Skip to contentSkip to conversation

Inverted NUMA test in setup_pcpu_remap()?

PATCH In review David Miller 3 participants 8 messages Apr 1, 2009 – Apr 2, 2009 x86

lore
Share & tools
lore
#0 lore raw

The test currently is:

	if (!cpu_has_pse || pcpu_need_numa())
		return -EINVAL;

Don't we really mean "!pcpu_need_numa()"?

The way I read the intent, setup_pcpu_remap() should be used in the NUMA case. But that's not what's happening because of how this test is coded.

In fact, the test here is identical to the one used in setup_pcpu_embed()

#1 lore raw
Quoted 8 lines · David Miller

The test currently is:

       if (!cpu_has_pse || pcpu_need_numa())
               return -EINVAL;

Don't we really mean "!pcpu_need_numa()"?

Quoted 7 lines

The way I read the intent, setup_pcpu_remap() should be used in the NUMA case.  But that's not what's happening because of how this test is coded.

In fact, the test here is identical to the one used in setup_pcpu_embed()
--

you are right. on one 4 sockets system got

[    0.000000] PERCPU: Allocated 471 4k pages, static data 1925472 bytes

that is from setup_pcpu_4k...

YH

#2 lore raw

PATCH 1/1 x86,percpu: fix inverted NUMA test in setup_pcpu_remap()

setup_percpu_remap() is for NUMA machines yet it bailed out with -EINVAL if pcpu_need_numa(). Fix the inverted condition.

This problem was reported by David Miller and verified by Yinhai Lu.

---
Oops, thanks for spotting it. I don't have a numa machine so the condition was never tested on actual machines. :-)

 arch/x86/kernel/setup_percpu.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/setup_percpu.c +1 -1
index 400331b..876b127 100644
@@ -162,7 +162,7 @@ static ssize_t __init setup_pcpu_remap(size_t static_size)
 	 * If large page isn't supported, there's no benefit in doing
 	 * this.  Also, on non-NUMA, embedding is better.
 	 */
-	if (!cpu_has_pse || pcpu_need_numa())
+	if (!cpu_has_pse || !pcpu_need_numa())
 		return -EINVAL;
 
 	last = NULL;
#3 lore raw

PATCH 1/1 x86: remove duplicated code with pcpu_need_numa()

Impact: clean up

those code pcpu_need_numa(), should be removed.

---
 arch/x86/kernel/setup_percpu.c |   14 --------------
 1 file changed, 14 deletions(-)
arch/x86/kernel/setup_percpu.c +0 -14
Index: linux-2.6/arch/x86/kernel/setup_percpu.c
===================================================================
@@ -153,7 +153,6 @@ static struct page * __init pcpur_get_pa
 static ssize_t __init setup_pcpu_remap(size_t static_size)
 {
 	static struct vm_struct vm;
-	pg_data_t *last;
 	size_t ptrs_size, dyn_size;
 	unsigned int cpu;
 	ssize_t ret;
@@ -165,19 +164,6 @@ static ssize_t __init setup_pcpu_remap(s
 	if (!cpu_has_pse || !pcpu_need_numa())
 		return -EINVAL;
 
-	last = NULL;
-	for_each_possible_cpu(cpu) {
-		int node = early_cpu_to_node(cpu);
-
-		if (node_online(node) && NODE_DATA(node) &&
-		    last && last != NODE_DATA(node))
-			goto proceed;
-
-		last = NODE_DATA(node);
-	}
-	return -EINVAL;
-
-proceed:
 	/*
 	 * Currently supports only single page.  Supporting multiple
 	 * pages won't be too difficult if it ever becomes necessary.
#6 lore raw

[tip:x86/urgent] x86: remove duplicated code with pcpu_need_numa()

In reply to Yinghai Lu
Commit-ID:  3de46fda4c104deef17ec70f85361f5c6b84ce0e
Gitweb:     http://git.kernel.org/tip/3de46fda4c104deef17ec70f85361f5c6b84ce0e
Author:     Yinghai Lu <[email protected]>
AuthorDate: Wed, 1 Apr 2009 00:27:44 -0700
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 2 Apr 2009 06:08:05 +0200

x86: remove duplicated code with pcpu_need_numa()

Impact: clean up

those code pcpu_need_numa(), should be removed.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/setup_percpu.c |   14 --------------
 1 files changed, 0 insertions(+), 14 deletions(-)
arch/x86/kernel/setup_percpu.c +0 -14
index 876b127..3a97a4c 100644
@@ -153,7 +153,6 @@ static struct page * __init pcpur_get_page(unsigned int cpu, int pageno)
 static ssize_t __init setup_pcpu_remap(size_t static_size)
 {
 	static struct vm_struct vm;
-	pg_data_t *last;
 	size_t ptrs_size, dyn_size;
 	unsigned int cpu;
 	ssize_t ret;
@@ -165,19 +164,6 @@ static ssize_t __init setup_pcpu_remap(size_t static_size)
 	if (!cpu_has_pse || !pcpu_need_numa())
 		return -EINVAL;
 
-	last = NULL;
-	for_each_possible_cpu(cpu) {
-		int node = early_cpu_to_node(cpu);
-
-		if (node_online(node) && NODE_DATA(node) &&
-		    last && last != NODE_DATA(node))
-			goto proceed;
-
-		last = NODE_DATA(node);
-	}
-	return -EINVAL;
-
-proceed:
 	/*
 	 * Currently supports only single page.  Supporting multiple
 	 * pages won't be too difficult if it ever becomes necessary.
#7 lore raw

[tip:x86/urgent] x86,percpu: fix inverted NUMA test in setup_pcpu_remap()

In reply to Tejun Heo
Commit-ID:  eb12ce60c81826a99eadbc56401e08ceb37a0cc2
Gitweb:     http://git.kernel.org/tip/eb12ce60c81826a99eadbc56401e08ceb37a0cc2
Author:     Tejun Heo <[email protected]>
AuthorDate: Wed, 1 Apr 2009 15:06:33 +0900
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 2 Apr 2009 06:08:05 +0200

x86,percpu: fix inverted NUMA test in setup_pcpu_remap()

setup_percpu_remap() is for NUMA machines yet it bailed out with -EINVAL if pcpu_need_numa(). Fix the inverted condition.

This problem was reported by David Miller and verified by Yinhai Lu.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/setup_percpu.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kernel/setup_percpu.c +1 -1
index 400331b..876b127 100644
@@ -162,7 +162,7 @@ static ssize_t __init setup_pcpu_remap(size_t static_size)
 	 * If large page isn't supported, there's no benefit in doing
 	 * this.  Also, on non-NUMA, embedding is better.
 	 */
-	if (!cpu_has_pse || pcpu_need_numa())
+	if (!cpu_has_pse || !pcpu_need_numa())
 		return -EINVAL;
 
 	last = NULL;