Hi
Op 01-06-2021 om 19:31 schreef Andy Shevchenko:
On Mon, May 31, 2021 at 11:22 PM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
On Mon, May 31, 2021 at 6:45 PM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
On Mon, May 31, 2021 at 5:49 PM Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
On Mon, May 31, 2021 at 05:33:02PM +0300, Andy Shevchenko wrote:
Hi!
The only patch that made v5.13-rc4 effectively broke USB on Intel Merrifield.
My reproducer:
1. Boot with switch in gadget mode
2. Enable USB ethernet
3. Switch to host
4. wait a bit for device enumeration, etc
5. Switch back
6. No gadget mode, kernel RCU stall detected (USB become unusable, so
does the system)
[ 115.792620] rcu: INFO: rcu_sched self-detected stall on CPU
[ 115.798410] rcu: 0-....: (24089 ticks this GP)
idle=886/1/0x4000000000000000 softirq=3796/3797 fqs=5240
[ 115.808333] (t=21000 jiffies g=6505 q=12158)
[ 115.812847] NMI backtrace for cpu 0
[ 115.816472] CPU: 0 PID: 23 Comm: kworker/0:1 Not tainted 5.13.0-rc4+ #213
...
[ 115.977913] add_dma_entry+0xd4/0x1d0
[ 115.981760] dma_map_page_attrs+0xd8/0x220
[ 115.986063] usb_hcd_map_urb_for_dma+0x3b6/0x4f0
[ 115.990895] usb_hcd_submit_urb+0x98/0xbf0
[ 115.995263] dln2_rx+0x1ae/0x280 [dln2]
...
100% Reproducibility, revert of the 25dda9fc56bd ("usb: dwc3: gadget:
Properly track pending and queued SG") fixes the issue.
Please, fix it properly or revert. Thanks!
Care to send the revert?
Sure, But I want to give Thinh a chance to react to this, maybe it
would be a better idea.
Let's say if nothing happens I will send it on Wednesday.
Please, hold on, I have to perform additional testing (something odd
is going on on my side).
Hi Andy I tested my 5.13-rc4 which has only 2 dwc3 related patches on
top: the good old "REVERTME: usb: dwc3: gadget: skip endpoints
ep[18]{in,out}" and "extcon: intel-mrfld: Sync hardware and software
state on init" (will hopefully go in 5.14).
I did not reproduce stalls or oopses.
Dwc3 host/gadget seems to be working nicely now and switching back and
forth without issues in my case (after considerable help from Thinh
earlier).
It appears that the reported patch has nothing to do with the issue.
It "luckily" helped, but under more tests I managed to dive as deep as
v5.11 and got the same issue. I will continue investigating and will
inform / send patch / etc when I find more robust reproducer. Anyway
it's not the reported patch for sure, sorry for the noise.
(It seems I'm looking in a wrong corner< I have some ideas, but it
requires time to check)
I'd be happy to apply additional patches to see if those trigger the
issue if you have suggestions.
In the meanwhile I'll be testing 5.13-rc's just in case.
