Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
1. Prepend a VenMedia Device Path in every initrd Device Path. In that case FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node - VenMedia - Initrd DP - end node - repeat
2. Prepend a VenMedia Device Path once. In that case FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
3. Use Vendor Defined Data of the VenMedia device path and copy the initrd device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
On Thu, 18 Feb 2021 at 10:04, Ilias Apalodimas ilias.apalodimas@linaro.org wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
this is from "10.3.2.4 Vendor Device Path" in UEFI spec right ?
1. Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would
look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias _______________________________________________ boot-architecture mailing list boot-architecture@lists.linaro.org https://lists.linaro.org/mailman/listinfo/boot-architecture
Hi Francois,
On Thu, Feb 18, 2021 at 11:31:42AM +0100, François Ozog wrote:
On Thu, 18 Feb 2021 at 10:04, Ilias Apalodimas ilias.apalodimas@linaro.org wrote:
[...]
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
this is from "10.3.2.4 Vendor Device Path" in UEFI spec right ?
Yes that's what I had in mind
[...]
Regards /Ilias
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
With 1 you can individually determine for each element its meaning by looking at the first node.
Therefore I prefer 1.
Best regards
Heinrich
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
+1
On Thu, 18 Feb 2021 at 12:15, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative
method
to load an initrd [1]. The kernel call to the firmware ends up calling
the
protocol with a Device Path End Structure, so the firmware must know
which
initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this
protocol,
in order to provide vertical distros a way of loading (kernel, initrd)
pairs
without GRUB. In that case we need a common way for firmware
implementations
to define and manage the initrd. User space applications that control
the boot
flow (e.g efibootmgr), should also be able to change the variable
accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can
use the
FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other
device
paths may optionally exist in the FilePathList, but their usage is OSV
specific.
Each element in the array is variable length, and ends at the device
path end
structure. Because the size of Description is arbitrary, this data
structure
is not guaranteed to be aligned on a natural boundary. This data
structure may
have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we
could
implement that. All 3 ways would allow us to specify multiple initrds
(and we
could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID.
We can
then use that GUID to identify the filetype and behavior of the device
paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that
case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[]
would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each
instance would
terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the
initrd
device path(s) in there. In that case the Vendor Defined Data will it
self
be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
With 1 you can individually determine for each element its meaning by looking at the first node.
Therefore I prefer 1.
Best regards
Heinrich
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
boot-architecture mailing list boot-architecture@lists.linaro.org https://lists.linaro.org/mailman/listinfo/boot-architecture
On Thu, Feb 18, 2021 at 12:15:30PM +0100, Heinrich Schuchardt wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
That interpretation is a bit different from what I have. The separation between initrd files is denoted by the end of instance, not the end node with end of a device path (Sub-Type 0xFF). The EFI spec says "Each element in the array is variable length, and ends at the device path end structure" so I assume we can use either? I initially thought it meant Sub-Type = 0xFF.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
Not really. The whole point is that you'll have a single VenMedia indicating the start of the initrd device paths. Those are not separated by the end node (0xFF), but by the end of instance. Once you see an end node it's not an initrd filepath anymore. So you dont really need the number of elements in there.
In any case (1) is easier to code as well, since it practically allows you to append a DP, instead of search - append instance - copy the remaining DP, which will makje our lives easier.
Regards /Ilias
With 1 you can individually determine for each element its meaning by looking at the first node.
Therefore I prefer 1.
Best regards
Heinrich
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
On 2/18/21 1:11 PM, Ilias Apalodimas wrote:
On Thu, Feb 18, 2021 at 12:15:30PM +0100, Heinrich Schuchardt wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
That interpretation is a bit different from what I have. The separation between initrd files is denoted by the end of instance, not the end node with end of a device path (Sub-Type 0xFF).
I explicitly wrote sub-type *0x01*.
The EFI spec says "Each element in the array is variable length, and ends at the device path end structure" so I assume we can use either? I initially thought it meant Sub-Type = 0xFF.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
Not really. The whole point is that you'll have a single VenMedia indicating the start of the initrd device paths. Those are not separated by the end node (0xFF), but by the end of instance. Once you see an end node it's not an initrd filepath anymore. So you dont really need the number of elements in there.
Both in suggestions 1 and 2 individual initrd device paths are separated by end nodes of sub-type 0x01.
You make the assumption that nothing follows the last initrd path. But we may want to add additional device paths, e.g. for the device-tree.
This is why we should not assume that an end node of sub-type 0xFF follows the last initrd device-patch.
Best regards
Heinrich
In any case (1) is easier to code as well, since it practically allows you to append a DP, instead of search - append instance - copy the remaining DP, which will makje our lives easier.
