Re: [PATCH] softmmu/memory: use memcpy for multi-byte accesses

[Peter Maydell]

On Wed, 15 Nov 2023 at 17:26, Patrick Venture <venture@google.com> wrote:
>
>
>
> On Wed, Nov 15, 2023 at 9:02 AM Richard Henderson 
> <richard.henderson@linaro.org> wrote:
>>
>> On 11/15/23 08:58, Patrick Venture wrote:
>> >
>> >
>> > On Wed, Nov 15, 2023 at 2:35 AM Peter Maydell <peter.maydell@linaro.org
>> > <mailto:peter.maydell@linaro.org>> wrote:
>> >
>> >     On Tue, 14 Nov 2023 at 20:55, Patrick Venture <venture@google.com
>> >     <mailto:venture@google.com>> wrote:
>> >      > Avoids unaligned pointer issues.
>> >      >
>> >
>> >     It would be nice to be more specific in the commit message here, by
>> >     describing what kind of guest behaviour or machine config runs into 
>> > this
>> >     problem, and whether this happens in a situation users are likely to
>> >     run into. If the latter, we should consider tagging the commit
>> >     with "Cc: qemu-stable@nongnu.org <mailto:qemu-stable@nongnu.org>" to 
>> > have it
>> >     backported to the
>> >     stable release branches.
>> >
>> >
>> > Thanks! I'll update the commit message with v2.  We were seeing this in our
>> > infrastructure with unaligned accesses using the pointer dereference as 
>> > there are no
>> > guarantees on alignment of the incoming values.
>>
>> Which host cpu, for reference?  There aren't many that generate unaligned 
>> traps these days...
>>
>
> Here's the sanitizer log/qemu log, the host-cpu was an amd64.
> third_party/qemu/softmmu/memory.c:1341:16: runtime error: load of misaligned 
> address 0x52500020b10d for type 'uint32_t' (aka 'unsigned int'), which 
> requires 4 byte alignment
> 0x52500020b10d: note: pointer points here
>  ab ab ab ab ab ab ab  ab ab ab ab ab ab ab ab  ab ab ab ab ab ab ab ab  ab 
> ab ab ab ab ab ab ab  ab
>              ^

> SUMMARY: UndefinedBehaviorSanitizer: misaligned-pointer-use 
> third_party/qemu/softmmu/memory.c:1341:16 in

Ah, right, so the clang/gcc undefined-behaviour sanitizers rather than
the actual host hardware barfing. (We definitely want to fix these
regardless.)

thanks
-- PMM