Hi Jagan,

Hi Lukasz,

On Wed, Nov 27, 2019 at 4:15 PM Lukasz Majewski lukma@denx.de wrote:

...

Hi Jagan,

...

Hi,

I have created GPT table start from 8MB for kernel, roots etc. something like

Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00004000 0x00023fff "boota" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: c12a7328-f81f-11d2-ba4b-00a0c93ec93b 2 0x00024000 0x00043fff "bootb" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 21686148-6449-6e6f-744e-656564454649 3 0x00044000 0x00243fff "rootfsa" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: b921b045-1df0-41c3-af44-4c6f280d3fae 4 0x00244000 0x00443fff "rootfsb" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 8da63339-0007-60c0-c436-083ac8230908 5 0x00444000 0x0070bfde "data" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 4f72ab70-69be-5948-81ff-4fc3daf24faa

I have not included SPL, U-Boot to the partition list since it start from 0x400 in i.MX6. So I'm writing SPL separately using fastboot(with offset) or ums.

But by doing this, the partition header seems overlapped so the output looks

GUID Partition Table Entry Array CRC is wrong: 0x6a1aba0a != 0x8e4fd548 find_valid_gpt: *** ERROR: Invalid GPT *** find_valid_gpt: *** Using Backup GPT *** Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00004000 0x00023fff "boota" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: c12a7328-f81f-11d2-ba4b-00a0c93ec93b 2 0x00024000 0x00043fff "bootb" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 21686148-6449-6e6f-744e-656564454649 3 0x00044000 0x00243fff "rootfsa" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: b921b045-1df0-41c3-af44-4c6f280d3fae 4 0x00244000 0x00443fff "rootfsb" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 8da63339-0007-60c0-c436-083ac8230908 5 0x00444000 0x0070bfde "data" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 4f72ab70-69be-5948-81ff-4fc3daf24faa

So, what I understand is If I create the GPT first and then write the SPL, the SPL writing process will destroy the GPT Table and if I write SPL first and then create GPT, the GPT creation process will destroy the SPL.

Is there any way to fix this? I remember we can prevent this overlap by preserving GPT Table som other or boot partition instead of having them at the beginning by default.

Any inputs?

On the diagram of GPT description [1] there is the info that partitions can start from LBA 34 (0x200 * 34) = 0x4400

From the above it seems like you starts from 0x4000 your boota partition, which then overwrites the "Entries 5-128" which shall be 0 and are protected by CRC written in the Primary GPT Header.

Please try to adjust your partition scheme to start from 0x4400.

I still see the overlap. I have created boota at 0x4400 and the write the SPL at 0x400

^^^^^^ - this overwrites the GPT header IMHO.

You may dump the eMMC and check if this is the case (even with u-boot's md.l utility)

I had similar problem on some Beagle Bone Black (AM33x) board.

The ROM wanted to boot from the fixed eMMC LBA offset which was clashing with GPT.

Fortunately, it was also possible to boot from FAT (it was checked before the raw offset from eMMC case) partition, so I used this option instead.

But hey, isn't it possible to store SPL/u-boot to the eMMC's boot (the separate HW partition) partition and store GPT to the user accessible one (the large one - e.g. 4 GiB)?

icorem6qdl-rqs> mmc part

Partition Map for MMC device 2 -- Partition Type: EFI

GUID Partition Table Entry Array CRC is wrong: 0xfca8e0bf != 0xc10fdc7b find_valid_gpt: *** ERROR: Invalid GPT *** find_valid_gpt: *** Using Backup GPT *** Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00004400 0x000243ff "bootA" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: c12a7328-f81f-11d2-ba4b-00a0c93ec93b 2 0x00024400 0x000443ff "bootB" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 21686148-6449-6e6f-744e-656564454649 3 0x00044400 0x002443ff "rootfsA" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: b921b045-1df0-41c3-af44-4c6f280d3fae 4 0x00244400 0x004443ff "rootfsB" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 8da63339-0007-60c0-c436-083ac8230908 5 0x00444400 0x0070bfde "data" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: d61c5b08-c376-f243-a883-15dd900766c3

Jagan.

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de

Hi Jagan,

Hi Lukasz,

On Wed, Nov 27, 2019 at 9:47 PM Lukasz Majewski lukma@denx.de wrote:

...

Hi Jagan,

...

