| Message ID | 20200907053426.1675646-1-takahiro.akashi@linaro.org |
|---|---|
| Headers | show |
| Series | efi_loader: add capsule update support | expand |
On 07.09.20 07:34, AKASHI Takahiro wrote: > Summary > ======= > 'UpdateCapsule' is one of runtime services defined in UEFI specification > and its aim is to allow a caller (OS) to pass information to the firmware, > i.e. U-Boot. This is mostly used to update firmware binary on devices by > instructions from OS. > > While 'UpdateCapsule' is a runtime services function, it is, at least > initially, supported only before exiting boot services alike other runtime > functions, [Get/]SetVariable. This is because modifying storage which may > be shared with OS must be carefully designed and there is no general > assumption that we can do it. > > Therefore, we practically support only "capsule on disk"; any capsule can > be handed over to UEFI subsystem as a file on a specific file system. > > In this patch series, all the related definitions and structures are given > as UEFI specification describes, and basic framework for capsule support > is provided. Currently supported is > * firmware update (Firmware Management Protocol or simply FMP) > > Most of functionality of firmware update is provided by FMP driver and > it can be, by nature, system/platform-specific. So you can and should > implement your own FMP driver(s) based on your system requirements. > Under the current implementation, we provide two basic but generic > drivers with two formats: > * FIT image format (as used in TFTP update and dfu) > * raw image format > > It's totally up to users which one, or both, should be used on users' > system depending on user requirements. > > Quick usage > =========== > 1. You can create a capsule file with the following host command: > > $ mkeficapsule [--fit <fit image> | --raw <raw image>] <output file> > > 2. Put the file under: > > /EFI/UpdateCapsule of UEFI system partition > > 3. Specify firmware storage to be updated in "dfu_alt_info" variable > (Please follow README.dfu for details.) > > ==> env set dfu_alt_info '...' > > 4. After setting up UEFI's OsIndications variable, reboot U-Boot: > > OsIndications <= EFI_OS_INDICATIONS_FILE_CAPSULE_DELIVERY_SUPPORTED > > Patch structure > =============== > Patch#1-#4,#12: preparatory patches > Patch#5-#11,#13: main part of implementation > Patch#14-#15: utilities > Patch#16-#17: pytests > > [1] https://git.linaro.org/people/takahiro.akashi/u-boot.git efi/capsule > > Prerequisite patches > ==================== > In this version, the DFU change by Heinrich[2] is, at least temporarily, > not a prerequisite due to the discussions[3]. > > [2] https://lists.denx.de/pipermail/u-boot/2020-July/420950.html > [3] https://lists.denx.de/pipermail/u-boot/2020-August/424716.html The end of the discussion was that the patch is postponed to v2021.01 like your series. > > Test > ==== > * passed all the pytests which are included in this patch series > on sandbox build locally. > > Please note that the capsule pytest itself won't be run in the CI > partly because some specific configuration for sandbox build is Shouldn't the sandbox_defconfig changes be added to your series? Is this the correct change: diff --git a/configs/sandbox_defconfig b/configs/sandbox_defconfig index 6e9f029cc9..da82afaa01 100644 --- a/configs/sandbox_defconfig +++ b/configs/sandbox_defconfig @@ -129,6 +129,7 @@ CONFIG_DM_DEMO_SIMPLE=y CONFIG_DM_DEMO_SHAPE=y CONFIG_BOARD=y CONFIG_BOARD_SANDBOX=y +CONFIG_DFU_SF=y CONFIG_DMA=y CONFIG_DMA_CHANNELS=y CONFIG_SANDBOX_DMA=y @@ -261,6 +262,10 @@ CONFIG_CMD_DHRYSTONE=y CONFIG_TPM=y CONFIG_LZ4=y CONFIG_ERRNO_STR=y +CONFIG_EFI_RUNTIME_UPDATE_CAPSULE=y +CONFIG_EFI_CAPSULE_ON_DISK=y +CONFIG_EFI_CAPSULE_FIRMWARE_FIT=y +CONFIG_EFI_CAPSULE_FIRMWARE_RAW=y CONFIG_EFI_SECURE_BOOT=y CONFIG_TEST_FDTDEC=y CONFIG_UNIT_TEST=y > required and partly because there is a problem with virt-make-fs. What problem with virt-make-fs exists? How will this be solved? > See test_efi_capsule_firmware.py. > > Issues > ====== > * Timing of executing capsules-on-disk > Currently, processing a capsule is triggered only as part of > UEFI subsystem initialization. This means that, for example, > firmware update, may not take place at system booting time and > will potentially be delayed until a first call of any UEFI functions. > => See patch#5 for my proposal > * A bunch of warnings like > WARNING: Use 'if (IS_ENABLED(CONFIG...))' instead of '#if or #ifdef' > where possible > I don't think that fixing those improves anything. Please, use 'if' instead of '#if' where possible so that the compiler checks the code syntax even if the configuration is not set. Like with '#if' the compiler will eliminate unused code. Best regards Heinrich > * Add a document in uefi.rst > > TODO's > ====== > (Won't be addressed in this series.) > * capsule authentication > * capsule dependency (dependency expression instruction set, or depex) > * loading drivers in a capsule > * handling RESET flag in a capsule and QeuryCapsuleCaps > * full semantics of ESRT (EFI System Resource Table) > * enabling capsule API at runtime > * json capsule > * recovery from update failure > > Changes > ======= > v6 (September 7, 2020) > * temporarily drop the prerequisite patch[2] > * add a missing field (dependencies) in efi_api.h (but never used) (Patch#10) > * add a missing field (image_capsule_support) and related definitions > in efi_api.h (Patch#10, #15) > * cosmetic changes on constant definitions in efi_api.h (Patch#10) > * strict check for INVALID_PARAMETER at GET_IMAGE_INFO api (Patch#11,#13) > * fix warnings in pytest (Patch#16,#17) > > v5 (August 3, 2020) > * removed superfluous type conversion at dfu_write_from_mem_addr() > (Patch#2) > * introduced a common helper function, efi_create_indexed_name() > (Patch#6,#7,#8) > * use efi_[get|set]_variable_int(), if necessary, with READ_ONLY > (Patch#7,#8) > * return EFI_UNSUPPORTED at