On Tue, Jan 29, 2019 at 09:22:43AM +0000, Guillaume Gardet wrote:
On Thu, 10 Jan 2019 10:50:59 +0100, Guillaume GARDET wrote:
as we need to have low latency on some systems.
Actually, it's an open question whether we still want to keep CONFIG_PREEMPT or not on other archs, too. CONFIG_PREEMPT hits performance significantly.
I don't know much about other distros, but IIRC, Fedora is using CONFIG_PREEMPT_VOLUNTARY instead. Ubuntu provide both non-
On Thursday, 10 January 2019 11:02 Takashi Iwai wrote: preempt and
CONFIG_PREEMPT versions? Correct me if wrong.
We used to have two flavors: -default (with preemption disabled) and - desktop (preemptive). There were also other differences (e.g. -desktop had HZ=1000, IIRC) but over the time, the only really important one was the preemption. At some point we decided to change -default to preemptive and drop -desktop.
Before changing anything we might want to check the discussions leading to this decision (in particular, Mike Galbraith's insight).
Could we move forward on this topic?
PREEMPT is the upstream default for aarch64 since Nov-13 and previous value was PREEMPT_VOLUNTARY. Aarch64 is not server only, but also desktop (and laptop). If PREEMPT is a problem for you, we should probably switch to PREEMPT_VOLUNTARY instead. But PREEMPT_NONE is not a good choice for aarch64, IMO.
I'm not opposed to changing the aarch64 config in principle, what I don't like is the reasoning. We rarely care about "upstream defaults" (i.e. the defconfigs in kernel git) and, to be honest, I'm a bit surprised that someone still does. If the reasoning is that we want the same value for all architectures, that makes perfect sense to me. If the reasoning is that aarch64 is specific in some way so that it should have a value different from other architectures, I'm fine with it as long as our ARM engineers ack such claim. Michal Kubecek -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org