Regards /Ilias
With 1 you can individually determine for each element its meaning by looking at the first node.
Therefore I prefer 1.
Best regards
Heinrich
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
On Thu, Feb 18, 2021 at 02:23:02PM +0100, Heinrich Schuchardt wrote:
On 2/18/21 1:11 PM, Ilias Apalodimas wrote:
On Thu, Feb 18, 2021 at 12:15:30PM +0100, Heinrich Schuchardt wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
That interpretation is a bit different from what I have. The separation between initrd files is denoted by the end of instance, not the end node with end of a device path (Sub-Type 0xFF).
I explicitly wrote sub-type *0x01*.
Yes and that's exactly what I am pointing out on the sentence above. The real point in on the sentence right after that and the interpretation of the EFI spec.
The EFI spec says "Each element in the array is variable length, and ends at the device path end structure" so I assume we can use either? I initially thought it meant Sub-Type = 0xFF.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
Not really. The whole point is that you'll have a single VenMedia indicating the start of the initrd device paths. Those are not separated by the end node (0xFF), but by the end of instance. Once you see an end node it's not an initrd filepath anymore. So you dont really need the number of elements in there.
Both in suggestions 1 and 2 individual initrd device paths are separated by end nodes of sub-type 0x01.
No they aren't. Here's what I wrote:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
The separator between the initrd instances is 0x01 while between different device paths 0xFF.
You make the assumption that nothing follows the last initrd path. But we may want to add additional device paths, e.g. for the device-tree.
This is why we should not assume that an end node of sub-type 0xFF follows the last initrd device-patch.
Why? The spec says that the FilePathList[] elements must end with and end node. the initrds themselves are separated by the end of instance.
Regards /Ilias
Best regards
Heinrich
In any case (1) is easier to code as well, since it practically allows you to append a DP, instead of search - append instance - copy the remaining DP, which will makje our lives easier.
Regards /Ilias
With 1 you can individually determine for each element its meaning by looking at the first node.
Therefore I prefer 1.
Best regards
Heinrich
Any preference on these? Is one of them closer to the EFI spec, so we could go ahead and try to standardize some of the GUIDs of the VenMedia?
[1] https://lkml.org/lkml/2020/2/16/105
Regards /Ilias
boot-architecture mailing list boot-architecture@lists.linaro.org https://lists.linaro.org/mailman/listinfo/boot-architecture
Hi Heinrich,
On Thu, 18 Feb 2021 at 04:15, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above.
The above is not readable in my opinion. I really wonder about this path that we are going down.
[..]
Regards, Simon
On 18.02.21 17:27, Simon Glass wrote:
Hi Heinrich,
On Thu, 18 Feb 2021 at 04:15, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2)
The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above.
The above is not readable in my opinion. I really wonder about this path that we are going down.
It is better readable then a hex string and conforms to the output format defined by the UEFI spec.
Just use 'efibootmgr -v' to display your Boot variables in Linux or use the UEFI shell command 'dh' to display handles and you will find the same format.
When defining the boot variable with efibootmgr you use a Unix style path and in U-Boot we use something like mmc 0:1 /vmlinux.
Best regards
Heinrich
On 18/02/2021 11:15, Heinrich Schuchardt wrote:
On 18.02.21 10:04, Ilias Apalodimas wrote:
Hi,
An arch agnostic way was recently added on the kernel, as an alternative method to load an initrd [1]. The kernel call to the firmware ends up calling the protocol with a Device Path End Structure, so the firmware must know which initrd to load on the buffer the kernel provides.
The protocol is currently implemented by U-boot and EDK2, which both define a way of specifying the initrd to load. We could use this protocol, in order to provide vertical distros a way of loading (kernel, initrd) pairs without GRUB. In that case we need a common way for firmware implementations to define and manage the initrd. User space applications that control the boot flow (e.g efibootmgr), should also be able to change the variable accordingly.
Looking at the EFI spec and specifically § 3.1.3 Load Options, we can use the FilePathList[] of the EFI_LOAD_OPTION, which is described as:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. The FilePathList[0] is specific to the device type. Other device paths may optionally exist in the FilePathList, but their usage is OSV specific. Each element in the array is variable length, and ends at the device path end structure. Because the size of Description is arbitrary, this data structure is not guaranteed to be aligned on a natural boundary. This data structure may have to be copied to an aligned natural boundary before it is used."
So FilePatrhList[1-n] are available for OS usage. There are 3 ways we could implement that. All 3 ways would allow us to specify multiple initrds (and we could extend the same logic to DTBs, but that's a different discussion). They all re-use the same idea, prepend a VenMedia DP, which has a GUID. We can then use that GUID to identify the filetype and behavior of the device paths.
