On Fri, Apr 16, 2010 at 9:57 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Fri, 16 Apr 2010, Jeremy Ramer wrote: > >> Hi, >> I am seeing a very strange kernel crashing issue with my linux driver. >> The strange thing is most of the time the driver works fine. I can >> read and write my USB device with no issues. Then usually one day a >> week, seeming to be on a rotating schedule of every 8 days or so, >> every time I try to read from the device with usb_bulk_msg() the >> kernel crashes hard. There is no response to anything. Removing the >> USB device does nothing. Magic SysRq commands do nothing. The only way >> to recover is turning off the power. This will happen for the full >> day. The next day reads work fine. > > Very strange. Does the behavior change right at 12 midnight? Do you > expect your device's behavior to change from day to day? I am not certain exactly when the behavior changes. I have seen some cases where it will not work in the morning and then does work in the afternoon. Also some cases where is will not work for two days in a row, then work for the rest of the week. The device behavior should not change from day to day. > >> I have seen this on 3 different computers. Two laptops running Kubuntu >> 9.10, kernel version 2.6.31-20 and a desktop running OpenSuse 11.0, >> kernel version 2.6.25.16. The HCD is EHCI. The host vendors are Intel >> 82801I (ICH9) and Intel ICH10. These are all 64-bit SMP systems. >> >> I have traced the position of the crash to be somewhere inside >> usb_bulk_msg(). > > You're sure it isn't somewhere in your driver following the call to > usb_bulk_msg()? Yes, I have a USB analyzer and have found that the USB read does not even hit the USB bus. Traces that I added immediately after the function never occur, even with long delays to ensure time for the messages to be written out. > >> I am working on tracing through that code but as the >> crash comes and goes I have not had the chance yet. One thing about my >> usage of usb_bulk_msg(). The read size is larger than the maximum >> packet size. This is to increase the efficiency of the read which is >> usually MBs in size. My review of source code shows that this is not >> usually done, but I have not seen any documentation that specifically >> forbids it. I could submit the URB myself but I would be just >> duplicating the code inside usb_bulk_msg(). Is there any reason the >> max packet size can not be exceeded with this function? > > It's perfectly acceptable to use large reads, larger than the maxpacket > size. That's good to know. -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
