Hello Michael:

Thanks for your review and reply, I sent the mail to qemud-devel about a
month ago but didn't receive any reply. Therefore, I send the mail to
qemu-trivial again and forget CC to qemu-devel. I will be careful next
time.

The consequence is that if the guest operating system want to check the
actual transfer length of a transfer, it may be regarded as a failed
transfer since the actual transfer length is zero originally.

Here is an example from FreeBSD:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=288968.
I have workarounded this issue by ignore any error from hid_write.

Interestingly, Parallel seems to share the same issue and can be
addressed with the same fix.

Michael Tokarev <[email protected]> writes:

> Please always send all patches to qemu-devel@ (cc'd)
> (in this case, in addition to qemu-trivial@).
>
> On 15.09.2025 20:57, ShengYi Hung wrote:
>> The return value of hid_keyboard_write is used to set the packet's 
>> actual_length
>> and pass to xhci directly to allow guest know how many byte actually 
>> processed.
>> Therefore, return 1 to indicate a successful transfer or it will be
>> considered as a wrong xfer.
>> Signed-off-by: ShengYi Hung <[email protected]>
>> ---
>>   hw/input/hid.c | 1 +
>>   1 file changed, 1 insertion(+)
>> diff --git a/hw/input/hid.c b/hw/input/hid.c
>> index 76bedc1844..de24cd0ef0 100644
>> --- a/hw/input/hid.c
>> +++ b/hw/input/hid.c
>> @@ -478,6 +478,7 @@ int hid_keyboard_write(HIDState *hs, uint8_t *buf, int 
>> len)
>>               ledstate |= QEMU_CAPS_LOCK_LED;
>>           }
>>           kbd_put_ledstate(ledstate);
>> +        return 1;
>>       }
>>       return 0;
>>   }
>
> Reviewed-by: Michael Tokarev <[email protected]>
>
> and queued up to the trivial-patches tree.
>
> It's an interesting one, I wonder what the consequences are -
> for having this at 0 all the time, and for actually making it 1.
> Did we miss a byte somewhere with current code?
>
> Thanks,
>
> /mjt

-- 
Best Regards.
ShengYi Hung.

Reply via email to