Re: [PATCH v1 3/3] partitions: Introduce NVIDIA Tegra Partition Table

[Dmitry Osipenko]

04.03.2020 19:36, Ulf Hansson ÐÐÑÐÑ:
> On Tue, 25 Feb 2020 at 01:20, Stephen Warren <swarren@xxxxxxxxxxxxx> wrote:
>>
>> On 2/24/20 4:18 PM, Dmitry Osipenko wrote:
>>> All NVIDIA Tegra devices use a special partition table format for the
>>> internal storage partitioning. Most of Tegra devices have GPT partition
>>> in addition to TegraPT, but some older Android consumer-grade devices do
>>> not or GPT is placed in a wrong sector, and thus, the TegraPT is needed
>>> in order to support these devices properly in the upstream kernel. This
>>> patch adds support for NVIDIA Tegra Partition Table format that is used
>>> at least by all NVIDIA Tegra20 and Tegra30 devices.
>>
>>> diff --git a/arch/arm/mach-tegra/tegra.c b/arch/arm/mach-tegra/tegra.c
>>
>>> +static void __init tegra_boot_config_table_init(void)
>>> +{
>>> +     void __iomem *bct_base;
>>> +     u16 pt_addr, pt_size;
>>> +
>>> +     bct_base = IO_ADDRESS(TEGRA_IRAM_BASE) + TEGRA_IRAM_BCT_OFFSET;
>>
>> This shouldn't be hard-coded. IIRC, the boot ROM writes a BIT (Boot
>> Information Table) to a fixed location in IRAM, and there's some value
>> in the BIT that points to where the BCT is in IRAM. In practice, it
>> might work out that the BCT is always at the same place in IRAM, but
>> this certainly isn't guaranteed. I think there's code in U-Boot which
>> extracts the BCT location from the BIT? Yes, see
>> arch/arm/mach-tegra/ap.c:get_odmdata().
> 
> So, have you considered using the command line partition option,
> rather than adding yet another partition scheme to the kernel?
> 
> In principle, you would let the boot loader scan for the partitions,
> likely from machine specific code in U-boot. Then you append these to
> the kernel command line and let block/partitions/cmdline.c scan for
> it.

The bootloader is usually locked-down on a consumer Tegra machines (it's
signed / encrypted).

Technically, it should be possible to chain-load some custom secondary
bootloader instead of a kernel image, but this is not very practical
because now:

1. There is a need to make a custom bootloader and it is quite a lot of
work.

2. You'll have to tell everybody that a custom booloader may need to be
used in order to get a working eMMC.

3. NVIDIA's bootloader already passes a command line parameter to kernel
for locating GPT entry, but this hack is not acceptable for the upstream
kernel.