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2) > The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
With 1 you can individually determine for each element its meaning by looking at the first node.
Attempting to paraphrase; does this mean the VenMedia component would be used to identify the file type? i.e., use one GUID for initrd, and another for DTB? Or is that worked out another way?
g. IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
Hi Grant,
On Fri, Feb 19, 2021 at 05:28:59PM +0000, Grant Likely wrote:
On 18/02/2021 11:15, Heinrich Schuchardt wrote:
[...]
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2) > The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
With 1 you can individually determine for each element its meaning by looking at the first node.
Attempting to paraphrase; does this mean the VenMedia component would be used to identify the file type? i.e., use one GUID for initrd, and another for DTB? Or is that worked out another way?
Yea exactly (in all 3 implementations). The difference in the 3 different proposals is how to interpret that VenMedia device path to identify all the device paths that refer to that specific GUID.
Regards /Ilias
g. IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
On 2/19/21 10:58 PM, Ilias Apalodimas wrote:
Hi Grant,
On Fri, Feb 19, 2021 at 05:28:59PM +0000, Grant Likely wrote:
On 18/02/2021 11:15, Heinrich Schuchardt wrote:
[...]
- Prepend a VenMedia Device Path in every initrd Device Path. In that case
FilePathList[] would look like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end node
- VenMedia - Initrd DP - end node - repeat
- Prepend a VenMedia Device Path once. In that case FilePathList[] would look
like this:
Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
In this case we could use the VenMedia Vendor Defined Data to indicate the number of device paths that follow, although it's redundant, since each instance would terminate on the Device Path End Structure.
- Use Vendor Defined Data of the VenMedia device path and copy the initrd
device path(s) in there. In that case the Vendor Defined Data will it self be in a device path format with all the initrds we want.
Loaded Image device path - end node - VenMedia - end node - other DPs
When passing the device path of the boot option to the EDK2 implementation of EFI_DEVICE_PATH_TO_TEXT_PROTOCOL.ConvertDevicePathToText(), it will print out all array elements as comma separated list like
HD(1,GPT,..,0x2000,0x200000)/File(\EFI\debian\shimaa64.efi),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd1),/VenMedia(00000001-0000-0000-0000-000000000000)/HD(2,GPT,..,0x2000,0x200000)/File(\initrd2) > The device path end nodes of sub-type 0x01 are rendered as commas.
With 1 and 2 this would show a readable output like above. With 3 you will just see a hex-string.
This excludes 3 for me.
If 2 does not add the number of initrds, it cannot be determined if a following array element starting with a VenMedia() node is an initrd or has a completely different meaning.
With 1 you can individually determine for each element its meaning by looking at the first node.
Attempting to paraphrase; does this mean the VenMedia component would be used to identify the file type? i.e., use one GUID for initrd, and another for DTB? Or is that worked out another way?
Yea exactly (in all 3 implementations). The difference in the 3 different proposals is how to interpret that VenMedia device path to identify all the device paths that refer to that specific GUID.
Regards /Ilias
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Best regards
Heinrich
-----Original Message----- From: Heinrich Schuchardt xypron.glpk@gmx.de Sent: Monday, March 1, 2021 2:39 PM To: Ilias Apalodimas ilias.apalodimas@linaro.org; Grant Likely Grant.Likely@arm.com Cc: Boot Architecture Mailman List boot-architecture@lists.linaro.org; Samer El-Haj-Mahmoud Samer.El-Haj-Mahmoud@arm.com; Ard Biesheuvel ardb@kernel.org; Leif Lindholm leif@nuviainc.com Subject: Re: EFI_LOAD_FILE2 for initrd standardization
...
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
Agree that this is not clear, and could be interpreted either way. And yes, agree the UEFI spec needs a clarification
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The EDK2 code seems to be incomplete, with a "TODO" to support the FilePathList[]. In fact, the code calls it "FilePath" to be clear that it is assuming a single DevicePath (which means a 0xFF sub-type termination) https://github.com/tianocore/edk2/blob/master/MdeModulePkg%2FLibrary%2FUefiB...
