Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

segfault on ppc64le when compiling master using nightly-2016-10-19 #37320

Closed
amboar opened this issue Oct 21, 2016 · 7 comments
Closed

segfault on ppc64le when compiling master using nightly-2016-10-19 #37320

amboar opened this issue Oct 21, 2016 · 7 comments

Comments

@amboar
Copy link

amboar commented Oct 21, 2016

Hi,

I see that nightlies and betas are now up for ppc{,64{,le}}! I've had a crack at building master on a ppc64le host using nightly-2016-10-19 as stage0, but no luck:

rustc: powerpc64le-unknown-linux-gnu/stage0/lib/rustlib/powerpc64le-unknown-linux-gnu/lib/libcore
compile: powerpc64le-unknown-linux-gnu/rustllvm/RustWrapper.o
compile: powerpc64le-unknown-linux-gnu/rustllvm/PassWrapper.o
compile: powerpc64le-unknown-linux-gnu/rustllvm/ArchiveWrapper.o
Segmentation fault
/home/arj/rust/mk/target.mk:191: recipe for target 'powerpc64le-unknown-linux-gnu/stage0/lib/rustlib/powerpc64le-unknown-linux-gnu/lib/stamp.core' failed
make: *** [powerpc64le-unknown-linux-gnu/stage0/lib/rustlib/powerpc64le-unknown-linux-gnu/lib/stamp.core] Error 139
make: *** Waiting for unfinished jobs....

My naive gdb bt gave:

$ CFG_LLVM_LINKAGE_FILE=/home/arj/rust/powerpc64le-unknown-linux-gnu/rt/llvmdeps.rs LD_LIBRARY_PATH=/home/arj/rust/powerpc64le-unknown-linux-gnu/st[30/1939]
/home/arj/rust/powerpc64le-unknown-linux-gnu/llvm/lib:$LD_LIBRARY_PATH  gdb powerpc64le-unknown-linux-gnu/stage0/bin/rustc
Reading symbols from powerpc64le-unknown-linux-gnu/stage0/bin/rustc...(no debugging symbols found)...done.
(gdb) run --cfg stage0  -O --cfg rtopt -C rpath -C prefer-dynamic -C no-stack-check --target=powerpc64le-unknown-linux-gnu   -L "powerpc64le-unknown-linux-gnu/rt" -L native="/h
ome/arj/rust/powerpc64le-unknown-linux-gnu/llvm/lib"     --out-dir powerpc64le-unknown-linux-gnu/stage0/lib/rustlib/powerpc64le-unknown-linux-gnu/lib -C extra-filename=-3d98e1d
d -C metadata=3d98e1dd src/libcore/lib.rs
Starting program: /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/rustc --cfg stage0  -O --cfg rtopt -C rpath -C prefer-dynamic -C no-stack-check --target=powerpc64le-u
nknown-linux-gnu   -L "powerpc64le-unknown-linux-gnu/rt" -L native="/home/arj/rust/powerpc64le-unknown-linux-gnu/llvm/lib"     --out-dir powerpc64le-unknown-linux-gnu/stage0/li
b/rustlib/powerpc64le-unknown-linux-gnu/lib -C extra-filename=-3d98e1dd -C metadata=3d98e1dd src/libcore/lib.rs
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/powerpc64le-linux-gnu/libthread_db.so.1".
[New Thread 0x3fffb2dfee80 (LWP 17275)]

Thread 2 "rustc" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x3fffb2dfee80 (LWP 17275)]
0x00003fffb7e25e48 in drop::hf6b6fd210b1b8022 () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
(gdb) bt
#0  0x00003fffb7e25e48 in drop::hf6b6fd210b1b8022 () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#1  0x00003fffb7e23fe8 in drop::he6b5c1de4328d7c4 () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#2  0x00003fffb7e1b144 in drop::ha6e76cd3e2aaf918 () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#3  0x00003fffb7e25ea4 in drop::hf6b6fd210b1b8022 () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#4  0x00003fffb7e145a8 in drop::h6c5bef721832e06f () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#5  0x00003fffb7eb4f4c in rustc_driver::driver::compile_input::h67df2cb32ed2d9fe ()
   from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#6  0x00003fffb7ed9574 in rustc_driver::run_compiler::hf62e42444e71fbda ()
   from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#7  0x00003fffb7e005bc in std::panicking::try::do_call::h0565df63f4851a92 ()
   from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#8  0x00003fffb7d088c8 in __rust_maybe_catch_panic () from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/libstd-17342542cc541012.so
#9  0x00003fffb7e26724 in _$LT$F$u20$as$u20$alloc..boxed..FnBox$LT$A$GT$$GT$::call_box::hebe081d66cfa068a ()
   from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/librustc_driver-1ed44d7390b9470e.so
#10 0x00003fffb7cfb3f8 in std::sys::thread::Thread::new::thread_start::hbaf2dd6b96963058 ()
   from /home/arj/rust/powerpc64le-unknown-linux-gnu/stage0/bin/../lib/libstd-17342542cc541012.so
#11 0x00003fffb34a84a0 in start_thread (arg=0x3fffb2dfee80) at pthread_create.c:335
#12 0x00003fffb7ba7e74 in clone () at ../sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S:96