Patch#7 > * changed the word, number, to 'index' (Patch#7,#8) > * removed 'ifdef CONFIG_EFI_CAPSULE_ON_DISK' from a header (Patch#8) > * initialize 'CapsuleLast' in efi_init_obj_list() (Patch#7,#8) > * added 'const' qualifier for filename argument at > efi_capsule_[read|delete]_file() (Patch#8) > > v4 (July 22, 2020) > * rebased to Heinrich's current efi-2020-10 > * rework dfu-related code to align with Heinrich's change (Patch#1,#3) > * change a type of 'addr' argument from int to 'void *' per Sughosh's > comment (Patch#2-#3,#11-#12) > * rework/simplify pytests (Patch#15-#16) > - utilize virt-make-fs > - drop Test Case 1 (updating U-Boot environment data) > - remove useless definitions (MNT_PNT, EFI_CAPSULE_IMAGE_NAME) > - apply autopep8 > > v3 (July 10, 2020) > * rebased to Heinrich's current efi-2020-10-rc1 > * refactor efi_firmware_[fit|raw]_get_image_info() (patch#11,#13) > > v2 (June 17, 2020) > * rebased to v2020.07-rc4 > * add preparatory patches for dfu (Patch#1-#5, #12) > * rework FIT capsule driver to utilize dfu_alt_info instead of CONFIG_xxx > (patch#11) > * extend get_image_info() to correspond to dfu_alt_info > (patch#11) > * add a 'raw binary' capsule support > (patch#13, #17) > * allow multiple capsule formats (with different GUIDs) to be installed > (patch#11, #13) > * extend mkeficapsule command to accept additional parameters, like > version/index/hardware instance for a capsule header info. > (patch#15) > * mkeficapsule can now also generate raw-binary capsule > (patch#16) > * add function descriptions > * apply autopep8 to pytests and fix more against pylint > > v1 (April 27, 2020) > * rebased to v2020.07-rc > * removed already-merged patches (RFC's #1 to #4) > * dropped 'variable update' capsule support (RFC's patch#10) > * dropped 'variable configuration table' support (RFC's patch#11) > (Those two should be discussed separately.) > * add preparatory patches (patch#1/#2) > * fix several build errors > * rename some Kconfig options to be aligned with UEFI specification's terms > (patch#3,4,6,7) > * enforce UpdateCapsule API to be disabled after ExitBootServices (patch#3) > * use config table, runtime_services_supported, instead of variable (patch#3) > * make EFI_CAPSULE_ON_DISK buildable even if UpdateCapsule API is disabled > (patch4) > * support OsIndications, invoking capsule-on-disk only if the variable > indicates so (patch#4) > * introduced EFI_CAPSULE_ON_DISK_EARLY to invoke capsule-on-disk in U-Boot > initialization (patch#4) > * detect capsule files only if they are on EFI system partition (patch#4) > * use printf, rather than EFI_PRINT, in error cases (patch#4) > * use 'header_size' field to retrieve capsule data, adding sanity checks > against capsule size (patch#6) > * call fmpt driver interfaces with EFI_CALL (patch#6) > * remove 'variable update capsule'-related code form mkeficapsule (patch#9) > * add a test case of OsIndications not being set properly (patch#10) > * adjust test scenario for EFI_CAPSULE_ON_DISK_EARLY (patch#10) > * revise pytest scripts (patch#10) > > Initial release as RFC (March 17, 2020) > > AKASHI Takahiro (17): > dfu: rename dfu_tftp_write() to dfu_write_by_name() > dfu: modify an argument type for an address > common: update: add a generic interface for FIT image > dfu: export dfu_list > efi_loader: add option to initialise EFI subsystem early > efi_loader: add efi_create_indexed_name() > efi_loader: define UpdateCapsule api > efi_loader: capsule: add capsule_on_disk support > efi_loader: capsule: add memory range capsule definitions > efi_loader: capsule: support firmware update > efi_loader: add firmware management protocol for FIT image > dfu: add dfu_write_by_alt() > efi_loader: add firmware management protocol for raw image > cmd: add "efidebug capsule" command > tools: add mkeficapsule command for UEFI capsule update > test/py: efi_capsule: test for FIT image capsule > test/py: efi_capsule: test for raw image capsule > > cmd/efidebug.c | 235 +++++ > common/Kconfig | 15 + > common/Makefile | 3 +- > common/board_r.c | 6 + > common/main.c | 4 + > common/update.c | 77 +- > drivers/dfu/Kconfig | 6 + > drivers/dfu/Makefile | 2 +- > drivers/dfu/dfu.c | 2 +- > drivers/dfu/dfu_alt.c | 125 +++ > drivers/dfu/dfu_tftp.c | 65 -- > include/dfu.h | 57 +- > include/efi_api.h | 166 ++++ > include/efi_loader.h | 30 + > include/image.h | 12 + > lib/efi_loader/Kconfig | 72 ++ > lib/efi_loader/Makefile | 2 + > lib/efi_loader/efi_capsule.c | 917 ++++++++++++++++++ > lib/efi_loader/efi_firmware.c | 403 ++++++++ > lib/efi_loader/efi_runtime.c | 104 +- > lib/efi_loader/efi_setup.c | 106 +- > test/py/tests/test_efi_capsule/conftest.py | 72 ++ > test/py/tests/test_efi_capsule/defs.py | 12 + > .../test_efi_capsule/test_capsule_firmware.py | 241 +++++ > .../tests/test_efi_capsule/uboot_bin_env.its | 36 + > tools/Makefile | 2 + > tools/mkeficapsule.c | 239 +++++ > 27 files changed, 2871 insertions(+), 140 deletions(-) > create mode 100644 drivers/dfu/dfu_alt.c > delete mode 100644 drivers/dfu/dfu_tftp.c > create mode 100644 lib/efi_loader/efi_capsule.c > create mode 100644 lib/efi_loader/efi_firmware.c > create mode 100644 test/py/tests/test_efi_capsule/conftest.py > create mode 100644 test/py/tests/test_efi_capsule/defs.py > create mode 100644 test/py/tests/test_efi_capsule/test_capsule_firmware.py > create mode 100644 test/py/tests/test_efi_capsule/uboot_bin_env.its > create mode 100644 tools/mkeficapsule.c >
On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > On 07.09.20 07:34, AKASHI Takahiro wrote: [snip] > > required and partly because there is a problem with virt-make-fs. > > What problem with virt-make-fs exists? How will this be solved? This I suspect is related to the difficulty of getting tests to run in all of our CI environments, where at least in one case they run and pass, rather than skip. This will require that the series get run through at least Travis and Azure before re-posting, at this point. Thanks.
