Hi Johan,
I just looked through the code, and slurp is only in detection. baud rate
change and error paths. Thus increasing the timeout is safe. From your
tests, it's also advisable.
You are right, we could keep owfs completely agnostic to the type of
serial/usb adapter. We do use libusb for the DS9490R, however, so we could
set USB parameters.
My thoughts are:
1. Keep the main code path ignorant of USB specifics. Just treat the device
path as a serial adapter whether hardware serial or USB-based. In fact, I
think figuring out the USB device from a serial port name would be
difficult. Scan /sys?
2. Add USB scanning to search for linkusb ( and hobbyboard masterhub soon )
as well as DS9490R as we do now. We'll want to do this in any case to make
configuration more seamless. In that case we'll know the type of USB
adapter and can send any tuning parameters at the start.
1-wire communication messages are pretty short, and often of variable
length. You test shows some nice speed increases with better timing
choices. I'd like to incorporate that eventually.
Paul
On Sun, Sep 21, 2014 at 9:13 AM, Johan Ström <jo...@stromnet.se> wrote:
> Hi!
>
> Another specific scenario where slurp is required, I guess, is in
> DS2480_set_baud, where we cannot rely on proper read-back, if the baud rate
> changed.
> So, as you say, slurp is probably required, and it's not worth trying to
> do proper reading, which is bound to fail anyway.
>
> Thus, my suggestion is to increase the slurp timeout so we can be sure
> that we actually any pending bytes.
> If slurp is only used during initial resets (which is the case for the
> DS2480 and the LINK code, I haven't studied any others), then it is no harm
> to increase this to say 100ms.
> HOWEVER, if any other device-code is using slurp during the normal
> communication flow, a longer timeout would of course hurt.
>
> The thing with the FTDI parameter is that you cannot really change that
> unless specific support for libftdi is added. Currently the DS2480 code
> only works with a generic serial device, with no information on the
> underlying hardware. So, hard to know when to apply ftdi changes, and even
> harder to identify which port to change..
>
> As discussed here previously, I've got some ftdi code in owfs working,
> although for the LinkUSB. Should be trivial to port to the DS2480-code, or
> probably most serial-based code. The question is if it makes sense to add
> chip-specific code, when this is only usable if you actually use a
> USB-serial-adapter built on that chip. In the LinkUSB case it makes more
> sense, since the actual "1-wire-product" LinkUSB is built around that chip.
>
> Anyway, by just raising the slurp timeout, we solve the main issue.
> Controlling the FTDI timeout would however give some speed increase, some
> simple tests shows that with a default of 16ms, the DS2480b reads
> /28.xxxxxxx/power in ~77ms, vs 27ms with 1ms timeout.
>
>
> Regarding 0-baud, I think it's only done from serial_powercycle, not sure
> why though. Power-cycle adapters powered directly by the serial-port?
> Since I don't know the use of the code, I cannot really say if this is an
> issue worth addressing. B0 is defined in termios.h so it should certainly
> be a valid value.
>
> Johan
>
>
> On 9/20/14 03:08 , Paul Alfille wrote:
>
> Thank you, Johan, for the detailed tests.
>
> The idea of "slurp" was to collect any pending serial traffic and allow
> communication with the bus master to be synchronized. I agree that matching
> expected response exactly is optimal, but that supposes that you start from
> a known clean state. I found in some of my early tests that old data was
> still delivered with the system was restarted.
>
> Obviously, testing is hardware-specific. The timing was adjusted based
> on my tests -- not very scientific.
>
> It sounds like you think we should adjust the slurp time-out, and
> perhaps some of the FTDI timing parameters. Also there is a 0-baud issue
> which we should look at fixing on our end as well. Is it the serial "BREAK"
> or baud changed that does it?
>
> Paul
>
> On Fri, Sep 19, 2014 at 2:57 AM, Johan Ström <jo...@stromnet.se> wrote:
>
>> Hi,
>>
>> I've noticed some problems using a FTDI based USB serial dongle together
>> with a DS2480B based adapter (also known as DS9097U).
>> On startup the device was not recognized at all, complaining about wrong
>> responses. Anyone else seen these issues?
>>
>> I've debugged the problem and found the cause.
>> On startup, this is what happens:
>>
>>
>> DEBUG: ow_ds9097U.c:(287) Attempt 0 of 3 to initialize the DS9097U
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: C1
>> <.>
>> DEBUG: ow_ds9097U.c:(381) Send the initial reset to the bus master.
