->shdr[j].sh_addr >= dso->phdr[i].p_vaddr
+ 	      && dso->shdr[j].sh_addr + dso->shdr[j].sh_size
+ 		 <= dso->phdr[i].p_vaddr + dso->phdr[i].p_memsz)
+ 	    {
+@@ -572,7 +573,8 @@ error_out:
+ 	  }
+ 
+ 	for (j = 1; j < dso->ehdr.e_shnum; ++j)
+-	  if (dso->shdr[j].sh_addr >= dso->phdr[i].p_vaddr
++	  if (dso->shdr[j].sh_size > 0
++	      && dso->shdr[j].sh_addr >= dso->phdr[i].p_vaddr
+ 	      && dso->shdr[j].sh_addr + dso->shdr[j].sh_size
+ 		 <= dso->phdr[i].p_vaddr + dso->phdr[i].p_memsz)
+ 	    {
diff --git a/meta/recipes-devtools/prelink/prelink_git.bb b/meta/recipes-devtools/prelink/prelink_git.bb
index 17141e9..aa5d9a4 100644
--- a/meta/recipes-devtools/prelink/prelink_git.bb
+++ b/meta/recipes-devtools/prelink/prelink_git.bb
@@ -10,7 +10,7 @@  LICENSE = "GPLv2"
 LIC_FILES_CHKSUM = "file://COPYING;md5=c93c0550bd3173f4504b2cbd8991e50b"
 SRCREV = "bb1b660c5e3859b6c5a2ac8d739713e9989a4dd7"
 PV = "1.0+git${SRCPV}"
-PR = "r8"
+PR = "r8.1"
 
 #
 # The cron script attempts to re-prelink the system daily -- on
@@ -31,7 +31,8 @@  SRC_URI = "git://git.yoctoproject.org/prelink-cross.git;protocol=git \
            file://prelink.conf \
            file://prelink.cron.daily \
            file://prelink.default \
-	   file://macros.prelink"
+	   file://macros.prelink \
+	   file://prelink-mips.patch"
 
 TARGET_OS_ORIG := "${TARGET_OS}"
 OVERRIDES_append = ":${TARGET_OS_ORIG}"