661800 - elfhack/elf.cpp:285: Elf::Elf(std::ifstream&): Assertion `segment->getFileSize() == phdr.p_filesz' failed.

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Octoploid]

User-Agent:       Mozilla/5.0 (X11; Linux x86_64; rv:7.0a1) Gecko/20100101 Firefox/7.0a1
Build Identifier: 

I've hit the following assertion in elfhack:

===
=== If you get failures below, please file a bug describing the error
=== and your environment (compiler and linker versions), and use
=== --disable-elf-hack until this is fixed.
===
cd ../../dist/bin; find . -name "*.so" | xargs ../../build/unix/elfhack/elfhack
./libnspr4.so: Already elfhacked. Skipping
./libplds4.so: No gain. Skipping
./libplc4.so: No gain. Skipping
./libmozsqlite3.so: Reduced by 8032 bytes
./libnssdbm3.so: No gain. Skipping
./libsmime3.so: No gain. Skipping
./libxpcom.so: No gain. Skipping
./libmozalloc.so: No gain. Skipping
./components/libmozgnome.so: No gain. Skipping
./components/libdbusservice.so: No gain. Skipping
./components/libnkgnomevfs.so: No gain. Skipping
./components/libbrowsercomps.so: Reduced by 12128 bytes
./libnssckbi.so: Already elfhacked. Skipping
./libsoftokn3.so: Already elfhacked. Skipping
./libnss3.so: Already elfhacked. Skipping
./libnssutil3.so: Already elfhacked. Skipping
./libssl3.so: Already elfhacked. Skipping
./libfreebl3.so: Already elfhacked. Skipping
elfhack: /var/tmp/mozilla-central/build/unix/elfhack/elf.cpp:285: Elf::Elf(std::ifstream&): Assertion `segment->getFileSize() == phdr.p_filesz' failed.
xargs: ../../build/unix/elfhack/elfhack: terminated by signal 6
make[3]: *** [elfhack] Error 125

This only happens with the latest gcc-4.7.0 during a profiledbuild.
A normal build works without problems.

Reproducible: Always

Comment 1

[Mike Hommey [:glandium]]

could you attach the file on which elfhack fails? (I guess it is libxul.so)

Comment 2

[Octoploid]

Attachment: libnspr4.so

Both libnspr4.so and libxul.so trigger the same assertion.
I've attached libnspr4.so, because it's smaller.

Comment 3

[Octoploid]

In case someone is wondering about the "Already elfhacked" messages
above. They appear because I forgot to delete the build directory (
that contained a conventional build) before starting the profiledbuild.

Comment 4

[Mike Hommey [:glandium]]

The problem is that elfhack believes the .dynamic section is in the TLS segment. What's weird is that I'd expect that to be caught by the test case. Could you attach $objdir/build/unix/elfhack/test.so as well?

Comment 5

[Octoploid]

Attachment: test.so

Comment 6

[Octoploid]

Attachment: test.so.bak

Comment 7

[Mike Hommey [:glandium]]

Attachment: Avoid elfhack considering non SHF_TLS sections as part of PT_TLS segment

Comment 8

[Octoploid]

This fixes the problem for me. Many thanks Mike.

Comment 9

[(dormant account)]

Comment on attachment 537290 [details] [diff] [review]
Avoid elfhack considering non SHF_TLS sections as part of PT_TLS segment

rubberstamp

Comment 10

[Mike Hommey [:glandium]]

http://hg.mozilla.org/mozilla-central/rev/d95b370f0f85

Comment 11

[Mike Hommey [:glandium]]

Backed out because of bug 664362
http://hg.mozilla.org/mozilla-central/rev/f6f801cfdf4c

Comment 12

[Mike Hommey [:glandium]]

Alternatively, landing with a clobber forcing thingy would work as well.

Comment 13

[Mike Hommey [:glandium]]

Attachment: part 0 - Add a dependency of host_elf*.o upon elfxx.h, to circumvent bug 664362.

In the end, it's much simpler to do it locally.

Comment 14

[(not currently active) Ted Mielczarek]

Comment on attachment 539454 [details] [diff] [review]
part 0 - Add a dependency of host_elf*.o upon elfxx.h, to circumvent bug 664362.

Review of attachment 539454 [details] [diff] [review]:
-----------------------------------------------------------------

Comment 15

[Mike Hommey [:glandium]]

http://hg.mozilla.org/integration/mozilla-inbound/rev/0640c9ecfba2
http://hg.mozilla.org/integration/mozilla-inbound/rev/6e3813929a57

Comment 16

[Mike Hommey [:glandium]]

Merged in m-c
http://hg.mozilla.org/mozilla-central/rev/0640c9ecfba2
http://hg.mozilla.org/mozilla-central/rev/6e3813929a57

Comment 17

[Tim Taubert [:ttaubert] (inactive)]

After this got merged to m-c, tip doesn't build for me:

=== If you get failures below, please file a bug describing the error
=== and your environment (compiler and linker versions), and use
=== --disable-elf-hack until this is fixed.
===
/home/tim/workspace/mozilla-central/objdir-ff-release/build/unix/elfhack/elfhack -b test.so
test.so: terminate called after throwing an instance of 'std::runtime_error'
  what():  PT_GNU_RELRO segment doesn't start on a section start

I guess this is related?

Comment 18

[Patrick McManus [:mcmanus]]

(In reply to comment #17)
> After this got merged to m-c, tip doesn't build for me:
> 

The tip does not (clobber) build for me either (5a46d17c72ca - 64bit linux)

===
=== If you get failures below, please file a bug describing the error
=== and your environment (compiler and linker versions), and use
=== --disable-elf-hack until this is fixed.
===
/home/mcmanus/src/mozilla2/wd/obj-debug-661036/build/unix/elfhack/elfhack -b test.so
test.so: terminate called after throwing an instance of 'std::runtime_error'
  what():  PT_GNU_RELRO segment doesn't start on a section start

Comment 19

[Mike Hommey [:glandium]]

this is bug 664366, and the most likely relationship between this and that bug is that the test case change triggers the ld bug.