Re: [PATCH v4 37/45] linux-user/aarch64: Do not allow duplicate or short sve records

[Richard Henderson <richard.henderson@linaro.org>]

On 7/5/22 09:00, Richard Henderson wrote:
> On 7/4/22 17:38, Peter Maydell wrote:
>>>           case TARGET_SVE_MAGIC:
>>> +            if (sve || size < sizeof(struct target_sve_context)) {
>>> +                goto err;
>>> +            }
>>>               if (cpu_isar_feature(aa64_sve, env_archcpu(env))) {
>>>                   vq = sve_vq(env);
>>>                   sve_size = QEMU_ALIGN_UP(TARGET_SVE_SIG_CONTEXT_SIZE(vq), 16);
>>> -                if (!sve && size == sve_size) {
>>> +                if (size == sve_size) {
>>>                       sve = (struct target_sve_context *)ctx;
>>>                       break;
>>>                   }
>>
>> On the other hand, the kernel seems to happily allow records
>> which are larger than the SVE_SIG_CONTEXT_SIZE, whereas we
>> ignore the record unless there's an exact size match.
> 
> Yeah, this gets fixed properly in patch 39.
> Perhaps I should simply squash this with that?

Bah!  No, those are two separate checks: the minimum size to contain vq and flags 
(target_sve_context) and the minimum size to contain all of the vector data 
(TARGET_SVE_SIG_CONTEXT_SIZE).

The latter *is* fixed in patch 39, but this one should stay as-is.


r~