On (09/17/15 08:21), Vlastimil Babka wrote:
> On 09/15/2015 06:22 AM, Sergey Senozhatsky wrote:
> >On (09/15/15 00:08), Dan Streetman wrote:
> >[..]
> >
> >correct. a bit of internals: we don't scan all the zspages every
> >time. each class has stats for allocated used objects, allocated
> >used objects, etc. so we 'compact' only classes that can be
> >compacted:
> >
> > static unsigned long zs_can_compact(struct size_class *class)
> > {
> > unsigned long obj_wasted;
> >
> > obj_wasted = zs_stat_get(class, OBJ_ALLOCATED) -
> > zs_stat_get(class, OBJ_USED);
> >
> > obj_wasted /= get_maxobj_per_zspage(class->size,
> > class->pages_per_zspage);
> >
> > return obj_wasted * class->pages_per_zspage;
> > }
> >
> >if we can free any zspages (which is at least one page), then we
> >attempt to do so.
> >
> >is compaction the root cause of the symptoms Vitaly observe?
>
> He mentioned the "compact_stalls" counter which in /proc/vmstat is for the
> traditional physical memory compaction, not the zsmalloc-specific one. Which
> would imply high-order allocations. Does zsmalloc try them first before
> falling back to the order-0 zspages linked together manually?
each zspage is a bunch (pages_per_zspage) of alloc_page() calls
for (i = 0; i < class->pages_per_zspage; i++) {
struct page *page;
page = alloc_page(flags);
if (!page)
goto cleanup;
INIT_LIST_HEAD(&page->lru);
if (i == 0) { /* first page */
SetPagePrivate(page);
set_page_private(page, 0);
first_page = page;
first_page->inuse = 0;
}
if (i == 1)
set_page_private(first_page, (unsigned long)page);
if (i >= 1)
set_page_private(page, (unsigned long)first_page);
if (i >= 2)
list_add(&page->lru, &prev_page->lru);
if (i == class->pages_per_zspage - 1) /* last page */
SetPagePrivate2(page);
prev_page = page;
}
-ss
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
