Re: [PATCH] eventfd: Enlarge recursion limit to allow vhost to work

[Juri Lelli]

Hi,

On 06/07/20 08:45, Juri Lelli wrote:
> On 03/07/20 19:11, He Zhe wrote:
> > 
> > 
> > On 7/3/20 4:12 PM, Juri Lelli wrote:
> > > Hi,
> > >
> > > On 10/04/20 19:47, zhe.he@xxxxxxxxxxxxx wrote:
> > >> From: He Zhe <zhe.he@xxxxxxxxxxxxx>
> > >>
> > >> commit b5e683d5cab8 ("eventfd: track eventfd_signal() recursion depth")
> > >> introduces a percpu counter that tracks the percpu recursion depth and
> > >> warn if it greater than zero, to avoid potential deadlock and stack
> > >> overflow.
> > >>
> > >> However sometimes different eventfds may be used in parallel. Specifically,
> > >> when heavy network load goes through kvm and vhost, working as below, it
> > >> would trigger the following call trace.
> > >>
> > >> -  100.00%
> > >>    - 66.51%
> > >>         ret_from_fork
> > >>         kthread
> > >>       - vhost_worker
> > >>          - 33.47% handle_tx_kick
> > >>               handle_tx
> > >>               handle_tx_copy
> > >>               vhost_tx_batch.isra.0
> > >>               vhost_add_used_and_signal_n
> > >>               eventfd_signal
> > >>          - 33.05% handle_rx_net
> > >>               handle_rx
> > >>               vhost_add_used_and_signal_n
> > >>               eventfd_signal
> > >>    - 33.49%
> > >>         ioctl
> > >>         entry_SYSCALL_64_after_hwframe
> > >>         do_syscall_64
> > >>         __x64_sys_ioctl
> > >>         ksys_ioctl
> > >>         do_vfs_ioctl
> > >>         kvm_vcpu_ioctl
> > >>         kvm_arch_vcpu_ioctl_run
> > >>         vmx_handle_exit
> > >>         handle_ept_misconfig
> > >>         kvm_io_bus_write
> > >>         __kvm_io_bus_write
> > >>         eventfd_signal
> > >>
> > >> 001: WARNING: CPU: 1 PID: 1503 at fs/eventfd.c:73 eventfd_signal+0x85/0xa0
> > >> ---- snip ----
> > >> 001: Call Trace:
> > >> 001:  vhost_signal+0x15e/0x1b0 [vhost]
> > >> 001:  vhost_add_used_and_signal_n+0x2b/0x40 [vhost]
> > >> 001:  handle_rx+0xb9/0x900 [vhost_net]
> > >> 001:  handle_rx_net+0x15/0x20 [vhost_net]
> > >> 001:  vhost_worker+0xbe/0x120 [vhost]
> > >> 001:  kthread+0x106/0x140
> > >> 001:  ? log_used.part.0+0x20/0x20 [vhost]
> > >> 001:  ? kthread_park+0x90/0x90
> > >> 001:  ret_from_fork+0x35/0x40
> > >> 001: ---[ end trace 0000000000000003 ]---
> > >>
> > >> This patch enlarges the limit to 1 which is the maximum recursion depth we
> > >> have found so far.
> > >>
> > >> Signed-off-by: He Zhe <zhe.he@xxxxxxxxxxxxx>
> > >> ---
> > > Not sure if this approch can fly, but I also encountered the same
> > > warning (which further caused hangs during VM install) and this change
> > > addresses that.
> > >
> > > I'd be interested in understanding what is the status of this problem/fix.
> > 
> > This is actually v2 of the patch and has not got any reply yet. Here is the v1. FYI.
> > https://lore.kernel.org/lkml/1586257192-58369-1-git-send-email-zhe.he@xxxxxxxxxxxxx/
> 
> I see, thanks. Hope this gets reviewed soon! :-)
> 
> > > On a side note, by looking at the code, I noticed that (apart from
> > > samples) all callers don't actually check eventfd_signal() return value
> > > and I'm wondering why is that the case and if is it safe to do so.
> > 
> > Checking the return value right after sending the signal can tell us if the
> > event counter has just overflowed, that is, exceeding ULLONG_MAX. I guess the
> > authors of the callers listed in the commit log just don't worry about that,
> > since they add only one to a dedicated eventfd.
> 
> OK. I was mostly wondering if returning early in case the WARN_ON_ONCE
> fires would cause a missing wakeup for the eventfd_ctx wait queue.

Gentle ping about this issue (mainly addressing relevant maintainers and
potential reviewers). It's easily reproducible with PREEMPT_RT.

Thanks,

Juri