Re: [PATCH 7/8] hw/9pfs: annotate V9fsTransport callbacks as coroutine_fn

On 21/7/26 10:06, Marc-André Lureau wrote:
> Hi
> 
> On Tue, Jul 21, 2026 at 11:58 AM Marc-André Lureau
> <[email protected]> wrote:
>>
>> Hi
>>
>> On Tue, Jul 21, 2026 at 12:54 AM Philippe Mathieu-Daudé
>> <[email protected]> wrote:
>>>
>>> On 20/7/26 22:42, Philippe Mathieu-Daudé wrote:
>>>> On 20/7/26 16:02, Christian Schoenebeck wrote:
>>>>> On Monday, 20 July 2026 09:55:36 CEST Marc-André Lureau wrote:
>>>>>> All V9fsTransport callbacks are invoked exclusively from coroutine
>>>>>> context (the v9fs_* PDU handlers). Annotate the function pointer
>>>>>> types in V9fsTransport and all implementations (virtio and xen
>>>>>> backends), as well as intermediate callers in 9p.c (pdu_marshal,
>>>>>> pdu_unmarshal, v9fs_init_qiov_from_pdu, etc.).
>>>>>>
>>>>>> Signed-off-by: Marc-André Lureau <[email protected]>
>>>>>> ---
>>>>>>    hw/9pfs/9p.h               | 26 +++++++++++++++-----------
>>>>>>    hw/9pfs/9p.c               | 31 +++++++++++++++++++------------
>>>>>>    hw/9pfs/virtio-9p-device.c | 24 +++++++++++++-----------
>>>>>>    hw/9pfs/xen-9p-backend.c   | 34 +++++++++++++++++-----------------
>>>>>>    4 files changed, 64 insertions(+), 51 deletions(-)
>>>>>>
>>>>>> diff --git a/hw/9pfs/9p.h b/hw/9pfs/9p.h
>>>>>> index 1a309664f6e..0e52ffbdf38 100644
>>>>>> --- a/hw/9pfs/9p.h
>>>>>> +++ b/hw/9pfs/9p.h
>>>>>> @@ -472,17 +472,21 @@ void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr);
>>>>>>    void v9fs_reset(V9fsState *s);
>>>>>>
>>>>>>
>>>>>>    struct V9fsTransport {
>>>>>> -    ssize_t     (*pdu_vmarshal)(V9fsPDU *pdu, size_t offset, const
>>>>>> char *fmt,
>>>>>> -                                va_list ap);
>>>>>> -    ssize_t     (*pdu_vunmarshal)(V9fsPDU *pdu, size_t offset, const
>>>>>> char *fmt,
>>>>>> -                                  va_list ap);
>>>>>> -    void        (*init_in_iov_from_pdu)(V9fsPDU *pdu, struct iovec
>>>>>> **piov,
>>>>>> -                                        unsigned int *pniov, size_t
>>>>>> size);
>>>>>> -    void        (*init_out_iov_from_pdu)(V9fsPDU *pdu, struct iovec
>>>>>> **piov,
>>>>>> -                                         unsigned int *pniov, size_t
>>>>>> size);
>>>>>> -    void        (*push_and_notify)(V9fsPDU *pdu);
>>>>>> -    size_t      (*msize_limit)(V9fsState *s);
>>>>>> -    size_t      (*response_buffer_size)(V9fsPDU *pdu);
>>>>>> +    ssize_t     coroutine_fn (*pdu_vmarshal)(V9fsPDU *pdu, size_t
>>>>>> offset,
>>>>>> +                                             const char *fmt,
>>>>>> va_list ap);
>>>>>> +    ssize_t     coroutine_fn (*pdu_vunmarshal)(V9fsPDU *pdu, size_t
>>>>>> offset,
>>>>>> +                                               const char *fmt,
>>>>>> va_list ap);
>>>>>> +    void        coroutine_fn (*init_in_iov_from_pdu)(V9fsPDU *pdu,
>>>>>> +                                                     struct iovec
>>>>>> **piov,
>>>>>> +                                                     unsigned int
>>>>>> *pniov,
>>>>>> +                                                     size_t size);
>>>>>> +    void        coroutine_fn (*init_out_iov_from_pdu)(V9fsPDU *pdu,
>>>>>> +                                                      struct iovec
>>>>>> **piov,
>>>>>> +                                                      unsigned int
>>>>>> *pniov,
>>>>>> +                                                      size_t size);
>>>>>> +    void        coroutine_fn (*push_and_notify)(V9fsPDU *pdu);
>>>>>> +    size_t      coroutine_fn (*msize_limit)(V9fsState *s);
>>>>>> +    size_t      coroutine_fn (*response_buffer_size)(V9fsPDU *pdu);
>>>>>>    };
>>>>>
>>>>> I wonder whether that could be made more diff/blame-friendly.
>>>>> Especially as
>>>>> this currently only serves documentation purposes.
>>>>>
>>>>> They way you are positioning coroutine_fn it usually resolves to an
>>>>> attribute
>>>>> attached to the function's return type, not to the function itself, even
>>>>> though clang/gcc handling it flexible. For attaching an attribute to the
>>>>> function itself the attribute should either be attached before the
>>>>> function
>>>>> declaration or after the function declaration.
>>>>
>>>> I tend to aggree it would be clearer to have these attributes /before/
>>>> the function prototype declaration.
>>>
>>> $ git grep -E '^(static )?coroutine_(mixed_)?fn' | wc -l
>>>         90
>>
>> More like
>> $  git grep -rnE '\b(coroutine_fn|co_wrapper[a-z_]*)\b' | wc -l
>> 1343
>>
> 
> That matching is not accurate, it's actually hard enough to match all
> annotations instances.. but this should be a rough estimation.
> 
>> most of them are "return-type co_annotation fn(..)"
>> we even have scripts/block-coroutine-wrapper.py that relies and
>> produce this order..
>> a bit late to change the style imho...
>>
>> At least, let's not make this a requirement for this series, please

Not a requirement on my side, but if we had to pick a style,
I'd vote for "attribute before return type".

>>
>>>> block/block-copy.c:47:static coroutine_fn int
>>> block_copy_task_entry(AioTask *task);
>>> block/copy-before-write.c:106:static coroutine_fn int
>>> cbw_do_copy_before_write(BlockDriverState *bs,
>>> block/nvme.c:1351:static coroutine_fn int nvme_co_flush(BlockDriverState
>>> *bs)
>>> block/ssh.c:1223:static coroutine_fn int ssh_flush(BDRVSSHState *s,
>>> BlockDriverState *bs)
>>> nbd/server.c:3077:static coroutine_fn void nbd_trip(void *opaque)
>>> block/block-copy.c:581:static coroutine_fn int
>>> block_copy_task_entry(AioTask *task)
>>> ...
>>>
>>>>
>>>>> In both cases you could then
>>>>> add it to its own, separate line, avoiding a lot of noise.
>>>>>
>>>>> /Christian
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
> 
>