Skip to contentSkip to conversation

usb-serial ipaq kernel problem

DISCUSSION Frank Gevaerts 2 participants 4 messages May 22, 2006 – May 23, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

We are having problems with the usb-serial ipaq driver in 2.6.16 (debian backports 2.6.16-1-686, but also reproducible with self-compiled kernel.org kernel)

Sometimes, we get the following on disconnect:

Unable to handle kernel paging request at virtual address 723d4ec3
 printing eip:
b01ea93d
*pde = 00000000
Oops: 0000 [#1]
Modules linked in: ppp_async crc_ccitt ppp_generic slhc ipaq usbserial uhci_hcd ohci_hcd usbhid ehci_hcd usbcore 8139too mii sr_mod sbp2 scsi_mod ieee1394 psmouse ide_generic ide_cd cdrom genrtc ext3 jbd mbcache ide_disk generic via82cxxx ide_core evdev mousedev
CPU:    0
EIP:    0060:[<b01ea93d>]    Not tainted VLI
EFLAGS: 00010246   (2.6.16-1-686 #2)

EIP is at check_tty_count+0x33/0x7a
eax: 723d4e4f ebx: 00000000 ecx: c97cb000 edx: c97cb170
esi: cbfe55a0 edi: 00000283 ebp: c97cb000 esp: cbe87f18
ds: 007b es: 007b ss: 0068
Process events/0 (pid: 4, threadinfo=cbe86000 task=cbfdfab0)

Stack: <0>c97cb000 b01eb558 c97cb000 b029719d 00000000 00000000 00000000 c97cb13c
       cbfe55a0 00000283 c97cb000 b012277d c97cb000 b01eb507 cbe86000 cbe87f84
       cbe87fa4 cbfe55a0 b01228a9 cbfe55a0 ffffffff ffffffff 00000001 00000000
Call Trace:
 [<b01eb558>] do_tty_hangup+0x51/0x294
 [<b012277d>] run_workqueue+0x6e/0xa2
 [<b01eb507>] do_tty_hangup+0x0/0x294
 [<b01228a9>] worker_thread+0xf8/0x12a
 [<b01138c3>] default_wake_function+0x0/0x12
 [<b026dabd>] schedule+0x45f/0x4cd
 [<b01138c3>] default_wake_function+0x0/0x12
 [<b01227b1>] worker_thread+0x0/0x12a
 [<b0124ece>] kthread+0x79/0xa3
 [<b0124e55>] kthread+0x0/0xa3
 [<b01012cd>] kernel_thread_helper+0x5/0xb
Code: 91 70 01 00 00 8b 02 0f 18 00 90 8d 81 70 01 00 00 39 c2 74 13 8b 12 43 8b 02 0f 18 00 90 8d 81 70 01 00 00 39 c2 75 ed 8b 41 04 <81> 78 74 04 00 02 00 75 17 8b 91 cc 00 00 00 85 d2 74 0d 83 ba

I don't know enough about the tty system to find the problem, but it seems to me that the tty structure might be released too soon. We have a hotplug script that starts ppp on connect, and we can reproduce this by repeatedly plugging and unplugging the ipaq.

Frank

Signature · 4 lines
Frank Gevaerts                                 [email protected]
fks bvba - Formal and Knowledge Systems        http://www.fks.be/
Stationsstraat 108                             Tel:  ++32-(0)11-21 49 11
B-3570 ALKEN                                   Fax:  ++32-(0)11-22 04 19
#1 lore raw
Quoted 7 lines · Frank Gevaerts

Hi,

We are having problems with the usb-serial ipaq driver in 2.6.16 (debian backports 2.6.16-1-686, but also reproducible with self-compiled kernel.org kernel)

Sometimes, we get the following on disconnect:

<snip>

Can you duplicate this on 2.6.17-rc4? A number of tty changes went into that release that should have fixed this issue.

thanks,

greg k-h

#2 lore raw
Quoted 7 lines · Greg KH
Quoted 6 lines · Frank Gevaerts

Hi,

We are having problems with the usb-serial ipaq driver in 2.6.16 (debian backports 2.6.16-1-686, but also reproducible with self-compiled kernel.org kernel)

Quoted 6 lines
Quoted 1 line

Sometimes, we get the following on disconnect:

<snip>

Can you duplicate this on 2.6.17-rc4? A number of tty changes went into that release that should have fixed this issue.

