Re: Regression: ftdi_sio is slow (since Wed Oct 10 15:05:06 2012)

[Peter Hurley <peter@hurleysoftware.com>]

On 05/04/2013 07:15 AM, Johan Hovold wrote:
> On Sat, May 04, 2013 at 01:50:42AM +0400, Stas Sergeev wrote:
>> 04.05.2013 00:34, Greg KH пишет:
>>> On Fri, May 03, 2013 at 10:27:18PM +0400, Stas Sergeev wrote:
>>>> 03.05.2013 21:16, Greg KH пишет:
>
> [...]
>
>>>>> There's no guarantee as to how long select or an ioctl will take, and
>>>>> now that we have fixed another bug, this device is slower.
>>>>>
>>>>> If you change hardware types to use a different usb to serial chip, that
>>>>> select call might take 4 times as long.  Are we somehow supposed to
>>>>> change the kernel to "fix" that?
>>>> Previously, the kernel was not calling to a device at all, so
>>>> select() was independent of the chip, and it was fast. I was
>>>> not aware you changed that willingly.
>>> I don't understand, what do you mean by this?  Some drivers just return
>>> the value of an internally held number, and don't query the device.
>>>
>>> The only way the FTDI driver can determine if the hardware buffer on the
>>> chip way out on the end of the USB cable is empty or not, is to query
>>> it.  So the driver now does so.
>> It does so only for one char. And the query takes longer than
>> to just xmit that char. So why do you think this even works as
>> expected?
>
> The query takes longer than the transmit at decent baudrates (>=38k)
> and under the assumption that flow control isn't causing any delays.
>
> But you do have a point, and I have been meaning to look into whether
> the added overhead of checking the hardware buffers could be mitigated
> by adding wait_until_sent support to usb-serial. This way the we would
> only query the hardware buffers on tty_wait_until_sent (e.g. at close)
> and select and TIOCMOUTQ would not suffer. This is also the way things
> are handled in serial_core.

Agreed. This is the correct solution.

> I'll prepare a series which adds wait_until_sent to usb-serial, but I
> doubt it would be stable material (even if it could get into 3.10).
>
> What do you think Greg, is this overhead to chars_in_buffer reason
> enough to disable it in the stable trees or should we simply fix it in
> 3.11 (or 3.10)? (The overhead is about 3-400 us per call when the port
> fifo is empty, which makes chars_in_buffer about 100 times slower on my
> test system.)

A better solution for stable would be to set port->drain_delay. It
won't help tcdrain() but at least the port won't shutdown on live
outbound data.

Regards,
Peter Hurley