[U-Boot] [PATCH] efi_loader: don't allocate unusable RAM

[Stephen Warren]

On 08/01/2018 11:57 PM, Heinrich Schuchardt wrote:
> On 08/01/2018 06:17 PM, Stephen Warren wrote:
>> On 08/01/2018 12:07 AM, Heinrich Schuchardt wrote:
>>> On 07/31/2018 09:44 PM, Stephen Warren wrote:
>>>> From: Stephen Warren <swarren at nvidia.com>
>>>>
>>>> Some boards define a maximum usable RAM top that's more restrictive than
>>>> the ranges defined by U-Boot's memory bank definitions[1]. In this case,
>>>> the unusable RAM isn't mapped in the page tables, and so the EFI code
>>>> must
>>>> not attempt to allocate RAM from outside the usable regions. Fix
>>>> efi_find_free_memory() to detect when max_addr is unconstrained or
>>>> out of
>>>> range, and substitue a valid value.
>>>>
>>>
>>> In this case the board has to call efi_add_memory_map() to mark the
>>> reserved region (cf. board/raspberrypi/rpi/rpi.c) or have to mark the
>>> memory region as reserved in the device tree (see
>>> efi_carve_out_dt_rsv()).
>>>
>>> Please, check if the tegra boards with which you had problems do so.
>>
>> I don't think this makes sense; memory should be managed the same way
>> within U-Boot no matter which part of U-Boot is consuming that memory.
>> Memory regions are currently represented by the content of the memory
>> bank definitions and gd->ram_top. Having different parts of the code
>> define legal RAM usage in different ways is horribly inconsistent.
> 
> Memory banks and gd->top together cannot reflect unusable memory regions
> in the middle of the RAM area.

Sure this is possible. For example, a common set of memory banks for 
Tegra is:

bank 0:
start:  0x080000000
length: 0x070000000
limit:  0x0f0000000
bank 1:
start:  0x100000000
length: 0x080000000
limit:  0x180000000

gd->ram_top = 0x0f0000000
That is set so that U-Boot itself (not just the UEFI code) doesn't use 
RAM above 4GB due to HW peripherals that can't access such RAM. The RAM 
above 4GB is left in the DRAM banks so that the kernel can use it for 
non-DMA purposes.

That represents a hole at 0xf0000000..0x100000000.

> To be consistent reflect all information in the device tree and
> calculate gd->ram_top from the device tree information.

Nothing in U-Boot that I'm aware of gets RAM information from DT. It all 
comes from the banks in gd. That's set up by some early low level code 
in arch/arm/mach-tegra/board2.c, which at least upstream hard-codes the 
size of the RAM gap below 4G rather than getting information from either 
DT or the earlier FW loads. So, I'm not sure what DT has to do with this.

>> At this point, I think the best thing is to revert aa909462d018
>> "efi_loader: efi_allocate_pages is too restrictive" since it causes a
>> regression on Jetson TX1, and we can work out the correct way to fix all
>> this once the system is working again.
>>
> 
> The situation before the patch was really buggy. It is sufficient if you
> get the Jetson device tree right.

What bugs specifically? Perhaps there's a better way to fix them.

Anyway, there was agreement in other messages in the thread to revert 
the problematic patch to avoid breaking the code while we work on a more 
complete solution. That's about all I can do right now since I'm heading 
off on vacation for a couple of weeks and would rather we have a working 
U-Boot running in the test farm during that time so that any other bugs 
that are introduced get caught rather than hidden. I'll send the revert.