Re: [edk2-devel] AArch64 with HeapGuard: page allocations wrongly aligned

["Oliver Smith-Denny" <osde@linux.microsoft.com>]

On 1/22/2024 2:06 PM, Rebecca Cran via groups.io wrote:
> On 1/19/2024 1:03 PM, Oliver Smith-Denny wrote:
> 
>> Thanks for trying. In lieu of being able to test myself, all I can offer
>> is adding some more prints, when the memory gets allocated, making sure
>> it is 64k aligned then. I'd be curious to see what the address is that
>> is attempting to be freed.
>>
>> My guess (as it was earlier) is that it is going to be aligned to
>> 64k but + 4k. I.e the guard page at the front is throwing it off. There
>> may have just been an error in my attempt to fix the check for that.
>>
>> If however that address is not 64k + 4k aligned, then something else is
>> afoot.
>>
>> Happy to look at some more data if you get it or can engineer an example
>> on an open source system (can you force the system to call this function
>> twice even without the extra SMBIOS entries, etc.).
> 
> These are the addresses it's allocating with and without HeapGuard (with 
> the original, upstream Page.c code).
> 
> 
> Without HeapGuard:
> 
> SmbiosCreate64BitTable() re-allocate SMBIOS 64-bit table
> Allocated 0x00000000EF110000 with gBS->AllocatePages (AllocateAnyPages, 
> EfiRuntimeServicesData, EFI_SIZE_TO_PAGES (55), 0x00000000FB7C0000)
> ...
> SmbiosCreate64BitTable() re-allocate SMBIOS 64-bit table
> SmbiosCreate64BitTable: calling FreePages (0x00000000EF110000, 1)
> Allocated 0x00000000EF110000 with gBS->AllocatePages (AllocateAnyPages, 
> EfiRuntimeServicesData, EFI_SIZE_TO_PAGES (4153), 0x00000000FB8AEC8E)
> 
> ------------------------------------------------------
> 
> WITH HeapGuard:
> 
> SmbiosCreate64BitTable() re-allocate SMBIOS 64-bit table
> Allocated 0x00000000ED36F000 with gBS->AllocatePages (AllocateAnyPages, 
> EfiRuntimeServicesData, EFI_SIZE_TO_PAGES (55), 0x00000000ED38F000)
> ...
> SmbiosCreate64BitTable() re-allocate SMBIOS 64-bit table
> SmbiosCreate64BitTable: calling FreePages (0x00000000ED36F000, 1)
> <ASSERT>
> 

I was able to repro your bug (by just turning on page guards on
ArmVirtQemu, allocating runtime mem and freeing it). I think you
are the first person to free runtime mem on ARM64 with page guards
enabled (and to care when it failed :).

The heap guard code is not written with ARM64 in mind (nor is much of
the codebase, of course). Specifically in this case the heap guard code
only wishes to preserve 4 KB alignment, it knows nothing of ARM64's
runtime page granularity required.

Let me take a look at this, I'm working on a solution here, but I want
to test this out further. I'll try to send a patch later this week or
next.

Oliver


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#114162): https://edk2.groups.io/g/devel/message/114162
Mute This Topic: https://groups.io/mt/103810212/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-