Re: usb: cdc-acm: BUG kmalloc-128 Poison overwritten

From: Bruno Thomsen
Date: Fri Feb 26 2021 - 09:16:45 EST


Den tor. 25. feb. 2021 kl. 10.57 skrev Oliver Neukum <oneukum@xxxxxxxx>:
>
> Am Mittwoch, den 24.02.2021, 16:21 +0100 schrieb Bruno Thomsen:
>
> Hi,
>
> > No, this is not a regression from 5.10. It seems that many attempts to
> > fix cdc-acm in the 5.x kernel series have failed to fix the root cause of
> > these oops. I have not seen this on 4.14 and 4.19, but I have observed
> > it on at least 5.3 and newer kernels in slight variations.
> > I guess this is because cdc-acm is very common in the embedded
> > ARM world and rarely used on servers or laptops. Combined with
> > ARM devices still commonly use 4.x LTS kernels. Not sure if
> > hardening options on the kernel has increased change of reproducing
> > oops.
>
> OK, so this is not an additional problem.
> According to your logs, an URB that should have been killed wasn't.

Thanks for looking into this bug rapport.

> > I am ready to test new patches and will continue to report oops
>
> Could you test the attached patches?

Yes, I am already running tests on the patches.
I have not seen any oops yet and it seems the USB cdc-acm driver is still
working as intended.

The only notable trace I have seen is this new error from the cdc-acm driver
but everything kept on working.
kernel: cdc_acm 1-1.1:1.7: acm_start_wb - usb_submit_urb(write bulk) failed: -19

Other then that I see this common error (should probably be a warning) during
device enumeration:
kernel: cdc_acm 1-1.2:1.0: failed to set dtr/rts

I will post an update next week when the patches have survived some
more runtime.

/Bruno