Re: [Qemu-devel] [PATCH v2 for-2.4 12/12] axienet: Flush queued packets when rx is done

[Jason Wang]

On 07/16/2015 11:32 AM, Fam Zheng wrote:
> On Thu, 07/16 10:58, Jason Wang wrote:
>>
>> On 07/15/2015 06:19 PM, Fam Zheng wrote:
>>> eth_can_rx checks s->rxsize and returns false if it is non-zero. Because
>>> of the .can_receive semantics change, this will make the incoming queue
>>> disabled by peer, until it is explicitly flushed. So we should flush it
>>> when s->rxsize is becoming zero.
>>>
>>> Squash eth_can_rx semantics into etx_rx and drop .can_receive()
>>> callback, also add flush when rx buffer becomes available again after a
>>> packet gets queued.
>>>
>>> The other conditions, "!axienet_rx_resetting(s) &&
>>> axienet_rx_enabled(s)" are OK because enet_write already calls
>>> qemu_flush_queued_packets when the register bits are changed.
>>>
>>> Signed-off-by: Fam Zheng <address@hidden>
>>> ---
>>>  hw/net/xilinx_axienet.c | 17 +++++++++++++----
>>>  1 file changed, 13 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/hw/net/xilinx_axienet.c b/hw/net/xilinx_axienet.c
>>> index 9205770..d63c423 100644
>>> --- a/hw/net/xilinx_axienet.c
>>> +++ b/hw/net/xilinx_axienet.c
>>> @@ -401,6 +401,9 @@ struct XilinxAXIEnet {
>>>  
>>>      uint8_t rxapp[CONTROL_PAYLOAD_SIZE];
>>>      uint32_t rxappsize;
>>> +
>>> +    /* Whether axienet_eth_rx_notify should flush incoming queue. */
>>> +    bool need_flush;
>>>  };
>>>  
>>>  static void axienet_rx_reset(XilinxAXIEnet *s)
>>> @@ -658,10 +661,8 @@ static const MemoryRegionOps enet_ops = {
>>>      .endianness = DEVICE_LITTLE_ENDIAN,
>>>  };
>>>  
>>> -static int eth_can_rx(NetClientState *nc)
>>> +static int eth_can_rx(XilinxAXIEnet *s)
>>>  {
>>> -    XilinxAXIEnet *s = qemu_get_nic_opaque(nc);
>>> -
>>>      /* RX enabled?  */
>>>      return !s->rxsize && !axienet_rx_resetting(s) && axienet_rx_enabled(s);
>>>  }
>>> @@ -701,6 +702,10 @@ static void axienet_eth_rx_notify(void *opaque)
>>>          s->rxpos += ret;
>>>          if (!s->rxsize) {
>>>              s->regs[R_IS] |= IS_RX_COMPLETE;
>>> +            if (s->need_flush) {
>>> +                s->need_flush = false;
>>> +                qemu_flush_queued_packets(qemu_get_queue(s->nic));
>>> +            }
>>>          }
>>>      }
>>>      enet_update_irq(s);
>>> @@ -721,6 +726,11 @@ static ssize_t eth_rx(NetClientState *nc, const 
>>> uint8_t *buf, size_t size)
>>>  
>>>      DENET(qemu_log("%s: %zd bytes\n", __func__, size));
>>>  
>>> +    if (!eth_can_rx(s)) {
>>> +        s->need_flush = true;
>>> +        return 0;
>>> +    }
>>> +
>> axienet_eth_rx_notify() was only called by eth_rx(). So when
>> s->need_flush is true, we won't ever reach axienet_eth_rx_notify()?
> We will.
>
> If we are here it measn a previous call to axienet_eth_rx_notify hasn't 
> drained
> the buffer:
>
>     static void axienet_eth_rx_notify(void *opaque)
>     {
>         ...
>
>         while (s->rxsize && stream_can_push(s->tx_data_dev,
>                                             axienet_eth_rx_notify, s)) {
>             size_t ret = stream_push(s->tx_data_dev, (void *)s->rxmem + 
> s->rxpos,
>                                      s->rxsize);
>             s->rxsize -= ret;
>             s->rxpos += ret;
>             if (!s->rxsize) {
>                 s->regs[R_IS] |= IS_RX_COMPLETE;
>             }
>         }
>         ...
>
>     }
>
> axienet_eth_rx_notify is passed to stream_can_push so it will be reached again
> once s->tx_data_dev can receive more data:
>
>     typedef struct StreamSlaveClass {
>         InterfaceClass parent;
>         /**
>          * can push - determine if a stream slave is capable of accepting at 
> least
>          * one byte of data. Returns false if cannot accept. If not 
> implemented, the
>          * slave is assumed to always be capable of receiving.
>          * @notify: Optional callback that the slave will call when the slave 
> is
>          * capable of receiving again. Only called if false is returned.
>          * @notify_opaque: opaque data to pass to notify call.
>          */
>         bool (*can_push)(StreamSlave *obj, StreamCanPushNotifyFn notify,
>                          void *notify_opaque);
>         ...
>
> Am I missing anything?
>
> Fam

Probably not. I misses the possible call in axidma_write(). So the patch
is ok :)