Re: [PATCH] net/skbuff: silence warnings under memory pressure

Cc printk maintainers

On Tue 03-09-19 17:42:28, Qian Cai wrote:

quoted

quoted

I suppose what happens is those skb_build() allocations are from softirq,
and
once one of them failed, it calls printk() which generates more interrupts.
Hence, the infinite loop.

Please elaborate more.

If you look at the original report, the failed allocation dump_stack() is,

 <IRQ>
 warn_alloc.cold.43+0x8a/0x148
 __alloc_pages_nodemask+0x1a5c/0x1bb0
 alloc_pages_current+0x9c/0x110
 allocate_slab+0x34a/0x11f0
 new_slab+0x46/0x70
 ___slab_alloc+0x604/0x950
 __slab_alloc+0x12/0x20
 kmem_cache_alloc+0x32a/0x400
 __build_skb+0x23/0x60
 build_skb+0x1a/0xb0
 igb_clean_rx_irq+0xafc/0x1010 [igb]
 igb_poll+0x4bb/0xe30 [igb]
 net_rx_action+0x244/0x7a0
 __do_softirq+0x1a0/0x60a
 irq_exit+0xb5/0xd0
 do_IRQ+0x81/0x170
 common_interrupt+0xf/0xf
 </IRQ>

Since it has no __GFP_NOWARN to begin with, it will call,

printk
  vprintk_default
    vprintk_emit
      wake_up_klogd
        irq_work_queue
          __irq_work_queue_local
            arch_irq_work_raise
              apic->send_IPI_self(IRQ_WORK_VECTOR)
                smp_irq_work_interrupt
                  exiting_irq
                    irq_exit

and end up processing pending net_rx_action softirqs again which are plenty due
to connected via ssh etc.

-- 
Michal Hocko
SUSE Labs