On 06.09.2024 20:06, Andrew Cooper wrote:
> On 04/09/2024 2:30 pm, Jan Beulich wrote:
>> @@ -1094,13 +1094,13 @@ static int hvmemul_linear_mmio_access(
>> if ( cache == NULL )
>> return X86EMUL_UNHANDLEABLE;
>>
>> - chunk = min_t(unsigned int, size, PAGE_SIZE - offset);
>> + ASSERT(size <= PAGE_SIZE - offset);
>
> Do we really want a plain assert, or should we go with
>
> if ( size > PAGE_SIZE - offset )
> {
> /* Callers should have arranged not to cross a page boundary */
> ASSERT_UNREACHABLE();
> return X86EMUL_UNHANDLEABLE;
> }
>
> This is hardly a fastpath, and it's rather safer.
I can switch, sure, yet to be honest it was already feeling a little
like going too far to have the assertion, considering the obviousness
of all callers guaranteeing this. The only reason I decided to add
one is the remaining concern of there, at some point, possibly being
single memory operands exceeding PAGE_SIZE. Yet nothing comes
anywhere near that right now; whole AMX tiles are 1k "only", and tile
rows / columns are even further restricted. Of course, if and when we
add XSAVE/XRSTORE emulation ...
Jan
