git.chromium.org Git - playground/mseaborn/native

Add IRT interface for random number generator

This adds the NACL_IRT_RANDOM_v0_1 interface table with one function
for exposing the SecureRandom facility, and a test for it.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2478
TEST= trybots + new test run_random_test_irt (--mode=opt-host,nacl_irt_test)

R=bsy@google.com

Review URL: http://codereview.chromium.org/8886002

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7422 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7421 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update chrome_rev in DEPS

Update to latest LKGR whose binaries are available.

BUG= none
TEST= trybots

R=bradchen@google.com,sehr@google.com

Review URL: http://codereview.chromium.org/8935007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7420 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Trivial fix to prevent recompile of toolchain .S files.

This change splits the step which generates crti.o and crtn.o into two
steps which correctly identify the output file. Previously the output
listed in the action was never actually created, so the action would
fire on any invocation.

In addition, this removed the creation of an output file when running
hello world and instead points the output to a bogus output file which
is never created, ensuring the smoke test always runs.

TEST= gyp compile (try)
R= ncbray@google.com
Review URL: http://codereview.chromium.org/8923013

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7419 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Fix a indenting bug in build/chromebinaries.py.

BUG= none
TEST= none

Review URL: http://codereview.chromium.org/8919027

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7418 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Filter -fstack-protector out of cflags for nacl_helper_bootstrap

Some Linux distributions want to build Chromium with
GYP_DEFINES="release_extra_cflags='-mthis -fstack-protector -mthat'".
This breaks the build of nacl_helper_bootstrap because we need to
ensure that -fstack-protector is never used for that file.

Filtering out the potentially incompatible options fixes this scenario.

BUG= http://code.google.com/p/chromium/issues/detail?id=104569
TEST= manually tested in chromium build with conflicting release_extra_cflags setting

R=noelallen@chromium.org

Review URL: http://codereview.chromium.org/8890043

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7417 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add test for wcstoll.c which is currently broken for glibc.

This is the second attempt to commit the CL. The first attempt is http://codereview.chromium.org/8885029

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2479

TEST= this is the test

Review URL: http://codereview.chromium.org/8922007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7416 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Newlib buildbots RED due to dependency on glibc toolchain.
We actually download and untar the toolchain, then proceed to delete it.
This change prevents the glibc toolchain from being deleted on the newlib bot.
Hopefully once we merge glibc and newlibc toolchains into one tarball we
can avoid these kinds of issues.

Pair review with bradn.

TEST= build
TRB= bradnelson@google.com

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7415 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Factor nameservice lookup out of manifest lookup code in IRT

Clean up this code a little and factor out the top-level lookup.
That will be reused for other nameservice cases such as SecureRandom.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2478
TEST= trybots

R=bsy@google.com

Review URL: http://codereview.chromium.org/8920007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7414 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Revert 7412 - Add test for wcstoll.c which is currently broken for glibc.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2479

TEST= this is the test

Review URL: http://codereview.chromium.org/8885029

TBR=krasin@chromium.org
Review URL: http://codereview.chromium.org/8917015

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7413 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add test for wcstoll.c which is currently broken for glibc.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2479

TEST= this is the test

Review URL: http://codereview.chromium.org/8885029

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7412 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7411 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Roll toolchain DEPs.

BUG= none
TEST= trybots

Review URL: http://codereview.chromium.org/8892018

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7410 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Implement exception handling for stand-alone sel_ldr on Win32.
sel_ldr runs itself under debugging and redirects exception in untrusted code
to untrusted handler.

Idea is to store address of nacl_user and exception_handler (in NaClApp) variables
in trampoline area. This area has fixed address relative the start of the cs and
so debugger can easily find it and read nacl_user and exception_handler address.
Then, debugger fetches NaClContext and reads exception_stack and exception_flag
fields in it. Fields exception_handler and exception_stack are filled
by two new syscalls which untrusted code can use. If exception_flag is set, debugger
aborts program execution. Otherwise, debugger sets exception_flag. At last, debugger
transfers control to exception_handler and sets stack to exception_stack address. Previous
values of eip and esp are pushed to the stack.

Exception handling is switched on with new sel_ldr flag "-e". If sel_ldr is under
debugger, this flag is ignored. This makes exception handling logic simplier
- we can use the same command line for the second sel_ldr and it will not run
the third sel_ldr (and so on) because it runs under debugger.

