USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[clameter]

I recently got a ricochet 128k GS wireless modem and I am running it with
kernel 2.4.4 and ppp 2.4.1.

Using the USB connection (configured to operatate at 460kbit)  I get up to
2kbyte per second. With serial(at 115kbit) this goes up to 8kbyte per
second.

Why is this?

(Note to all prospective Ricochet/Earthlink buyers: Beware. It takes 10-20
successful connects before you can establish a connection that lets data
through. Tested on multiple OSe and my ricochet repeater is across the
street on a lamp).

May  9 10:42:55 piii syslogd 1.4.1#2: restart.
May  9 10:42:55 piii kernel: klogd 1.4.1#2, log source = /proc/kmsg started.
May  9 10:42:55 piii kernel: Inspecting /boot/System.map-2.4.4-ac4
May  9 10:42:55 piii kernel: Loaded 14111 symbols from /boot/System.map-2.4.4-ac4.
May  9 10:42:55 piii kernel: Symbols match kernel version 2.4.4.
May  9 10:42:55 piii kernel: Loaded 122 symbols from 8 modules.
May  9 10:42:55 piii kernel: port... done.
May  9 10:42:55 piii kernel: Checking 'hlt' instruction... OK.

May  9 10:42:55 piii kernel: Serial driver version 5.05a (2001-03-20) with MANY_PORTS SHARE_IRQ SERIAL_PCI enabled
May  9 10:42:55 piii kernel: ttyS00 at 0x03f8 (irq = 4) is a 16550A

May  9 10:42:55 piii kernel: usb.c: registered new driver usbdevfs
May  9 10:42:55 piii kernel: usb.c: registered new driver hub
May  9 10:42:55 piii kernel: usb-ohci.c: USB OHCI at membase 0xd004a000, IRQ 3
May  9 10:42:55 piii kernel: usb-ohci.c: usb-00:01.2, Silicon Integrated  Systems [SiS] 7001
May  9 10:42:55 piii kernel: usb.c: new USB bus registered, assigned bus number 1
May  9 10:42:55 piii kernel: Product: USB OHCI Root Hub
May  9 10:42:55 piii kernel: SerialNumber: d004a000
May  9 10:42:55 piii kernel: hub.c: USB hub found
May  9 10:42:55 piii kernel: hub.c: 2 ports detected
May  9 10:42:55 piii kernel: hub.c: USB new device connect on bus1/1, assigned device number 2
May  9 10:42:55 piii kernel: Manufacturer: Metricom, Inc.
May  9 10:42:55 piii kernel: Product: Metricom GS Modem
May  9 10:42:55 piii kernel: SerialNumber: SN00a079038f68
May  9 10:42:55 piii kernel: usb.c: USB device 2 (vend/prod 0x870/0x1) is not claimed by any active driver.

May  9 10:42:55 piii kernel: usb-uhci.c: $Revision: 1.259 $ time 12:14:15 May  4 2001
May  9 10:42:55 piii kernel: usb-uhci.c: High bandwidth mode enabled
May  9 10:42:55 piii kernel: usb.c: registered new driver acm
May  9 10:42:55 piii kernel: ttyACM0: USB ACM device
May  9 10:42:55 piii kernel: CDCEther.c: CDCEther.c: v0.98.2 28 March 2001 Brad Hards and another
May  9 10:42:55 piii kernel: usb.c: registered new driver CDCEther

May  9 17:39:29 piii pppd[4388]: pppd 2.4.1 started by root, uid 0
May  9 17:39:31 piii chat[4392]: abort on (BUSY)
May  9 17:39:31 piii chat[4392]: abort on (VOICE)
May  9 17:39:31 piii chat[4392]: abort on (NO DIALTONE)
May  9 17:39:31 piii chat[4392]: send (ATDT3333^M)
May  9 17:39:31 piii chat[4392]: expect (CONNECT)
May  9 17:39:31 piii chat[4392]: ATDT3333^M^M
May  9 17:39:31 piii chat[4392]: CONNECT
May  9 17:39:31 piii chat[4392]:  -- got it
May  9 17:39:31 piii pppd[4388]: Serial connection established.
May  9 17:39:31 piii pppd[4388]: Using interface ppp0
May  9 17:39:31 piii pppd[4388]: Connect: ppp0 <--> /dev/ttyS0
May  9 17:39:46 piii pppd[4388]: Remote message: CHAP authentication success, unit 713
May  9 17:39:47 piii pppd[4388]: local  IP address 63.13.222.189
May  9 17:39:47 piii pppd[4388]: remote IP address 63.13.222.1
May  9 17:39:47 piii pppd[4388]: primary   DNS address 198.6.1.195
May  9 17:39:47 piii pppd[4388]: secondary DNS address 198.6.1.5