On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > On 07.09.20 07:34, AKASHI Takahiro wrote: > > Summary > > ======= > > 'UpdateCapsule' is one of runtime services defined in UEFI specification > > and its aim is to allow a caller (OS) to pass information to the firmware, > > i.e. U-Boot. This is mostly used to update firmware binary on devices by > > instructions from OS. > > > > While 'UpdateCapsule' is a runtime services function, it is, at least > > initially, supported only before exiting boot services alike other runtime > > functions, [Get/]SetVariable. This is because modifying storage which may > > be shared with OS must be carefully designed and there is no general > > assumption that we can do it. > > > > Therefore, we practically support only "capsule on disk"; any capsule can > > be handed over to UEFI subsystem as a file on a specific file system. > > > > In this patch series, all the related definitions and structures are given > > as UEFI specification describes, and basic framework for capsule support > > is provided. Currently supported is > > * firmware update (Firmware Management Protocol or simply FMP) > > > > Most of functionality of firmware update is provided by FMP driver and > > it can be, by nature, system/platform-specific. So you can and should > > implement your own FMP driver(s) based on your system requirements. > > Under the current implementation, we provide two basic but generic > > drivers with two formats: > > * FIT image format (as used in TFTP update and dfu) > > * raw image format > > > > It's totally up to users which one, or both, should be used on users' > > system depending on user requirements. > > > > Quick usage > > =========== > > 1. You can create a capsule file with the following host command: > > > > $ mkeficapsule [--fit <fit image> | --raw <raw image>] <output file> > > > > 2. Put the file under: > > > > /EFI/UpdateCapsule of UEFI system partition > > > > 3. Specify firmware storage to be updated in "dfu_alt_info" variable > > (Please follow README.dfu for details.) > > > > ==> env set dfu_alt_info '...' > > > > 4. After setting up UEFI's OsIndications variable, reboot U-Boot: > > > > OsIndications <= EFI_OS_INDICATIONS_FILE_CAPSULE_DELIVERY_SUPPORTED > > > > Patch structure > > =============== > > Patch#1-#4,#12: preparatory patches > > Patch#5-#11,#13: main part of implementation > > Patch#14-#15: utilities > > Patch#16-#17: pytests > > > > [1] https://git.linaro.org/people/takahiro.akashi/u-boot.git efi/capsule > > > > Prerequisite patches > > ==================== > > In this version, the DFU change by Heinrich[2] is, at least temporarily, > > not a prerequisite due to the discussions[3]. > > > > [2] https://lists.denx.de/pipermail/u-boot/2020-July/420950.html > > [3] https://lists.denx.de/pipermail/u-boot/2020-August/424716.html > > The end of the discussion was that the patch is postponed to v2021.01 > like your series. My understanding was that the patch will be deferred until v2021.01 and we need more discussions. Right? > > > > Test > > ==== > > * passed all the pytests which are included in this patch series > > on sandbox build locally. > > > > Please note that the capsule pytest itself won't be run in the CI > > partly because some specific configuration for sandbox build is > > Shouldn't the sandbox_defconfig changes be added to your series? I won't hesitate to add them if Simon agrees. Since the current test removed the dependency on ENV_IS_IN_SPI_FLASH which was required in v3 or early, it would be more acceptable. > Is this the correct change: It would be fine. > diff --git a/configs/sandbox_defconfig b/configs/sandbox_defconfig > index 6e9f029cc9..da82afaa01 100644 > --- a/configs/sandbox_defconfig > +++ b/configs/sandbox_defconfig > @@ -129,6 +129,7 @@ CONFIG_DM_DEMO_SIMPLE=y > CONFIG_DM_DEMO_SHAPE=y > CONFIG_BOARD=y > CONFIG_BOARD_SANDBOX=y > +CONFIG_DFU_SF=y > CONFIG_DMA=y > CONFIG_DMA_CHANNELS=y > CONFIG_SANDBOX_DMA=y > @@ -261,6 +262,10 @@ CONFIG_CMD_DHRYSTONE=y > CONFIG_TPM=y > CONFIG_LZ4=y > CONFIG_ERRNO_STR=y > +CONFIG_EFI_RUNTIME_UPDATE_CAPSULE=y > +CONFIG_EFI_CAPSULE_ON_DISK=y > +CONFIG_EFI_CAPSULE_FIRMWARE_FIT=y > +CONFIG_EFI_CAPSULE_FIRMWARE_RAW=y > CONFIG_EFI_SECURE_BOOT=y > CONFIG_TEST_FDTDEC=y > CONFIG_UNIT_TEST=y > > > required and partly because there is a problem with virt-make-fs. > > What problem with virt-make-fs exists? How will this be solved? See my reply to Tom. -Takahiro Akashi > > See test_efi_capsule_firmware.py. > > > > Issues > > ====== > > * Timing of executing capsules-on-disk > > Currently, processing a capsule is triggered only as part of > > UEFI subsystem initialization. This means that, for example, > > firmware update, may not take place at system booting time and > > will potentially be delayed until a first call of any UEFI functions. > > => See patch#5 for