BUG= none
TEST= scons -jN --mode=dbg-win,nacl --platform=x86-32 run_exception_test
Review URL: http://codereview.chromium.org/8341018

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7409 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add -Wno-array-bounds to SConstruct for PNaCl GlibC build.

GlibC's strpbrk is implemented by a macro which uses short-circuiting with
the ternary operator to quickly evaluate the case for 0, 1, 2, or 3 length strings.
For some reason, this construction triggers Clang's array-bounds detection.

I've filed an LLVM bug for this:
http://llvm.org/bugs/show_bug.cgi?id=11536

TBR= jvoung
BUG= http://llvm.org/bugs/show_bug.cgi?id=11536
TEST= PNaCl GLibC bot
Review URL: http://codereview.chromium.org/8911001

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7408 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Merge with LLVM up to r146310.

Merge had two oddities:

1) Upstream added a new "BUNDLE" opcode (r145975) which groups
MachineInstr's together for VLIW targets. This is unrelated to
our notion of bundling, but it can be confusing because the code
has similar labels. (https://groups.google.com/group/llvm-dev/browse_thread/thread/4464740bfdf923ce/074893d23227520b)

2) MachineBasicBlock::setAlignment was changed to use powers-of-2
instead of bytes (r145889). I did a search and replace through localmods to
change MBB.setAlignment(16) to MBB.setAlignment(4). (and 32 to 5).

TBR= jvoung
Review URL: http://codereview.chromium.org/8909001

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7407 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Revert 7405, due to broken gyp build.

TBR= mseaborn
Review URL: http://codereview.chromium.org/8895019

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7406 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Tidy: Make NACL_x86/arm #defines more robust against NACL_BUILD_ARCH not being defined

We discovered that header files were silently assuming x86 if
NACL_BUILD_ARCH was not defined. For PNaCl, we don't want to
#define NACL_BUILD_ARCH.

BUG=none
TEST=trybots

Review URL: http://codereview.chromium.org/8883036

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7405 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Make elf_constants.h #includable without NACL_BUILD_ARCH/SUBARCH defined

Move the definition of EM_EXPECTED_BY_NACL into elf_util.c, which
is the only place where it is used.

This is a step towards not #defining NACL_BUILD_ARCH/SUBARCH for
most code compiled with PNaCl.

elf_constants.h is referenced by elf32.h which defines part of
our portable ABI.

BUG=none
TEST=trybots

Review URL: http://codereview.chromium.org/8885028

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7404 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add ability to run test_hello_world_nexe for GLIBC

This CL adds the targets for building NEXE and NLIB object for glibc.
It updates test_build.py to run multiple tests and supports glibc.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2485
TEST= try
R= jvoung@google.com
Review URL: http://codereview.chromium.org/8799002

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7403 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Augment initfini test by adding 3 function pointers to each array. This will catch bugs in which preinit/init/fini_array elements are skipped due to pointer alignment/size issues.

Also augment initfini test by adding a shared library test. The dynamic loader is responsible for running init_array when loading a shared object, so this exercises a different code path.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2482
Review URL: http://codereview.chromium.org/8892010

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7402 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Remove unused next_outgoing_request_id member from struct NaClSrpcChannel

It's always zero. We don't need a word of memory for that.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=1504
TEST= trybots

R=sehr@google.com

Review URL: http://codereview.chromium.org/8899013

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7401 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Pass arch argument along to pre-req checker in the all()
convenience function in the llvm-test-suite script.

R=dschuff@google.com
BUG=
TEST=

Review URL: http://codereview.chromium.org/8873043

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7400 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Disable sel_universal tests broken by Chrome.

Tests in Chrome, but not run there, only in NaCl, making it difficult to fix.
Disabling by removed reference in SConstruct.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2480
TEST= try
TRB= robertm@google.com
Review URL: http://codereview.chromium.org/8894010

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7399 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7398 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Bump Chrome 113806

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7397 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

llvm-mc compatibility fix for nacl_text_pad_test.S

BUG= http://code.google.com/p/nativeclient/issues/detail?id=1968
Review URL: http://codereview.chromium.org/8890025

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7396 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Disable nacl_semaphore_test test when run under Qemu/arm because of flakiness.
BUG= http://code.google.com/p/nativeclient/issues/detail?id=2142
Review URL: http://codereview.chromium.org/8872018

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7395 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add a quick-n-dirty nacl64-gdb attacher.

