On Tue, Dec 31, 2019 at 12:42:36PM +0200, Cristian Ciocaltea wrote:
On Mon, Dec 30, 2019 at 09:03:38PM +0100, Heinrich Schuchardt wrote:
On 12/30/19 8:32 PM, Stephen Warren wrote:
On 12/30/19 12:05 PM, Heinrich Schuchardt wrote:
On 12/30/19 5:38 PM, Stephen Warren wrote:
On 12/30/19 3:52 AM, Heinrich Schuchardt wrote:
Provide dictionary env__efi_fit_tftp_file describing the file used for the UEFI FIT image test.
diff --git a/py/travis-ci/travis_tftp.py b/py/travis-ci/travis_tftp.py
+def efifit2env(addr=None): + """Create dictionary describing file for EFI fit image test
+ @addr: address used for loading the file as int (e.g. 0x40400000) + Return: dictionary describing the file with entries + * fn - filename + * addr - loading address, optional + * dn - tftp directory + """ + tftp_dir = os.environ['UBOOT_TRAVIS_BUILD_DIR']
+ ret = { + "fn": "test-efi-fit.img",
If this function were to exist, then the filename shouldn't be hard-coded; it should be a parameter.
Hello Stephen,
thanks for reviewing.
This is the name of a generated file. It does not depend on the board.
What generates the file and when/why?
Generated files should generally be put into u_boot_console.config.persistent_data_dir, and presumably the name hard-coded into the test that uses it.
Hello Stephen,
this is the test case:
https://lists.denx.de/pipermail/u-boot/2019-December/394957.html test/py: Create a test for launching UEFI binaries from FIT images
The test can be run in different styles:
- A complete FIT image can be supplied. In this case the dictionary must contain a "size" entry.
- The test can generate a FIT image from lib/efi_loader/helloworld.efi. In this case no "size" entry shall be supplied. The "fn" field provides the name of the generated file. The file is generated in cons.config.build_dir. The "dn" field" describes the tFTP root directory to which the generated file is copied.
A small correction here: if the "size" entry is not provided in the dictionary, the test generates a FIT image using a hardcoded file name (test-efi-fit-helloworld.fit), so any "fn" entry provided in the dictionary is ignored in this case.
The tFTP directory "dn" is only known in uboot-test-hooks.git.
Yes, and this is actually the only mandatory information to be provided in the dictionary when the test is supposed to generate the FIT image. We could get rid of this, too, if we make the assumption that the tFTP root directory is u_boot_console.config.build_dir by default.
test/py/README.md says --persistent-data-dir is used for data that may be re-used across test runs. Currently the FIT file is generated in every run. In principal it could be reused. So Christian could consider adjusting his patch.
Actually the FIT image file can be generated multiple times per test run, with different content, which is determined by various combinations of flags. E.g. for sandbox we have: enable/disable GZIP compression, enable/disable FDT usage. Each combination is used in a subtest, that (re)generates the corresponding FIT image.
To re-use the generated data we need to introduce additional complexity to handle FIT image identification for each subtest and to invalidate those images when they become out of sync with the parametrized input data samples (ITS, FDT) used to generate their content.
Best regards
Heinrich
I would propose the following:
def efifit2env(file_name=None, addr=None): """Create dictionary describing file for EFI fit image test
@filename: name of an existing file, optional @addr: address used for loading the file as int (e.g. 0x40400000) Return: dictionary describing the file with entries * dn - tftp directory * fn - filename, optional * size - file size in bytes, optional * crc32 - checksum using CRC-32 algorithm, optional * addr - loading address, optional """ tftp_dir = os.environ['UBOOT_TRAVIS_BUILD_DIR']
ret = { 'dn': tftp_dir, }
if addr is not None: ret['addr'] = addr
if file_name: file_full = tftp_dir + '/' + file_name if os.path.isfile(file_full): ret['fn'] = file_name ret['size'] = os.path.getsize(file_full) with open(file_full, 'rb') as f: ret['crc32'] = hex(binascii.crc32(f.read()) & 0xffffffff)[2:]
return ret
This allows the test to run in both scenarios described by Heinrich: * the FIT image is available outside the test -> efifit2env() is called with file_name pointing to an existing FIT image file, optionally pass also the addr * the FIT image is generated by the test itself -> efifit2env() is called without arguments, or optionally pass only the addr