my proposal > > * A bunch of warnings like > > WARNING: Use 'if (IS_ENABLED(CONFIG...))' instead of '#if or #ifdef' > > where possible > > I don't think that fixing those improves anything. > > Please, use 'if' instead of '#if' where possible so that the compiler > checks the code syntax even if the configuration is not set. Like with > '#if' the compiler will eliminate unused code. > > Best regards > > Heinrich > > > * Add a document in uefi.rst > > > > TODO's > > ====== > > (Won't be addressed in this series.) > > * capsule authentication > > * capsule dependency (dependency expression instruction set, or depex) > > * loading drivers in a capsule > > * handling RESET flag in a capsule and QeuryCapsuleCaps > > * full semantics of ESRT (EFI System Resource Table) > > * enabling capsule API at runtime > > * json capsule > > * recovery from update failure > > > > Changes > > ======= > > v6 (September 7, 2020) > > * temporarily drop the prerequisite patch[2] > > * add a missing field (dependencies) in efi_api.h (but never used) (Patch#10) > > * add a missing field (image_capsule_support) and related definitions > > in efi_api.h (Patch#10, #15) > > * cosmetic changes on constant definitions in efi_api.h (Patch#10) > > * strict check for INVALID_PARAMETER at GET_IMAGE_INFO api (Patch#11,#13) > > * fix warnings in pytest (Patch#16,#17) > > > > v5 (August 3, 2020) > > * removed superfluous type conversion at dfu_write_from_mem_addr() > > (Patch#2) > > * introduced a common helper function, efi_create_indexed_name() > > (Patch#6,#7,#8) > > * use efi_[get|set]_variable_int(), if necessary, with READ_ONLY > > (Patch#7,#8) > > * return EFI_UNSUPPORTED at Patch#7 > > * changed the word, number, to 'index' (Patch#7,#8) > > * removed 'ifdef CONFIG_EFI_CAPSULE_ON_DISK' from a header (Patch#8) > > * initialize 'CapsuleLast' in efi_init_obj_list() (Patch#7,#8) > > * added 'const' qualifier for filename argument at > > efi_capsule_[read|delete]_file() (Patch#8) > > > > v4 (July 22, 2020) > > * rebased to Heinrich's current efi-2020-10 > > * rework dfu-related code to align with Heinrich's change (Patch#1,#3) > > * change a type of 'addr' argument from int to 'void *' per Sughosh's > > comment (Patch#2-#3,#11-#12) > > * rework/simplify pytests (Patch#15-#16) > > - utilize virt-make-fs > > - drop Test Case 1 (updating U-Boot environment data) > > - remove useless definitions (MNT_PNT, EFI_CAPSULE_IMAGE_NAME) > > - apply autopep8 > > > > v3 (July 10, 2020) > > * rebased to Heinrich's current efi-2020-10-rc1 > > * refactor efi_firmware_[fit|raw]_get_image_info() (patch#11,#13) > > > > v2 (June 17, 2020) > > * rebased to v2020.07-rc4 > > * add preparatory patches for dfu (Patch#1-#5, #12) > > * rework FIT capsule driver to utilize dfu_alt_info instead of CONFIG_xxx > > (patch#11) > > * extend get_image_info() to correspond to dfu_alt_info > > (patch#11) > > * add a 'raw binary' capsule support > > (patch#13, #17) > > * allow multiple capsule formats (with different GUIDs) to be installed > > (patch#11, #13) > > * extend mkeficapsule command to accept additional parameters, like > > version/index/hardware instance for a capsule header info. > > (patch#15) > > * mkeficapsule can now also generate raw-binary capsule > > (patch#16) > > * add function descriptions > > * apply autopep8 to pytests and fix more against pylint > > > > v1 (April 27, 2020) > > * rebased to v2020.07-rc > > * removed already-merged patches (RFC's #1 to #4) > > * dropped 'variable update' capsule support (RFC's patch#10) > > * dropped 'variable configuration table' support (RFC's patch#11) > > (Those two should be discussed separately.) > > * add preparatory patches (patch#1/#2) > > * fix several build errors > > * rename some Kconfig options to be aligned with UEFI specification's terms > > (patch#3,4,6,7) > > * enforce UpdateCapsule API to be disabled after ExitBootServices (patch#3) > > * use config table, runtime_services_supported, instead of variable (patch#3) > > * make EFI_CAPSULE_ON_DISK buildable even if UpdateCapsule API is disabled > > (patch4) > > * support OsIndications, invoking capsule-on-disk only if the variable > > indicates so (patch#4) > > * introduced EFI_CAPSULE_ON_DISK_EARLY to invoke capsule-on-disk in U-Boot > > initialization (patch#4) > > * detect capsule files only if they are on EFI system partition (patch#4) > > * use printf, rather than EFI_PRINT, in error cases (patch#4) > > * use 'header_size' field to retrieve capsule data, adding sanity checks > > against capsule size (patch#6) > > * call fmpt driver interfaces with EFI_CALL (patch#6) > > * remove 'variable update capsule'-related code form mkeficapsule (patch#9) > > * add a test case of OsIndications not being set properly (patch#10) > > * adjust test scenario for EFI_CAPSULE_ON_DISK_EARLY (patch#10) > > * revise pytest scripts (patch#10) > > > > Initial release as RFC (March 17, 2020) > > > > AKASHI Takahiro (17): > > dfu: rename dfu_tftp_write() to dfu_write_by_name() > > dfu: modify an argument type for an address > > common: update: add a generic interface for FIT image > > dfu: export dfu_list > > efi_loader: add option to initialise EFI subsystem early > > efi_loader: add efi_create_indexed_name() > > efi_loader: define UpdateCapsule api > > efi_loader: capsule: add capsule_on_disk support > > efi_loader: capsule: add memory range capsule definitions > > efi_loader: capsule: support firmware update > > efi_loader: add firmware management protocol for FIT image > > dfu: add dfu_write_by_alt() > > efi_loader: add firmware management protocol for raw image > > cmd: add "efidebug capsule" command > > tools: add mkeficapsule command for UEFI capsule update > > test/py: efi_capsule: test for FIT image capsule > > test/py: efi_capsule: test for raw image capsule > > > > cmd/efidebug.c | 235 +++++ > > common/Kconfig | 15 + > > common/Makefile | 3 +- > > common/board_r.c | 6 + > > common/main.c | 4 + > > common/update.c | 77 +- > > drivers/dfu/Kconfig | 6 + > > drivers/dfu/Makefile | 2 +- > > drivers/dfu/dfu.c | 2 +- > > drivers/dfu/dfu_alt.c | 125 +++ > > drivers/dfu/dfu_tftp.c | 65 -- > > include/dfu.h | 57 +- > > include/efi_api.h | 166 ++++ > > include/efi_loader.h | 30 + > > include/image.h | 12 + > > lib/efi_loader/Kconfig | 72 ++ > > lib/efi_loader/Makefile | 2 + > > lib/efi_loader/efi_capsule.c | 917 ++++++++++++++++++ > > lib/efi_loader/efi_firmware.c | 403 ++++++++ > > lib/efi_loader/efi_runtime.c | 104 +- > > lib/efi_loader/efi_setup.c | 106 +- > > test/py/tests/test_efi_capsule/conftest.py | 72 ++ > > test/py/tests/test_efi_capsule/defs.py | 12 + > > .../test_efi_capsule/test_capsule_firmware.py | 241 +++++ > > .../tests/test_efi_capsule/uboot_bin_env.its | 36 + > > tools/Makefile | 2 + > > tools/mkeficapsule.c | 239 +++++ > > 27 files changed, 2871 insertions(+), 140 deletions(-) > > create mode 100644 drivers/dfu/dfu_alt.c > > delete mode 100644 drivers/dfu/dfu_tftp.c > > create mode 100644 lib/efi_loader/efi_capsule.c > > create mode 100644 lib/efi_loader/efi_firmware.c > > create mode 100644 test/py/tests/test_efi_capsule/conftest.py > > create mode 100644 test/py/tests/test_efi_capsule/defs.py > > create mode 100644 test/py/tests/test_efi_capsule/test_capsule_firmware.py > > create mode 100644 test/py/tests/test_efi_capsule/uboot_bin_env.its > > create mode 100644 tools/mkeficapsule.c > > >
On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > On 07.09.20 07:34, AKASHI Takahiro wrote: > [snip] > > > required and partly because there is a problem with virt-make-fs. > > > > What problem with virt-make-fs exists? How will this be solved? > > This I suspect is related to the difficulty of getting tests to run in > all of our CI environments, where at least in one case they run and > pass, rather than skip. This will require that the series get run > through at least Travis and Azure before re-posting, at this point. > Thanks. I think that we discussed this issue several times before. https://lists.denx.de/pipermail/u-boot/2020-July/419430.html https://lists.denx.de/pipermail/u-boot/2020-July/420712.html I have not implemented a fallback handling in case of virt-make-fs failure as Heinrich doesn't like it (even dropped the code from my patches.) The only solution is Heinrich's patch: https://lists.denx.de/pipermail/u-boot/2020-July/419976.html -Takahiro Akashi > -- > Tom
On Thu, Sep 10, 2020 at 11:52:37AM +0900, AKASHI Takahiro wrote: > On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > > On 07.09.20 07:34, AKASHI Takahiro wrote: > > [snip] > > > > required and partly because there is a problem with virt-make-fs. > > > > > > What problem with virt-make-fs exists? How will this be solved? > > > > This I suspect is related to the difficulty of getting tests to run in > > all of our CI environments, where at least in one case they run and > > pass, rather than skip. This will require that the series get run > > through at least Travis and Azure before re-posting, at this point. > > Thanks. > > I think that we discussed this issue several times before. > > https://lists.denx.de/pipermail/u-boot/2020-July/419430.html > https://lists.denx.de/pipermail/u-boot/2020-July/420712.html > > I have not implemented a fallback handling in case of virt-make-fs failure > as Heinrich doesn't like it (even dropped the code from my patches.) > > The only solution is Heinrich's patch: > https://lists.denx.de/pipermail/u-boot/2020-July/419976.html The problem is that Heinrich's patch breaks things. A complete solution is required. That patch might be part of the solution, along with perhaps a rework of the logic to try and gracefully catch and re-try when we have a problem with one of the possible tools.