Searching the code further, I see there is support for parsing multi-instance device path (separated by END_INSTANCE_DEVICE_PATH_SUBTYPE, or 0x1) in things like UefiDevicePathLib , parsing code, etc... But it does not seem to be supported at all in the UEFI Boot Manager implemented in EDK2
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
Looking at EDK2, ConXDev are treated as multi-instance devices, with ConPlatformDxe using AppendDevicePathInstance() and GetNextDevicePathInstance() to construct / navigate the multiple-instances
The spec should have defined these as a "multi-instance device path of all possible console X devices". See my notes below on multi-instance DP.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
I am leaning towards this conclusion as well. This is further supported by the following evidence:
* EFI_DEVICE_PATH_UTILITIES_PROTOCOL has a function called IsDevicePathMultiInstance() along with walker functions AppendDevicePathInstance(), GetNextDevicePathInstance(). The description of these functions make it clear that the intention is to treat 0x01 as a separator between multiple instances of device paths in a multi-instance device path structure, with 0xff as the final end of DP indicator. For example, AppendDevicePathInstance() is defined as:
" This function creates a new device path by appending a copy of the specified device path instance to a copy of the specified device path in an allocated buffer. The end-of-device-path device node is moved after the end of the appended device node and a new end-of-device-path-instance node is inserted between."
* In EFI_USER_INFO , there are user info policy types that leverage the device path definitions (such as EFI_USER_INFO_ACCESS_FORBID_LOAD), and they are clear on distinguishing the term "multi-instance device path" from a "series of UEFI device paths":
"The record is a series of normal UEFI device paths (not multi-instance device paths)."
* The "multi-instance" usage also shows up in the language of LocateDevicePath()
So if the spec intended for FilePathList[] to use the 0x1 separator, they should have used the term "multi-instance" device path, which is not the case.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Yes. Or could simply say: "A packed array of UEFI device paths (not multi-instance device paths). The first element of the array is a device path that describes the device and location of the Image for this load option."
Either way, we can do a "code first" ECR in the UEFI spec. I say code first to ensure that it remains in public domain, and not blocked by UEFI Forum NDA until the publication of the next version of the UEFI spec. If we agree, I can get this process started and propose the language.
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
One suggestion here is to take into consideration: whether we standardize this proposal (for loading additional kernel files from element[i], i>0 ) or not, the FW implementation must still be compliant with the UEFI rules around booting from a "short form device path" (UEFI spec, section 3.1.2)
Best regards
Heinrich
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
On 3/1/21 9:44 PM, Samer El-Haj-Mahmoud wrote:
-----Original Message----- From: Heinrich Schuchardt xypron.glpk@gmx.de Sent: Monday, March 1, 2021 2:39 PM To: Ilias Apalodimas ilias.apalodimas@linaro.org; Grant Likely Grant.Likely@arm.com Cc: Boot Architecture Mailman List boot-architecture@lists.linaro.org; Samer El-Haj-Mahmoud Samer.El-Haj-Mahmoud@arm.com; Ard Biesheuvel ardb@kernel.org; Leif Lindholm leif@nuviainc.com Subject: Re: EFI_LOAD_FILE2 for initrd standardization
...
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
Agree that this is not clear, and could be interpreted either way. And yes, agree the UEFI spec needs a clarification
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The EDK2 code seems to be incomplete, with a "TODO" to support the FilePathList[]. In fact, the code calls it "FilePath" to be clear that it is assuming a single DevicePath (which means a 0xFF sub-type termination) https://github.com/tianocore/edk2/blob/master/MdeModulePkg%2FLibrary%2FUefiB...
Searching the code further, I see there is support for parsing multi-instance device path (separated by END_INSTANCE_DEVICE_PATH_SUBTYPE, or 0x1) in things like UefiDevicePathLib , parsing code, etc... But it does not seem to be supported at all in the UEFI Boot Manager implemented in EDK2
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
Looking at EDK2, ConXDev are treated as multi-instance devices, with ConPlatformDxe using AppendDevicePathInstance() and GetNextDevicePathInstance() to construct / navigate the multiple-instances
The spec should have defined these as a "multi-instance device path of all possible console X devices". See my notes below on multi-instance DP.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
I am leaning towards this conclusion as well. This is further supported by the following evidence:
- EFI_DEVICE_PATH_UTILITIES_PROTOCOL has a function called IsDevicePathMultiInstance() along with walker functions AppendDevicePathInstance(), GetNextDevicePathInstance(). The description of these functions make it clear that the intention is to treat 0x01 as a separator between multiple instances of device paths in a multi-instance device path structure, with 0xff as the final end of DP indicator. For example, AppendDevicePathInstance() is defined as:
" This function creates a new device path by appending a copy of the specified device path instance to a copy of the specified device path in an allocated buffer. The end-of-device-path device node is moved after the end of the appended device node and a new end-of-device-path-instance node is inserted between."
- In EFI_USER_INFO , there are user info policy types that leverage the device path definitions (such as EFI_USER_INFO_ACCESS_FORBID_LOAD), and they are clear on distinguishing the term "multi-instance device path" from a "series of UEFI device paths":
"The record is a series of normal UEFI device paths (not multi-instance device paths)."
