serial performance

serial performance

[Trifle Menot]

If I start dosemu -s with this:

> $_ports = "fast range 0x3F8,0x3FF"
> $_irqpassing = "4"

A speed of 57600 looks fast, like it should be.  But if I omit the
direct hardware access, and try this:

> $_com1 = "/dev/ttyS0 irq 4"

57600 still works, but looks much slower on the remote side of a null
modem cable.  Seems like I read where the code traps serial port writes
and buffers data to /dev/ttyS0.  Is the trapping causing the slowdown?
Is there a cure?

I don't want to use direct hardware access.  I tested it that way, only
to see how things work.

I'm using the latest dosemu svn code.  I first tried a 2.6.32 kernel on
arch linux, but got a kernel panic, then tried a suse 2.6.11 kernel, and
got these results.  Both were i386 only, never x86_64.



-- 
Web mail, POP3, and SMTP
http://www.beewyz.com/freeaccounts.php
 

Re: serial performance

[Trifle Menot]

On Fri, 12 Mar 2010 15:57:44 +0000, Trifle Menot
<triflemenot@beewyz.com> wrote:

>If I start dosemu -s with this:
>
>> $_ports = "fast range 0x3F8,0x3FF"
>> $_irqpassing = "4"
>
>A speed of 57600 looks fast, like it should be.  But if I omit the
>direct hardware access, and try this:
>
>> $_com1 = "/dev/ttyS0 irq 4"
>
>57600 still works, but looks much slower on the remote side of a null
>modem cable.  Seems like I read where the code traps serial port writes
>and buffers data to /dev/ttyS0.  Is the trapping causing the slowdown?
>Is there a cure?

Digging deeper, I noticed that my PCBoard BBS recognized a 16550 when
using direct hardware access, but only a 16450 with /dev/ttyS0.

I looked at the dosemu source to try and understand why.  Being new to
the code, I'm not 100% sure of what I'm saying here, but it looks like
FIFO mode is turned on according to whatever bits are poked into the
virtual UART.  So then I looked closer my PCBoard configuration, and I
noticed a setting to "Force 16550A usage" and I turned it on.

With that setting, PCBoard now sees a 16550 when using /dev/ttyS0, and
my remote output looks fast, like 57600 should.  So poking the necessary
bits into the dosemu virtual UART fixed things up.


-- 
Web mail, POP3, and SMTP
http://www.beewyz.com/freeaccounts.php
 

Re: serial performance

[Paul Crawford]

Trifle Menot wrote:
> With that setting, PCBoard now sees a 16550 when using /dev/ttyS0, and
> my remote output looks fast, like 57600 should.  So poking the necessary
> bits into the dosemu virtual UART fixed things up.

Can you elaborate on which bits you poked to achieve this?

We have also had some performance issues with the virtual UART and 
anything that makes it more like real-time would be appreciated.

Regards,
Paul
-- 
  Dr. Paul S. Crawford
  c/o Satellite Station
  University of Dundee
  Small's Wynd, Dundee, DD1 4HN
  Email: psc@sat.dundee.ac.uk
  Tel: +44 (0)1382 38 4687
  The University of Dundee is a Scottish Registered Charity, No. SC015096

Re: serial performance

[Trifle Menot]

On Sat, 13 Mar 2010 07:39:00 +0000, Paul Crawford <psc@sat.dundee.ac.uk>
wrote:

>Trifle Menot wrote:
>> With that setting, PCBoard now sees a 16550 when using /dev/ttyS0, and
>> my remote output looks fast, like 57600 should.  So poking the necessary
>> bits into the dosemu virtual UART fixed things up.

>Can you elaborate on which bits you poked to achieve this?

>We have also had some performance issues with the virtual UART and 
>anything that makes it more like real-time would be appreciated.

I'm not sure, but probably UART FCR0=1.  I got the idea from reading
PCBoard's help text for the "Force 16550A Usage" setting:

> Some multi-tasking environments, such as OS/2 or Windows, use device
> drivers to virtualize the comm port.  In the process of doing so, they
> may hide the 16550A UART, making it appear to be a 16450 UART when
> accessed by a DOS application.

> Setting the answer to 'Y' tells PCBoard to operate AS IF it had
> detected a 16550A chip.

From a quick read of the dosemu source, it seems that the virutal UART
defaults to 16450 mode, until you put it into 16550 mode, presumably by
poking FCR0=1 into the (virtual) UART.

If the app cannot do that on its own, maybe a small C program that does
it, prior to launching the app, would achieve the same result.


-- 
Web mail, POP3, and SMTP
http://www.beewyz.com/freeaccounts.php