On Wed, Sep 09, 2020 at 10:58:53PM -0400, Tom Rini wrote: > On Thu, Sep 10, 2020 at 11:52:37AM +0900, AKASHI Takahiro wrote: > > On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > > > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > > > On 07.09.20 07:34, AKASHI Takahiro wrote: > > > [snip] > > > > > required and partly because there is a problem with virt-make-fs. > > > > > > > > What problem with virt-make-fs exists? How will this be solved? > > > > > > This I suspect is related to the difficulty of getting tests to run in > > > all of our CI environments, where at least in one case they run and > > > pass, rather than skip. This will require that the series get run > > > through at least Travis and Azure before re-posting, at this point. > > > Thanks. > > > > I think that we discussed this issue several times before. > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419430.html > > https://lists.denx.de/pipermail/u-boot/2020-July/420712.html > > > > I have not implemented a fallback handling in case of virt-make-fs failure > > as Heinrich doesn't like it (even dropped the code from my patches.) > > > > The only solution is Heinrich's patch: > > https://lists.denx.de/pipermail/u-boot/2020-July/419976.html > > The problem is that Heinrich's patch breaks things. A complete solution > is required. What kind of issue did you see? > That patch might be part of the solution, along with > perhaps a rework of the logic to try and gracefully catch and re-try > when we have a problem with one of the possible tools. Anyway, I can't do anything here as Heinrich rejects any sudo-based solution. See: https://lists.denx.de/pipermail/u-boot/2020-July/419951.html So he is responsible for fixing it. -Takahiro Akashi > -- > Tom
On Fri, Sep 11, 2020 at 09:52:10AM +0900, AKASHI Takahiro wrote: > On Wed, Sep 09, 2020 at 10:58:53PM -0400, Tom Rini wrote: > > On Thu, Sep 10, 2020 at 11:52:37AM +0900, AKASHI Takahiro wrote: > > > On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > > > > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > > > > On 07.09.20 07:34, AKASHI Takahiro wrote: > > > > [snip] > > > > > > required and partly because there is a problem with virt-make-fs. > > > > > > > > > > What problem with virt-make-fs exists? How will this be solved? > > > > > > > > This I suspect is related to the difficulty of getting tests to run in > > > > all of our CI environments, where at least in one case they run and > > > > pass, rather than skip. This will require that the series get run > > > > through at least Travis and Azure before re-posting, at this point. > > > > Thanks. > > > > > > I think that we discussed this issue several times before. > > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419430.html > > > https://lists.denx.de/pipermail/u-boot/2020-July/420712.html > > > > > > I have not implemented a fallback handling in case of virt-make-fs failure > > > as Heinrich doesn't like it (even dropped the code from my patches.) > > > > > > The only solution is Heinrich's patch: > > > https://lists.denx.de/pipermail/u-boot/2020-July/419976.html > > > > The problem is that Heinrich's patch breaks things. A complete solution > > is required. > > What kind of issue did you see? I believe it's in the thread. But in short, we then stop running the other filesystem tests as they get skipped. > > That patch might be part of the solution, along with > > perhaps a rework of the logic to try and gracefully catch and re-try > > when we have a problem with one of the possible tools. > > Anyway, I can't do anything here as Heinrich rejects any sudo-based > solution. See: > > https://lists.denx.de/pipermail/u-boot/2020-July/419951.html > > So he is responsible for fixing it. No, it's on you to provide a solution that runs in both CI and developer machines. It's not sudo or make-virt-fs, it's get the try/catch logic correct in the test framework such that if we can't run sudo we do run make-virt-fs, and if we're able to run the tests somehow, we run the tests somehow.
Tom, On Thu, Sep 10, 2020 at 09:02:59PM -0400, Tom Rini wrote: > On Fri, Sep 11, 2020 at 09:52:10AM +0900, AKASHI Takahiro wrote: > > On Wed, Sep 09, 2020 at 10:58:53PM -0400, Tom Rini wrote: > > > On Thu, Sep 10, 2020 at 11:52:37AM +0900, AKASHI Takahiro wrote: > > > > On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > > > > > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > > > > > On 07.09.20 07:34, AKASHI Takahiro wrote: > > > > > [snip] > > > > > > > required and partly because there is a problem with virt-make-fs. > > > > > > > > > > > > What problem with virt-make-fs exists? How will this be solved? > > > > > > > > > > This I suspect is related to the difficulty of getting tests to run in > > > > > all of our CI environments, where at least in one case they run and > > > > > pass, rather than skip. This will require that the series get run > > > > > through at least Travis and Azure before re-posting, at this point. > > > > > Thanks. > > > > > > > > I think that we discussed this issue several times before. > > > > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419430.html > > > > https://lists.denx.de/pipermail/u-boot/2020-July/420712.html > > > > > > > > I have not implemented a fallback handling in case of virt-make-fs failure > > > > as Heinrich doesn't like it (even dropped the code from my patches.) > > > > > > > > The only solution is Heinrich's patch: > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419976.html > > > > > > The problem is that Heinrich's patch breaks things. A complete solution > > > is required. > > > > What kind of issue did you see? > > I believe it's in the thread. But in short, we then stop running the > other filesystem tests as they get skipped. I believe that the reason of failures in filesystem tests is the same; /boot/vmlinuz* have not 'read' permission to users and any commands in libguestfs-tools, including guestmount and virt-make-fs, cannot set up a virtual machine internally used in those commands. So again, the only solution is to apply Heinrich's patch. > > > That patch might be part of the solution, along with > > > perhaps a rework of the logic to try and gracefully catch and re-try > > > when we have a problem with one of the possible tools. > > > > Anyway, I can't do anything here as Heinrich rejects any sudo-based > > solution. See: > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419951.html > > > > So he is responsible for fixing it. > > No, it's on you to provide a solution that runs in both CI and developer > machines. It's not sudo or make-virt-fs, it's get the try/catch logic > correct in the test framework such that if we can't run sudo The point is that Heinrich doesn't accept any sudo solution and that he modified my tests to solely use virt-make-fs by the patch above. So what can I do here? > we do run > make-virt-fs, and if we're able to run the tests somehow, we run the > tests somehow. Again, Heinrich's patch will fix it. -Takahiro Akashi > > -- > Tom
