Public bug reported:
Ubuntu 24.04 kernel 6.8.0-139.139, running on Linux Mint 22.3, which ships the
Ubuntu
kernel unmodified.
## Symptom
On every boot, before Plymouth takes over, the screen shows:
EFI stub: WARNING: Failed to measure data for event 1:
0x800000000000000b
The system then boots normally and nothing is broken. The problem is that the
user is shown
this at all.
Expected behaviour on a machine booting with `quiet splash`: the Plymouth
splash, or a black
screen — no kernel text whatsoever. What actually happens is a lone line saying
WARNING on an
otherwise black screen, which a non-technical user cannot tell apart from a
real failure. The
diagnostic belongs somewhere a technical user can go look for it, not on the
boot console of
every machine with this firmware. On an LTS release this is a first-impression
bug.
## Why `quiet` does not suppress it — the actual defect
The message comes from `efi_measure_tagged_event()`,
`drivers/firmware/efi/libstub/efi-stub-helper.c:271` (v6.8).
Event 1 is `EFISTUB_EVT_LOAD_OPTIONS`, and its call site is inside
`efi_convert_cmdline()`:
efi-stub-helper.c:292
if (options_size > 0)
efi_measure_tagged_event((unsigned long)options, options_size,
EFISTUB_EVT_LOAD_OPTIONS);
`efi_convert_cmdline()` necessarily runs *before* `efi_parse_options()` — the
command line has
to be converted from UTF-16 before anything can parse it (`efi-stub.c:124` then
`:141`;
`x86-stub.c:487` then `:880`).
`quiet` is precisely what sets `efi_loglevel = CONSOLE_LOGLEVEL_QUIET`
(`efi-stub-helper.c:72-73`), and `efi_printk()` filters only on `efi_loglevel`
(`printk.c:135`). At the moment this warning is emitted, `efi_loglevel` is still
`CONSOLE_LOGLEVEL_DEFAULT`.
So `quiet splash` cannot suppress this particular warning, by construction.
Neither the user
nor the distro can turn it off without rebuilding the kernel.
Still present in mainline v6.17: same call site, `efi-stub-helper.c:320`
/ `:341`.
## Why the firmware refuses
`0x800000000000000b` = `EFI_VOLUME_FULL` (bit 63 + 11, UEFI spec appendix D).
The firmware's
`EFI_TCG2_PROTOCOL.HashLogExtendEvent()` rejects the event because its TCG
event log has no
room left. Others have reported the same status correlating with the number of
UEFI boot
entries: https://lkml.iu.edu/hypermail/linux/kernel/2411.2/01195.html
On this machine Secure Boot is disabled and nothing consumes the PCR
measurements, so the
failed measurement has no practical consequence. It *would* matter for
TPM-sealed secrets
(`systemd-cryptenroll --tpm2-device`), but a console warning during a splash
boot is not a
useful way to signal that either — the affected tooling checks the event log
itself.
## Requested
The user must not see this line. Whether the measurement itself is fixed is
secondary — a
firmware event log that is full is not actionable by the person watching the
boot.
Any one of:
1. Perform the LOAD_OPTIONS measurement after option parsing (or parse `quiet`
before it), so
the documented way of silencing early boot output also covers this path and
the splash is
all the user sees.
2. Downgrade firmware refusals such as `EFI_VOLUME_FULL` to `efi_debug()` for
this event, so
it is reachable with `efi=debug` by whoever actually wants it and invisible
otherwise.
3. Ubuntu-side: carry (1) or (2) as a patch. 24.04 is supported into 2029 and
users on
affected firmware see this on every single boot until then.
## System
Linux Mint 22.3 Zena (Ubuntu 24.04 noble base), Cinnamon
linux-image-6.8.0-139-generic 6.8.0-139.139
/proc/version_signature: Ubuntu 6.8.0-139.139-generic 6.8.12
Lenovo ThinkPad T14 Gen 2i (20W100DLBO), BIOS N34ET61W (1.61), 2024-03-08
11th Gen Intel Core i5-1145G7
TPM: Infineon 2.0 (tpm_tis IFX0785:00, device-id 0x1B, rev-id 22)
Secure Boot: disabled
/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-6.8.0-139-generic
root=UUID=<redacted> ro quiet splash
25 UEFI boot entries (mostly Lenovo firmware built-ins)
plymouth 24.004.60-1ubuntu7.2
Note: the message is absent from dmesg/journalctl — EFI stub output goes to the
EFI console
before ExitBootServices. It can only be observed on screen, or with
`earlycon=efifb`.
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
** Tags: efistub noble tpm uefi
** Description changed:
Ubuntu 24.04 kernel 6.8.0-139.139, running on Linux Mint 22.3, which ships
the Ubuntu
kernel unmodified.
