Re: "usbip: Implement SG support to vhci-hcd and stub driver" causes a deadlock

["Marek Marczykowski-Górecki" <marmarek@invisiblethingslab.com>]

[-- Attachment #1: Type: text/plain, Size: 4274 bytes --]

On Sat, Dec 07, 2019 at 01:58:09AM +0100, Marek Marczykowski-Górecki wrote:
> On Fri, Dec 06, 2019 at 02:12:08PM -0700, Shuah Khan wrote:
> > On 12/6/19 1:57 PM, Marek Marczykowski-Górecki wrote:
> > > On Fri, Dec 06, 2019 at 03:50:58PM +0900, Suwan Kim wrote:
> > > > On Fri, Dec 06, 2019 at 04:24:06AM +0100, Marek Marczykowski-Górecki wrote:
> > > > > Hello,
> > > > > 
> > > > > I've hit an issue with recent 4.19 and 5.4 kernels. In short: if I
> > > > > connect Yubikey 4 and use its CCID interface (for example `ykman oath
> > > > > list` command), the client side hangs (100% reliably). After 60s I get a
> > > > > message that a CPU hangs waiting for a spinlock (see below).
> > > > > 
> > > > > I've bisected it to a ea44d190764b4422af ("usbip: Implement SG support
> > > > > to vhci-hcd and stub driver") commit. Which indeed is also backported to
> > > > > 4.19.
> > > > > 
> > > > > Any idea what is going on here? I can easily provide more information,
> > > > > if you tell me how to get it.
> > > > > 
> > > > 
> > > > Hi,
> > > > 
> > > > Thanks for reporting. Could you turn on lockdep and USBIP_DEBUG
> > > > in kernel config and send dmesg log? It will be helpful to figure
> > > > out lock dependency in vhci_hcd.
> > > 
> > > Hmm, I've tried, but I don't see much more information there (see
> > > below). I've just enabled PROVE_LOCKING and USBIP_DEBUG. Do I need to do
> > > anything more, like some boot option?
> > > 
> > > Also, this one (as the previous one) is from 4.19.84. Interestingly, on
> > > 4.19.87 I don't get the message at all.
> > > 
> > > Hmm, I've done also another test: got 4.19.84 with "usbip: Implement SG
> > > support to vhci-hcd and stub driver" reverted and it still hangs...
> > > 
> > > I'm going for another bisect round (4.19.81 works, 4.19.84 doesn't).
> > 
> > Does 4.19.82 work?
> 
> Yes, it does, or at least something after 4.19.82 works (9da271c1c).

I've done bisect again and again got into the same commit. Most of the
time I didn't get any message about the deadlock, it simply hanged.

Full bisect log:

git bisect start
# good: [ef244c3088856cf048c77231653b4c92a7b2213c] Linux 4.19.81
git bisect good ef244c3088856cf048c77231653b4c92a7b2213c
# bad: [c555efaf14026c7751fa68d87403a5eb5ae7dcaf] Linux 4.19.84
git bisect bad c555efaf14026c7751fa68d87403a5eb5ae7dcaf
# good: [9da271c1cdc14839b694e23889a653c1ed0b5f8f] net: usb: lan78xx: Disable interrupts before calling generic_handle_irq()
git bisect good 9da271c1cdc14839b694e23889a653c1ed0b5f8f
# bad: [8181146cd7de890cdfdda68ddc3730250887d7fc] PCI: tegra: Enable Relaxed Ordering only for Tegra20 & Tegra30
git bisect bad 8181146cd7de890cdfdda68ddc3730250887d7fc
# good: [8e6bf4bc3a88e4b84e5c4ec50143a71a61503336] mm: memcontrol: fix network errors from failing __GFP_ATOMIC charges
git bisect good 8e6bf4bc3a88e4b84e5c4ec50143a71a61503336
# good: [0327c7818da27b018464c0b9f541c5d276c57172] can: flexcan: disable completely the ECC mechanism
git bisect good 0327c7818da27b018464c0b9f541c5d276c57172
# good: [5e36cf8edb5812e378b57511263d1a0a9172eeb9] configfs: fix a deadlock in configfs_symlink()
git bisect good 5e36cf8edb5812e378b57511263d1a0a9172eeb9
# good: [4f6c5200269998f0066e1bd4db931297b7b2b906] ALSA: usb-audio: Fix possible NULL dereference at create_yamaha_midi_quirk()
git bisect good 4f6c5200269998f0066e1bd4db931297b7b2b906
# good: [502bd151448c2c76a927b26783e5538875c534ff] sched/fair: Fix low cpu usage with high throttling by removing expiration of cpu-local slices
git bisect good 502bd151448c2c76a927b26783e5538875c534ff
# good: [f865ae473c16fb2b8b8601fa04f4f6517ad557b3] usbip: Fix vhci_urb_enqueue() URB null transfer buffer error path
git bisect good f865ae473c16fb2b8b8601fa04f4f6517ad557b3
# bad: [e2dd254bde5cdac24e7774584d6f3c2c61fe09e5] usbip: Implement SG support to vhci-hcd and stub driver
git bisect bad e2dd254bde5cdac24e7774584d6f3c2c61fe09e5
# first bad commit: [e2dd254bde5cdac24e7774584d6f3c2c61fe09e5] usbip: Implement SG support to vhci-hcd and stub driver


-- 
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]