- The "multi-instance" usage also shows up in the language of LocateDevicePath()
So if the spec intended for FilePathList[] to use the 0x1 separator, they should have used the term "multi-instance" device path, which is not the case.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Yes. Or could simply say: "A packed array of UEFI device paths (not multi-instance device paths). The first element of the array is a device
Each array element except index 0 may be a multi-instance device path.
path that describes the device and location of the Image for this load option."
Either way, we can do a "code first" ECR in the UEFI spec. I say code first to ensure that it remains in public domain, and not blocked by UEFI Forum NDA until the publication of the next version of the UEFI spec. If we agree, I can get this process started and propose the language.
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
One suggestion here is to take into consideration: whether we standardize this proposal (for loading additional kernel files from element[i], i>0 ) or not, the FW implementation must still be compliant with the UEFI rules around booting from a "short form device path" (UEFI spec, section 3.1.2)
It is not my intent to change this. The requirements of section 3.1.2 should be equally applied to initrd and dtb.
U-Boot's implementation of LoadImage() allows short paths to be used. See function efi_dp_find_obj() in lib/efi_loader/efi_device_path.c.
https://source.denx.de/u-boot/u-boot/-/blob/master/lib/efi_loader/efi_device...
Best regards
Heinric
Hi Samer,
Thanks for having a look at this
On Mon, Mar 01, 2021 at 08:44:00PM +0000, Samer El-Haj-Mahmoud wrote:
-----Original Message----- From: Heinrich Schuchardt xypron.glpk@gmx.de Sent: Monday, March 1, 2021 2:39 PM To: Ilias Apalodimas ilias.apalodimas@linaro.org; Grant Likely Grant.Likely@arm.com Cc: Boot Architecture Mailman List boot-architecture@lists.linaro.org; Samer El-Haj-Mahmoud Samer.El-Haj-Mahmoud@arm.com; Ard Biesheuvel ardb@kernel.org; Leif Lindholm leif@nuviainc.com Subject: Re: EFI_LOAD_FILE2 for initrd standardization
...
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
Agree that this is not clear, and could be interpreted either way. And yes, agree the UEFI spec needs a clarification
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The EDK2 code seems to be incomplete, with a "TODO" to support the FilePathList[]. In fact, the code calls it "FilePath" to be clear that it is assuming a single DevicePath (which means a 0xFF sub-type termination) https://github.com/tianocore/edk2/blob/master/MdeModulePkg%2FLibrary%2FUefiB...
Searching the code further, I see there is support for parsing multi-instance device path (separated by END_INSTANCE_DEVICE_PATH_SUBTYPE, or 0x1) in things like UefiDevicePathLib , parsing code, etc... But it does not seem to be supported at all in the UEFI Boot Manager implemented in EDK2
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
Looking at EDK2, ConXDev are treated as multi-instance devices, with ConPlatformDxe using AppendDevicePathInstance() and GetNextDevicePathInstance() to construct / navigate the multiple-instances
The spec should have defined these as a "multi-instance device path of all possible console X devices". See my notes below on multi-instance DP.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
I am leaning towards this conclusion as well. This is further supported by the following evidence:
Yes that's exactly what I thought as well and that's what option (2) was trying to describe on my first mail.
What I had in there was: Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
- EFI_DEVICE_PATH_UTILITIES_PROTOCOL has a function called IsDevicePathMultiInstance() along with walker functions AppendDevicePathInstance(), GetNextDevicePathInstance(). The description of these functions make it clear that the intention is to treat 0x01 as a separator between multiple instances of device paths in a multi-instance device path structure, with 0xff as the final end of DP indicator. For example, AppendDevicePathInstance() is defined as:
" This function creates a new device path by appending a copy of the specified device path instance to a copy of the specified device path in an allocated buffer. The end-of-device-path device node is moved after the end of the appended device node and a new end-of-device-path-instance node is inserted between."
- In EFI_USER_INFO , there are user info policy types that leverage the device path definitions (such as EFI_USER_INFO_ACCESS_FORBID_LOAD), and they are clear on distinguishing the term "multi-instance device path" from a "series of UEFI device paths":
"The record is a series of normal UEFI device paths (not multi-instance device paths)."
- The "multi-instance" usage also shows up in the language of LocateDevicePath()
So if the spec intended for FilePathList[] to use the 0x1 separator, they should have used the term "multi-instance" device path, which is not the case.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Yes. Or could simply say: "A packed array of UEFI device paths (not multi-instance device paths). The first element of the array is a device path that describes the device and location of the Image for this load option."
