Re: Urgent Bug report: PPPoE ioctl(PPPIOCCONNECT): Transport endpoint is not connected

and this : 

  PerfTop:   26378 irqs/sec  kernel:61.4%  exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles],  (all, 12 CPUs)
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

     5.65%  libfrr.so.0.0.0   [.] 0x00000000000ce970
     5.56%  [kernel]          [k] osq_lock
     5.22%  [kernel]          [k] mutex_spin_on_owner
     3.66%  [pppoe]           [k] pppoe_flush_dev
     3.01%  libfrr.so.0.0.0   [.] 0x00000000000ce94e
     1.98%  libc.so.6         [.] 0x00000000000965a2
     1.84%  libc.so.6         [.] 0x0000000000186afa
     1.55%  libc.so.6         [.] 0x0000000000186e57
     1.54%  zebra             [.] 0x00000000000b9271
     1.46%  zebra             [.] 0x00000000000b91f1
     1.46%  libteam.so.5.6.1  [.] 0x0000000000006470
     1.44%  libc.so.6         [.] 0x00000000000965a0
     1.30%  libteam.so.5.6.1  [.] 0x0000000000009e7d
     1.08%  [kernel]          [k] fib_table_flush
     1.02%  libc.so.6         [.] 0x0000000000186eac
     0.93%  [kernel]          [k] do_poll.constprop.0
     0.85%  libc.so.6         [.] 0x0000000000186afe
     0.80%  dtvbras           [.] 0x0000000000014be8
     0.78%  [kernel]          [k] queued_read_lock_slowpath
     0.72%  [kernel]          [k] next_uptodate_page
     0.64%  [kernel]          [k] zap_pte_range
     0.64%  bgpd              [.] 0x0000000000070a46
     0.61%  [kernel]          [k] fib_table_insert

On 15 Sep 2021, at 17:25, Martin Zaharinov [off-list ref] wrote:

Hey Florian

make test in lab and look much better that before.

see this perf 

PerfTop:    6551 irqs/sec  kernel:77.8%  exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles],  (all, 12 CPUs)
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

   15.70%  [ixgbe]           [k] ixgbe_read_reg
   13.33%  [kernel]          [k] mutex_spin_on_owner
    7.65%  [kernel]          [k] osq_lock
    2.85%  libfrr.so.0.0.0   [.] 0x00000000000ce970
    1.94%  libfrr.so.0.0.0   [.] 0x00000000000ce94e
    1.19%  libc.so.6         [.] 0x0000000000186afa
    1.15%  [kernel]          [k] do_poll.constprop.0
    0.99%  [kernel]          [k] inet_dump_ifaddr
    0.94%  libteam.so.5.6.1  [.] 0x0000000000006470
    0.79%  libc.so.6         [.] 0x0000000000186e57
    0.71%  [ixgbe]           [k] ixgbe_update_mc_addr_list_generic
    0.65%  [kernel]          [k] __fget_files
    0.61%  [kernel]          [k] sock_poll
    0.57%  libteam.so.5.6.1  [.] 0x0000000000009e7d
    0.51%  perf              [.] 0x00000000000bc7b3
    0.51%  libteam.so.5.6.1  [.] 0x0000000000006501
    0.48%  [kernel]          [k] next_uptodate_page
    0.46%  [kernel]          [k] _raw_read_lock_bh
    0.43%  libc.so.6         [.] 0x0000000000186eac
    0.42%  bgpd              [.] 0x0000000000070a46
    0.41%  [pppoe]           [k] pppoe_flush_dev
    0.39%  [kernel]          [k] zap_pte_range


This happened when remove and add new interface on time of drop and reconnect users.


now : ip a command work fine !


Martin

quoted

On 14 Sep 2021, at 14:00, Florian Westphal [off-list ref] wrote:

Martin Zaharinov [off-list ref] wrote:

[ Trimming CC list ]

quoted

Florian: 

If you make patch send to test please.

Attached.  No idea if it helps, but 'ip' should stay responsive
even when masquerade processes netdevice events.
<defer_masq_work.diff>