Re: [v5 11/15] arm64/kasan: explicitly zero kasan shadow memory

From: Pasha Tatashin
Date: Fri Aug 04 2017 - 10:02:27 EST

Hi Ard,

Thank you very much for reviewing this. I will fix the bug you found in the next iteration.

Typo here: vemmap -> vmemmap

Yeap, will rename here, and in Intel variant.

+ struct memblock_region *reg;
+ u64 start, end;
+ for_each_memblock(memory, reg) {
+ start = __phys_to_virt(reg->base);
+ end = __phys_to_virt(reg->base + reg->size);
+ if (start >= end)
How would this ever be true? And why is it a stop condition?

Yes this is a stop condition. Also look at the way kasan allocates its shadow memory in this file kasan_init():

187 for_each_memblock(memory, reg) {
188 void *start = (void *)__phys_to_virt(reg->base);
189 void *end = (void *)__phys_to_virt(reg->base + reg->size);
191 if (start >= end)
192 break;
200 vmemmap_populate(...)


Are you missing a couple of kasan_mem_to_shadow() calls here? I can't
believe your intention is to wipe all of DRAM.

True. Thank you for catching this bug. I have not really tested on arm, only compiled for sanity checking. Need to figure out how to configure qemu to run most generic arm code. I tested on x86 and sparc both real and qemu hardware.

KASAN uses vmemmap_populate as a convenience: kasan has nothing to do
with vmemmap, but the function already existed and happened to do what
KASAN requires.

Given that that will no longer be the case, it would be far better to
stop using vmemmap_populate altogether, and clone it into a KASAN
specific version (with an appropriate name) with the zeroing folded
into it.

I agree, but this would be outside of the scope of this project.