[PATCH] part: efi: Rely on a definition to specify the primary GPT position

[Roman Stratiienko]

Hello Heinrich,

Sorry, I'm having a hard time answering your questions.

> When writing the partition table shouldn't the offset be reflected in
the protective MBR fields StartingCHS and StartingLBA? See Table 5.4
"Protective MBR Partition Record protecting the entire disk*" of the
UEFI 2.10 specification.
> Does Linux support this?

Linux doesn't follow the protective MBR to find the GPT location if
that was your question.

> For testing remove the backup partition table!

Testing what? The backup partition table is always restored by Linux
if missing or corrupted.
I tested it previously and it worked.

That was the only question I could find. If your other statements
intended to ask the question, please let me know.

BR,
Roman

нд, 14 лип. 2024 р. о 09:39 Heinrich Schuchardt <xypron.glpk at gmx.de> пише:
>
> On 7/13/24 23:16, Roman Stratiienko wrote:
> > Hello Heinrich,
> >
> > Thank you for your comments.
> >
> > Just a few thoughts about this patch.
> >
> > 1. I was also sending this patch as a noninvasive, semi-cosmetic
> > change. Relying on hardcoded values is usually not a good practice.
> > 2. Shifting of primary GPT works in u-boot and Linux kernel with minor
> > tweaks. See patches [1] and [2] for more context. I'm not going to
> > upstream them.
> > 3. Accepting this patch upstream helps me to reduce the dependency
> > chain and potential conflicts during the rebase.
>
> Hello Roman,
>
> could you, please, answer my questions below.
>
> Best regards
>
> Heinrich
>
> >
> > BR, Roman.
> >
> > [1]: https://github.com/GloDroidCommunity/amlogic-series/blob/8cc5ee7df1358092f77f43aa0c9661883473e5af/patches-aosp/glodroid/bootloader/u-boot/0005-GLODROID-Shift-GPT-position-to-support-Amlogic-platf.patch
> > [2]: https://github.com/GloDroidCommunity/amlogic-series/blob/8cc5ee7df1358092f77f43aa0c9661883473e5af/patches-aosp/glodroid/kernel/common-android14-6.1-lts/0001-GLODROID-Allow-placing-primary-GPT-to-2MiB-offset-to.patch
> >
> > пн, 10 черв. 2024 р. о 15:22 Heinrich Schuchardt <xypron.glpk at gmx.de> пише:
> >>
> >> On 19.05.24 17:35, Roman Stratiienko wrote:
> >>> Use GPT_PRIMARY_PARTITION_TABLE_LBA as the only source for a primary GPT
> >>> location instead of hardcoding it every time.
> >>>
> >>> Sometimes, we need to shift the primary GPT location.
> >>> Having the location in the single place simplifies such shifting.
> >>
> >> In gdisk there is an expert command "j - move the main partition table".
> >> This corresponds to CONFIG_EFI_PARTITION_ENTRIES_OFF. It does not move
> >> the GPT primary header but only the partition entries. So sector 1 is
> >> still occupied.
> >>
> >> If I read you correctly, CONFIG_EFI_PARTITION_ENTRIES_OFF is not doing
> >> what you need. Instead of only moving the partition entries of the
> >> primary partition table, you want to move the complete primary GPT table.
> >>
> >> Somehow the operating system must find the primary GPT table.
> >>
> >> When writing the partition table shouldn't the offset be reflected in
> >> the protective MBR fields StartingCHS and StartingLBA? See Table 5.4
> >> "Protective MBR Partition Record protecting the entire disk*" of the
> >> UEFI 2.10 specification.
> >>
> >> Does Linux support this? For testing remove the backup partition table!
> >>
> >> GPT_PRIMARY_PARTITION_TABLE_LBA should be made configurable in this case.
> >>
> >> Best regards
> >>
> >> Heinrich
> >>
> >>>
> >>> Signed-off-by: Roman Stratiienko <r.stratiienko at gmail.com>
> >>> ---
> >>>    disk/part_efi.c | 10 +++++-----
> >>>    1 file changed, 5 insertions(+), 5 deletions(-)
> >>>
> >>> diff --git a/disk/part_efi.c b/disk/part_efi.c
> >>> index 4ce9243ef2..b55e251ea1 100644
> >>> --- a/disk/part_efi.c
> >>> +++ b/disk/part_efi.c
> >>> @@ -387,7 +387,7 @@ int write_gpt_table(struct blk_desc *desc, gpt_header *gpt_h, gpt_entry *gpt_e)
> >>>        gpt_h->header_crc32 = cpu_to_le32(calc_crc32);
> >>>
> >>>        /* Write the First GPT to the block right after the Legacy MBR */
> >>> -     if (blk_dwrite(desc, 1, 1, gpt_h) != 1)
> >>> +     if (blk_dwrite(desc, (lbaint_t)le64_to_cpu(gpt_h->my_lba), 1, gpt_h) != 1)
> >>>                goto err;
> >>>
> >>>        if (blk_dwrite(desc, le64_to_cpu(gpt_h->partition_entry_lba),
> >>> @@ -534,7 +534,7 @@ int gpt_fill_pte(struct blk_desc *desc,
> >>>
> >>>    static uint32_t partition_entries_offset(struct blk_desc *desc)
> >>>    {
> >>> -     uint32_t offset_blks = 2;
> >>> +     uint32_t offset_blks = GPT_PRIMARY_PARTITION_TABLE_LBA + 1;
> >>>        uint32_t __maybe_unused offset_bytes;
> >>>        int __maybe_unused config_offset;
> >>>
> >>> @@ -572,8 +572,8 @@ static uint32_t partition_entries_offset(struct blk_desc *desc)
> >>>         * The earliest LBA this can be at is LBA#2 (i.e. right behind
> >>>         * the (protective) MBR and the GPT header.
> >>>         */
> >>> -     if (offset_blks < 2)
> >>> -             offset_blks = 2;
> >>> +     if (offset_blks < (GPT_PRIMARY_PARTITION_TABLE_LBA + 1))
> >>> +             offset_blks = GPT_PRIMARY_PARTITION_TABLE_LBA + 1;
> >>>
> >>>        return offset_blks;
> >>>    }
> >>> @@ -584,7 +584,7 @@ int gpt_fill_header(struct blk_desc *desc, gpt_header *gpt_h, char *str_guid,
> >>>        gpt_h->signature = cpu_to_le64(GPT_HEADER_SIGNATURE_UBOOT);
> >>>        gpt_h->revision = cpu_to_le32(GPT_HEADER_REVISION_V1);
> >>>        gpt_h->header_size = cpu_to_le32(sizeof(gpt_header));
> >>> -     gpt_h->my_lba = cpu_to_le64(1);
> >>> +     gpt_h->my_lba = cpu_to_le64(GPT_PRIMARY_PARTITION_TABLE_LBA);
> >>>        gpt_h->alternate_lba = cpu_to_le64(desc->lba - 1);
> >>>        gpt_h->last_usable_lba = cpu_to_le64(desc->lba - 34);
> >>>        gpt_h->partition_entry_lba =
> >>
>