Re: [PATCH v6] xen/console: introduce domain_console struct

[Jan Beulich <jbeulich@xxxxxxxx>]

On 11.07.2025 09:25, dmkhn@xxxxxxxxx wrote:
> On Fri, Jul 11, 2025 at 07:59:27AM +0200, Jan Beulich wrote:
>> On 11.07.2025 03:07, dmkhn@xxxxxxxxx wrote:
>>> On Thu, Jul 10, 2025 at 01:16:24PM +0200, Jan Beulich wrote:
>>>> On 10.07.2025 03:35, dmkhn@xxxxxxxxx wrote:
>>>>> @@ -877,6 +873,16 @@ struct domain *domain_create(domid_t domid,
>>>>>
>>>>>      /* All error paths can depend on the above setup. */
>>>>>
>>>>> +    BUILD_BUG_ON(DOMAIN_CONSOLE_BUF_SIZE <= 0);
>>>>
>>>> While the "equals 0" case can in principle happen, the "less than" part
>>>> is dead code (and hence this needs checking differently): The type of
>>>> DOMAIN_CONSOLE_BUF_SIZE is an unsigned one, so wrapping through 0 will
>>>> yield huge positive values.
>>>>
>>>>> +    err = -ENOMEM;
>>>>> +    d->console = xzalloc_bytes(DOMAIN_CONSOLE_SIZE);
>>>>
>>>> As previously indicated, new code ought to use the xv*alloc family of
>>>> functions, which deliberately doesn't include any ..._bytes() forms.
>>>> Note how instead there is xvzalloc_flex_struct() for situations like
>>>> the one here.
>>>
>>> Looks like xvzalloc_flex_struct() is not used anywhere in the code base...
>>
>> And what do you mean to tell me by that? xvmalloc_flex_struct() is used,
>> and x[mz]alloc_flex_struct() uses are still in need of conversion (which
>> is going to be a slow-going process).
> 
> Just an observation; I was not aware of these APIs and I was not aware that
> there's some slow-going process of switching to preferred use of
> xvzalloc_flex_struct().

The ongoing conversion is (more generally) from x[mz]alloc family of functions
to the xv[mz]alloc one, as is made explicit by the comment near the top of
xvmalloc.h.

Jan