Either way, we can do a "code first" ECR in the UEFI spec. I say code first to ensure that it remains in public domain, and not blocked by UEFI Forum NDA until the publication of the next version of the UEFI spec. If we agree, I can get this process started and propose the language.
That's fine by me. I have most of the u-boot code ready, I'll finish it up and send patches.
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
One suggestion here is to take into consideration: whether we standardize this proposal (for loading additional kernel files from element[i], i>0 ) or not, the FW implementation must still be compliant with the UEFI rules around booting from a "short form device path" (UEFI spec, section 3.1.2)
Ok that's something I'll have to change since I am inserting the entire DP, but as Heinrich pointed out, u-boot already has support for short form device paths so the change shouldn't be too hard.
Thanks /Ilias
On Tue, 2 Mar 2021 at 00:31, Ilias Apalodimas ilias.apalodimas@linaro.org wrote:
Hi Samer,
Thanks for having a look at this
On Mon, Mar 01, 2021 at 08:44:00PM +0000, Samer El-Haj-Mahmoud wrote:
-----Original Message----- From: Heinrich Schuchardt xypron.glpk@gmx.de Sent: Monday, March 1, 2021 2:39 PM To: Ilias Apalodimas ilias.apalodimas@linaro.org; Grant Likely Grant.Likely@arm.com Cc: Boot Architecture Mailman List boot-architecture@lists.linaro.org; Samer El-Haj-Mahmoud Samer.El-Haj-Mahmoud@arm.com; Ard Biesheuvel ardb@kernel.org; Leif Lindholm leif@nuviainc.com Subject: Re: EFI_LOAD_FILE2 for initrd standardization
...
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
Agree that this is not clear, and could be interpreted either way. And yes, agree the UEFI spec needs a clarification
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The EDK2 code seems to be incomplete, with a "TODO" to support the FilePathList[]. In fact, the code calls it "FilePath" to be clear that it is assuming a single DevicePath (which means a 0xFF sub-type termination) https://github.com/tianocore/edk2/blob/master/MdeModulePkg%2FLibrary%2FUefiB...
Searching the code further, I see there is support for parsing multi-instance device path (separated by END_INSTANCE_DEVICE_PATH_SUBTYPE, or 0x1) in things like UefiDevicePathLib , parsing code, etc... But it does not seem to be supported at all in the UEFI Boot Manager implemented in EDK2
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
Looking at EDK2, ConXDev are treated as multi-instance devices, with ConPlatformDxe using AppendDevicePathInstance() and GetNextDevicePathInstance() to construct / navigate the multiple-instances
The spec should have defined these as a "multi-instance device path of all possible console X devices". See my notes below on multi-instance DP.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
I am leaning towards this conclusion as well. This is further supported by the following evidence:
Yes that's exactly what I thought as well and that's what option (2) was trying to describe on my first mail.
What I had in there was: Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
- EFI_DEVICE_PATH_UTILITIES_PROTOCOL has a function called IsDevicePathMultiInstance() along with walker functions AppendDevicePathInstance(), GetNextDevicePathInstance(). The description of these functions make it clear that the intention is to treat 0x01 as a separator between multiple instances of device paths in a multi-instance device path structure, with 0xff as the final end of DP indicator. For example, AppendDevicePathInstance() is defined as:
" This function creates a new device path by appending a copy of the specified device path instance to a copy of the specified device path in an allocated buffer. The end-of-device-path device node is moved after the end of the appended device node and a new end-of-device-path-instance node is inserted between."
- In EFI_USER_INFO , there are user info policy types that leverage the device path definitions (such as EFI_USER_INFO_ACCESS_FORBID_LOAD), and they are clear on distinguishing the term "multi-instance device path" from a "series of UEFI device paths":
"The record is a series of normal UEFI device paths (not multi-instance device paths)."
- The "multi-instance" usage also shows up in the language of LocateDevicePath()
So if the spec intended for FilePathList[] to use the 0x1 separator, they should have used the term "multi-instance" device path, which is not the case.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Yes. Or could simply say: "A packed array of UEFI device paths (not multi-instance device paths). The first element of the array is a device path that describes the device and location of the Image for this load option."
Either way, we can do a "code first" ECR in the UEFI spec. I say code first to ensure that it remains in public domain, and not blocked by UEFI Forum NDA until the publication of the next version of the UEFI spec. If we agree, I can get this process started and propose the language.
That's fine by me. I have most of the u-boot code ready, I'll finish it up and send patches.
