mbox series

[v6,00/17] efi_loader: add capsule update support

Message ID 20200907053426.1675646-1-takahiro.akashi@linaro.org
Headers show
Series efi_loader: add capsule update support | expand

Message

AKASHI Takahiro Sept. 7, 2020, 5:34 a.m. UTC
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

-- 
2.28.0

Comments

Heinrich Schuchardt Sept. 9, 2020, 2:48 p.m. UTC | #1
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

>
Tom Rini Sept. 9, 2020, 4:56 p.m. UTC | #2
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.

-- 
Tom
AKASHI Takahiro Sept. 10, 2020, 2:43 a.m. UTC | #3
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

> >

>
AKASHI Takahiro Sept. 10, 2020, 2:52 a.m. UTC | #4
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
Tom Rini Sept. 10, 2020, 2:58 a.m. UTC | #5
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.

-- 
Tom
AKASHI Takahiro Sept. 11, 2020, 12:52 a.m. UTC | #6
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
Tom Rini Sept. 11, 2020, 1:02 a.m. UTC | #7
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
AKASHI Takahiro Sept. 11, 2020, 2:31 a.m. UTC | #8
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
Tom Rini Sept. 11, 2020, 2:44 a.m. UTC | #9
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.

-- 
Tom