Re: deadlock on flt_noram5

[yamt%mwd.biglobe.ne.jp@localhost (YAMAMOTO Takashi)]

> Hi
> 
> I experience kernel hangs, which seems to be related to memory getting scarce.
> getty on console is unresponsive, but I least I can drop into ddb. 

"show uvm" might be interesting.

> 
> db> ps
> PID    LID S CPU     FLAGS       STRUCT LWP *               NAME WAIT
> 27030    1 3   0         4           cbc6fa80               cron flt_noram5
> 24850    1 3   0         4           cbc8d840                 sh flt_noram5
> 24643    1 3   0         4           cbc6f300                 ld flt_noram5
> (...)
> 8058     1 3   0         4           cb29c280                 sh flt_noram5
> (...)
> 516      1 3   0         4           cb0470c0               cron flt_noram5
> (...)
> 229      1 3   0         4           cac4f580           perfused flt_noram5

it can be a deadlock if this "perfused" thread is effectively blocking
the pagedaemon's work.

> 
> Reading the sources, it seems that we sleep on flt_noram5 when we are out of
> memory, but that nothing in the kernel is there to awake us from flt_noram5,
> both on -current and netbsd-5

"wakeup(&uvmexp.free)" will wake it up.

YAMAMOTO Takashi

> 
> Is that a known problem?
> 
> -- 
> Emmanuel Dreyfus
> http://hcpnet.free.fr/pubz
> manu%netbsd.org@localhost