On Fri, Sep 11, 2020 at 11:31:11AM +0900, AKASHI Takahiro wrote: > Tom, > > On Thu, Sep 10, 2020 at 09:02:59PM -0400, Tom Rini wrote: > > On Fri, Sep 11, 2020 at 09:52:10AM +0900, AKASHI Takahiro wrote: > > > On Wed, Sep 09, 2020 at 10:58:53PM -0400, Tom Rini wrote: > > > > On Thu, Sep 10, 2020 at 11:52:37AM +0900, AKASHI Takahiro wrote: > > > > > On Wed, Sep 09, 2020 at 12:56:28PM -0400, Tom Rini wrote: > > > > > > On Wed, Sep 09, 2020 at 04:48:30PM +0200, Heinrich Schuchardt wrote: > > > > > > > On 07.09.20 07:34, AKASHI Takahiro wrote: > > > > > > [snip] > > > > > > > > required and partly because there is a problem with virt-make-fs. > > > > > > > > > > > > > > What problem with virt-make-fs exists? How will this be solved? > > > > > > > > > > > > This I suspect is related to the difficulty of getting tests to run in > > > > > > all of our CI environments, where at least in one case they run and > > > > > > pass, rather than skip. This will require that the series get run > > > > > > through at least Travis and Azure before re-posting, at this point. > > > > > > Thanks. > > > > > > > > > > I think that we discussed this issue several times before. > > > > > > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419430.html > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/420712.html > > > > > > > > > > I have not implemented a fallback handling in case of virt-make-fs failure > > > > > as Heinrich doesn't like it (even dropped the code from my patches.) > > > > > > > > > > The only solution is Heinrich's patch: > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419976.html > > > > > > > > The problem is that Heinrich's patch breaks things. A complete solution > > > > is required. > > > > > > What kind of issue did you see? > > > > I believe it's in the thread. But in short, we then stop running the > > other filesystem tests as they get skipped. > > I believe that the reason of failures in filesystem tests is the same; > /boot/vmlinuz* have not 'read' permission to users and any commands > in libguestfs-tools, including guestmount and virt-make-fs, cannot > set up a virtual machine internally used in those commands. > > So again, the only solution is to apply Heinrich's patch. No, I believe I fixed that as part of digging in to this at the time. > > > > That patch might be part of the solution, along with > > > > perhaps a rework of the logic to try and gracefully catch and re-try > > > > when we have a problem with one of the possible tools. > > > > > > Anyway, I can't do anything here as Heinrich rejects any sudo-based > > > solution. See: > > > > > > https://lists.denx.de/pipermail/u-boot/2020-July/419951.html > > > > > > So he is responsible for fixing it. > > > > No, it's on you to provide a solution that runs in both CI and developer > > machines. It's not sudo or make-virt-fs, it's get the try/catch logic > > correct in the test framework such that if we can't run sudo > > The point is that Heinrich doesn't accept any sudo solution and that > he modified my tests to solely use virt-make-fs by the patch above. > So what can I do here? > > > we do run > > make-virt-fs, and if we're able to run the tests somehow, we run the > > tests somehow. > > Again, Heinrich's patch will fix it. Please put together a series that passes and runs on CI and doesn't make other tests stop running. That's the requirement. If it doesn't run in CI, it doesn't get run frequently enough. If it makes some CI tests stop running, that's a problem too.
Summary ======= 'UpdateCapsule' is one of runtime services defined in UEFI specification and its aim is to allow a caller (OS) to pass information to the firmware, i.e. U-Boot. This is mostly used to update firmware binary on devices by instructions from OS. While 'UpdateCapsule' is a runtime services function, it is, at least initially, supported only before exiting boot services alike other runtime functions, [Get/]SetVariable. This is because modifying storage which may be shared with OS must be carefully designed and there is no general assumption that we can do it. Therefore, we practically support only "capsule on disk"; any capsule can be handed over to UEFI subsystem as a file on a specific file system. In this patch series, all the related definitions and structures are given as UEFI specification describes, and basic framework for capsule support is provided. Currently supported is * firmware update (Firmware Management Protocol or simply FMP) Most of functionality of firmware update is provided by FMP driver and it can be, by nature, system/platform-specific. So you can and should implement your own FMP driver(s) based on your system requirements. Under the current implementation, we provide two basic but generic drivers with two formats: * FIT image format (as used in TFTP update and dfu) * raw image format It's totally up to users which one, or both, should be used on users' system depending on user requirements. Quick usage =========== 1. You can create a capsule file with the following host command: $ mkeficapsule [--fit <fit image> | --raw <raw image>] <output file> 2. Put the file under: /EFI/UpdateCapsule of UEFI system partition 3. Specify firmware storage to be updated in "dfu_alt_info" variable (Please follow README.dfu for details.) ==> env set dfu_alt_info '...' 4. After setting up UEFI's OsIndications variable, reboot U-Boot: OsIndications <= EFI_OS_INDICATIONS_FILE_CAPSULE_DELIVERY_SUPPORTED Patch structure =============== Patch#1-#4,#12: preparatory patches Patch#5-#11,#13: main part of implementation Patch#14-#15: utilities Patch#16-#17: pytests [1] https://git.linaro.org/people/takahiro.akashi/u-boot.git efi/capsule Prerequisite patches ==================== In this version, the DFU change by Heinrich[2] is, at least temporarily, not a prerequisite due to the discussions[3]. [2] https://lists.denx.de/pipermail/u-boot/2020-July/420950.html [3] https://lists.denx.de/pipermail/u-boot/2020-August/424716.html Test ==== * passed all the pytests which are included in this patch series on sandbox build locally. Please note that the capsule pytest itself won't be run in the CI partly because some specific configuration for sandbox build is required and partly because there is a problem with virt-make-fs. See test_efi_capsule_firmware.py. Issues ====== * Timing of executing capsules-on-disk Currently, processing a capsule is triggered