[PATCH] efi_loader : Suppress error print message

[Heinrich Schuchardt]

On 1/23/24 19:53, Bhumkar, Tejas Arvind wrote:
> [AMD Official Use Only - General]
>
> Hi Ilias & Heinrich,
>
>> -----Original Message-----
>> From: Heinrich Schuchardt <xypron.glpk at gmx.de>
>> Sent: Tuesday, January 23, 2024 3:30 PM
>> To: Ilias Apalodimas <ilias.apalodimas at linaro.org>; Bhumkar, Tejas Arvind
>> <tejas.arvind.bhumkar at amd.com>
>> Cc: u-boot at lists.denx.de; xypron.glpk at gmx.de; trini at konsulko.com;
>> sjg at chromium.org; Simek, Michal <michal.simek at amd.com>; Abbarapu,
>> Venkatesh <venkatesh.abbarapu at amd.com>; git at xilinx.com
>> Subject: Re: [PATCH] efi_loader : Suppress error print message
>>
>> Caution: This message originated from an External Source. Use proper caution
>> when opening attachments, clicking links, or responding.
>>
>>
>> On 1/23/24 09:33, Ilias Apalodimas wrote:
>>> Hi Tejas,
>>>
>>> On Mon, 22 Jan 2024 at 21:12, Tejas Bhumkar
>>> <tejas.arvind.bhumkar at amd.com> wrote:
>>>>
>>>> Currently, on certain Xilinx platforms, an issue has been identified,
>>>> manifesting as follows:
>>
>> Hello Tejas,
>>
>> thank you for bringing forward this issue.
>>
>> Which defconfig are you relating to?
>
> [Tejas]:
> Checking with xilinx_zynqmp_virt_defconfig
>
>>
>>>>
>>>> Starting kernel ...
>>>>
>>>> efi_free_pool: illegal free 0x0000000077830040
>>>> efi_free_pool: illegal free 0x000000007782d040
>>>> efi_free_pool: illegal free 0x000000007782c040
>>>>
>>>> The issue arises when the ramdisk image is relocated, placing it
>>>> within the previously allocated EFI memory region( as EFI is
>>>> established quite early in U-Boot).
>>
>> Which version of U-Boot are you on? Commit 06d514d77c37 ("lmb: consider EFI
>> memory map") was meant to avoid such a situation.
>
> [Tejas] :
> I'm verifying against the latest changes in the master branch. The introduction of commit 06d514d77c37 ("lmb: consider EFI memory map") has resolved the occurrence of the "efi_free_pool: illegal free" error. but, it leads to a new error, as detailed in the following patch: https://lore.kernel.org/all/20230212150706.2967007-1-sjoerd@collabora.com/

Could you, please, try to reproduce your issues with origin/master as of
today and provide the full boot log. Please, also provide the output of
the bdinfo and the printenv command as well as the sizes of the kernel
and the RAM disk.

Best regards

Heinrich

>
>>
>>>
>>> I don't mind suppressing the print for some time, but out of
>>> curiosity, how is the ramdisk relocated? LMB should be aware of the
>>> EFI regions by then, so I assume the relocation code doesn't check
>>> those?
>
>
> [Tejas] :  I observe that the relocation of the RAM disk is taking place in the line below.
> https://github.com/u-boot/u-boot/blob/master/lib/lmb.c#L480-L491
> Yes, the relocated code specifically examines the LMB region and does not consider the EFI region.
>
>>
>> The indicated situation is serious. If the EFI sub-system is using the same memory
>> area as the ramdisk, this may lead to corruption of the ramdisk. Suppressing the
>> messages is by no means adequate to solve the problem.
>
> [Tejas] :
> The challenge arises when relocating the ramdisk image, inserting it into the previously assigned EFI memory region,
> established early in U-Boot. Consequently, when attempting to release memory in the EFI region during the handover
> process to the kernel in the efi_free_pool function, we first verify if the memory was allocated by efi_allocate_pool().
> The issue originates from a checksum mismatch because, during the ramdisk relocation, we overwrite memory allocated by EFI.
> This leads to the appearance of the error message: efi_free_pool: illegal free 0x0000000077830040.
> Crucially, there is no corruption of the ramdisk seen since we do not actually releasing memory due to the checksum mismatch.
> In our testing, this issue was observed only when the ramdisk size exceeded approximately 24 MB.
>
>>
>> Best regards
>>
>> Heinrich
>>
>>
>>>
>>> Thanks
>>> /Ilias
>>>
>>>> Consequently, when
>>>> attempting to release memory in the EFI memory region during the
>>>> handover process to the kernel,we encounter memory violations.
>>>>
>>>> Highlighting that EFI remains active primarily during the booting of
>>>> an EFI application, and the lmb persists while configuring images for
>>>> the boot process. Since we aren't utilizing the EFI memory region
>>>> during the boot process, there is no adverse impact even in the event
>>>> of a violation.
>>>>
>>>> Currently, there is an ongoing discussion regarding the handling
>>>> strategies of three memory allocators: malloc, lmb, and EFI. This
>>>> discussion is documented in the email chain titled "Proposal: U-Boot
>>>> memory management."
>>>>
>>>> Therefore, it is advisable to suppress the print message during the
>>>> boot process for now.
>>>>
>>>> Signed-off-by: Tejas Bhumkar <tejas.arvind.bhumkar at amd.com>
>>>> ---
>>>>    lib/efi_loader/efi_memory.c | 2 +-
>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/lib/efi_loader/efi_memory.c
>>>> b/lib/efi_loader/efi_memory.c index edfad2d95a..821fe7616e 100644
>>>> --- a/lib/efi_loader/efi_memory.c
>>>> +++ b/lib/efi_loader/efi_memory.c
>>>> @@ -713,7 +713,7 @@ efi_status_t efi_free_pool(void *buffer)
>>>>           /* Check that this memory was allocated by efi_allocate_pool() */
>>>>           if (((uintptr_t)alloc & EFI_PAGE_MASK) ||
>>>>               alloc->checksum != checksum(alloc)) {
>>>> -               printf("%s: illegal free 0x%p\n", __func__, buffer);
>>>> +               debug("%s: illegal free 0x%p\n", __func__, buffer);
>>>>                   return EFI_INVALID_PARAMETER;
>>>>           }
>>>>           /* Avoid double free */
>>>> --
>>>> 2.27.0
>>>>
>