[U-Boot] [PATCH v1] fastboot: handle flash write to GPT partition

[Steve Rae <srae@broadcom.com>]

Hi Lukasz,

On 14-12-08 03:21 AM, Lukasz Majewski wrote:
> Hi Steve,
>
>> Implement a feature to allow fastboot to write the downloaded image
>> to the space reserved for the Protective MBR and the Primary GUID
>> Partition Table.
>>
>> Signed-off-by: Steve Rae <srae@broadcom.com>
>> ---
>>
>>   README          |  7 +++++++
>>   common/fb_mmc.c | 19 ++++++++++++++++---
>>   2 files changed, 23 insertions(+), 3 deletions(-)
>>
>> diff --git a/README b/README
>> index 66770b6..3b6ef7f 100644
>> --- a/README
>> +++ b/README
>> @@ -1769,6 +1769,13 @@ The following options need to be configured:
>>   		regarding the non-volatile storage device. Define
>> this to the eMMC device that fastboot should use to store the image.
>>
>> +		CONFIG_FASTBOOT_GPT_NAME
>> +		The fastboot "flash" command supports writing the
>> downloaded
>> +		image to the Protective MBR and the Primary GUID
>> Partition
>> +		Table. This occurs when the specified "partition
>> name" on the
>> +		"fastboot flash" command line matches this value.
>> +		Default is GPT_ENTRY_NAME (currently "gpt") if
>> undefined. +
>>   - Journaling Flash filesystem support:
>>   		CONFIG_JFFS2_NAND, CONFIG_JFFS2_NAND_OFF,
>> CONFIG_JFFS2_NAND_SIZE, CONFIG_JFFS2_NAND_DEV
>> diff --git a/common/fb_mmc.c b/common/fb_mmc.c
>> index fb06d8a..89fbf23 100644
>> --- a/common/fb_mmc.c
>> +++ b/common/fb_mmc.c
>> @@ -4,12 +4,17 @@
>>    * SPDX-License-Identifier:	GPL-2.0+
>>    */
>>
>> +#include <config.h>
>>   #include <common.h>
>>   #include <fb_mmc.h>
>>   #include <part.h>
>>   #include <aboot.h>
>>   #include <sparse_format.h>
>>
>> +#ifndef CONFIG_FASTBOOT_GPT_NAME
>> +#define CONFIG_FASTBOOT_GPT_NAME GPT_ENTRY_NAME
>> +#endif
>> +
>>   /* The 64 defined bytes plus the '\0' */
>>   #define RESPONSE_LEN	(64 + 1)
>>
>> @@ -62,9 +67,9 @@ static void write_raw_image(block_dev_desc_t
>> *dev_desc, disk_partition_t *info, void fb_mmc_flash_write(const char
>> *cmd, void *download_buffer, unsigned int download_bytes, char
>> *response) {
>> -	int ret;
>>   	block_dev_desc_t *dev_desc;
>>   	disk_partition_t info;
>> +	lbaint_t blksz;
>>
>>   	/* initialize the response buffer */
>>   	response_str = response;
>> @@ -76,8 +81,16 @@ void fb_mmc_flash_write(const char *cmd, void
>> *download_buffer, return;
>>   	}
>>
>> -	ret = get_partition_info_efi_by_name(dev_desc, cmd, &info);
>> -	if (ret) {
>> +	if (strcmp(cmd, CONFIG_FASTBOOT_GPT_NAME) == 0) {
>> +		printf("%s: updating GUID Partition Table (including
>> MBR)\n",
>> +		       __func__);
>> +		/* start at Protective MBR */
>> +		info.start = (GPT_PRIMARY_PARTITION_TABLE_LBA - 1);
>> +		blksz = dev_desc->blksz;
>> +		info.blksz = blksz;
>> +		/* assume that the Partition Entry Array starts in
>> LBA 2 */
>> +		info.size = (2 + (GPT_ENTRY_NUMBERS *
>> GPT_ENTRY_SIZE) / blksz);
>> +	} else if (get_partition_info_efi_by_name(dev_desc, cmd,
>> &info)) { error("cannot find partition: '%s'\n", cmd);
>>   		fastboot_fail("cannot find partition");
>>   		return;
>
> Sorry for a late reply. I've just come back from a short holidays.
>
> I'm curious if you have encountered any problems with GPT replaced in
> that way?

No -- this "technique" seems to be fine (for the Primary GPT)....

>
> It seems strange to me that you only change primary GPT partition
> without taking care of the secondary (backup) one.

It seems that the device operates correctly with or without the Backup 
GPT, and it doesn't seem to matter if they are the same or not.
Thus, we have gone back and forth on this one - should we automatically 
update the Backup GPT whenever the Primary GPT is updated, or should 
there be a second step (possibly a "fastboot oem" command) to update the 
Backup GPT... (currently, we are proposing the latter)
What would you suggest?

>
>  From my experience when you export your eMMC to Host PC via UMS, host's
> PC tools will complain about mismatch in the GPT tables.

( I have never done this - what tools are you using? Could you provide 
instructions for me to try? Thanks! )

>
> Moreover, I would suggest transactional update of GPT by checking GPT
> image CRC before writing. In this way you can always perform recovery if
> needed.

This is a good idea - I'll look into it - Thanks!
>

Thanks, Steve