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Greg KH]

On Wed, May 09, 2001 at 05:52:04PM -0700, clameter@lameter.com wrote:
> I recently got a ricochet 128k GS wireless modem and I am running it with
> kernel 2.4.4 and ppp 2.4.1.
> 
> Using the USB connection (configured to operatate at 460kbit)  I get up to
> 2kbyte per second. With serial(at 115kbit) this goes up to 8kbyte per
> second.
> 
> Why is this?

Because currently the USB acm driver is not tuned for speed, reliability
up to now has been more important :)

greg k-h

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[clameter]

On Wed, 9 May 2001, Greg KH wrote:
> Because currently the USB acm driver is not tuned for speed, reliability
> up to now has been more important :)

Allright then you should first check why the ACM driver is unable to
handle an MTU of 1500. I had to set it to 232 or 500 to make it work at
all. With an MTU of 1500 it does ICMP but not long tcp packets. There is
some issue with long packets that might exceed some buffer size(?).

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Greg KH]

On Wed, May 09, 2001 at 11:09:36PM -0700, clameter@lameter.com wrote:
> 
> Allright then you should first check why the ACM driver is unable to
> handle an MTU of 1500. I had to set it to 232 or 500 to make it work at
> all. With an MTU of 1500 it does ICMP but not long tcp packets. There is
> some issue with long packets that might exceed some buffer size(?).

I don't see anything in the ACM driver that would cause a problem for
large MTU settings.  It is probably a device limitation, not the driver.

greg k-h

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[clameter]

On Wed, 9 May 2001, Greg KH wrote:

> On Wed, May 09, 2001 at 11:09:36PM -0700, clameter@lameter.com wrote:
> > 
> > Allright then you should first check why the ACM driver is unable to
> > handle an MTU of 1500. I had to set it to 232 or 500 to make it work at
> > all. With an MTU of 1500 it does ICMP but not long tcp packets. There is
> > some issue with long packets that might exceed some buffer size(?).
> 
> I don't see anything in the ACM driver that would cause a problem for
> large MTU settings.  It is probably a device limitation, not the driver.

The Richochet USB stuff uses generic serial I/O. No special driver. And it
works fine under Win/ME. Have you run a regular PPP connection over the
ACM driver with an MTU of 1500?

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Greg KH]

On Wed, May 09, 2001 at 11:25:26PM -0700, clameter@lameter.com wrote:
> 
> The Richochet USB stuff uses generic serial I/O. No special driver. And it
> works fine under Win/ME. Have you run a regular PPP connection over the
> ACM driver with an MTU of 1500?

The Linux USB ACM driver is the same "generic serial I/O" driver that
you speak of Win/ME having.  What is the MTU setting under Win/ME for
the device?

And no, I haven't run a PPP connection over a ACM device, as I do not
have an ACM device (otherwise I would have fixed the speed issues by now :)

But again, there are no packet size limitations in the driver, except as
such is specified by the specific device (endpoint size is determined by
the device, not the driver.)  So that would point to either a PPP
problem (maybe), or a device problem (probably, would have to have a
large buffer to handle a MTU of that size, and silicon isn't cheap for
tiny devices like modems.)

Hope this helps,

greg k-h

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Drew Bertola]

On Wed, May 09, 2001 at 11:25:26PM -0700, clameter@lameter.com wrote:
> On Wed, 9 May 2001, Greg KH wrote:
> 
> > On Wed, May 09, 2001 at 11:09:36PM -0700, clameter@lameter.com wrote:
> > > 
> > > Allright then you should first check why the ACM driver is unable to
> > > handle an MTU of 1500. I had to set it to 232 or 500 to make it work at
> > > all. With an MTU of 1500 it does ICMP but not long tcp packets. There is
> > > some issue with long packets that might exceed some buffer size(?).
> > 
> > I don't see anything in the ACM driver that would cause a problem for
> > large MTU settings.  It is probably a device limitation, not the driver.
> 
> The Richochet USB stuff uses generic serial I/O. No special driver. And it
> works fine under Win/ME. Have you run a regular PPP connection over the
> ACM driver with an MTU of 1500?