Hi Lukasz,

On Wed, Nov 27, 2019 at 4:15 PM Lukasz Majewski lukma@denx.de wrote:

...

Hi Jagan,

...

Hi,

I have created GPT table start from 8MB for kernel, roots etc. something like

Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00004000 0x00023fff "boota" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: c12a7328-f81f-11d2-ba4b-00a0c93ec93b 2 0x00024000 0x00043fff "bootb" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 21686148-6449-6e6f-744e-656564454649 3 0x00044000 0x00243fff "rootfsa" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: b921b045-1df0-41c3-af44-4c6f280d3fae 4 0x00244000 0x00443fff "rootfsb" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 8da63339-0007-60c0-c436-083ac8230908 5 0x00444000 0x0070bfde "data" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 4f72ab70-69be-5948-81ff-4fc3daf24faa

I have not included SPL, U-Boot to the partition list since it start from 0x400 in i.MX6. So I'm writing SPL separately using fastboot(with offset) or ums.

But by doing this, the partition header seems overlapped so the output looks

GUID Partition Table Entry Array CRC is wrong: 0x6a1aba0a != 0x8e4fd548 find_valid_gpt: *** ERROR: Invalid GPT *** find_valid_gpt: *** Using Backup GPT *** Part Start LBA End LBA Name Attributes Type GUID Partition GUID 1 0x00004000 0x00023fff "boota" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: c12a7328-f81f-11d2-ba4b-00a0c93ec93b 2 0x00024000 0x00043fff "bootb" attrs: 0x0000000000000004 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 21686148-6449-6e6f-744e-656564454649 3 0x00044000 0x00243fff "rootfsa" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: b921b045-1df0-41c3-af44-4c6f280d3fae 4 0x00244000 0x00443fff "rootfsb" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 8da63339-0007-60c0-c436-083ac8230908 5 0x00444000 0x0070bfde "data" attrs: 0x0000000000000000 type: ebd0a0a2-b9e5-4433-87c0-68b6b72699c7 guid: 4f72ab70-69be-5948-81ff-4fc3daf24faa

So, what I understand is If I create the GPT first and then write the SPL, the SPL writing process will destroy the GPT Table and if I write SPL first and then create GPT, the GPT creation process will destroy the SPL.

Is there any way to fix this? I remember we can prevent this overlap by preserving GPT Table som other or boot partition instead of having them at the beginning by default.

Any inputs?

On the diagram of GPT description [1] there is the info that partitions can start from LBA 34 (0x200 * 34) = 0x4400

From the above it seems like you starts from 0x4000 your boota partition, which then overwrites the "Entries 5-128" which shall be 0 and are protected by CRC written in the Primary GPT Header.

Please try to adjust your partition scheme to start from 0x4400.

I still see the overlap. I have created boota at 0x4400 and the write the SPL at 0x400

         ^^^^^^ - this overwrites the GPT header IMHO.  

True, this overlap GPT. and we don't have an option since ROM expecting the SPL at 0x400.

The ROM expects some LBA from eMMC being accessible.

Maybe it would be possible to add some specific for your application DCD code to enable eMMC's boot{01} partition?

Also IIRC i.MX6 allows playing around with some fuses, so maybe you can specify eMMC to use boot{01} HW partition?

...

You may dump the eMMC and check if this is the case (even with u-boot's md.l utility)

I had similar problem on some Beagle Bone Black (AM33x) board.

The ROM wanted to boot from the fixed eMMC LBA offset which was clashing with GPT.

Fortunately, it was also possible to boot from FAT (it was checked before the raw offset from eMMC case) partition, so I used this option instead.

You mean create FAT partition and copy SPL/U-Boot binaries on the directory?

Yes, this worked with BBB.

...

But hey, isn't it possible to store SPL/u-boot to the eMMC's boot (the separate HW partition) partition and store GPT to the user accessible one (the large one - e.g. 4 GiB)?

Yes, this is what I'm thinking of. But how to preserve the GPT at another partition or user accessible area than beginning (which is done by default).

Maybe you can use DCD? Or play around with fuses?

ROM booting options are quite verbose - and I don't believe that NXP did not consider the use case with GPT and single eMMC as the memory medium.

(All my boards use SPI-NOR / NOR-FLASH for boot up, and then switch to eMMC if needed).

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de