BUG=none
TEST=none

Review URL: http://codereview.chromium.org/8832007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7394 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Clean out use_pnacl_irt from run.py now that bitcode=1 doesn't affect the IRT build.

R=pdox@google.com
BUG= none
TEST= none

Review URL: http://codereview.chromium.org/8890051

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7393 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Scons: Remove unused IrtIsBroken() function.

BUG=none
TEST=trybots
Review URL: http://codereview.chromium.org/8890045

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7392 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Re-enable multiple-sandboxes test on Windows

This test used to be flaky, but it is very likely that this was a
result of calling exit(), which runs process-global destructors and
might interfere with NaCl threads that are exiting concurrently. The
test stopped being flaky on Linux when I changed it to use NaClExit().

If the test turns out to be flaky on Windows still, this change can be
reverted.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=1485
TEST=run_multidomain_test

Review URL: http://codereview.chromium.org/8890028

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7391 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Enable run_strto{,u}ll_test for x86-64. They were recently fixed.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2444

TEST=this is the test

Review URL: http://codereview.chromium.org/8889035

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7390 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Some additional torture test failures with O3 not caught before
and pass --verbose to the torture test when running on a bot to see invocation

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2204
TEST=self

Review URL: http://codereview.chromium.org/8890047

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7389 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Fix bootstrap loading for toolchain torture tests
The tests ran, but the sanity check for existence of the tools wasn't right.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2197
TEST=

Review URL: http://codereview.chromium.org/8885014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7388 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Modify ARM .S code so it can be processed by both gnu-as and llvm-mc.
NOTE: there is now only one assert keeping us from switchingover to llvm-mc

BUG= http://code.google.com/p/nativeclient/issues/detail?id=1968
Review URL: http://codereview.chromium.org/8826003

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7387 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7386 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update to get default command line toolchain for sandboxed translators.
BUG=none
TEST=none
Review URL: http://codereview.chromium.org/8848002

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7385 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Buildbot script for running LLVM test-suite and torture tests,
pnacl/DEPS update to bring in the testsuite TEST files,
Update to run.py and toolchain tester to work with the bootstrap loader and qemu

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2197
TEST=this is a test

Review URL: http://codereview.chromium.org/8855008

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7384 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add a DescWrapper factory method for random number generation.
BUG=none
TEST=none
Review URL: http://codereview.chromium.org/8852001

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7383 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Assimilate the names "between_builders" files between regular and trybots.

In particular:
* Drop "arm/" prefix for trybots since "arm"
is implicit in the botnames
* use BUILDBOT_BUILDERNAME/BUILDBOT_TRIGGERED_BY_BUILDERNAME
vars for regular bots instead of hardcoded names
which have no connection to the machine name
* drop rev_ prefix for regular bots
Review URL: http://codereview.chromium.org/8828007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7382 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Convert some crash tests to using command_tester.py's --declares_exit_status

As before, this removes the need for the tests to have golden files.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2442
TEST=trybots

Review URL: http://codereview.chromium.org/8776051

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7381 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Scons: Use booleans in CommandTest() arguments where possible

It turns out that booleans already work. "True" is passed as a string
which is treated as true in command_tester.py, and "False" is skipped
by CommandTest().

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2442
TEST=trybots

Review URL: http://codereview.chromium.org/8825001

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7380 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add --pnacl-default-command-line flag to the driver. This flag controls
whether the sandboxed translators use the command lines baked into llc and ld.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2440
TEST=none
Review URL: http://codereview.chromium.org/8824014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7379 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Script and gclient DEPS to sync and run llvm testsuite with pnacl/sel_ldr

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2197
TEST= this is a test

Review URL: http://codereview.chromium.org/8824008

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7378 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Script to parse LLVM testsuite report output and list of known failures,
report unexpected failures and passes,
and dump the compilation and run logs for failed tests

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2197
TEST=

Review URL: http://codereview.chromium.org/8770055

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7377 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Stop using ppapi_ppb_core inside of PyAuto.

