Re: [PATCH v3] extcon: Fixed sysfs duplicate filename issue

["Diederik de Haas" <diederik@cknow-tech.com>]

On Tue Dec 30, 2025 at 12:26 PM CET, Andy Shevchenko wrote:
> On Tue, Dec 30, 2025 at 12:31:17AM +0100, Diederik de Haas wrote:
>> On Sun Dec 28, 2025 at 6:26 PM CET, Andy Shevchenko wrote:
>> > On Sun, Dec 28, 2025 at 12:01:54PM +0100, Diederik de Haas wrote:
>> >> On Sat Dec 27, 2025 at 4:08 PM CET, Andy Shevchenko wrote:
>> >> > On Sun, Dec 21, 2025 at 03:01:11PM +0100, Diederik de Haas wrote:
>> >> >> On Fri Oct 24, 2025 at 4:49 AM CEST, Michael Wu wrote:
>> >> >> > With current extcon_dev_unregister() timing, ida_free is before
>> >> >> > device_unregister(), that may cause current id re-alloc to another
>> >> >> > device in extcon_dev_register() context but sysfs filename path not
>> >> >> > removal completed yet.
>> >> >> 
>> >> >> I periodically get errors like this:
>> >> >> 
>> >> >> [    7.116152] rockchip-usb2phy fe8a0000.usb2phy: error -EEXIST: failed to register extcon device
>> >> >> [    7.117005] rockchip-usb2phy fe8a0000.usb2phy: probe with driver rockchip-usb2phy failed with error -17
>> >> >> 
>> >> >> This was today on a NanoPi R5S (rk3568), but I have seen it before and
>> >> >> on multiple devices. They are all Rockchip based, but that's (quite)
>> >> >> possible because that's what I use the most (and where I pay quite a bit
>> >> >> of attention to dmesg).
>> >> >> 
>> >> >> Slightly fuller dmesg output of the above error is here:
>> >> >> https://paste.sr.ht/~diederik/42c6b3405386c823cd9f837d73a9a32e810361be
>> >> >> And via 'journalctl' I hopefully got the full dmesg output:
>> >> >> https://paste.sr.ht/~diederik/7a6109115b1ad85290de482db091dad3759ec159
>> >> >> 
>> >> >> Here are 2 similar logs, this time on PineTab2 and Quartz64-A (both rk3566)
>> >> >> https://paste.sr.ht/~diederik/cfe606801d3dab0267bea7049687d24c0d6e8d71
>> >> >> 
>> >> >> Those are all the logs I have saved, but it has happened several times
>> >> >> besides that. But most times I just rebooted and didn't save the log.
>> >> >> Today I looked a bit further, found commit 7bba9e81a6fb, searched on
>> >> >> lore.k.o and found this patch.
>> >> >> 
>> >> >> So I'm wondering whether this patch would also fix 'my' issue?
>> >> >> Or is this a different issue?
>> >> >
>> >> > But have you had a chance to test it? The code looks correct after this patch
>> >> > as the first we remove the struct device and all associated data (including
>> >> > sysfs nodes) and then remove the ID from the list. Without that there is a
>> >> > window to get the same id for the devices that is pending to be removed.
>> >> 
>> >> I've build a kernel with it and it will be part of my kernels going
>> >> forward. But it can take many months before the issue shows up again.
>> >> So if I don't get it in a year, then I'm going to *assume* it's probably
>> >> fixed.
>> >
>> > I believe you may speed up the test case by creating two shell scripts that
>> > trying insmod / rmmod the same module in parallel.
>> 
>> I checked on several devices and the phy_rockchip_inno_usb2 module is
>> used by various other modules on all of them, so I can't just 'rmmod' it.
>> I guess it should technically be possible to remove all its dependencies
>> and then remove that module. I don't know if my skills are sufficient
>> for that and it feels like I'm mostly testing the robustness of the USB
>> subsystem.
>> 
>> As I regularly already get the following kernel warning
>> ``Warning! ehci_hcd should always be loaded before uhci_hcd and ohci_hcd, not after``
>> which often results in USB 2 ports not working, I have my doubts about
>> the robustness anyway.
>
> This is interesting issue, have you reported it to USB people (such as Alan Stern)?

Not yet. I was reasonably sure I had mentioned this issue before, but
when doing a search on lore.k.o I couldn't find it. I guess I only
'reported' it on IRC.

Interestingly enough, I did find another very recent mail/patch:
https://lore.kernel.org/all/20251230080014.3934590-1-chenhuacai@loongson.cn/

but looks like Greg doesn't like it ... and his suggestion (make the
kernel module built-in) was exactly the thing I wanted to try before
reporting it. If it's tristate it _should_ be able to be built as
module, but it's extremely unlikely I'll spend time on that though.

>> So I'll just keep an eye out if the ``sysfs: cannot create duplicate
>> filename`` error pops up again 'naturally' and report if/when it does.
>
> Good, please monitor it and try to collect logs for the further investigations,
> if required.

Will do. FWIW, Dragan asked me to enable DEBUG_DEVRES and share the
logs. Maybe they're useful for you too.
On Rock 5B (RK3588): https://paste.sr.ht/~diederik/2d8e0beced6477cc1abd28cd870ea76a1e413a9e
On NanoPi R5S (RK3568): https://paste.sr.ht/~diederik/1c4026d0c277b92391136bab4e80f64a989006f5

Cheers,
  Diederik

_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-rockchip