[OE-core][PATCH] grub: update search parameter

[OE-core][PATCH] grub: update search parameter

[Xiangyu Chen]

From: Xiangyu Chen <xiangyu.chen@windriver.com>

Grub would report an error message in boot stage as below:

   "error: no such device: ((hd0,gpt1)/EFI/BOOT)/EFI/BOOT/grub.cfg"

Consequently, the root variable is not set, and the intended protection
against cross-device configuration loading (the purpose of the original 2014 commit)
is lost.

The most robust fix is to use the --hint parameter.
This separates the search target from the device hint, avoiding
fragile string concatenation and supporting both prefixed and
non-prefixed $cmdpath formats.

Fixes: 5ce73b6055ac ("grub: add cmdpath to grub configuration file")
Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
---
 meta/recipes-bsp/grub/files/cfg | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/meta/recipes-bsp/grub/files/cfg b/meta/recipes-bsp/grub/files/cfg
index 8ca53d24d7..8fe1a8d0c7 100644
--- a/meta/recipes-bsp/grub/files/cfg
+++ b/meta/recipes-bsp/grub/files/cfg
@@ -1,2 +1,2 @@
-search.file ($cmdpath)/EFI/BOOT/grub.cfg root
+search --file --set=root --hint-efi=$cmdpath /EFI/BOOT/grub.cfg
 set prefix=($root)/EFI/BOOT
-- 
2.39.4

Re: [OE-core][PATCH] grub: update search parameter

[Paul Barker]

[-- Attachment #1: Type: text/plain, Size: 1541 bytes --]

On Thu, 2026-02-26 at 16:19 +0800, Xiangyu Chen via
lists.openembedded.org wrote:
> From: Xiangyu Chen <xiangyu.chen@windriver.com>
> 
> Grub would report an error message in boot stage as below:
> 
>    "error: no such device: ((hd0,gpt1)/EFI/BOOT)/EFI/BOOT/grub.cfg"
> 
> Consequently, the root variable is not set, and the intended protection
> against cross-device configuration loading (the purpose of the original 2014 commit)
> is lost.
> 
> The most robust fix is to use the --hint parameter.
> This separates the search target from the device hint, avoiding
> fragile string concatenation and supporting both prefixed and
> non-prefixed $cmdpath formats.
> 
> Fixes: 5ce73b6055ac ("grub: add cmdpath to grub configuration file")
> Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
> ---
>  meta/recipes-bsp/grub/files/cfg | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/meta/recipes-bsp/grub/files/cfg b/meta/recipes-bsp/grub/files/cfg
> index 8ca53d24d7..8fe1a8d0c7 100644
> --- a/meta/recipes-bsp/grub/files/cfg
> +++ b/meta/recipes-bsp/grub/files/cfg
> @@ -1,2 +1,2 @@
> -search.file ($cmdpath)/EFI/BOOT/grub.cfg root
> +search --file --set=root --hint-efi=$cmdpath /EFI/BOOT/grub.cfg

I'm not sure that this can work for all use cases. The grub
documentatation for hints says "First try the device HINT" (i.e. it's
explicitly a device), whereas the documentation for cmdpath says that it
is "an absolute directory name".

Best regards,

-- 
Paul Barker


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 252 bytes --]

Re: [OE-core][PATCH] grub: update search parameter

[Xiangyu Chen]

Hi Paul,


On 3/5/26 19:49, Paul Barker wrote:
> I'm not sure that this can work for all use cases. The grub
> documentatation for hints says "First try the device HINT" (i.e. it's
> explicitly a device), whereas the documentation for cmdpath says that it
> is "an absolute directory name".
While the documentation defines $cmdpath as a directory and hints as 
devices, grub’s internal implementation of
grub_disk_open (grub-core/kern/disk.c [1]) is designed to handle this 
exact "device+path" format.