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
One suggestion here is to take into consideration: whether we standardize this proposal (for loading additional kernel files from element[i], i>0 ) or not, the FW implementation must still be compliant with the UEFI rules around booting from a "short form device path" (UEFI spec, section 3.1.2)
Ok that's something I'll have to change since I am inserting the entire DP, but as Heinrich pointed out, u-boot already has support for short form device paths so the change shouldn't be too hard.
If I am understanding this correctly, the intent is to use the first devicepath in the Boot#### variable as the boot image, as is the case today, and use subsequent pairs of devicepaths as a key/value store, where the key is an abstract device path which describes how uboot should expose the actual device path in the value, using LoadFile2 and the existing Linux initrd vendor GUID + an filepath specifier.
I am fine with all this as long as there are no changes to the way the kernel consumes the LoadFile2 protocol. The kernel does not reason about Boot#### variables in the first places, and so how uboot uses them to keep track of boot entries is of no relevance to the loader.
I think repurposing LoadFile2 for other OSes or hypervisors like this makes sense, and there was already some pull from the Xen folks to implement GRUB multiboot in this manner. The only hard requirement I have from the Linux side is that loading LoadFile2+end_node yields the Linux initrd as it does today. What it does for LoadFile2+whatever (including where whatever == /initrd) does not impact Linux so it is free to do whatever is needed for the other use cases you have in mind.
If
Hi Ard,
ilias.apalodimas@linaro.org wrote:
Hi Samer,
Thanks for having a look at this
On Mon, Mar 01, 2021 at 08:44:00PM +0000, Samer El-Haj-Mahmoud wrote:
-----Original Message----- From: Heinrich Schuchardt xypron.glpk@gmx.de Sent: Monday, March 1, 2021 2:39 PM To: Ilias Apalodimas ilias.apalodimas@linaro.org; Grant Likely Grant.Likely@arm.com Cc: Boot Architecture Mailman List boot-architecture@lists.linaro.org; Samer El-Haj-Mahmoud Samer.El-Haj-Mahmoud@arm.com; Ard Biesheuvel ardb@kernel.org; Leif Lindholm leif@nuviainc.com Subject: Re: EFI_LOAD_FILE2 for initrd standardization
...
The UEFI spec knows two types of separators for device paths. Both have type 0x7F (End of Hardware Device Path) but differ by the sub-type:
Sub-Type 0xff – End Entire Device Path Sub-Type 0x01 – End Instance of a Device Path
Field EFI_LOAD_OPTION.FilePathList[] is described in the UEFI spec as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option."
It is not immediately clear if the separators between the array elements are of sub-type 0xff or 0x01. The description in the UEFI spec should be reworked for more clarity.
Agree that this is not clear, and could be interpreted either way. And yes, agree the UEFI spec needs a clarification
The current EDK II coding requires that the device path identifying the UEFI binary (i.e. FilePathList[0] is terminated by a sub-type 0xff end node.
The EDK2 code seems to be incomplete, with a "TODO" to support the FilePathList[]. In fact, the code calls it "FilePath" to be clear that it is assuming a single DevicePath (which means a 0xFF sub-type termination) https://github.com/tianocore/edk2/blob/master/MdeModulePkg%2FLibrary%2FUefiB...
Searching the code further, I see there is support for parsing multi-instance device path (separated by END_INSTANCE_DEVICE_PATH_SUBTYPE, or 0x1) in things like UefiDevicePathLib , parsing code, etc... But it does not seem to be supported at all in the UEFI Boot Manager implemented in EDK2
The UEFI variable ConDev is decribed as "The device path of all possible console input devices". The spec does not refer to it as an array of device paths.
Looking at EDK2, ConXDev are treated as multi-instance devices, with ConPlatformDxe using AppendDevicePathInstance() and GetNextDevicePathInstance() to construct / navigate the multiple-instances
The spec should have defined these as a "multi-instance device path of all possible console X devices". See my notes below on multi-instance DP.
So it seems that the UEFI spec editors mean by array of device paths that an element of the array is separated by sub-type 0xff. Each individual array element may be a collection of device paths instances separated by 0x01 sub-type end nodes.
I am leaning towards this conclusion as well. This is further supported by the following evidence:
Yes that's exactly what I thought as well and that's what option (2) was trying to describe on my first mail.
What I had in there was: Loaded Image device path - end node - VenMedia - Initrd DP - end instance - (repeat) - Initrd DP - end node - other DPs
- EFI_DEVICE_PATH_UTILITIES_PROTOCOL has a function called IsDevicePathMultiInstance() along with walker functions AppendDevicePathInstance(), GetNextDevicePathInstance(). The description of these functions make it clear that the intention is to treat 0x01 as a separator between multiple instances of device paths in a multi-instance device path structure, with 0xff as the final end of DP indicator. For example, AppendDevicePathInstance() is defined as:
" This function creates a new device path by appending a copy of the specified device path instance to a copy of the specified device path in an allocated buffer. The end-of-device-path device node is moved after the end of the appended device node and a new end-of-device-path-instance node is inserted between."
