weird DirectMap2M accounting.

weird DirectMap2M accounting.

[Dave Jones]

I just spotted this in /proc/meminfo on an old Core2 machine with 4G.

DirectMap2M:    18446744073709543424 kB

Looks like we subtracted 8192 from 0 somewhere.

Should split_page_count() be checking that direct_pages_count > 0 ?

	Dave

Re: weird DirectMap2M accounting.

[Dave Jones]

On Wed, Jan 06, 2016 at 06:55:27PM -0500, Dave Jones wrote:
 > I just spotted this in /proc/meminfo on an old Core2 machine with 4G.
 > 
 > DirectMap2M:    18446744073709543424 kB
 > 
 > Looks like we subtracted 8192 from 0 somewhere.
 > 
 > Should split_page_count() be checking that direct_pages_count > 0 ?

Ok, this diff makes that number print out as 0.

If this looks ok, I'll submit it properly, though I'd like to better
understand what's happening here. Shouldn't I have 2M pages ?

	Dave


diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c
index a3137a4feed1..ff0e0c6c350e 100644
--- a/arch/x86/mm/pageattr.c
+++ b/arch/x86/mm/pageattr.c
@@ -66,6 +66,9 @@ void update_page_count(int level, unsigned long pages)
 
 static void split_page_count(int level)
 {
+	if (direct_pages_count[level] == 0)
+		return;
+
 	direct_pages_count[level]--;
 	direct_pages_count[level - 1] += PTRS_PER_PTE;
 }
 

Re: weird DirectMap2M accounting.

[Dave Jones]

On Wed, Jan 06, 2016 at 07:11:41PM -0500, Dave Jones wrote:
 > On Wed, Jan 06, 2016 at 06:55:27PM -0500, Dave Jones wrote:
 >  > I just spotted this in /proc/meminfo on an old Core2 machine with 4G.
 >  > 
 >  > DirectMap2M:    18446744073709543424 kB
 >  > 
 >  > Looks like we subtracted 8192 from 0 somewhere.
 >  > 
 >  > Should split_page_count() be checking that direct_pages_count > 0 ?
 > 
 > Ok, this diff makes that number print out as 0.
 > 
 > If this looks ok, I'll submit it properly, though I'd like to better
 > understand what's happening here. Shouldn't I have 2M pages ?

Ah. We disable them when compiling with DEBUG_PAGEALLOC.
So this is only needed in that case.
If people think it's worth an ifdef for those two added lines I'll add it,
but it seems pretty benign to just always check.

thoughts?

	Dave

Re: weird DirectMap2M accounting.

[Thomas Gleixner]

On Wed, 6 Jan 2016, Dave Jones wrote:
> On Wed, Jan 06, 2016 at 07:11:41PM -0500, Dave Jones wrote:
>  > On Wed, Jan 06, 2016 at 06:55:27PM -0500, Dave Jones wrote:
>  >  > I just spotted this in /proc/meminfo on an old Core2 machine with 4G.
>  >  > 
>  >  > DirectMap2M:    18446744073709543424 kB
>  >  > 
>  >  > Looks like we subtracted 8192 from 0 somewhere.
>  >  > 
>  >  > Should split_page_count() be checking that direct_pages_count > 0 ?
>  > 
>  > Ok, this diff makes that number print out as 0.
>  > 
>  > If this looks ok, I'll submit it properly, though I'd like to better
>  > understand what's happening here. Shouldn't I have 2M pages ?
> 
> Ah. We disable them when compiling with DEBUG_PAGEALLOC.
> So this is only needed in that case.
> If people think it's worth an ifdef for those two added lines I'll add it,
> but it seems pretty benign to just always check.

Agreed.

	tglx