When a hint like (hd0,gpt1)/EFI/BOOT/ is passed, grub_disk_open calls 
find_part_sep to locate the partition/path delimiter.
It then uses grub_memcpy to extract only the device part (e.g., (hd0) 
into the raw buffer for the disk driver.

The remaining string (gpt1)/EFI/BOOT/) is passed to 
grub_partition_probe, which identifies the partition and gracefully
ignores the trailing directory path.

The current syntax ($cmdpath) results in double parentheses ((hd0...)) 
if the variable is already populated.
In grub_disk_open, this extracts a malformed device name ((hd0 which no 
driver can open, triggering a hard no such device error.


Ref:

[1] 
https://elixir.bootlin.com/grub/grub-2.14/source/grub-core/kern/disk.c#L192


Br,

Xiangyu

Re: [OE-core][PATCH] grub: update search parameter

[Xiangyu Chen]

Hi Richard,


Since the behavior was found in OE scarthgap branch, could you please 
help to cherry-pick this

commit: 42b530581f72 ("grub: update search parameter") to whinlatter and 
scarthgap branch?


Thanks.


Br,

Xiangyu

在 2026/3/31 10:57, Xiangyu Chen 写道:
> Hi Paul,
>
>
> On 3/5/26 19:49, Paul Barker wrote:
>> I'm not sure that this can work for all use cases. The grub
>> documentatation for hints says "First try the device HINT" (i.e. it's
>> explicitly a device), whereas the documentation for cmdpath says that it
>> is "an absolute directory name".
> While the documentation defines $cmdpath as a directory and hints as 
> devices, grub’s internal implementation of
> grub_disk_open (grub-core/kern/disk.c [1]) is designed to handle this 
> exact "device+path" format.
>
> When a hint like (hd0,gpt1)/EFI/BOOT/ is passed, grub_disk_open calls 
> find_part_sep to locate the partition/path delimiter.
> It then uses grub_memcpy to extract only the device part (e.g., (hd0) 
> into the raw buffer for the disk driver.
>
> The remaining string (gpt1)/EFI/BOOT/) is passed to 
> grub_partition_probe, which identifies the partition and gracefully
> ignores the trailing directory path.
>
> The current syntax ($cmdpath) results in double parentheses ((hd0...)) 
> if the variable is already populated.
> In grub_disk_open, this extracts a malformed device name ((hd0 which 
> no driver can open, triggering a hard no such device error.
>
>
> Ref:
>
> [1] 
> https://elixir.bootlin.com/grub/grub-2.14/source/grub-core/kern/disk.c#L192
>
>
> Br,
>
> Xiangyu
>

Re: [OE-core][PATCH] grub: update search parameter

[Yoann Congal]

On Thu Apr 23, 2026 at 8:04 AM CEST, Xiangyu Chen via lists.openembedded.org wrote:
> Hi Richard,
>
>
> Since the behavior was found in OE scarthgap branch, could you please 
> help to cherry-pick this
>
> commit: 42b530581f72 ("grub: update search parameter") to whinlatter and 
> scarthgap branch?

Hello,

For whinlatter, that will be too late (the last build has already left for
QA).

For scarthgap, I would need to look into it (change does not look
trivial). Can you send a rebased patch targeting scarthgap?

Thanks!

>
>
> Thanks.
>
>
> Br,
>
> Xiangyu
>
> 在 2026/3/31 10:57, Xiangyu Chen 写道:
>> Hi Paul,
>>
>>
>> On 3/5/26 19:49, Paul Barker wrote:
>>> I'm not sure that this can work for all use cases. The grub
>>> documentatation for hints says "First try the device HINT" (i.e. it's
>>> explicitly a device), whereas the documentation for cmdpath says that it
>>> is "an absolute directory name".
>> While the documentation defines $cmdpath as a directory and hints as 
>> devices, grub’s internal implementation of
>> grub_disk_open (grub-core/kern/disk.c [1]) is designed to handle this 
>> exact "device+path" format.
>>
>> When a hint like (hd0,gpt1)/EFI/BOOT/ is passed, grub_disk_open calls 
>> find_part_sep to locate the partition/path delimiter.
>> It then uses grub_memcpy to extract only the device part (e.g., (hd0) 
>> into the raw buffer for the disk driver.
>>
>> The remaining string (gpt1)/EFI/BOOT/) is passed to 
>> grub_partition_probe, which identifies the partition and gracefully
>> ignores the trailing directory path.
>>
>> The current syntax ($cmdpath) results in double parentheses ((hd0...)) 
>> if the variable is already populated.
>> In grub_disk_open, this extracts a malformed device name ((hd0 which 
>> no driver can open, triggering a hard no such device error.
>>
>>
>> Ref:
>>
>> [1] 
>> https://elixir.bootlin.com/grub/grub-2.14/source/grub-core/kern/disk.c#L192
>>
>>
>> Br,
>>
>> Xiangyu
>>


-- 
Yoann Congal
Smile ECS

Re: [OE-core][PATCH] grub: update search parameter

[Xiangyu Chen]

在 2026/4/23 14:38, Yoann Congal via lists.openembedded.org 写道:
> CAUTION: This email comes from a non Wind River email account!
> Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
> On Thu Apr 23, 2026 at 8:04 AM CEST, Xiangyu Chen via lists.openembedded.org wrote:
>> Hi Richard,
>>
>>
>> Since the behavior was found in OE scarthgap branch, could you please
>> help to cherry-pick this
>>
>> commit: 42b530581f72 ("grub: update search parameter") to whinlatter and
>> scarthgap branch?
> Hello,
>
> For whinlatter, that will be too late (the last build has already left for
> QA).
>
> For scarthgap, I would need to look into it (change does not look
> trivial). Can you send a rebased patch targeting scarthgap?
>
> Thanks!

Thanks for your info, a rebased patch will be sent later.


Br,

Xiangyu

>
>>
>> Thanks.
>>
>>
>> Br,
>>
>> Xiangyu
>>
>> 在 2026/3/31 10:57, Xiangyu Chen 写道:
>>> Hi Paul,
>>>
>>>
>>> On 3/5/26 19:49, Paul Barker wrote:
>>>> I'm not sure that this can work for all use cases. The grub
>>>> documentatation for hints says "First try the device HINT" (i.e. it's
>>>> explicitly a device), whereas the documentation for cmdpath says that it
>>>> is "an absolute directory name".
>>> While the documentation defines $cmdpath as a directory and hints as
>>> devices, grub’s internal implementation of
>>> grub_disk_open (grub-core/kern/disk.c [1]) is designed to handle this
>>> exact "device+path" format.
>>>
>>> When a hint like (hd0,gpt1)/EFI/BOOT/ is passed, grub_disk_open calls
>>> find_part_sep to locate the partition/path delimiter.
>>> It then uses grub_memcpy to extract only the device part (e.g., (hd0)
>>> into the raw buffer for the disk driver.
>>>
>>> The remaining string (gpt1)/EFI/BOOT/) is passed to
>>> grub_partition_probe, which identifies the partition and gracefully
>>> ignores the trailing directory path.
>>>
>>> The current syntax ($cmdpath) results in double parentheses ((hd0...))
>>> if the variable is already populated.
>>> In grub_disk_open, this extracts a malformed device name ((hd0 which
>>> no driver can open, triggering a hard no such device error.
>>>
>>>
>>> Ref:
>>>
>>> [1]
>>> https://elixir.bootlin.com/grub/grub-2.14/source/grub-core/kern/disk.c#L192
>>>
>>>
>>> Br,
>>>
>>> Xiangyu
>>>
>
> --
> Yoann Congal
> Smile ECS
>
>
> -=-=-=-=-=-=-=-=-=-=-=-
> Links: You receive all messages sent to this group.
> View/Reply Online (#235740): https://lists.openembedded.org/g/openembedded-core/message/235740
> Mute This Topic: https://lists.openembedded.org/mt/118009422/7175143
> Group Owner: openembedded-core+owner@lists.openembedded.org
> Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [xiangyu.chen@eng.windriver.com]
> -=-=-=-=-=-=-=-=-=-=-=-
>

