Re: kern/57199: Pure PVH i386 guests hang on disk activity

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Jul 12, 2024 at 01:17:50PM -0400, Brad Spencer wrote:
> 
> This is still 100% reproducable in 10.0 built from source as of
> 2024-03-28.
> 
> Config file:
> kernel = "/lhome/xen/kernels/NetBSD_10.x/NetBSD_10_branch_20240328/i386/netbsd-GENERIC"
> memory = 512
> cpu_weight = 32
> name = "nbsd10test32bit"
> type="pvh"
> vcpus = 1
> vif = [ 'mac=aa:00:00:50:07:f4, bridge=bridge4' ]
> disk = [ 'phy:/dev/mapper/rustvg0-nbsd10test32bitlv0,0x01,w' ]
> root = "xbd0"
> 
> 
> nbsd10test32bit# uname -a
> NetBSD nbsd10test32bit 10.0 NetBSD 10.0 (GENERIC) #0: Sat Mar 30 22:01:46 EDT 2024  brad%samwise.nat.eldar.org@localhost:/lhome/NetBSD_10_branch_20240328/i386/OBJ/sys/arch/i386/compile/GENERIC i386
> 
> Create a tar archive that is fairly big.  It is likely that the larger
> sets will work, but what I used was a tar archive from the userland of
> the system itself (something like: cd /;tar -cf /var/tmp/distball.tar
> ./bin ./lib ./sbin ./usr ./stand).  I placed this in /var/tmp and called
> it distball.tar
> 
> Create a directory in the path /var/tmp/Q and cd there.
> 
> Untar the archive..  something like:  tar -xvpf ../distball.tar
> 
> The untar will work for a short time and then hang up.  A CTRL-T
> produces this:
> x ./usr/libexec/cc1obj[ 115.0723697] load: 0.15  cmd: tar 1854 [uvnfp2] 0.04u 0.14s 0% 5340k
> [ 116.9523882] load: 0.13  cmd: tar 1854 [uvnfp2] 0.04u 0.14s 0% 5340k
> [ 118.0724972] load: 0.13  cmd: tar 1854 [uvnfp2] 0.04u 0.14s 0% 5340k
> [ 118.6725079] load: 0.13  cmd: tar 1854 [uvnfp2] 0.04u 0.14s 0% 5340k
> 
> 
> If I ran the same test with 1GB of memory it get a little further along,
> but still hangs up:
> x ./usr/libexec/postfix/scache[  66.3212543] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  67.1712694] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  67.3612609] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  68.2412022] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  68.4512385] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  68.6312665] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  68.8012961] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  68.9713106] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> [  69.1213030] load: 0.34  cmd: tar 1977 [biowait] 0.04u 0.20s 0% 5344k
> 
> 
> When the system is hung up, trying to log into the console will stick
> after entering the username.  You can get into ddb from the console.
> 
> A PV+PVSHIM kernel of the same generation of source will work just fine
> with the same test.

I can reproduce a hang of the network interface, which may have the same
cause. Looks like an event is missed by the backend.

What is your dom0 ?

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--