I'll try it in the morning. In the meantime, we found some other problems in ipaq.c : apparently pocketpc accepts usb enumeration long before it accepts usb-serial commands (sometimes 50 or more seconds), which makes ipaq_open fail. When it fails, the read urb is not killed, while the associated structures are freed, which gives a panic when the urb completes. The following patch solves that :
Since changing this, I also have not seen the original problem anymore, but I will do some more testing tomorrow.

drivers/usb/serial/ipaq.c 2006-05-23 00:00:33.000000000 +0200 renamed from drivers/usb/serial/ipaq.c 2006-03-20 06:53:29.000000000 +0100 +8 -7
@@ -59,7 +59,7 @@
 #include "usb-serial.h"
 #include "ipaq.h"
 
-#define KP_RETRIES     100
+#define KP_RETRIES     1000
 
 /*
  * Version Information
@@ -652,12 +652,6 @@
                      usb_rcvbulkpipe(serial->dev, port->bulk_in_endpointAddress),
                      port->read_urb->transfer_buffer, port->read_urb->transfer_buffer_length,
                      ipaq_read_bulk_callback, port);
-       result = usb_submit_urb(port->read_urb, GFP_KERNEL);
-       if (result) {
-               err("%s - failed submitting read urb, error %d", __FUNCTION__, result);
-               goto error;
-       }
-
        /*
         * Send out control message observed in win98 sniffs. Not sure what
         * it does, but from empirical observations, it seems that the device
@@ -671,8 +665,15 @@
                                usb_sndctrlpipe(serial->dev, 0), 0x22, 0x21,
                                0x1, 0, NULL, 0, 100);
                if (result == 0) {
+                       result = usb_submit_urb(port->read_urb, GFP_KERNEL);
+                       if (result) {
+                               err("%s - failed submitting read urb, error %d", __FUNCTION__, result);
+                               goto error;
+                       }
+
                        return 0;
                }
+                msleep(100);
        }
        err("%s - failed doing control urb, error %d", __FUNCTION__, result);
        goto error;
Quoted 5 lines

thanks,

greg k-h

Signature · 4 lines
Frank Gevaerts                                 [email protected]
fks bvba - Formal and Knowledge Systems        http://www.fks.be/
Stationsstraat 108                             Tel:  ++32-(0)11-21 49 11
B-3570 ALKEN                                   Fax:  ++32-(0)11-22 04 19
#3 lore raw
Quoted 15 lines · Frank Gevaerts
Quoted 7 lines · Greg KH
Quoted 6 lines · Frank Gevaerts

Hi,

We are having problems with the usb-serial ipaq driver in 2.6.16 (debian backports 2.6.16-1-686, but also reproducible with self-compiled kernel.org kernel)

Quoted 6 lines
Quoted 1 line

Sometimes, we get the following on disconnect:

<snip>

Can you duplicate this on 2.6.17-rc4? A number of tty changes went into that release that should have fixed this issue.

Quoted 8 lines

I'll try it in the morning. In the meantime, we found some other problems in ipaq.c : apparently pocketpc accepts usb enumeration long before it accepts usb-serial commands (sometimes 50 or more seconds), which makes ipaq_open fail. When it fails, the read urb is not killed, while the associated structures are freed, which gives a panic when the urb completes. The following patch solves that :
Since changing this, I also have not seen the original problem anymore, but I will do some more testing tomorrow.

I cannot reproduce the original problem anymore with my patch applied

Frank

Quoted 55 lines
drivers/usb/serial/ipaq.c 2006-05-23 00:00:33.000000000 +0200 renamed from drivers/usb/serial/ipaq.c 2006-03-20 06:53:29.000000000 +0100 +8 -7
@@ -59,7 +59,7 @@
 #include "usb-serial.h"
 #include "ipaq.h"
 
-#define KP_RETRIES     100
+#define KP_RETRIES     1000
 
 /*
  * Version Information
@@ -652,12 +652,6 @@
                      usb_rcvbulkpipe(serial->dev, port->bulk_in_endpointAddress),
                      port->read_urb->transfer_buffer, port->read_urb->transfer_buffer_length,
                      ipaq_read_bulk_callback, port);
-       result = usb_submit_urb(port->read_urb, GFP_KERNEL);
-       if (result) {
-               err("%s - failed submitting read urb, error %d", __FUNCTION__, result);
-               goto error;
-       }
-
        /*
         * Send out control message observed in win98 sniffs. Not sure what
         * it does, but from empirical observations, it seems that the device
@@ -671,8 +665,15 @@
                                usb_sndctrlpipe(serial->dev, 0), 0x22, 0x21,
                                0x1, 0, NULL, 0, 100);
                if (result == 0) {
+                       result = usb_submit_urb(port->read_urb, GFP_KERNEL);
+                       if (result) {
+                               err("%s - failed submitting read urb, error %d", __FUNCTION__, result);
+                               goto error;
+                       }
+
                        return 0;
                }
+                msleep(100);
        }
        err("%s - failed doing control urb, error %d", __FUNCTION__, result);
        goto error;
Quoted 6 lines

thanks,

greg k-h

Signature · 4 lines
Frank Gevaerts                                 [email protected]
fks bvba - Formal and Knowledge Systems        http://www.fks.be/
Stationsstraat 108                             Tel:  ++32-(0)11-21 49 11
B-3570 ALKEN                                   Fax:  ++32-(0)11-22 04 19
Signature · 4 lines
Frank Gevaerts                                 [email protected]
fks bvba - Formal and Knowledge Systems        http://www.fks.be/
Stationsstraat 108                             Tel:  ++32-(0)11-21 49 11
B-3570 ALKEN                                   Fax:  ++32-(0)11-22 04 19