Re: [OE-core][PATCH] grub: update search parameter

[Yoann Congal]

On Thu Apr 23, 2026 at 9:18 AM CEST, Xiangyu Chen wrote:
>
> 在 2026/4/23 14:38, Yoann Congal via lists.openembedded.org 写道:
>> CAUTION: This email comes from a non Wind River email account!
>> Do not click links or open attachments unless you recognize the sender and know the content is safe.
>>
>> On Thu Apr 23, 2026 at 8:04 AM CEST, Xiangyu Chen via lists.openembedded.org wrote:
>>> Hi Richard,
>>>
>>>
>>> Since the behavior was found in OE scarthgap branch, could you please
>>> help to cherry-pick this
>>>
>>> commit: 42b530581f72 ("grub: update search parameter") to whinlatter and
>>> scarthgap branch?
>> Hello,
>>
>> For whinlatter, that will be too late (the last build has already left for
>> QA).
>>
>> For scarthgap, I would need to look into it (change does not look
>> trivial). Can you send a rebased patch targeting scarthgap?
>>
>> Thanks!
>
> Thanks for your info, a rebased patch will be sent later.

Received, thanks!

I will consider it for the next review cycle.

>
>
> Br,
>
> Xiangyu
>
>>
>>>
>>> Thanks.
>>>
>>>
>>> Br,
>>>
>>> Xiangyu
>>>
>>> 在 2026/3/31 10:57, Xiangyu Chen 写道:
>>>> Hi Paul,
>>>>
>>>>
>>>> On 3/5/26 19:49, Paul Barker wrote:
>>>>> I'm not sure that this can work for all use cases. The grub
>>>>> documentatation for hints says "First try the device HINT" (i.e. it's
>>>>> explicitly a device), whereas the documentation for cmdpath says that it
>>>>> is "an absolute directory name".
>>>> While the documentation defines $cmdpath as a directory and hints as
>>>> devices, grub’s internal implementation of
>>>> grub_disk_open (grub-core/kern/disk.c [1]) is designed to handle this
>>>> exact "device+path" format.
>>>>
>>>> When a hint like (hd0,gpt1)/EFI/BOOT/ is passed, grub_disk_open calls
>>>> find_part_sep to locate the partition/path delimiter.
>>>> It then uses grub_memcpy to extract only the device part (e.g., (hd0)
>>>> into the raw buffer for the disk driver.
>>>>
>>>> The remaining string (gpt1)/EFI/BOOT/) is passed to
>>>> grub_partition_probe, which identifies the partition and gracefully
>>>> ignores the trailing directory path.
>>>>
>>>> The current syntax ($cmdpath) results in double parentheses ((hd0...))
>>>> if the variable is already populated.
>>>> In grub_disk_open, this extracts a malformed device name ((hd0 which
>>>> no driver can open, triggering a hard no such device error.
>>>>
>>>>
>>>> Ref:
>>>>
>>>> [1]
>>>> https://elixir.bootlin.com/grub/grub-2.14/source/grub-core/kern/disk.c#L192
>>>>
>>>>
>>>> Br,
>>>>
>>>> Xiangyu
>>>>
>>
>> --
>> Yoann Congal
>> Smile ECS
>>
>>
>> -=-=-=-=-=-=-=-=-=-=-=-
>> Links: You receive all messages sent to this group.
>> View/Reply Online (#235740): https://lists.openembedded.org/g/openembedded-core/message/235740
>> Mute This Topic: https://lists.openembedded.org/mt/118009422/7175143
>> Group Owner: openembedded-core+owner@lists.openembedded.org
>> Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [xiangyu.chen@eng.windriver.com]
>> -=-=-=-=-=-=-=-=-=-=-=-
>>


-- 
Yoann Congal
Smile ECS