Vendor merge review: ZFS I/O deadman thread

[Xin Li]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 2/23/13 4:57 AM, Martin Matuska wrote:
> Please review the following proposed merge from vendor.
> 
> The patch uses callout_drain(), I am not sure if callout_stop() is 
> safe in this case. As of the patch, the deadman thread is disabled 
> by default. Should we enable it?
> 
> Thank you.
> 
> http://people.freebsd.org/~mm/patches/zfs/zfs_deadman.patch
> 
> Link to vendor ZFS issue: 3246 ZFS I/O deadman thread 
> https://www.illumos.org/issues/3246

(In the patch I didn't see callout_stop, I think you have already
changed them to callout_drain per Davide's comments?  I don't think
this requires high resolution callout as it's a watchdog type thing)

I think this should only be enabled when the system is on physical
system so I think we could probably leave the default there and check
it on module load (something like, if (zfs_deadman_enabled == -1)
zfs_deadman_enabled = (vm_guest == 0)? 1:0;)

The other parts seems fine.

Cheers,
-----BEGIN PGP SIGNATURE-----

iQEcBAEBCAAGBQJRKpKcAAoJEG80Jeu8UPuz3M0H/RDaXuocJ3DYiqomGCnuwBwd
5XBjtpCKTdo2JfVPi0Xz4miqk5Cxd6ET27clmNgohgGGOrbBgretr9mwiVboO+nf
V1Qt67ULabtAtOCElMO9ZaI5T2autt0ecrlMiHxXkMs29gpRAvREU7gPqoL16p8r
LeCNBpbOIiXJXWxRJyY8x4FbmbgFI4SK4Co7t5dXz+UQJM5MfA5S4qGHBhqJfJXx
x3+m2J99QNKeCvUDrp18DI2Y7xpv4nTiX7Xx81mfFE8PbMG0ItVjwkknQElxF8ZL
H4CZq/OWB6tEcF9K3hMvOI4yodO9sIG5+s+jKUwKJU/l5TYGvBPc140bL9JisWk=
=CMDj
-----END PGP SIGNATURE-----