- In EFI_USER_INFO , there are user info policy types that leverage the device path definitions (such as EFI_USER_INFO_ACCESS_FORBID_LOAD), and they are clear on distinguishing the term "multi-instance device path" from a "series of UEFI device paths":
"The record is a series of normal UEFI device paths (not multi-instance device paths)."
- The "multi-instance" usage also shows up in the language of LocateDevicePath()
So if the spec intended for FilePathList[] to use the 0x1 separator, they should have used the term "multi-instance" device path, which is not the case.
In the device path spec a sentence could be added as follows:
"A packed array of UEFI device paths. The first element of the array is a device path that describes the device and location of the Image for this load option. *Each array element is terminated by a sub-type 0xff, End Entire Device Path node.*"
Yes. Or could simply say: "A packed array of UEFI device paths (not multi-instance device paths). The first element of the array is a device path that describes the device and location of the Image for this load option."
Either way, we can do a "code first" ECR in the UEFI spec. I say code first to ensure that it remains in public domain, and not blocked by UEFI Forum NDA until the publication of the next version of the UEFI spec. If we agree, I can get this process started and propose the language.
That's fine by me. I have most of the u-boot code ready, I'll finish it up and send patches.
Adding initrd or device tree device paths could be implemented as follows:
Array element [0]: device path of the binary (e.g. the Linux kernel) terminated by 0xff.
Array element [i], i > 0: device paths of the different initial RAM disks separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as initrds we can prepend a VenMedia() device path node with a specific GUID to the entire path.
Array element [j], j > 0: device path of the device-tree possibly followed by instances of device paths of device-tree overlays separated by 0x01 instance end nodes and terminated by an 0xff entire path node.
To identify the device paths with all its instances as device trees and device tree overlays we can prepend a VenMedia() device path node with a specific GUID to the entire path.
One suggestion here is to take into consideration: whether we standardize this proposal (for loading additional kernel files from element[i], i>0 ) or not, the FW implementation must still be compliant with the UEFI rules around booting from a "short form device path" (UEFI spec, section 3.1.2)
Ok that's something I'll have to change since I am inserting the entire DP, but as Heinrich pointed out, u-boot already has support for short form device paths so the change shouldn't be too hard.
If I am understanding this correctly, the intent is to use the first devicepath in the Boot#### variable as the boot image, as is the case today, and use subsequent pairs of devicepaths as a key/value store, where the key is an abstract device path which describes how uboot should expose the actual device path in the value, using LoadFile2 and the existing Linux initrd vendor GUID + an filepath specifier.
You do understand it correctly, there's slight variation on what we can do to store those device paths on the FilePathList[] but that's more an implementation detail. Both would still have the same result and it only affects how the bootloader will parse those (and possibly concat multiple initrds) before copying them to the kernel provided buffer.
I am fine with all this as long as there are no changes to the way the kernel consumes the LoadFile2 protocol. The kernel does not reason about Boot#### variables in the first places, and so how uboot uses them to keep track of boot entries is of no relevance to the loader.
I think repurposing LoadFile2 for other OSes or hypervisors like this makes sense, and there was already some pull from the Xen folks to implement GRUB multiboot in this manner. The only hard requirement I have from the Linux side is that loading LoadFile2+end_node yields the Linux initrd as it does today. What it does for LoadFile2+whatever (including where whatever == /initrd) does not impact Linux so it is free to do whatever is needed for the other use cases you have in mind.
I think what's hard for people to understand here, and part of the confusion comes from the part that the efi-stub will eventually call the protocol with the end node. So the discussions I've been having around is "why does the kernel call that with and end node and not the initrd GUID" (by initrd GUID I mean the "struct efi_vendor_dev_path vendor" defined in the kernel).
But what we need to make clear here, is that this device path is *already used* to locate the protocol. After that and following the EFI spec, the 'front' part is removed and the remaining portion is returned. So the returned device path (which is now just the end node) is used by the kernel to call that protocol. This identifies the 'initrd GUID uniquely' and a similar method can be used for other files.
Hope that makes sense now
Regards /Ilias
If
boot-architecture@lists.linaro.org
-
Ard Biesheuvel -
François Ozog -
Grant Likely -
Heinrich Schuchardt -
Ilias Apalodimas -
Samer El-Haj-Mahmoud -
Simon Glass