>>> This e-mail is a follow-up as requested in this bug[1]. I will repost
>>> everything so far in this e-mail. Please CC me as I'm not subscribed
>>> to your list.
>>>
>>> Current head gives this when I plug a 'Mass Storage Device' into a 2.0 hub:
>>>
>>> [ 842.760400] hub 1-0:1.0: unable to enumerate USB device on port 3
>>> [ 843.080058] usb 1-3: new high-speed USB device number 48 using ehci-pci
>>> [ 858.230072] usb 1-3: device descriptor read/64, error -110
>>> [ 873.490070] usb 1-3: device descriptor read/64, error -110
>>>
>>> Reverting the following commit makes it work again:
>>>
>>> commit 55bcdce8a8228223ec4d17d8ded8134ed265d2c5
>>> Author: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
>>> Date: Fri Jan 25 16:52:45 2013 -0500
>>>
>>> USB: EHCI: remove ASS/PSS polling timeout
>>>
>>> This patch (as1647) attempts to work around a problem that seems to
>>> affect some nVidia EHCI controllers. They sometimes take a very long
>>> time to turn off their async or periodic schedules. I don't know if
>>> this is a result of other problems, but in any case it seems wise not
>>> to depend on schedule enables or disables taking effect in any
>>> specific length of time.
>>>
>>> The patch removes the existing 20-ms timeout for enabling and
>>> disabling the schedules. The driver will now continue to poll the
>>> schedule state at 1-ms intervals until the controller finally decides
>>> to obey the most recent command issued by the driver. Just in case
>>> this hides a problem, a debugging message will be logged if the
>>> controller takes longer than 20 polls.
>>>
>>> I don't know if this will actually fix anything, but it can't hurt.
>>>
>>> Signed-off-by: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
>>> Tested-by: Piergiorgio Sartor <piergiorgio.sartor@xxxxxxxx>
>>> CC: <stable@xxxxxxxxxxxxxxx>
>>> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>>>
>>> diff --git a/drivers/usb/host/ehci-timer.c b/drivers/usb/host/ehci-timer.c
>>> index 20dbdcb..f904071 100644
>>> --- a/drivers/usb/host/ehci-timer.c
>>> +++ b/drivers/usb/host/ehci-timer.c
>>> @@ -113,14 +113,15 @@ static void ehci_poll_ASS(struct ehci_hcd *ehci)
>>>
>>> if (want != actual) {
>>>
>>> - /* Poll again later, but give up after about 20 ms */
>>> - if (ehci->ASS_poll_count++ < 20) {
>>> - ehci_enable_event(ehci, EHCI_HRTIMER_POLL_ASS, true);
>>> - return;
>>> - }
>>> - ehci_dbg(ehci, "Waited too long for the async schedule
>>> status(%x/%x), giving up\n",
>>> - want, actual);
>>> + /* Poll again later */
>>> + ehci_enable_event(ehci, EHCI_HRTIMER_POLL_ASS, true);
>>> + ++ehci->ASS_poll_count;
>>> + return;
>>> }
>>> +
>>> + if (ehci->PSS_poll_count > 20)
>>> + ehci_dbg(ehci, "PSS poll count reached %d\n",
>>> + ehci->PSS_poll_count);
>>> ehci->PSS_poll_count = 0;
>>>
>>> /* The status is up-to-date; restart or stop the schedule as needed */
>>>
>>> Please note, that I'm using the 'irqpoll' cmdline to improve system
>>> stability. What I forgot to mention in the bug was the chipset:
>>>
>>> 00:10.3 USB controller: VIA Technologies, Inc. USB 2.0 (rev 82)
>>> (prog-if 20 [EHCI])
>>> Subsystem: Micro-Star International Co., Ltd. KT4AV motherboard
>>> Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV+ VGASnoop- ParErr-
>>> Stepping- SERR- FastB2B- DisINTx-
>>> Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=medium >TAbort-
>>> <TAbort- <MAbort- >SERR- <PERR- INTx-
>>> Latency: 32, Cache Line Size: 32 bytes
>>> Interrupt: pin D routed to IRQ 21
>>> Region 0: Memory at dffeff00 (32-bit, non-prefetchable) [size=256]
>>> Capabilities: [80] Power Management version 2
>>> Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=375mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
>>> Status: D0 NoSoftRst- PME-Enable- DSel=0 DScale=0 PME-
>>> Kernel driver in use: ehci-pci
>>>
>>> Yes, it's 10 years old, and no I'm not scrooge. We are waiting a while
>>> for computer prices to plummet mkay?
>>>
>>> [1]: https://bugzilla.kernel.org/show_bug.cgi?id=54031
>>
>> Would like to add that while searching the mailing lists I stumbled on this:
>>
>> http://marc.info/?l=linux-usb&m=136045531311402&w=4
>>
>> It's an entirely seperate USB issue in this cycle. The person is doing
>> git bisect to find the regression. I did another approach as
>> recompiling the full kernel on a 1,25GHz isn't even remotely funny
>> anymore.
>>
>> On top of HEAD I started reverting groups of USB EHCI patches one by
>> one. I'm just mentioning it since I'm not sure if this procedure is
>> accepted here.
>
> Did some more testing this morning. It seems like it's a race
> condition, which somewhat confirms that this patch is involved. Just
> had an occurance where the kernel with this patch *not* reverted
> handled the USB just fine. But subsequent attempts failed like this:
>
> # attempt 2
> [ 382.370377] hub 1-0:1.0: unable to enumerate USB device on port 3
> [ 382.690046] usb 1-3: new high-speed USB device number 7 using ehci-pci
> [ 397.840031] usb 1-3: device descriptor read/64, error -110
> # attempt 3
> [ 413.040329] hub 1-0:1.0: unable to enumerate USB device on port 3
> [ 413.360069] usb 1-3: new high-speed USB device number 8 using ehci-pci
> [ 428.510049] usb 1-3: device descriptor read/64, error -110
>
> Please note the ~15 second time-out between detection and error. This
> also explains what made ordinary bisect somewhat 'tedious'... . Kernel
> with this patch reverted handles the USB solidly everytime and on
> every subsequent occurance so far.
>
> The 'bad' kernel handles the USB correctly every once in a while. But
> subsequent occurances always failed so far under these conditions.
One final observation, sometimes the kernel without the patch reverted
rejects the usb-stick right away (i.e. no initial onetime success).
Dmesg then looks like this:
[ 146.980077] usb 1-3: new high-speed USB device number 4 using ehci-pci
[ 152.140713] usb 1-3: device descriptor read/all, error -110
[ 152.200338] hub 1-0:1.0: unable to enumerate USB device on port 3
[ 152.520050] usb 1-3: new high-speed USB device number 6 using ehci-pci
[ 167.670035] usb 1-3: device descriptor read/64, error -110
[ 182.930046] usb 1-3: device descriptor read/64, error -110
[ 183.100306] hub 1-0:1.0: unable to enumerate USB device on port 3
[ 183.420047] usb 1-3: new high-speed USB device number 8 using ehci-pci
[ 198.570063] usb 1-3: device descriptor read/64, error -110
[ 213.830047] usb 1-3: device descriptor read/64, error -110
[ 214.000248] hub 1-0:1.0: unable to enumerate USB device on port 3
# above 4 lines repeat until stick get's unplugged
Notice the first error to be related to 'read/all' and subsequent
errors are related to 'read/64'. Kernel with patch reverted still
works without a hitch so far.
Conclusion: So I'm pretty confident this patch is the 'first bad commit'.
--
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
