Novatel Merlin V620 CDMA PC Card: How To?

gpetme · 06-25-2005, 02:17 PM

Guys -

I've got my V620 working on Gentoo 2005.0 (running 2.6.11). I'm using pppd. I had it working for about 2 weeks, and then (without any changes to my config) I started getting the timeout issues after a couple minutes being connected. Here' the relevant config info from my now working setup:

# File: /etc/ppp/peers/verizon
#
/dev/ttyUSB0 # device
# The following two settings need a corresponding entry in
# /etc/ppp/chap-secrets.
user 2125551212@vzw3g.com
remotename verizon
230400 # speed
defaultroute # use the cellular network for the default route
usepeerdns # use the DNS servers from the remote network
nodetach # keep pppd in the foreground
crtscts # hardware flow control
lock # lock the serial port
noauth # don't expect the modem to authenticate itself
# the 2 below entries were commented out to try to fix the timeout issue <-------------------------------------
#novj
#novjccomp
# scripts for connect/disconnect
connect "/usr/sbin/chat -v -f /etc/ppp/chatscripts/verizon-connect"
disconnect "/usr/sbin/chat -v -f /etc/ppp/chatscripts/verizon-disconnect"

# File: /etc/chatscripts/verizon-connect
#
TIMEOUT 10
ABORT 'BUSY'
ABORT 'NO ANSWER'
ABORT 'NO CARRIER'
SAY 'Starting Verizon connect script\n'

# Get the modem's attention and reset it.
'' 'ATZ'

# E0=No echo, V1=English result codes
OK 'ATE0V1'

# Dial the number
SAY 'Dialing...\n'
OK 'ATD#777'
CONNECT ''

# File: /etc/chatscripts/verizon-disconnect
#
"" "\K"
"" "+++ATH0"
SAY "Disconnected from Verizon."

Let me know if you'd like to see any other configuration info. FYI, 'novj' and 'novjcomp' worked fine for me for a while, but then I commented them out. This was the only thing I changed to get rid of the timeout issue (which is contrary to what others have said, unfortunately).

Greg

sipples · 06-27-2005, 10:15 AM

There have been a number of USB-related fixes in the latest Linux kernels, according to the kernel announcements. Upthread there's a report of some limited success with 2.6.11. Anyone tried anything more recent?

Also, for those of you experiencing 2.5 minute timeouts, any luck just leaving a ping going (about every 30 seconds) in the background? I'm guessing that's the network "dormant" event, and Linux doesn't know what to do with it.

mjolnir · 06-30-2005, 12:57 PM

This is a little off topic but I am dual booting xp and an h/d install of knoppix. I have a pc3220 and have decided to upgrade the card. I would love to access my data plan with linux! One of the advertised strengths of the v620 is better reception in low signal strength areas(EV-DO). Has anyone noticed better reception when using 1xrtt?

gpetme · 07-02-2005, 01:02 PM

Quote:

Originally posted by sipples

Also, for those of you experiencing 2.5 minute timeouts, any luck just leaving a ping going (about every 30 seconds) in the background? I'm guessing that's the network "dormant" event, and Linux doesn't know what to do with it.

I tried the a ping keepalive every 15 seconds, but it didn't help. I also tried a continuous ping all to no avail... I'll try updating my kernel once I can get to a hotspot somewhere - 1xEVDO is all I've got right now so I'm forced to boot into Windoze to do my work.

Greg

gpetme · 07-04-2005, 08:40 PM

I compiled a vanilla 2.6.12.2 kernel. I can get a continuous ping to run (for ~ 15000 pings), but when I try to put any traffic through it (open Gmail, use Google for a few searches, connect to a Cisco VPN, etc) the tunnel fails. So, it's gotten a bit better. I'm going to adjust some other options and see how those help.

Greg

mjolnir · 07-13-2005, 07:38 AM

Any progress guys? I have 3 days left in my 15 day trial and I am thinking of taking the v620 back and trying the 5220 or just waiting a while. I haven't gotten past a connect script error so far, my fault not the cards, but the point is mute if I finally get a connection and can't pass data.

On the other hand I haven't run any tests but the v620 SEEMS to have better reception than the PC3220 I used to use(1xRTT). I haven't been to an ev-do area yet. On a side note if I tether my v710 to my laptop it FEELS considerably faster than either card.

gpetme · 07-14-2005, 02:04 PM

Sorry, haven't had a chance to test much further. 2.6.12 has improved the situation slightly, but the connection drops under any major load.

Greg

sipples · 07-14-2005, 07:38 PM

Has anyone contacted Novatel Wireless? Any info from them?

junxiondeveloper · 07-19-2005, 10:38 AM

Cheers everyone.

