Linux EFI development
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Ard Biesheuvel" <ardb+git@google.com>, linux-efi@vger.kernel.org
Cc: "Ard Biesheuvel" <ardb@kernel.org>, "Kees Cook" <kees@kernel.org>,
	"Guilherme G. Piccoli" <gpiccoli@igalia.com>
Subject: Re: [PATCH 2/2] efi: pstore: Convert into a platform driver for 'efivars'
Date: Fri, 20 Nov 2026 13:39:48 +0000	[thread overview]
Message-ID: <sashiko-outbox-225258@kernel.org> (raw)
In-Reply-To: <20261120131827.480055-6-ardb+git@google.com>

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] efi: pstore: missing efivars device for non-generic efivars
  backends
- [Medium] efi: pstore: stale efivars_pstore_autoprobe_disable breaks
  manual bind
It also noticed 1 related pre-existing issue this patch does not
introduce:
- [Critical] efi: use-after-free in __efi_queue_work() [1]
--

commit ff714ad400a8da4c7c0e25c05fa514c713686cad
Author: Ard Biesheuvel <ardb@kernel.org>

efi: pstore: Convert into a platform driver for 'efivars'

This commit converts efi-pstore into a platform driver matching the
"efivars" platform device and only registers that device when
EFI_RT_SUPPORTED_SET_VARIABLE is advertised.

> diff --git a/drivers/firmware/efi/efi-pstore.c b/drivers/firmware/efi/efi-pstore.c
> index 19a7e4fe5b524..44571fa1dd885 100644
> --- a/drivers/firmware/efi/efi-pstore.c
> +++ b/drivers/firmware/efi/efi-pstore.c
[ ... ]
> @@ -228,13 +231,17 @@ static struct pstore_info efi_pstore_info = {
>  	.erase		= efi_pstore_erase,
>  };
>  
> -static __init int efivars_pstore_init(void)
> +static int efi_pstore_probe(struct platform_device *pdev)
>  {
> +	int ret;
> +
>  	if (!efivar_supports_writes())
> -		return 0;
> +		return -ENODEV;
>  
> -	if (efivars_pstore_disable)
> -		return 0;
> +	if (efivars_pstore_autoprobe_disable) {
> +		efivars_pstore_autoprobe_disable = false;
> +		return -ENODEV;
> +	}

[ ... ]

> +static __init int efivars_pstore_init(void)
> +{
> +	efivars_pstore_autoprobe_disable = efivars_pstore_disable;
> +	return platform_driver_register(&efi_pstore_driver);
> +}

[Severity: Medium]
Can efivars_pstore_autoprobe_disable stay true after
efivars_pstore_init() returns, causing the first manual bind via sysfs to
fail with -ENODEV?

If pstore_disable is set and platform_driver_register() does not reach the
efivars_pstore_autoprobe_disable check in efi_pstore_probe() (for example
when drivers_autoprobe is disabled on the platform bus, or when
efivar_supports_writes() is false during initial registration before a
late efivars provider registers), efivars_pstore_autoprobe_disable is not
cleared.  Resetting efivars_pstore_autoprobe_disable to false in
efivars_pstore_init() after platform_driver_register() returns would avoid
leaving it set past initial registration.

> diff --git a/drivers/firmware/efi/efi.c b/drivers/firmware/efi/efi.c
> index 9bc22de0d67ee..e98436a9bb9bd 100644
> --- a/drivers/firmware/efi/efi.c
> +++ b/drivers/firmware/efi/efi.c
> @@ -483,7 +483,8 @@ static int __init efisubsys_init(void)
>  		error = efivar_ssdt_load();
>  		if (error)
>  			pr_err("efi: failed to load SSDT, error %d.\n", error);
> -		platform_device_register_simple("efivars", 0, NULL, 0);
> +		if (efi_rt_services_supported(EFI_RT_SUPPORTED_SET_VARIABLE))
> +			platform_device_register_simple("efivars", 0, NULL, 0);
>  	}
>  
>  	BLOCKING_INIT_NOTIFIER_HEAD(&efivar_ops_nh);

[Severity: High]
Does gating efi-pstore on the "efivars" platform device registered here
prevent efi-pstore from probing on platforms that provide writable
variables through non-generic efivars_register() backends?

Before this change, efivars_pstore_init() checked efivar_supports_writes()
and registered with pstore directly without requiring the "efivars"
platform device.  With this change, platforms where generic EFI runtime
services do not advertise EFI_RT_SUPPORTED_SET_VARIABLE at boot, or where
EFI_BOOT is not set (such as Coreboot GSMI in
drivers/firmware/coreboot/gsmi.c, OP-TEE StandaloneMM in
drivers/firmware/efi/stmm/tee_stmm_efi.c, or Qualcomm UEFI Secure App in
drivers/firmware/qcom/qcom_qseecom_uefisecapp.c), never register the
"efivars" platform device, so efi-pstore cannot probe or be bound via
sysfs.

Conversely, if EFI_RT_SUPPORTED_SET_VARIABLE is set in
efi.runtime_supported_mask but generic_ops_supported() returns false in
generic_ops_register() (such as on Qualcomm systems where
efi.get_next_variable() returns EFI_UNSUPPORTED), efisubsys_init() still
registers the "efivars" platform device while __efivars is NULL, causing
efi_pstore_probe() to see !efivar_supports_writes() and fail with -ENODEV
before the platform-specific efivars driver probes.

[Severity: Critical]
This problem wasn't introduced by this patch, but if efisubsys_init()
fails [1], can destroying efi_rts_wq without clearing
EFI_RUNTIME_SERVICES in efi.flags cause a use-after-free in
__efi_queue_work()?

drivers/firmware/efi/efi.c:efisubsys_init() {
    ...
err_destroy_wq:
	if (efi_rts_wq)
		destroy_workqueue(efi_rts_wq);

	return error;
}

When a failure in efisubsys_init() reaches err_destroy_wq, efi_rts_wq is
destroyed while EFI_RUNTIME_SERVICES remains set, so a later runtime
service call through __efi_queue_work() passes
efi_enabled(EFI_RUNTIME_SERVICES) and passes the freed efi_rts_wq pointer
to queue_work().

[1] https://sashiko.dev/#/bug/linux-a1f302ad-3fdd-4e74-a0f8-c73ca89f1ff3

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261120131827.480055-4-ardb+git@google.com?part=2

      reply	other threads:[~2026-11-20 13:39 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-20 13:18 [PATCH 0/2] efi: Fix pstore runtime en/disable Ard Biesheuvel
2026-11-20 13:18 ` [PATCH 1/2] efi: pstore: Make pstore_disable parameter read-only again Ard Biesheuvel
2026-11-20 13:18 ` [PATCH 2/2] efi: pstore: Convert into a platform driver for 'efivars' Ard Biesheuvel
2026-11-20 13:39   ` sashiko-bot [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=sashiko-outbox-225258@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=ardb+git@google.com \
    --cc=ardb@kernel.org \
    --cc=gpiccoli@igalia.com \
    --cc=kees@kernel.org \
    --cc=linux-efi@vger.kernel.org \
    --cc=sashiko-reviews@lists.linux.dev \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox