Re: [PATCH 1/5] serial: imx: remove unneeded imx_transmit_buffer() from imx_start_tx()

[Dean Jenkins <Dean_Jenkins@mentor.com>]

On 12/05/14 07:53, Huang Shijie wrote:
> 于 2014年05月12日 14:40, Dirk Behme 写道:
>> On 12.05.2014 08:30, Huang Shijie wrote:
>>> 于 2014年05月12日 13:45, Dirk Behme 写道:
>>>> On 12.05.2014 05:40, Huang Shijie wrote:
>>>>> 于 2014年05月09日 23:19, dean_jenkins@mentor.com 写道:
>>>>>> Use imx_start_tx() just to enable the TX interrupt. It's the job 
>>>>>> of the
>>>>>> TX interrupt ISR to fill the transmit buffer, then. If the transmit
>>>>>> buffer
>>>   From the Documentation/serial/driver, we can see:
>>>     -----------------------------------------
>>>      start_tx(port)
>>>       Start transmitting characters.
>>>      -----------------------------------------
>>>
>>> It tells us we can transmit data in the imx_start_tx.
>>
>> Well, it depends how you read 'Start transmitting', no?
>>
>> It doesn't have to mean 'actually transmit data'. It talks about 
>> 'start'. And this could also mean 'start the transmission (by 
>> enabling the interrupt)'.
>>
> Ok :)
> I do not object this patch.
>
> My opinion is : If the third patch is redundant, could this patch 
> still needed?
>
>
>>> But this patch moves it to the interrupt handler,
>>
>> It doesn't 'move' any code to the interrupt handler. The code in the 
>> interrupt handler is already there.
>>
> i knew it.
>
>>> this patch makes the
>>> interrupt handler do more jobs.
>>
>> ... to get the locking in the correct order.
>>
> again, the third patch can be ignored.
> The locking is correct for the imx.c.
>
Well, the locking might now be OK but due to an existing issue the TX 
will be woken up excessively with no new characters to transmit so there 
will be excessive context switching with the new work queue that results 
in no useful work done. There is an API issue in HCI LDISC because 
TTY_DO_WRITE_WAKEUP is set BEFORE checking whether any characters remain 
pending to be sent. This means most of the time hci_uart_tx_wakeup() is 
unnecessarily called will no chance of seeing any remain pending 
characters to be process. This weakness contributed to triggering lockup 
issue.

In my opinion, TTY_DO_WRITE_WAKEUP should only be set when the UART 
character circular holding buffer becomes full which for BCSP traffic is 
unlikely I think (4kB buffer versus traffic < 1kB BCSP frames in 
length). The current API needs fixing so that TTY_DO_WRITE_WAKEUP is set 
after filling up the holding buffer but BEFORE enabling the TX 
interrupts. I think TTY_DO_WRITE_WAKEUP needs to be set in a callback 
function called from the bound layer below HCI LDISC.

Best regards,
Dean

--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html