[PATCH] x86: cache_info: Fix setup of l2/l3 ids

[PATCH] x86: cache_info: Fix setup of l2/l3 ids

[Ido Yariv]

From: Shai Fultheim <shai@scalemp.com>

On some architectures (such as vSMP), it is possible to have CPUs with a
different number of cores sharing the same cache.

The current implementation implicitly assumes that all CPUs will have
the same number of cores sharing caches, and as a result, different CPUs
can end up with the same l2/l3 ids.

Fix this by masking out the shared cache bits, instead of shifting the
APICID. By doing so, it is guaranteed that the generated cache ids are
always unique.

Signed-off-by: Shai Fultheim <shai@scalemp.com>
[ido@wizery.com: rebased, simplified, and reworded the commit message]
Signed-off-by: Ido Yariv <ido@wizery.com>
---
 arch/x86/kernel/cpu/intel_cacheinfo.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index 73d08ed..caa6cb0 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -615,14 +615,14 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
 					new_l2 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(num_threads_sharing);
-					l2_id = c->apicid >> index_msb;
+					l2_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				case 3:
 					new_l3 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(
 							num_threads_sharing);
-					l3_id = c->apicid >> index_msb;
+					l3_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				default:
 					break;
-- 
1.7.7.6

[PATCH RESEND] x86: cache_info: Fix setup of l2/l3 ids

[Ido Yariv]

From: Shai Fultheim <shai@scalemp.com>

On some architectures (such as vSMP), it is possible to have CPUs with a
different number of cores sharing the same cache.

The current implementation implicitly assumes that all CPUs will have
the same number of cores sharing caches, and as a result, different CPUs
can end up with the same l2/l3 ids.

Fix this by masking out the shared cache bits, instead of shifting the
APICID. By doing so, it is guaranteed that the generated cache ids are
always unique.

Signed-off-by: Shai Fultheim <shai@scalemp.com>
[ido@wizery.com: rebased, simplified, and reworded the commit message]
Signed-off-by: Ido Yariv <ido@wizery.com>
---
 arch/x86/kernel/cpu/intel_cacheinfo.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index 73d08ed..caa6cb0 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -615,14 +615,14 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
 					new_l2 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(num_threads_sharing);
-					l2_id = c->apicid >> index_msb;
+					l2_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				case 3:
 					new_l3 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(
 							num_threads_sharing);
-					l3_id = c->apicid >> index_msb;
+					l3_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				default:
 					break;
-- 
1.7.7.6

Re: [PATCH RESEND] x86: cache_info: Fix setup of l2/l3 ids

[Ido Yariv]

On Fri, Apr 20, 2012 at 1:09 AM, Ido Yariv <ido@wizery.com> wrote:
> From: Shai Fultheim <shai@scalemp.com>
>
> On some architectures (such as vSMP), it is possible to have CPUs with a
> different number of cores sharing the same cache.
>
> The current implementation implicitly assumes that all CPUs will have
> the same number of cores sharing caches, and as a result, different CPUs
> can end up with the same l2/l3 ids.
>
> Fix this by masking out the shared cache bits, instead of shifting the
> APICID. By doing so, it is guaranteed that the generated cache ids are
> always unique.
>
> Signed-off-by: Shai Fultheim <shai@scalemp.com>
> [ido@wizery.com: rebased, simplified, and reworded the commit message]
> Signed-off-by: Ido Yariv <ido@wizery.com>

Ping?

[tip:x86/cpu] x86/cache_info: Fix setup of l2/l3 ids

[tip-bot for Shai Fultheim]

Commit-ID:  ddc5681ed33a279fdc188e98e71f0c539f08c6e6
Gitweb:     http://git.kernel.org/tip/ddc5681ed33a279fdc188e98e71f0c539f08c6e6
Author:     Shai Fultheim <shai@scalemp.com>
AuthorDate: Fri, 20 Apr 2012 01:09:11 +0300
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 7 May 2012 15:27:37 +0200

x86/cache_info: Fix setup of l2/l3 ids

On some architectures (such as vSMP), it is possible to have
CPUs with a different number of cores sharing the same cache.

The current implementation implicitly assumes that all CPUs will
have the same number of cores sharing caches, and as a result,
different CPUs can end up with the same l2/l3 ids.

Fix this by masking out the shared cache bits, instead of
shifting the APICID. By doing so, it is guaranteed that the
generated cache ids are always unique.

Signed-off-by: Shai Fultheim <shai@scalemp.com>
[ rebased, simplified, and reworded the commit message]
Signed-off-by: Ido Yariv <ido@wizery.com>
Cc: Borislav Petkov <borislav.petkov@amd.com>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Mike Travis <travis@sgi.com>
Cc: Dave Jones <davej@redhat.com>
Link: http://lkml.kernel.org/r/1334873351-31142-1-git-send-email-ido@wizery.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/kernel/cpu/intel_cacheinfo.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/intel_cacheinfo.c b/arch/x86/kernel/cpu/intel_cacheinfo.c
index b8f3653..9a7c90d 100644
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c
@@ -615,14 +615,14 @@ unsigned int __cpuinit init_intel_cacheinfo(struct cpuinfo_x86 *c)
 					new_l2 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(num_threads_sharing);
-					l2_id = c->apicid >> index_msb;
+					l2_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				case 3:
 					new_l3 = this_leaf.size/1024;
 					num_threads_sharing = 1 + this_leaf.eax.split.num_threads_sharing;
 					index_msb = get_count_order(
 							num_threads_sharing);
-					l3_id = c->apicid >> index_msb;
+					l3_id = c->apicid & ~((1 << index_msb) - 1);
 					break;
 				default:
 					break;