# Copyright 2014 The Chromium Authors. All rights reserved. # Use of this source code is governed by a BSD-style license that can be # found in the LICENSE file. # pylint: disable=W0201 """Default flavor, used for running code on desktop machines.""" WIN_TOOLCHAIN_DIR = 't' class DeviceDirs(object): def __init__(self, bin_dir, dm_dir, perf_data_dir, resource_dir, images_dir, lotties_dir, skp_dir, svg_dir, tmp_dir): self._bin_dir = bin_dir self._dm_dir = dm_dir self._perf_data_dir = perf_data_dir self._resource_dir = resource_dir self._images_dir = images_dir self._lotties_dir = lotties_dir self._skp_dir = skp_dir self._svg_dir = svg_dir self._tmp_dir = tmp_dir @property def bin_dir(self): return self._bin_dir @property def dm_dir(self): """Where DM writes.""" return self._dm_dir @property def perf_data_dir(self): return self._perf_data_dir @property def resource_dir(self): return self._resource_dir @property def images_dir(self): return self._images_dir @property def lotties_dir(self): return self._lotties_dir @property def skp_dir(self): """Holds SKP files that are consumed by RenderSKPs and BenchPictures.""" return self._skp_dir @property def svg_dir(self): return self._svg_dir @property def tmp_dir(self): return self._tmp_dir class DefaultFlavor(object): def __init__(self, module): # Store a pointer to the parent recipe module (SkiaFlavorApi) so that # FlavorUtils objects can do recipe module-like things, like run steps or # access module-level resources. self.module = module # self.m is just a shortcut so that Flavor objects can use the same # syntax as regular recipe modules to run steps, eg: self.m.step(...) self.m = module.m self._chrome_path = None self.device_dirs = DeviceDirs( bin_dir=self.m.vars.build_dir, dm_dir=self.m.vars.swarming_out_dir, perf_data_dir=self.m.vars.swarming_out_dir, resource_dir=self.m.path['start_dir'].join('skia', 'resources'), images_dir=self.m.path['start_dir'].join('skimage'), lotties_dir=self.m.path['start_dir'].join('lottie-samples'), skp_dir=self.m.path['start_dir'].join('skp'), svg_dir=self.m.path['start_dir'].join('svg'), tmp_dir=self.m.vars.tmp_dir) self.host_dirs = self.device_dirs def device_path_join(self, *args): """Like os.path.join(), but for paths on a connected device.""" return self.m.path.join(*args) def copy_directory_contents_to_device(self, host_dir, device_dir): """Like shutil.copytree(), but for copying to a connected device.""" # For "normal" builders who don't have an attached device, we expect # host_dir and device_dir to be the same. if str(host_dir) != str(device_dir): raise ValueError('For builders who do not have attached devices, copying ' 'from host to device is undefined and only allowed if ' 'host_path and device_path are the same (%s vs %s).' % ( str(host_dir), str(device_dir))) def copy_directory_contents_to_host(self, device_dir, host_dir): """Like shutil.copytree(), but for copying from a connected device.""" # For "normal" builders who don't have an attached device, we expect # host_dir and device_dir to be the same. if str(host_dir) != str(device_dir): raise ValueError('For builders who do not have attached devices, copying ' 'from device to host is undefined and only allowed if ' 'host_path and device_path are the same (%s vs %s).' % ( str(host_dir), str(device_dir))) def copy_file_to_device(self, host_path, device_path): """Like shutil.copyfile, but for copying to a connected device.""" # For "normal" builders who don't have an attached device, we expect # host_dir and device_dir to be the same. if str(host_path) != str(device_path): raise ValueError('For builders who do not have attached devices, copying ' 'from host to device is undefined and only allowed if ' 'host_path and device_path are the same (%s vs %s).' % ( str(host_path), str(device_path))) def create_clean_device_dir(self, path): """Like shutil.rmtree() + os.makedirs(), but on a connected device.""" self.create_clean_host_dir(path) def create_clean_host_dir(self, path): """Convenience function for creating a clean directory.""" self.m.run.rmtree(path) self.m.file.ensure_directory( 'makedirs %s' % self.m.path.basename(path), path) def install(self): """Run device-specific installation steps.""" pass def cleanup_steps(self): """Run any device-specific cleanup steps.""" pass def _run(self, title, cmd, infra_step=False, **kwargs): return self.m.run(self.m.step, title, cmd=cmd, infra_step=infra_step, **kwargs) def _py(self, title, script, infra_step=True, args=()): return self.m.run(self.m.python, title, script=script, args=args, infra_step=infra_step) def