Re: 3.4.4/amd64 full interrupt hangs under big nfs copies

[Eric Dumazet <eric.dumazet@gmail.com>]

On Sun, 2012-07-15 at 14:59 -0700, Marc MERLIN wrote:
> On Tue, Apr 10, 2012 at 10:27:33PM -0700, Marc MERLIN wrote:
> > On Tue, Apr 10, 2012 at 08:11:03AM +0200, Eric Dumazet wrote:
> > > Please try following patch, as it solved the problem for me (no more
> > > order-1 allocations in tx path)
> > 
> > I applied our patch to 3.3.1 and cannot reproduce the problem anymore.
> > 
> > I'll leave a big wireless copy running overnight just in case, but I think
> > you fixed it.
> 
> Mmmh, so I'm running 3.4.4 and I had another full machine hang while copying
> big files (gigabytes) over wireless via NFS.
> The laptop self recovered after 5mn or so (mouse cursor would not even
> move) and I was able to kill -9 the process (midnight commander).
> mc did not actually stop for another 4mn or so (i.e. it took that long for
> the process to come out of kernel hung state), but the machine was usable
> during that time.
> Note that copying the same data with scp works fine.
> NFS mount looks like this:
> gargamel:/mnt/dshelf2/ /net/gargamel/mnt/dshelf2 nfs4 rw,nosuid,nodev,relatime,vers=4.0,rsize=524288,wsize=524288,namlen=255,hard,proto=tcp,port=0,timeo=600,retrans=2,sec=sys,clientaddr=192.168.205.7,local_lock=none,addr=192.168.205.3 0 0
> 
> I didn't have anything like last time in the kernel logs, and more
> annoyingly, ps -elf does not show anything for any process in WCHAN,
> making pointing the finger a bit harder (procps-ng 3.3.3 does not show
> anything other than '-' in WCHAN for any process with 3.4.4).
> 
> My understanding is that user space calling drivers that shut off all
> interrupts for extended periods of time (as least I think so since my mouse
> cursor would not move), is still a kernel bug.
> 
> For what it's worth, copying 1GB of data in lots of small files does not
> cause problems, it seems that it's big files that cause a problem since they
> likely fill a buffer somewhere while interrupts are disabled.
> 
> Do you have an idea of how I can find out where my mc process is stuck in
> the kernel?
> Should I reproduce with specific sysrq output?

Just to clarify, you get this freeze when transferring a big file from a
remote NFS server to your PC, (aka a download), not the reverse way ?

If so, you might hit OOM condition because iwlwifi uses big/fat RX
buffers, I never understood why...

(amsdu_size_8K = 1)

Storing an MTU=1500 frams in 8KB of memory sounds really bad.


diff --git a/drivers/net/wireless/iwlwifi/iwl-drv.c b/drivers/net/wireless/iwlwifi/iwl-drv.c
index cc41cfa..434b924 100644
--- a/drivers/net/wireless/iwlwifi/iwl-drv.c
+++ b/drivers/net/wireless/iwlwifi/iwl-drv.c
@@ -1006,7 +1006,7 @@ void iwl_drv_stop(struct iwl_drv *drv)
 
 /* shared module parameters */
 struct iwl_mod_params iwlwifi_mod_params = {
-	.amsdu_size_8K = 1,
+	.amsdu_size_8K = 0,
 	.restart_fw = 1,
 	.plcp_check = true,
 	.bt_coex_active = true,