The stress test keeps flaking out.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2448
TEST= none

Review URL: http://codereview.chromium.org/8824004

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7376 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Revert r7374: x86-64 Windows: Make stack unwinding test stricter still

This change failed to compile on the 64-bit Windows Buildbots, which
apparently differ from the trybots in the version of the Windows
headers that are installed. UNWIND_HISTORY_TABLE and
RtlLookupFunctionEntry() are not defined. The errors indicate that
the header being used is:
C:\Program Files\Microsoft SDKs\Windows\v6.0A\include\winnt.h

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2414
TEST=none

Review URL: http://codereview.chromium.org/8825006

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7375 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

x86-64 Windows: Make stack unwinding test stricter still

Add a check that the return address has unwind info. When I committed
the initial version of this test I had not found the right API for
doing this.

SymFunctionTableAccess64() ought to work for this purpose too, though
it returns a different address -- presumably a different copy of the
unwind info, since this function works across processes and is
provided by dbghelp.lib rather than the runtime library.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2414
TEST=run_windows_crash_in_syscall_test

Review URL: http://codereview.chromium.org/8769015

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7374 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Augmented pnacl-as to handle .S files, so we can remove .s/.S support from pnacl-clang.
Also, make import statements style guide compliant

BUG= http://code.google.com/p/nativeclient/issues/detail?id=1966
Review URL: http://codereview.chromium.org/8769054

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7373 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Really disable x86_64 hello_world_test on the Mac/ASan bot.

TBR=khim
Review URL: http://codereview.chromium.org/8816014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7372 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Disable x86_64 hello_world_test on the Mac/ASan bot.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2473
TBR=khim
Review URL: http://codereview.chromium.org/8821014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7371 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Switch ASan bots to opt-host.

TBR=bradnelson
Review URL: http://codereview.chromium.org/8799008

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7370 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Change -fasan to -faddress-sanitizer. Add a sanity test.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2467
TEST=run small_tests with --asan
Review URL: http://codereview.chromium.org/8799007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7369 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Move ncval usage text to its own variable.

This change fixes Clang -O2 build.

printf() in optimized builds turns into a macro; using "#if" inside
macro arguments is not allowed (Clang, -pedantic).
Review URL: http://codereview.chromium.org/8803012

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7368 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Merge ... LLVM up to r145883. Merge was trivial except for conflicts caused by adjacent code changes.

TBR= jvoung
Review URL: http://codereview.chromium.org/8817003

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7367 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Get rid of upload to "latest" directory for PNaCl toolchain builders.

R= bradn
Review URL: http://codereview.chromium.org/8817004

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7366 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Scons: Remove the unused argument 'nocpp'

BUG=none
TEST=trybots

Review URL: http://codereview.chromium.org/8785012

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7365 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Build GLibC's crt1 as bitcode.

The benefit is that the PNaCl coordinator/translator doesn't need to handle crt1.o
as a special case.

This CL adds the nacl-glibc repository as a checkout in pnacl/git/glibc.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2423
R= Roland McGrath, sehr
Review URL: http://codereview.chromium.org/8659034

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7364 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Expose LLVM changes to get __builtin_nacl_target_arch intrinsic.
BUG=none
TEST=none
Review URL: http://codereview.chromium.org/8804028

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7363 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Need to excise -faddress-sanitizer now

I failed to notice r110925 on the chromium/src side.
Duplicate that change here so Chrome's ASan builders won't freak out.

BUG= none
TEST= none

R=sehr@google.com

Review URL: http://codereview.chromium.org/8799021

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7362 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Convert pnacl_abi.o into pnacl_abi.bc (a bitcode stub).

Changes:
1) Roll UPSTREAM_REV to pull in: http://codereview.chromium.org/8787015/
2) Renamed pnacl_abi.S to pnacl_abi.c, and modified to make it produce C.
3) Added a compiler flag to produce stubs from C.
4) Adjust build.sh to produce pnacl_abi.bc
5) Change the driver to use pnacl_abi.bc

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2454
Review URL: http://codereview.chromium.org/8785019

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7361 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Tidy: Remove sel_ldr-inl.c and sel_util-inl.c

These were necessary when we were using gnu_inline, which can cause
gcc to omit the definition of an inline function such that it
generates an object file with an undefined symbol.

However, we are now using the normal "static inline", which does not
have this problem.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=697
TEST=compile

Review URL: http://codereview.chromium.org/8774020

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7360 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Make the timestamps reading/writing match in download_toolchain.py.

