lkml.org 
[lkml]   [2012]   [Jun]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2] KSM: numa awareness sysfs knob
On Fri, 29 Jun 2012, Andrew Morton wrote:

> On Fri, 29 Jun 2012 13:49:52 +0200
> Petr Holasek <pholasek@redhat.com> wrote:
>
> > Introduces new sysfs boolean knob /sys/kernel/mm/ksm/merge_nodes
> > which control merging pages across different numa nodes.
> > When it is set to zero only pages from the same node are merged,
> > otherwise pages from all nodes can be merged together (default behavior).
> >
> > Typical use-case could be a lot of KVM guests on NUMA machine
> > and cpus from more distant nodes would have significant increase
> > of access latency to the merged ksm page. Sysfs knob was choosen
> > for higher scalability.
> >
> > Every numa node has its own stable & unstable trees because
> > of faster searching and inserting. Changing of merge_nodes
> > value is possible only when there are not any ksm shared pages in system.
>
> It would be neat to have a knob which enables KSM for all anon
> mappings. ie: pretend that MADV_MERGEABLE is always set. For testing
> coverage purposes.

Interesting idea, I'll try to add it in next release if /sys/kernel/mm/ksm
directory is the right place for such debug knob.

> > --- a/Documentation/vm/ksm.txt
> > +++ b/Documentation/vm/ksm.txt
> > @@ -58,6 +58,12 @@ sleep_millisecs - how many milliseconds ksmd should sleep before next scan
> > e.g. "echo 20 > /sys/kernel/mm/ksm/sleep_millisecs"
> > Default: 20 (chosen for demonstration purposes)
> >
> > +merge_nodes - specifies if pages from different numa nodes can be merged.
> > + When set to 0, ksm merges only pages which physically
> > + resides in the memory area of same NUMA node. It brings
> > + lower latency to access to shared page.
> > + Default: 1
>
> s/resides/reside/.
>
> This doc should mention that /sys/kernel/mm/ksm/run should be zeroed to
> alter merge_nodes. Otherwise confusion will reign.
>

Oh, forgot to mention it. I'll fix it.

> >
> > ...
> >
> > +static ssize_t merge_nodes_store(struct kobject *kobj,
> > + struct kobj_attribute *attr,
> > + const char *buf, size_t count)
> > +{
> > + int err;
> > + unsigned long knob;
> > +
> > + err = kstrtoul(buf, 10, &knob);
> > + if (err)
> > + return err;
> > + if (knob > 1)
> > + return -EINVAL;
> > +
> > + if (ksm_run & KSM_RUN_MERGE)
> > + return -EBUSY;
> > +
> > + mutex_lock(&ksm_thread_mutex);
> > + if (ksm_merge_nodes != knob) {
> > + if (ksm_pages_shared > 0)
> > + return -EBUSY;
> > + else
> > + ksm_merge_nodes = knob;
> > + }
> > + mutex_unlock(&ksm_thread_mutex);
> > +
> > + return count;
> > +}
>
> Seems a bit racy. Shouldn't the test of ksm_run be inside the locked
> region?
>

Agreed.

Thanks for your review!


\
 
 \ /
  Last update: 2012-06-30 15:01    [W:0.094 / U:2.736 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site