Re: [PATCH 11/15] tty: msm_serial: use dmaengine_prep_slave_sg()

[Jiri Slaby <jirislaby@kernel.org>]

On 17. 04. 24, 14:45, Marek Szyprowski wrote:
> On 17.04.2024 12:50, Jiri Slaby wrote:
>> On 17. 04. 24, 12:15, Marek Szyprowski wrote:
>>> On 16.04.2024 12:23, Jiri Slaby wrote:
>>>> On 15. 04. 24, 23:17, Marek Szyprowski wrote:
>>>>> On 05.04.2024 08:08, Jiri Slaby (SUSE) wrote:
>>>>>> This is a preparatory for the serial-to-kfifo switch. kfifo
>>>>>> understands
>>>>>> only scatter-gatter approach, so switch to that.
>>>>>>
>>>>>> No functional change intended, it's just
>>>>>> dmaengine_prep_slave_single()
>>>>>> inline expanded.
>>>>>>
>>>>>> And in this case, switch from dma_map_single() to dma_map_sg() too.
>>>>>> This
>>>>>> needs struct msm_dma changes. I split the rx and tx parts into an
>>>>>> union.
>>>>>> TX is now struct scatterlist, RX remains the old good phys-virt-count
>>>>>> triple.
>>>>>>
>>>>>> Signed-off-by: Jiri Slaby (SUSE) <jirislaby@kernel.org>
>>>>>> Cc: Bjorn Andersson <andersson@kernel.org>
>>>>>> Cc: Konrad Dybcio <konrad.dybcio@linaro.org>
>>>>>> Cc: linux-arm-msm@vger.kernel.org
>>>>>
>>>>> I've just found that this patch broke UART operation on DragonBoard
>>>>> 410c. I briefly checked and didn't notice anything obviously wrong
>>>>> here,
>>>>> but the board stops transmitting any data from its serial port
>>>>> after the
>>>>> first message. I will try to analyze this issue a bit more tomorrow.
>>>>
>>>> I double checked, but I see no immediate issues in the patch too. So
>>>> please, if you can analyze this more…
>>>
>>> I've spent some time digging into this issue and frankly speaking I
>>> still have no idea WHY it doesn't work (or I seriously mixed something
>>> in the scatterlist principles). However I found a workaround to make it
>>> working. Maybe it will help a bit guessing what happens there.
>> ...
>>> @@ -434,7 +436,7 @@ static void msm_start_tx(struct uart_port *port)
>>>            struct msm_dma *dma = &msm_port->tx_dma;
>>>
>>>            /* Already started in DMA mode */
>>> -       if (sg_dma_len(&dma->tx_sg))
>>> +       if (dma->mapped)
>>
>> Thanks for looking into this.
>>
>> I was hesitant if I should use a flag. I should have, apparently.
>>
>> Quick question:
>> What's value of CONFIG_NEED_SG_DMA_LENGTH in your .config?
> 
> 
> CONFIG_NEED_SG_DMA_LENGTH=y
> 
> 
> I alse tried to change the "if (dma->mapped)" check in msm_start_tx() to:
> 
> 1. if (dma->tx_sg.length)
> 
> 2. if (dma->tx_sg.page_link & ~SG_PAGE_LINK_MASK)
> 
> but none of the above worked what is really strange and incomprehensible
> for me.

Thanks. Neither for me. Could you add:
         {
                 static DEFINE_RATELIMIT_STATE(rs, 
DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
                 if (dma->mapped != !!sg_dma_len(&dma->tx_sg) && 
__ratelimit(&rs))
                         printk_deferred(KERN_DEBUG "%s (%d): mapped=%u 
dma_len=%u\n",
                                         __func__, __LINE__, 
dma->mapped, sg_dma_len(&dma->tx_sg));
         }

before each of your 'if (dma->mapped)' to see where sg_dma_len() is 
wrong and what is its value in the bad case. I hope I did the logic right.

thanks,
-- 
js
suse labs