R=noelallen@google.com
BUG= none
TEST= none

Review URL: http://codereview.chromium.org/8800028

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7359 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Disable dynamic_load_test on arm hw

This test is flaky on the ARM hw bots.
See bug for details. Disable it there until it can be properly investigated.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2472
TEST= none

R=sehr@google.com

Review URL: http://codereview.chromium.org/8804025

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7358 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Harmonize flags for Linux builds

Always build PIE, always use -z relro and -z now, always use _FORTIFY_SOURCE.

Since we're now using a PIE everywhere, use the bootstrap loader everywhere.

BUG= none
TEST= trybots

R=sehr@google.com

Review URL: http://codereview.chromium.org/8800013

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7357 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Avoid linking in pnacl irt shim for the hello world commandline test.
That test has a size check (must be below 64KB with glibc).
The IRT shim itself is ~50KB and may grow, so linking it in could make
this awfully close to 64KB.

BUG= none -- build breakage on pnacl glibc toolchain bot.
TEST= run_hello_world_test, run_hello_world_size_test

Review URL: http://codereview.chromium.org/8798002

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7356 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Don't misalign the trusted stack on ARM

We can't figure out why this + 4 was there in the first place. It seems
always to have been wrong. It only matters in practice when a confluence
of circumstances winds up with something in the trusted code depending on
the normal ABI stack alignment, so it has gone unnoticed for a long time.

BUG= none
TEST= trybots

R=bsy@google.com

Review URL: http://codereview.chromium.org/8790007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7355 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Retry Windows commands in CygTar.

Attempt to make cygtar more robust on Windows where occationally access to a fail
is prevented by what appears to be stale file locks.

TBR= jvoung@google.com
Review URL: http://codereview.chromium.org/8795005

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7354 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Switch the nop pnacl x86-64 IRT shim to the real one generated from IDL.

We add pnacl_irt_shim to the Implicit Libs for pnacl x86-64 runs
so that we get the freshest version instead of whatever is in the toolchain
tarball.

R=robertm@google.com,sehr@google.com
BUG= http://code.google.com/p/nativeclient/issues/detail?id=2413
TEST= ./scons bitcode=1 platform=x86-64 chrome_browser_tests

Review URL: http://codereview.chromium.org/8776023

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7353 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7352 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

bump chrome deps and bin rev to next lkgr

BUG= none
TEST= trybots
Review URL: http://codereview.chromium.org/8770047

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7351 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Refactor download scripts to use cygtar.

Refactor download scripts to use cygtar, so that we use it end to end
from chrome to nacl to sdk. Also consolidate toolchainbinaries into
the toolchain download script simplifying imports and logic, and
sync_file, sync_zip into download_utils and download_chrome as
appropriate.

TBR= bradnelson@google.com
Review URL: http://codereview.chromium.org/8788012

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7350 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Merge with LLVM up to r145760.

Merge was trivial, except for conflict caused by code moving around.

TBR= jvoung
Review URL: http://codereview.chromium.org/8787014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7349 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add a python (tarfile) based tar/untar

This CL adds a module which can be used to tar and untar files such as the toolchain which may contain symlinks and hardlinks in a platform independent way. Symlinks are windows are treated as Cygwin style links to enable the toolchain.
Review URL: http://codereview.chromium.org/8786012

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7348 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Add test tarfile.

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7347 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Fix a library path bias accidentally introduced into the newlib-based build.
It was introduced when the library path (NACL_SDK_LIB) was changed for the GlibC-based build.

SCons is not supposed to give pnacl-clang -L to the native directories when the front-end
is operating in pure bitcode mode.

Also, re-enable the pnacl_client_translator test, except for missing_ld and missing_llc which
are still failing.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2470
Review URL: http://codereview.chromium.org/8787008

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7346 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Improve the browser tester's web server.

This change should reduce the chance that the web server will drop connections
when dealing with simultaneous requests.

BUG= http://code.google.com/p/chromium/issues/detail?id=106217
TEST= none

Review URL: http://codereview.chromium.org/8787001

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7345 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

command_tester.py: Add new option for testing exit statuses for crash tests

Allow test programs to declare their intended exit status by printing
it to stderr with a line of the form "** intended_exit_status=X".
With the option --declares_exit_status=1, command_tester.py will
require such a line in stderr and will check it against the actual
exit status.

This allows crash tests to be less coupled to Scons because the exit
status won't have to be declared in Scons. It also removes the need
for crash tests to have golden files.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2442
TEST=I've converted run_execute_data_test

Review URL: http://codereview.chromium.org/8565031

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7344 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Initial cut at making pnacl shim available to translator example test.
We should eventually make this available in a different way.
E.g., if we run
pnacl/pnacl_packaging/pnacl_component_crx_gen.py -u 0.1.2.3
then all files in lib-${arch} are copied over (vs maintaining this list).