## Symptom
On every boot, before Plymouth takes over, the screen shows:
EFI stub: WARNING: Failed to measure data for event 1:
0x800000000000000b
- The system then boots normally and nothing is broken. The problem is
presentational: a raw
- kernel warning is displayed on a machine booting with `quiet splash`. It is
the only thing on
- an otherwise black screen, it says WARNING, and a non-technical user has no
way to tell it
- apart from a real failure. On an LTS release that is a first-impression bug.
+ The system then boots normally and nothing is broken. The problem is that the
user is shown
+ this at all.
+
+ Expected behaviour on a machine booting with `quiet splash`: the Plymouth
splash, or a black
+ screen — no kernel text whatsoever. What actually happens is a lone line
saying WARNING on an
+ otherwise black screen, which a non-technical user cannot tell apart from a
real failure. The
+ diagnostic belongs somewhere a technical user can go look for it, not on the
boot console of
+ every machine with this firmware. On an LTS release this is a
first-impression bug.
## Why `quiet` does not suppress it — the actual defect
The message comes from `efi_measure_tagged_event()`,
`drivers/firmware/efi/libstub/efi-stub-helper.c:271` (v6.8).
Event 1 is `EFISTUB_EVT_LOAD_OPTIONS`, and its call site is inside
`efi_convert_cmdline()`:
efi-stub-helper.c:292
if (options_size > 0)
efi_measure_tagged_event((unsigned long)options, options_size,
EFISTUB_EVT_LOAD_OPTIONS);
`efi_convert_cmdline()` necessarily runs *before* `efi_parse_options()` — the
command line has
to be converted from UTF-16 before anything can parse it (`efi-stub.c:124`
then `:141`;
`x86-stub.c:487` then `:880`).
`quiet` is precisely what sets `efi_loglevel = CONSOLE_LOGLEVEL_QUIET`
(`efi-stub-helper.c:72-73`), and `efi_printk()` filters only on `efi_loglevel`
(`printk.c:135`). At the moment this warning is emitted, `efi_loglevel` is
still
`CONSOLE_LOGLEVEL_DEFAULT`.
So `quiet splash` cannot suppress this particular warning, by construction.
Neither the user
nor the distro can turn it off without rebuilding the kernel.
Still present in mainline v6.17: same call site, `efi-stub-helper.c:320`
/ `:341`.
## Why the firmware refuses
`0x800000000000000b` = `EFI_VOLUME_FULL` (bit 63 + 11, UEFI spec appendix D).
The firmware's
`EFI_TCG2_PROTOCOL.HashLogExtendEvent()` rejects the event because its TCG
event log has no
room left. Others have reported the same status correlating with the number
of UEFI boot
entries: https://lkml.iu.edu/hypermail/linux/kernel/2411.2/01195.html
On this machine Secure Boot is disabled and nothing consumes the PCR
measurements, so the
failed measurement has no practical consequence. It *would* matter for
TPM-sealed secrets
(`systemd-cryptenroll --tpm2-device`), but a console warning during a splash
boot is not a
useful way to signal that either — the affected tooling checks the event log
itself.
## Requested
+ The user must not see this line. Whether the measurement itself is fixed is
secondary — a
+ firmware event log that is full is not actionable by the person watching the
boot.
+
Any one of:
1. Perform the LOAD_OPTIONS measurement after option parsing (or parse
`quiet` before it), so
- the documented way of silencing early boot output also covers this path.
- 2. Downgrade firmware refusals such as `EFI_VOLUME_FULL` to `efi_debug()` for
this event — a
- full firmware event log is not something a user can act on.
+ the documented way of silencing early boot output also covers this path
and the splash is
+ all the user sees.
+ 2. Downgrade firmware refusals such as `EFI_VOLUME_FULL` to `efi_debug()` for
this event, so
+ it is reachable with `efi=debug` by whoever actually wants it and
invisible otherwise.
3. Ubuntu-side: carry (1) or (2) as a patch. 24.04 is supported into 2029 and
users on
affected firmware see this on every single boot until then.
+
## System
Linux Mint 22.3 Zena (Ubuntu 24.04 noble base), Cinnamon
linux-image-6.8.0-139-generic 6.8.0-139.139
/proc/version_signature: Ubuntu 6.8.0-139.139-generic 6.8.12
Lenovo ThinkPad T14 Gen 2i (20W100DLBO), BIOS N34ET61W (1.61), 2024-03-08
11th Gen Intel Core i5-1145G7
TPM: Infineon 2.0 (tpm_tis IFX0785:00, device-id 0x1B, rev-id 22)
Secure Boot: disabled
/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-6.8.0-139-generic
root=UUID=<redacted> ro quiet splash
25 UEFI boot entries (mostly Lenovo firmware built-ins)
plymouth 24.004.60-1ubuntu7.2
Note: the message is absent from dmesg/journalctl — EFI stub output goes to
the EFI console
before ExitBootServices. It can only be observed on screen, or with
`earlycon=efifb`.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2167519
Title:
efistub: "Failed to measure data for event 1" warning bypasses `quiet`
and is shown on splash boot
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167519/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
