[U-Boot] [PATCH] gpt: add part-uuid and part-num subcommands
Andrey Yurovsky
yurovsky at gmail.com
Tue Dec 5 19:00:55 UTC 2017
On Tue, Dec 5, 2017 at 7:10 AM, Lukasz Majewski <l.majewski at majess.pl> wrote:
> Hi Andrey,
>
>> On Mon, Dec 4, 2017 at 1:12 PM, Lukasz Majewski
>> <l.majewski at majess.pl> wrote:
>> > Hi Andrey,
>> >
>> >> Hi Lukasz,
>> >>
>> >> On Tue, Nov 14, 2017 at 1:45 AM, Lukasz Majewski
>> >> <l.majewski at majess.pl> wrote:
>> >> > Hi Andrey,
>> >> >
>> >> >> Hi Lukasz,
>> >> >>
>> >> >> On Thu, Nov 9, 2017 at 2:28 PM, Lukasz Majewski
>> >> >> <l.majewski at majess.pl> wrote:
>> >> >> > On Thu, 9 Nov 2017 07:34:44 -0800
>> >> >> > Andrey Yurovsky <yurovsky at gmail.com> wrote:
>> >> >> >
>> >> >> >> On Thu, Nov 9, 2017 at 1:55 AM, Lukasz Majewski
>> >> >> >> <l.majewski at majess.pl> wrote:
>> >> >> >> > Hi Andrey,
>> >> >> >> >
>> >> >> >> >> Hi Otavio,
>> >> >> >> >>
>> >> >> >> >> On Wed, Nov 8, 2017 at 2:47 AM, Otavio Salvador
>> >> >> >> >> <otavio.salvador at ossystems.com.br> wrote:
>> >> >> >> >> > On Tue, Nov 7, 2017 at 10:43 PM, your name
>> >> >> >> >> > <yurovsky at gmail.com> wrote:
>> >> >> >> >> >> From: Andrey Yurovsky <yurovsky at gmail.com>
>> >> >> >> >> >>
>> >> >> >> >> >> It is useful to be able to retrieve a partition UUID or
>> >> >> >> >> >> number given the partition label, for instance some
>> >> >> >> >> >> systems use the partition label to indicate the purpose
>> >> >> >> >> >> of the partition (such as "rootfs0" being the 0th root
>> >> >> >> >> >> file system in an A/B image scheme).
>> >> >> >> >> >>
>> >> >> >> >> >> Add "gpt part-uuid" to retrieve the partition UUID for
>> >> >> >> >> >> a given label and "gpt part-num" to retrieve the
>> >> >> >> >> >> partition number for a given label along with some
>> >> >> >> >> >> documentation.
>> >> >> >> >> >>
>> >> >> >> >> >> Signed-off-by: Andrey Yurovsky <yurovsky at gmail.com>
>> >> >> >> >> >
>> >> >> >> >> > Why not use the 'part' cmd? it provides it.
>> >> >> >> >>
>> >> >> >> >> Sorry, I missed the part cmd, it doesn't seem to be
>> >> >> >> >> documented in doc/ and it's unclear what <dev> means
>> >> >> >> >> there.
>> >> >> >> >
>> >> >> >> > If I may ask - Andrey, if you are now on this "topic" -
>> >> >> >> > would you dare to add some ./doc entry for 'part'
>> >> >> >> > command?
>> >> >> >>
>> >> >> >> Yes, I will do that.
>> >> >> >
>> >> >> > Thanks :-)
>> >> >>
>> >> >> On further investigation I am not sure that it's possible to
>> >> >> extend the part command to retrieve UUIDs by label because of
>> >> >> the design of the partition type drivers. Here is how I
>> >> >> understand it to work: 1. the "part" command uses
>> >> >> part_get_info() and in turn gets a partition driver and can
>> >> >> call print() there (which is how EFI/GPT disks are printed with
>> >> >> "part list"). The right information (including label) is
>> >> >> printed but it's not tied to the caller in any way.
>> >> >
>> >> > Maybe you can set some env variable with proper data?
>> >> >
>> >> > For example, please refer to ./cmd/part.c do_part_start()
>> >> > function.
>> >> >
>> >> > Example call from envs (include/configs/display5.h):
>> >> > "part start mmc ${mmcdev} ${kernel_part} lba_start; " \
>> >> >
>> >>
>> >> Again that assumes the partition is referred to by number, I need
>> >> it to be by label, and the part/disk interface does not seem to
>> >> have any way to utilize labels. Unfortunately it looks like my
>> >> original approach with the gpt command is the only way to
>> >> implement this with the current design (at least from what I see
>> >> here). Please let me know if I've missed something. Thanks!
>> >
>> > Please correct me if I'm wrong - you need the starting LBA of the
>> > partition named e.g. "FOO" in gpt ?
>> >
>> > Conceptually it would be correct to have:
>> >
>> > part start <interface> <dev> <NAME - e.g.'FOO'> <env to set>
>> > gpt start <interface> <dev> <NAME- e.g. 'FOO'> <env to set>
>> >
>> > If your code is really _small_ and can be used only with GPT, then
>> > lets go for the second option.
>>
>> The use cases I have in mind:
>>
>> 1. determine which root file system to use by label, let's say
>> "rootfs1" and pass its UUID to the Linux kernel via the command line,
>> ex: "PART-UUID=${uuid}". To do this we need a way to ask for a UUID
>> corresponding to a label in the partition table (given duplicate
>> labels, assume it gives us the first or last match).
>
> I see. I thought that you need start LBA.
>
> From your use case it seems like extending the 'gpt' command is the
> right thing to do - since already some uuid handling is done there.
In that case would you please let me know if the patch I sent is reasonable?
>> 2. determine which file system to load a file from (ex: fatload) given
>> a label.
>
> There is a group of generic commands - like load, ls, etc.
It seems that the lowest common denominator for these is indeed the
partition number (then they can talk to the disk driver) so adding a
way to map labels to that number (as in my original patch) may be
useful, the user's script can then use these commands once it learns
that information via the gpt command.
>> I'm not sure that the starting LBA is helpful here, we really
>> are looking to map a label to a partition number in the table.
>
> If you are using GPT, then you also may want to extend the 'gpt'
> command -> iterate through PTEs and when label matched, return the part
> number. With it you can use e.g. load mmc 0 <part> file.
Right, exactly. For that I added "gpt part-num" in the patch.
>>
>> The implementation of "part start" interprets the argument as a
>> partition number, so I can get the stating LBA if I know the partition
>> number but I don't see a way (via cmd/disk.c) to get anything useful
>> if all I know is a label, and that is what I'm trying to solve.
>
>
>
> --
> Best regards,
>
> Ćukasz Majewski
More information about the U-Boot
mailing list