However, even with that the shim is pinned to the toolchain tarball rev,
and may be stale (but it shouldn't be stale for this hello world test
which uses M14 interfaces).
Review URL: http://codereview.chromium.org/8785010

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7343 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Use bootstrap loader for trusted mmap test

This test program is basically a variant of sel_ldr, so it needs
to be run via the bootstrap loader.

BUG= none
TEST= trybots

R=bsy@google.com

Review URL: http://codereview.chromium.org/8775069

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7342 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

* make the big jail the default.
* add some helper scripts
* some cleanup.
Review URL: http://codereview.chromium.org/8760007

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7341 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Scons treats .s and .S differently:
.s files are routed through $AS while .S files are routed
through the $CC driver, which will also pre-process them with cpp.

This change routes .S files through a new Scons variable $ASPP.
For now this will be identical to $CC.
Eventually, this will make it possible to route .S files through a different driver for PNaCl where we do not want $CC to handle assembly files.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=1966

TESTS= trybots
Review URL: http://codereview.chromium.org/8775055

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7340 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Fix the browser tester so run_inbrowser_crash_in_syscall_test retries on flake.

This test imports the browser tester as a library and has an entry point that
previously bypassed the flake retry mechanism.

BUG= none
TEST= none

Review URL: http://codereview.chromium.org/8772051

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7339 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7338 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update DEPS to get argz changes for sandboxed translators.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2440
TEST=none
Review URL: http://codereview.chromium.org/8771038

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7337 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

dyncode_delete: Fix NULL pointer dereference when enumerating threads

The threads array can contain NULL entries where threads have exited
-- handle that. Use the correct array size.

The 'i == 0' check would no longer be correct, so use INT_MAX to
handle the first-thread case instead.

This is best tested via a unit test rather than an integration test.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2190
TEST=run_service_runtime_tests

Review URL: http://codereview.chromium.org/8776021

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7336 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Update .DEPS.git

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7335 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Second attempt at rolling arm trusted CL.
Initial one was:
http://codereview.chromium.org/8756002

which missing a gyp change
Review URL: http://codereview.chromium.org/8776014

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7334 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Revert r7329 which broke arm HW bots. Note: the trybots seem to reproduce the failures as well.
Review URL: http://codereview.chromium.org/8772043

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7333 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Convert sandboxed translator driver to use argz command lines.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2440
TEST=none
Review URL: http://codereview.chromium.org/8770030

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7332 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Merge with LLVM up to r145642.

Also, rolls DEPS to activate CL:

http://codereview.chromium.org/8776036/

Merge was trivial, except that llvm.memory.barrier was removed.
Review URL: http://codereview.chromium.org/8772032

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7331 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Make "driver" build option appear in the help documentation rather than
help-full
Review URL: http://codereview.chromium.org/8774028

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7330 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Harmonize flags for Linux builds

Always build PIE, always use -z relro and -z now, always use _FORTIFY_SOURCE.

Since we're now using a PIE everywhere, use the bootstrap loader everywhere.

BUG= none
TEST= trybots

Review URL: http://codereview.chromium.org/8374013

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7329 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Browser tests are retried when known flake occurs.

Currently the only flake handled is when Chrome crashes on launch.

BUG= http://code.google.com/p/nativeclient/issues/detail?id=2461
TEST= none

Review URL: http://codereview.chromium.org/8682019

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7328 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

sync_tgz.py: Fix unchecked subprocess.call() call

Also fix some minor Python style issues at the same time. Remove an
unused import.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=1745
TEST=run download_toolchains.py on Windows

Review URL: http://codereview.chromium.org/8772011

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7327 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2

Enable tests/windows_crash_in_syscall to build with PNaCl

Use '-nostdlib' for linking the test program, bringing this into line
with tests/minnacl. Moving the test program to above the 'Return()'
allows us to test building it with PNaCl on Linux.

BUG=none
TEST=build with "scons test_guest bitcode=1"

Review URL: http://codereview.chromium.org/8775015

git-svn-id: svn://svn.chromium.org/native_client/trunk/src/native_client@7326 fcba33aa-ac0c-11dd-b9e7-8d5594d729c2