2023-10-10 14:56:13

by Matthew Wilcox

Subject: [PATCH] bounds: Support non-power-of-two CONFIG_NR_CPUS

ilog2() rounds down, so for example when PowerPC 85xx sets CONFIG_NR_CPUS
to 24, we will only allocate 4 bits to store the number of CPUs instead
of 5. Use bits_per() instead, which rounds up. Found by code inspection.
The effect of this would probably be a misaccounting when doing NUMA
balancing, so to a user, it would only be a performance penalty.
The effects may be more wide-spread; it's hard to tell.

Signed-off-by: Matthew Wilcox (Oracle) <[email protected]>
Fixes: 90572890d202 ("mm: numa: Change page last {nid,pid} into {cpu,pid}")
Cc: Peter Zijlstra <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Rik van Riel <[email protected]>
---
kernel/bounds.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/bounds.c b/kernel/bounds.c
index b529182e8b04..c5a9fcd2d622 100644
--- a/kernel/bounds.c
+++ b/kernel/bounds.c
@@ -19,7 +19,7 @@ int main(void)
DEFINE(NR_PAGEFLAGS, __NR_PAGEFLAGS);
DEFINE(MAX_NR_ZONES, __MAX_NR_ZONES);
#ifdef CONFIG_SMP
- DEFINE(NR_CPUS_BITS, ilog2(CONFIG_NR_CPUS));
+ DEFINE(NR_CPUS_BITS, bits_per(CONFIG_NR_CPUS));
#endif
DEFINE(SPINLOCK_SIZE, sizeof(spinlock_t));
#ifdef CONFIG_LRU_GEN
--
2.40.1

2023-10-10 15:20:06

by Rik van Riel

Subject: Re: [PATCH] bounds: Support non-power-of-two CONFIG_NR_CPUS

On Tue, 2023-10-10 at 15:55 +0100, Matthew Wilcox (Oracle) wrote:
> ilog2() rounds down, so for example when PowerPC 85xx sets
> CONFIG_NR_CPUS
> to 24, we will only allocate 4 bits to store the number of CPUs
> instead
> of 5.  Use bits_per() instead, which rounds up.  Found by code
> inspection.
> The effect of this would probably be a misaccounting when doing NUMA
> balancing, so to a user, it would only be a performance penalty.
> The effects may be more wide-spread; it's hard to tell.
>
> Signed-off-by: Matthew Wilcox (Oracle) <[email protected]>
> Fixes: 90572890d202 ("mm: numa: Change page last {nid,pid} into
> {cpu,pid}")
> Cc: Peter Zijlstra <[email protected]>
> Cc: Mel Gorman <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Rik van Riel <[email protected]>

Reviewed-by: Rik van Riel <[email protected]>

--
All Rights Reversed.

2023-10-10 15:31:45

by Mel Gorman

Subject: Re: [PATCH] bounds: Support non-power-of-two CONFIG_NR_CPUS

On Tue, Oct 10, 2023 at 03:55:49PM +0100, Matthew Wilcox (Oracle) wrote:
> ilog2() rounds down, so for example when PowerPC 85xx sets CONFIG_NR_CPUS
> to 24, we will only allocate 4 bits to store the number of CPUs instead
> of 5. Use bits_per() instead, which rounds up. Found by code inspection.
> The effect of this would probably be a misaccounting when doing NUMA
> balancing, so to a user, it would only be a performance penalty.
> The effects may be more wide-spread; it's hard to tell.
>
> Signed-off-by: Matthew Wilcox (Oracle) <[email protected]>
> Fixes: 90572890d202 ("mm: numa: Change page last {nid,pid} into {cpu,pid}")
> Cc: Peter Zijlstra <[email protected]>
> Cc: Mel Gorman <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Rik van Riel <[email protected]>

Acked-by: Mel Gorman <[email protected]>

--
Mel Gorman
SUSE Labs