Joey Hess had a problem similar to what you described, though he noticed
it while using the pcmcia ricochet modem.  He passed along this patch:


--- Serial.c.orig       Fri Feb  2 12:55:44 2001
+++ serial.c    Fri Feb  2 12:56:43 2001
@@ -569,10 +569,16 @@

        icount = &info->state->icount;
        do {
-
+               /*
+                * Check if flip buffer is full -- if it is, try to
flip,
+                * and if flipping got queued, return immediately
+                */
+               if (tty->flip.count >= TTY_FLIPBUF_SIZE) {
+                       tty->flip.tqueue.routine((void *) tty);
+                       if (tty->flip.count >= TTY_FLIPBUF_SIZE)
+                               return;
+               }
                ch = serial_inp(info, UART_RX);
-               if (tty->flip.count >= TTY_FLIPBUF_SIZE)
-                       goto ignore_char;
                *tty->flip.char_buf_ptr = ch;
                icount->rx++;


-- 
Drew Bertola  | Send a text message to my pager or cell ... 
              |   http://jpager.com/Drew

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Greg KH]

[-- Attachment #1: Type: text/plain, Size: 466 bytes --]

On Thu, May 10, 2001 at 08:07:50PM -0700, Drew Bertola wrote:
> 
> Joey Hess had a problem similar to what you described, though he noticed
> it while using the pcmcia ricochet modem.  He passed along this patch:

Doh!  I've only fixed this same kind of problem about 3 different times
in the usb-serial drivers.  clameter, could you try the attached patch
against 2.4.4 and see if that fixes the MTU issue for you?

Thanks Drew for reminding me of this.

greg k-h


[-- Attachment #2: usb-acm-2.4.4.patch --]
[-- Type: text/plain, Size: 620 bytes --]

--- linux-2.4.4/drivers/usb/acm.c	Fri Feb 16 16:06:17 2001
+++ linux-2.4/drivers/usb/acm.c	Thu May 10 21:29:29 2001
@@ -233,8 +240,14 @@
 		dbg("nonzero read bulk status received: %d", urb->status);
 
 	if (!urb->status & !acm->throttle)  {
-		for (i = 0; i < urb->actual_length && !acm->throttle; i++)
+		for (i = 0; i < urb->actual_length && !acm->throttle; i++) {
+			/* if we insert more than TTY_FLIPBUF_SIZE characters, 
+			 * we drop them. */
+			if (tty->flip.count >= TTY_FLIPBUF_SIZE) {
+				tty_flip_buffer_push(tty);
+			}
 			tty_insert_flip_char(tty, data[i], 0);
+		}
 		tty_flip_buffer_push(tty);
 	}
 

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[clameter]

On Thu, 10 May 2001, Drew Bertola wrote:

> > The Richochet USB stuff uses generic serial I/O. No special driver. And it
> > works fine under Win/ME. Have you run a regular PPP connection over the
> > ACM driver with an MTU of 1500?
> 
> Joey Hess had a problem similar to what you described, though he noticed
> it while using the pcmcia ricochet modem.  He passed along this patch:

I tried the patch a couple of days ago and it did not do anything for me.

Re: USB broken in 2.4.4? Serial Ricochet works, USB performance sucks.

[Drew Bertola]

Damn bad luck.  I love my ricochet, but have Metricom as the provider.
I've used it in the bay area, LA and OC area, Boulder, and NYC.

It was a snap to setup with the acm module, and I get almost full
115kbit throughput.

On Wed, May 09, 2001 at 05:52:04PM -0700, clameter@lameter.com wrote:
> I recently got a ricochet 128k GS wireless modem and I am running it with
> kernel 2.4.4 and ppp 2.4.1.
> 
> Using the USB connection (configured to operatate at 460kbit)  I get up to
> 2kbyte per second. With serial(at 115kbit) this goes up to 8kbyte per
> second.
> 
> Why is this?
> 
> (Note to all prospective Ricochet/Earthlink buyers: Beware. It takes 10-20
> successful connects before you can establish a connection that lets data
> through. Tested on multiple OSe and my ricochet repeater is across the
> street on a lamp).

-- 
Drew Bertola  | Send a text message to my pager or cell ... 
              |   http://jpager.com/Drew