step(self, name, cmd): app = self.device_dirs.bin_dir.join(cmd[0]) cmd = [app] + cmd[1:] env = self.m.context.env path = [] ld_library_path = [] slave_dir = self.m.vars.slave_dir clang_linux = str(slave_dir.join('clang_linux')) extra_tokens = self.m.vars.extra_tokens if self.m.vars.is_linux: if (self.m.vars.builder_cfg.get('cpu_or_gpu', '') == 'GPU' and 'Intel' in self.m.vars.builder_cfg.get('cpu_or_gpu_value', '')): dri_path = slave_dir.join('mesa_intel_driver_linux') ld_library_path.append(dri_path) env['LIBGL_DRIVERS_PATH'] = str(dri_path) env['VK_ICD_FILENAMES'] = str(dri_path.join('intel_icd.x86_64.json')) if 'Vulkan' in extra_tokens: path.append(slave_dir.join('linux_vulkan_sdk', 'bin')) ld_library_path.append(slave_dir.join('linux_vulkan_sdk', 'lib')) if 'OpenCL' in extra_tokens: ld_library_path.append(slave_dir.join('opencl_ocl_icd_linux')) # TODO(dogben): Limit to the appropriate GPUs when we start running on # GPUs other than IntelIris640. # Skylake and later use the NEO driver. neo_path = slave_dir.join('opencl_intel_neo_linux') ld_library_path.append(neo_path) # Generate vendors dir contaning the ICD file pointing to the NEO OpenCL # library. vendors_dir = self.m.vars.tmp_dir.join('OpenCL', 'vendors') self.m.file.ensure_directory('mkdirs OpenCL/vendors', vendors_dir) self.m.file.write_raw('write NEO OpenCL ICD', vendors_dir.join('neo.icd'), '%s\n' % neo_path.join('libigdrcl.so')) env['OPENCL_VENDOR_PATH'] = vendors_dir if 'SwiftShader' in extra_tokens: ld_library_path.append(self.host_dirs.bin_dir.join('swiftshader_out')) if 'MSAN' in extra_tokens: # Find the MSAN-built libc++. ld_library_path.append(clang_linux + '/msan') if any('SAN' in t for t in extra_tokens): # Sanitized binaries may want to run clang_linux/bin/llvm-symbolizer. path.append(clang_linux + '/bin') # We find that testing sanitizer builds with libc++ uncovers more issues # than with the system-provided C++ standard library, which is usually # libstdc++. libc++ proactively hooks into sanitizers to help their # analyses. We ship a copy of libc++ with our Linux toolchain in /lib. ld_library_path.append(clang_linux + '/lib') elif self.m.vars.is_linux: cmd = ['catchsegv'] + cmd elif 'ProcDump' in extra_tokens: dumps_dir = self.m.path.join(self.m.vars.swarming_out_dir, 'dumps') self.m.file.ensure_directory('makedirs dumps', dumps_dir) procdump = str(self.m.vars.slave_dir.join('procdump_win', 'procdump64.exe')) # Full docs for ProcDump here: # https://docs.microsoft.com/en-us/sysinternals/downloads/procdump # -accepteula automatically accepts the license agreement # -mp saves a packed minidump to save space # -e 1 tells procdump to dump once # -x <dump dir> <exe> <args> launches exe and writes dumps to the # specified dir cmd = [procdump, '-accepteula', '-mp', '-e', '1', '-x', dumps_dir] + cmd if 'ASAN' in extra_tokens or 'UBSAN' in extra_tokens: # Note: if you see "<unknown module>" in stacktraces for xSAN warnings, # try adding "fast_unwind_on_malloc=0" to xSAN_OPTIONS. if 'Mac' in self.m.vars.builder_cfg.get('os', ''): env['ASAN_OPTIONS'] = 'symbolize=1' # Mac doesn't support detect_leaks. else: env['ASAN_OPTIONS'] = 'symbolize=1 detect_leaks=1' env[ 'LSAN_OPTIONS'] = 'symbolize=1 print_suppressions=1' env['UBSAN_OPTIONS'] = 'symbolize=1 print_stacktrace=1' if 'TSAN' in extra_tokens: # We don't care about malloc(), fprintf, etc. used in signal handlers. # If we're in a signal handler, we're already crashing... env['TSAN_OPTIONS'] = 'report_signal_unsafe=0' if 'Coverage' in extra_tokens: # This is the output file for the coverage data. Just running the binary # will produce the output. The output_file is in the swarming_out_dir and # thus will be an isolated output of the Test step. profname = '%s.profraw' % self.m.vars.builder_cfg.get('test_filter','o') env['LLVM_PROFILE_FILE'] = self.m.path.join(self.m.vars.swarming_out_dir, profname) if path: env['PATH'] = '%%(PATH)s:%s' % ':'.join('%s' % p for p in path) if ld_library_path: env['LD_LIBRARY_PATH'] = ':'.join('%s' % p for p in ld_library_path) to_symbolize = ['dm', 'nanobench'] if name in to_symbolize and self.m.vars.is_linux: # Convert path objects or placeholders into strings such that they can # be passed to symbolize_stack_trace.py args = [slave_dir] + [str(x) for x in cmd] with self.m.context(cwd=self.m.path['start_dir'].join('skia'), env=env): self._py('symbolized %s' % name, self.module.resource('symbolize_stack_trace.py'), args=args, infra_step=False) else: with self.m.context(env=env): self._run(name, cmd)