URL:
http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.6/2.6.5-aa5.gz
http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.6/2.6.5-aa5/
Changelog diff between 2.6.5-aa4 and 2.6.5-aa5:
Only in 2.6.5-aa5: nonlinear-can-do-mlock
Avoid truncate vs nonlinear BUG mentioned by Hugh Dickins on l-k.
Give nonlinear only to privilegied users that knows how to avoid
breaking the kernel. If local security doesn't matter disable-cap-mlock
can be enabled and it'll give mlock and remap-file-pages too to every
user. truncate must be fixed anyways to increase the strength of
disable-cap-mlock but it gets low priority after moving the thing under
sysctl, and I plan to keep the nonlinear hack forever under the sysctl
for VM robusteness in providing _fair_ swapping to all processes.
BTW, the mmap(MAP_POPULATE) API is horrible, there's no way to catch
a failure with it, it doesn't return any retval, and I cannot fix it
in my tree or I become ABI/API incompatible with linux 2.6, that's just
a broken API that wants to be fixed properly in mainline despite the
break of backwards compatibility.
> Changelog diff between 2.6.5-aa4 and 2.6.5-aa5:
But its aa4, yet. ;-)
+EXTRAVERSION =-aa4
I will do today latency tests with this aa5, mc3 (with cfq and antic.
scheduler) and the vanilla 2.6.5 kernel with latencytest-0.5.2 from
http://www.alsa-project.org/~iwai/alsa.html#LatencyTest
I love after the stress test browsing the graphical output with firefox. :-)
best regards,
Marcus