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 71
>> <q>
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 0F
>> <.>
>> DEBUG: ow_tcp_read.c:(64) attempt 1 bytes Time: 5.000000 seconds
>> TRAFFIC IN <NETREAD> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 70
>> <p>
>> DEBUG: ow_tcp_read.c:(114) read: 1 - 0 = 1
>> DEBUG: ow_ds9097U.c:(459) wrong response (70 not 00)
>>
>>
>> After each TRAFFIC OUT byte above, a COM_Slurp() is issued to read and
>> remove the response.
>>
>> The response to the 71 command should be 70.. Which is what we see..
>> except that it fails to slurp those bytes, it's actually
>> reading them as the response to the next command.
>>
>> After some further debugging and added printf statements, I realize that
>> COM_slurp fails to read the bytes, and instead just timeouts (1ms).
>> I've been working on my pure-ftdi-branch for the LinkUSB, where I've
>> learned that the FTDI chips by default have a 16ms timeout for small
>> transfers, making them not very efficient when dealing with few bytes
>> like this.
>> Manually changing my USB-serial dongle's setting to 1ms with libftdi,
>> together with a slurp timeout of 2ms, seems to fix the problem. However,
>> the slurp still times out, so I guess the accompanying COM_flush does
>> the trick.. Running with default 16ms FTDI setting, and a slurp timeout
>> of 100ms, DOES succeed with reading the slurped data [1], and all works
>> fine.
>>
>> The "proper" way to solve this is probably to actually read the bytes we
>> expect, not just hope that slurp catches them. However, this might be
>> problematic when changing bitrates etc?
>> The solution used above would be to give the slurp a longer timeout.
>> Besides that device init would take a few ms longer, would this have any
>> other side effects?
>> >From the DS2480B code at least, it does not seem to slurp anywhere
>> except on device init.
>>
>> For the record, the DS2480B works perfectly fine with my STLab 2 port
>> dongle (MosChip mos78x0 chip).. except that the stable FreeBSD-10 kernel
>> panics when setting baud rate 0, which happens sometimes in OWFS when
>> trying to forcefully reset. This was patched in
>>
>> https://github.com/freebsd/freebsd/commit/120a212c4b16b5137d6acc436b8f5702a5f5bf35
>> but until the fix hits the mainline kernel, I'll have to go with another
>> dongle.
>>
>> Regards
>> Johan
>>
>>
>> [1]
>> This is how a proper session looks, with the responses slurped (FTDI
>> chip 16ms, COM_slurp timeout set to 100ms):
>>
>> DEBUG: ow_ds9097U.c:(287) Attempt 0 of 3 to initialize the DS9097U
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: C1
>> <.>
>> TRAFFIC IN <slurp> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: CD
>> <.>
>> Slurp timeout 100000 us..
>> DEBUG: ow_ds9097U.c:(381) Send the initial reset to the bus master.
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 71
>> <q>
>> TRAFFIC IN <slurp> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 70
>> <p>
>> Slurp timeout 100000 us..
>> TRAFFIC OUT <write> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 0F
>> <.>
>> DEBUG: ow_tcp_read.c:(64) attempt 1 bytes Time: 5.000000 seconds
>> TRAFFIC IN <NETREAD> bus=0 (/dev/cua-labdesk)
>> Byte buffer anonymous, length=1
>> --000: 00
>> <.>
>> DEBUG: ow_tcp_read.c:(114) read: 1 - 0 = 1
>> DEBUG: ow_com_read.c:(83) COM_read, read 1 bytes.
>> read_get=16302.000000us, tcdrain=0.000000us: 1
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Slashdot TV. Video for Nerds. Stuff that Matters.
>>
>> http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
>> _______________________________________________
>> Owfs-developers mailing list
>> Owfs-developers@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/owfs-developers
>>
>
>
>
> ------------------------------------------------------------------------------
> Slashdot TV. Video for Nerds. Stuff that
> Matters.http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
>
>
>
> _______________________________________________
> Owfs-developers mailing
> listOwfs-developers@lists.sourceforge.nethttps://lists.sourceforge.net/lists/listinfo/owfs-developers
>
>
>
>
> ------------------------------------------------------------------------------
> Slashdot TV. Video for Nerds. Stuff that Matters.
>
> http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
> _______________________________________________
> Owfs-developers mailing list
> Owfs-developers@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/owfs-developers
>
>
------------------------------------------------------------------------------
Slashdot TV. Video for Nerds. Stuff that Matters.
http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
_______________________________________________
Owfs-developers mailing list
Owfs-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/owfs-developers
