2019-11-02 14:20:33

by hui yang

Subject: [PATCH] mm: There should have an unit (kB)

From: YangHui <[email protected]>

- printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
+ printk(KERN_CONT " %ldkB", zone->lowmem_reserve[i]);
Make it look more perfect

Signed-off-by: YangHui <[email protected]>
---
mm/page_alloc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index ecc3dba..ee5043a 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5389,7 +5389,7 @@ void show_free_areas(unsigned int filter, nodemask_t *nodemask)
K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
printk("lowmem_reserve[]:");
for (i = 0; i < MAX_NR_ZONES; i++)
- printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
+ printk(KERN_CONT " %ldkB", zone->lowmem_reserve[i]);
printk(KERN_CONT "\n");
}

--
2.7.4

2019-11-02 20:57:39

by Matthew Wilcox

Subject: Re: [PATCH] mm: There should have an unit (kB)

On Sat, Nov 02, 2019 at 10:18:07PM +0800, hui yang wrote:
> From: YangHui <[email protected]>
>
> - printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
> + printk(KERN_CONT " %ldkB", zone->lowmem_reserve[i]);
> Make it look more perfect

I don't think this is accounted in kilobytes though. Isn't it the
number of pages?

2019-11-02 23:56:06

by David Rientjes

Subject: Re: [PATCH] mm: There should have an unit (kB)

On Sat, 2 Nov 2019, Matthew Wilcox wrote:

> On Sat, Nov 02, 2019 at 10:18:07PM +0800, hui yang wrote:
> > From: YangHui <[email protected]>
> >
> > - printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
> > + printk(KERN_CONT " %ldkB", zone->lowmem_reserve[i]);
> > Make it look more perfect
>
> I don't think this is accounted in kilobytes though. Isn't it the
> number of pages?
>

Yes, it's not in KB.

Speaking of lowmem_reserve, however, the current default of 1/256 for both
ZONE_DMA and ZONE_DMA32 seem too large as memory capacities of nodes
increase. We tune this to be much smaller so that we don't have as much
memory set aside only for GFP_DMA or GFP_DMA32 allocations for that reason
and because there is less reliance on lowmem for our configurations.

I'm wondering if the default should either be 1/256 up to a bounded memory
capacity and then the excess is disregarded or whether the default itself
should be changed to, say, 1024.

Looping in Johannes who may also have an opinion on this.