Happy to dig in further if the debug symbols are available somewhere.
@alexcrichton
Copy link
Member

Note that for the most part we're just producing cross compilers, not actually running them. In that sense it may be possible that they just segfault on startup (as this appears to be doing) and it may be a legit bug in trans or something like that (unsure)

@nagisa
Copy link
Member

nagisa commented Oct 22, 2016

T-compiler. Might also want a powerpc tag of some sort.

@amboar
Copy link
Author

amboar commented Oct 24, 2016

The cross-build I created a while back didn't have this issue. I plan on cross-building again when I have a moment to see if I can reproduce. Off the top of my head two differences are they were built from different commits on master, and my initial cross-build was done with the standard build system, not rustbuild. I'll try to iron out whether either of those has had an effect.

@cuviper
Copy link
Member

cuviper commented Oct 25, 2016

I believe PPC has multiple possible page sizes -- perhaps this is a case like #36994? If your page size is larger than what jemalloc was compiled with, then madvise(MADV_DONTNEED) behaves badly, and you may be able to get around this with MALLOC_CONF=lg_dirty_mult:-1. If your actual page size is smaller than jemalloc's expectations, it's still possible for other errors to crop up. You can check your own with getconf PAGESIZE, but I'm not sure there's an easy way to check how some binary was compiled.

@amboar
Copy link
Author

amboar commented Oct 25, 2016

@cuviper thanks, exporting MALLOC_CONF=lg_dirty_mult:-1 appears to work around the crash. Page size on the ppc64le host was 64K.

@alexcrichton alexcrichton mentioned this issue Oct 25, 2016
Merged
@alexcrichton
Copy link
Member

Whoa thanks for the diagnosis @cuviper! I've opened a PR to disable jemalloc on powerpc: #37392

alexcrichton added a commit to alexcrichton/rust that referenced this issue Oct 27, 2016
@alexcrichton
Disable jemalloc on aarch64/powerpc
de80670 
Sounds like jemalloc is broken on systems which differ in page size than the
host it was compiled on (unless an option was passed). This unfortunately
reduces the portability of binaries created and can often make Rust segfault by
default. For now let's patch over this by disabling jemalloc until we can figure
out a better solution.

Closes rust-lang#36994
Closes rust-lang#37320
cc jemalloc/jemalloc#467
bors added a commit that referenced this issue Oct 29, 2016
@bors
Auto merge of #37392 - alexcrichton:more-disable-jemalloc, r=brson
62bdf88 
Disable jemalloc on aarch64/powerpc

Sounds like jemalloc is broken on systems which differ in page size than the
host it was compiled on (unless an option was passed). This unfortunately
reduces the portability of binaries created and can often make Rust segfault by
default. For now let's patch over this by disabling jemalloc until we can figure
out a better solution.

Closes #36994
Closes #37320
cc jemalloc/jemalloc#467
bors added a commit that referenced this issue Oct 30, 2016
@bors
Auto merge of #37392 - alexcrichton:more-disable-jemalloc, r=brson
aef5ca5 
Disable jemalloc on aarch64/powerpc

Sounds like jemalloc is broken on systems which differ in page size than the
host it was compiled on (unless an option was passed). This unfortunately
reduces the portability of binaries created and can often make Rust segfault by
default. For now let's patch over this by disabling jemalloc until we can figure
out a better solution.

Closes #36994
Closes #37320
cc jemalloc/jemalloc#467
@infinity0 infinity0 mentioned this issue Dec 28, 2016
Closed
@infinity0
Copy link
Contributor

After setting this envvar for the whole build on powerpc64le (and making other fixes) I'm getting a failure in tests::test_typed_arena_drop_small_count, could it be related?

        thread 'tests::test_typed_arena_drop_small_count' panicked at 'assertion failed: `(left == right)` (left: `0`, right: `100`)', src/libarena/lib.rs:465

infinity0 added a commit to infinity0/rust that referenced this issue Jan 12, 2017
@infinity0
Disable jemalloc on s390x as well (closes rust-lang#38596)
417953f 
See also rust-lang#37320 and rust-lang#37392
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
5 participants
@cuviper @alexcrichton @infinity0 @amboar @nagisa