On Wed, Apr 11, 2012 at 12:54 PM, Christian Lamparter <chunkeey@xxxxxxxxxxxxxx> wrote: > On Wednesday, April 11, 2012 03:20:24 PM J Igrap wrote: >> On Wed, Sep 14, 2011 at 1:00 PM, Christian Lamparter >> <chunkeey@xxxxxxxxxxxxxx> wrote: >> > On Wednesday, September 14, 2011 03:09:14 PM J Igrap wrote: >> >> On Wed, Sep 14, 2011 at 5:50 AM, Christian Lamparter >> >>> On Tuesday, September 13, 2011 11:18:23 PM J Igrap wrote: >> >>>> While using the past days the carl9170 firmware with a USB card under a >> >>>> linux guest running different kernel and driver versions I kept running into >> >>>> the issue of a usb disconnect when the card was put under load: >> >>> linux guest? You are not using carl9170 in a VM, are you? >> >>> >> >> It is in a VM. I was working on a VM for ease of debugging the issue. >> > So, I presume you have already tried running the driver natively >> > and it fails in a similar fashion, right? [I just want to rule out >> > a bug in the VM layers] >> > >> >> Yes, it appears to fail in slow/older physicals too. > > what's "slower/older"? I know a few people reported problems with embedded USB > HCDs but that's understandable. However, I occasionally test carl9170 on a 7 > year old laptop and so far it runs ok. > A physical pentium 4 single core machine. Also pretty much every vmware version (workstation and fusion). The card gets reset randomly. >> >>>> usb 1-1: no command feedback received (-110). >> >>>> carl9170 cmd: 08 01 00 00 f0 36 1c 00 00 24 00 00 .....6...$.. >> >>>> usb 1-1: restart device (6) >> >>>> >> >>>> No matter what kernel driver/firmware I tried I will still get it. I decided >> >>>> to look into it a bit more and I narrowed it down to be a firmware issue >> >>>> with the following code snippet: >> >>> Or it could be a problem with the USB PHY. However, the driver >> >>> seems to be able to handle the situation and restarts the device >> >>> accordingly. >> >>> >> >> I tried with several physical devices and they all seemed to have the >> >> same behavior. The device does restart however you lose connectivity >> >> and state of what you were doing. >> > Any transmission protocol should be able to handle loss of connectivity. >> > Furthermore all connection managers [wpa_supplicant, NetworkManager, wicd] >> > will automatically reestablish the link if the device was put out by a >> > catastrophic failure. >> > >> >>>> void handle_cmd(struct carl9170_rsp *resp) in src/cmd.c >> >>>> >> >>>> case CARL9170_CMD_WREG: >> >>>> esp->hdr.len = 0; >> >>>> for (i = 0; i < (cmd->hdr.len / 8); i++) >> >>>> set(cmd->wreg.regs[i].addr, cmd->wreg.regs[i].val); >> >>>> break; >> >>>> >> >>>> That code appears to handle event 1 which is a write into a register. In >> >>>> some cases that write appeared to cause a failure and a reset into the card. >> >>>> I added a simple delay loop before the switch statement and that seemed to >> >>>> fix the issue and I don't lose the card anymore even under a lot of load. >> >>>> Obviously that's not a real fix and something else more reliable needs to be >> >>>> in place. >> >>> Any idea what this "something" else might be? >> >> >> >> I'm not very familiar with how USB works, maybe someone with more >> >> experience can shed some light here? It seems to me that the event >> >> handling needs to be slowed down a little or add some kind of >> >> verification. >> > the usb protocol already incorporates some verification >> > http://www.beyondlogic.org/usbnutshell/usb3.shtml >> > >> > furthermore, the driver counts each command frame, therefore it can >> > detect whenever a frame was lost. >> >> It does detect it and restarwts the device, hoever when running under >> VMware os r slower physicalit gets in the way since the device resets >> very often. >> >> I tried build various firmware's as an experiment and I noticed that >> the higher I set the Max Frame Length the more resistant it becomes. >> The script I wrote basically sends >> a beacon packet multiple times in monitor mode and has a counter of >> how many packets went through before the card resets. Any ideas what >> could be causing this? > no. Unlike ath9k_htc fw, carl9170 does not have "space" to add any > sophisticated tracer. So, you are stuck with the 1/2-bit LEDs as a > window to the device. > > Regards, > Christian -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