Our Junxion Box product uses PCMCIA cell cards, such as the Novatel
Merlin V620, Kyocera KPC 650, and Sierra Wireless AC 580, as part of a
wireless wan router device. These cards expose a standard serial modem
device that require the generic usb serial driver in Linux to work. We
have a workaround for the connection stalls experienced during high bandwidth
utilization on the EVDO network. The workaround can be viewed at
http://www.junxion.com/opensource/li...usbserial.html .

As usual your comments and suggestions are welcome. We will continue to
monitor these forums and provide assistance whenever possible.
thanks,
H.T.

davidh.davidh · 07-19-2005, 11:42 AM

I hacked* this fix into my 2.6.12 usbserial drive and it worked like a charm. Good speed and no hangs. WooHoo EVDO!!!

Thanks Junxion.

*I do mean hack, I hard coded maxSize=2048 in usb-serial, added the buffer_size line from the 2.4 patch and compiled a version just for this card.

Bini2 · 07-20-2005, 09:43 PM

Hey, great post.

I was able toget my Sprint Vision working with Novatel Merlin V620 CDMA.

Question, on Win 2k I get 1.5 to 1.7 Mbps....on Linux I get 200-400kbps, any suggestions for tuning? Is there a specific command that I need to send? Something I ammissing?

Thanks in advance.

sipples · 07-20-2005, 10:03 PM

Has this patch -- or something like it -- been submitted to kernel.org yet for incorporation into the mainline? If not, I'd like to try and help with that.

Ideally that parameter would be maxsize (all lowercase) for consistency, by the way.

Thanks for the good work!

sipples · 07-20-2005, 11:13 PM

Here's a slightly modified version of the patch suggested by Junxion. Call this Version 0.2, I suppose. :-) I've made the following changes:

1. Changed the parameter from MaxSize (mixed case, vague) to pktsize (i.e. "packet size"). Better name, I think, but suggestions welcome.
2. Changed the int to a __u16 for platform portability. (Shouldn't use int in drivers. And it should be unsigned anyway.)
3. Made the parameter description a little more understandable.

I have not tested the original nor these changes yet.

Code:

--- drivers/usb/serial/usbserial.c      Thu Nov 28 15:53:15 2002
+++ drivers/usb/serial/usbserial_junxion.c      Wed Jul 20 18:00:00 2005
@@ -331,6 +331,7 @@
 #ifdef CONFIG_USB_SERIAL_GENERIC
 static __u16   vendor  = 0x05f9;
 static __u16   product = 0xffff;
+static __u16   pktsize = 0;

 static struct usb_device_id generic_device_ids[2]; /* Initially all zeroes. */

@@ -1255,6 +1256,9 @@
                        goto probe_error;
                }
                buffer_size = endpoint->wMaxPacketSize;
+#ifdef CONFIG_USB_SERIAL_GENERIC
+               buffer_size = (endpoint->wMaxPacketSize > pktsize)?endpoint->wMaxPacketSize:pktsize;
+#endif
                port->bulk_in_endpointAddress = endpoint->bEndpointAddress;
                port->bulk_in_buffer = kmalloc (buffer_size, GFP_KERNEL);
                if (!port->bulk_in_buffer) {
@@ -1608,4 +1612,7 @@

 MODULE_PARM(product, "h");
 MODULE_PARM_DESC(product, "User specified USB idProduct");
+
+MODULE_PARM(pktsize,"i");
+MODULE_PARM_DESC(pktsize,"User specified USB endpoint maximum packet size");
 #endif

On edit: Why is this code in an #ifdef, by the way? It would seem to be harmless enough for setting buffer_size generally since the driver is going to pick the higher of the device-reported max packet size or the user-specified size. Is there any harm to having a "too large" packet size?

On 2nd edit: Should pktsize default to 1? Are there any devices that would report 0 -- and which should use 0? Or is there some real-world minimum that this code should set?

I think the best way to handle this would be to look for the product/vendor IDs for the Novatel and Kyocera cards and then set the packet size correctly specifically for those devices, PLUS this user-specified parameter for generic devices. But I've got to dig into the code for that, and I'm not near the usbserial.c source at the moment. Anyone know what the Kyocera product/vendor IDs are?

georgecmu · 07-21-2005, 09:10 AM

IMHO this code would make more sense:

if(serial->vendor == 0x1410 && serial->product == 0x1110) {
buffer_size = 2048;
} else {
buffer_size = endpoint->wMaxPacketSize;
}

This way there's no need for an extra module parameter and an extra usbserial module, since the buffer_size is only hardcoded for the specific device Novatel V620.

gpetme · 07-21-2005, 02:15 PM

Does anyone have the ability to port this patch to 2.6?

Thanks,

Greg