only as part of UEFI subsystem initialization. This means that, for example, firmware update, may not take place at system booting time and will potentially be delayed until a first call of any UEFI functions. => See patch#5 for my proposal * A bunch of warnings like WARNING: Use 'if (IS_ENABLED(CONFIG...))' instead of '#if or #ifdef' where possible I don't think that fixing those improves anything. * Add a document in uefi.rst TODO's ====== (Won't be addressed in this series.) * capsule authentication * capsule dependency (dependency expression instruction set, or depex) * loading drivers in a capsule * handling RESET flag in a capsule and QeuryCapsuleCaps * full semantics of ESRT (EFI System Resource Table) * enabling capsule API at runtime * json capsule * recovery from update failure Changes ======= v6 (September 7, 2020) * temporarily drop the prerequisite patch[2] * add a missing field (dependencies) in efi_api.h (but never used) (Patch#10) * add a missing field (image_capsule_support) and related definitions in efi_api.h (Patch#10, #15) * cosmetic changes on constant definitions in efi_api.h (Patch#10) * strict check for INVALID_PARAMETER at GET_IMAGE_INFO api (Patch#11,#13) * fix warnings in pytest (Patch#16,#17) v5 (August 3, 2020) * removed superfluous type conversion at dfu_write_from_mem_addr() (Patch#2) * introduced a common helper function, efi_create_indexed_name() (Patch#6,#7,#8) * use efi_[get|set]_variable_int(), if necessary, with READ_ONLY (Patch#7,#8) * return EFI_UNSUPPORTED at Patch#7 * changed the word, number, to 'index' (Patch#7,#8) * removed 'ifdef CONFIG_EFI_CAPSULE_ON_DISK' from a header (Patch#8) * initialize 'CapsuleLast' in efi_init_obj_list() (Patch#7,#8) * added 'const' qualifier for filename argument at efi_capsule_[read|delete]_file() (Patch#8) v4 (July 22, 2020) * rebased to Heinrich's current efi-2020-10 * rework dfu-related code to align with Heinrich's change (Patch#1,#3) * change a type of 'addr' argument from int to 'void *' per Sughosh's comment (Patch#2-#3,#11-#12) * rework/simplify pytests (Patch#15-#16) - utilize virt-make-fs - drop Test Case 1 (updating U-Boot environment data) - remove useless definitions (MNT_PNT, EFI_CAPSULE_IMAGE_NAME) - apply autopep8 v3 (July 10, 2020) * rebased to Heinrich's current efi-2020-10-rc1 * refactor efi_firmware_[fit|raw]_get_image_info() (patch#11,#13) v2 (June 17, 2020) * rebased to v2020.07-rc4 * add preparatory patches for dfu (Patch#1-#5, #12) * rework FIT capsule driver to utilize dfu_alt_info instead of CONFIG_xxx (patch#11) * extend get_image_info() to correspond to dfu_alt_info (patch#11) * add a 'raw binary' capsule support (patch#13, #17) * allow multiple capsule formats (with different GUIDs) to be installed (patch#11, #13) * extend mkeficapsule command to accept additional parameters, like version/index/hardware instance for a capsule header info. (patch#15) * mkeficapsule can now also generate raw-binary capsule (patch#16) * add function descriptions * apply autopep8 to pytests and fix more against pylint v1 (April 27, 2020) * rebased to v2020.07-rc * removed already-merged patches (RFC's #1 to #4) * dropped 'variable update' capsule support (RFC's patch#10) * dropped 'variable configuration table' support (RFC's patch#11) (Those two should be discussed separately.) * add preparatory patches (patch#1/#2) * fix several build errors * rename some Kconfig options to be aligned with UEFI specification's terms (patch#3,4,6,7) * enforce UpdateCapsule API to be disabled after ExitBootServices (patch#3) * use config table, runtime_services_supported, instead of variable (patch#3) * make EFI_CAPSULE_ON_DISK buildable even if UpdateCapsule API is disabled (patch4) * support OsIndications, invoking capsule-on-disk only if the variable indicates so (patch#4) * introduced EFI_CAPSULE_ON_DISK_EARLY to invoke capsule-on-disk in U-Boot initialization (patch#4) * detect capsule files only if they are on EFI system partition (patch#4) * use printf, rather than EFI_PRINT, in error cases (patch#4) * use 'header_size' field to retrieve capsule data, adding sanity checks against capsule size (patch#6) * call fmpt driver interfaces with EFI_CALL (patch#6) * remove 'variable update capsule'-related code form mkeficapsule (patch#9) * add a test case of OsIndications not being set properly (patch#10) * adjust test scenario for EFI_CAPSULE_ON_DISK_EARLY (patch#10) * revise pytest scripts (patch#10) Initial release as RFC (March 17, 2020) AKASHI Takahiro (17): dfu: rename dfu_tftp_write() to dfu_write_by_name() dfu: modify an argument type for an address common: update: add a generic interface for FIT image dfu: export dfu_list efi_loader: add option to initialise EFI subsystem early efi_loader: add efi_create_indexed_name() efi_loader: define UpdateCapsule api efi_loader: capsule: add capsule_on_disk support efi_loader: capsule: add memory range capsule definitions efi_loader: capsule: support firmware update efi_loader: add firmware management protocol for FIT image dfu: add dfu_write_by_alt() efi_loader: add firmware management protocol for raw image cmd: add "efidebug capsule" command tools: add mkeficapsule command for UEFI capsule update test/py: efi_capsule: test for FIT image capsule test/py: efi_capsule: test for raw image capsule cmd/efidebug.c | 235 +++++ common/Kconfig | 15 + common/Makefile | 3 +- common/board_r.c | 6 + common/main.c | 4 + common/update.c | 77 +- drivers/dfu/Kconfig | 6 + drivers/dfu/Makefile | 2 +- drivers/dfu/dfu.c | 2 +- drivers/dfu/dfu_alt.c | 125 +++ drivers/dfu/dfu_tftp.c | 65 -- include/dfu.h | 57 +- include/efi_api.h | 166 ++++ include/efi_loader.h | 30 + include/image.h | 12 + lib/efi_loader/Kconfig | 72 ++ lib/efi_loader/Makefile | 2 + lib/efi_loader/efi_capsule.c | 917 ++++++++++++++++++ lib/efi_loader/efi_firmware.c | 403 ++++++++ lib/efi_loader/efi_runtime.c | 104 +- lib/efi_loader/efi_setup.c | 106 +- test/py/tests/test_efi_capsule/conftest.py | 72 ++ test/py/tests/test_efi_capsule/defs.py | 12 + .../test_efi_capsule/test_capsule_firmware.py | 241 +++++ .../tests/test_efi_capsule/uboot_bin_env.its | 36 + tools/Makefile | 2 + tools/mkeficapsule.c | 239 +++++ 27 files changed, 2871 insertions(+), 140 deletions(-) create mode 100644 drivers/dfu/dfu_alt.c delete mode 100644 drivers/dfu/dfu_tftp.c create mode 100644 lib/efi_loader/efi_capsule.c create mode 100644 lib/efi_loader/efi_firmware.c create mode 100644 test/py/tests/test_efi_capsule/conftest.py create mode 100644 test/py/tests/test_efi_capsule/defs.py create mode 100644 test/py/tests/test_efi_capsule/test_capsule_firmware.py create mode 100644 test/py/tests/test_efi_capsule/uboot_bin_env.its create mode 100644 tools/mkeficapsule.c