mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] x86, amd: Probe CPUs by APICID instead of initial APICID
@ 2012-04-17 10:06 Vlad Zolotarov
  2012-04-17 17:26 ` Andreas Herrmann
  0 siblings, 1 reply; 2+ messages in thread
From: Vlad Zolotarov @ 2012-04-17 10:06 UTC (permalink / raw)
  To: linux-kernel, tglx, mingo, hpa
  Cc: andreas.herrmann3, borislav.petkov, Ravikiran.Thirumalai,
	Shai Fultheim (Shai@ScaleMP.com)

From: Shai Fultheim <shai@scalemp.com>

Make AMD identify processors using APICID (32 bit) rather than the initial 
APICID (8 bit).  This is critical to make sure the last level cache (llc), 
which is used for sibling detection, will be different between boards - on 
aggregated systems with more than 8 processors.


Signed-off-by: Shai Fultheim <shai@scalemp.com>
Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>
---
 arch/x86/kernel/cpu/amd.c |   10 +++++-----
 1 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 0bab2b1..33d9502 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -277,7 +277,7 @@ static void __cpuinit amd_get_topology(struct cpuinfo_x86 *c)
 
 		cpuid(0x8000001e, &eax, &ebx, &ecx, &edx);
 		nodes = ((ecx >> 8) & 7) + 1;
-		node_id = ecx & 7;
+		node_id = (ecx & 7) + (c->phys_proc_id << 3);
 
 		/* get compute unit information */
 		smp_num_siblings = ((ebx >> 8) & 3) + 1;
@@ -288,7 +288,7 @@ static void __cpuinit amd_get_topology(struct cpuinfo_x86 *c)
 
 		rdmsrl(MSR_FAM10H_NODE_ID, value);
 		nodes = ((value >> 3) & 7) + 1;
-		node_id = value & 7;
+		node_id = (value & 7) + (c->phys_proc_id << 3);
 	} else
 		return;
 
@@ -323,9 +323,9 @@ static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
 
 	bits = c->x86_coreid_bits;
 	/* Low order bits define the core id (index of core in socket) */
-	c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
-	/* Convert the initial APIC ID into the socket ID */
-	c->phys_proc_id = c->initial_apicid >> bits;
+	c->cpu_core_id = c->apicid & ((1 << bits)-1);
+	/* Convert the APIC ID into the socket ID */
+	c->phys_proc_id = c->apicid >> bits;
 	/* use socket ID also for last level cache */
 	per_cpu(cpu_llc_id, cpu) = c->phys_proc_id;
 	amd_get_topology(c);
-- 
1.7.7.6


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] x86, amd: Probe CPUs by APICID instead of initial APICID
  2012-04-17 10:06 [PATCH] x86, amd: Probe CPUs by APICID instead of initial APICID Vlad Zolotarov
@ 2012-04-17 17:26 ` Andreas Herrmann
  0 siblings, 0 replies; 2+ messages in thread
From: Andreas Herrmann @ 2012-04-17 17:26 UTC (permalink / raw)
  To: Vlad Zolotarov
  Cc: linux-kernel, tglx, mingo, hpa, borislav.petkov,
	Ravikiran.Thirumalai, Shai Fultheim (Shai@ScaleMP.com)

On Tue, Apr 17, 2012 at 01:06:36PM +0300, Vlad Zolotarov wrote:
> From: Shai Fultheim <shai@scalemp.com>
> 
> Make AMD identify processors using APICID (32 bit) rather than the initial 
> APICID (8 bit).  This is critical to make sure the last level cache (llc), 
> which is used for sibling detection, will be different between boards - on 
> aggregated systems with more than 8 processors.
> 
> 
> Signed-off-by: Shai Fultheim <shai@scalemp.com>
> Signed-off-by: Vlad Zolotarov <vlad@scalemp.com>

I think instead of fiddling with the generic code you rather should do
what was implemented for numascale support. See apic_numachip.c

Especially changing node_id will break other stuff.


> ---
>  arch/x86/kernel/cpu/amd.c |   10 +++++-----
>  1 files changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
> index 0bab2b1..33d9502 100644
> --- a/arch/x86/kernel/cpu/amd.c
> +++ b/arch/x86/kernel/cpu/amd.c
> @@ -277,7 +277,7 @@ static void __cpuinit amd_get_topology(struct cpuinfo_x86 *c)
>  
>  		cpuid(0x8000001e, &eax, &ebx, &ecx, &edx);
>  		nodes = ((ecx >> 8) & 7) + 1;
> -		node_id = ecx & 7;
> +		node_id = (ecx & 7) + (c->phys_proc_id << 3);
>  
>  		/* get compute unit information */
>  		smp_num_siblings = ((ebx >> 8) & 3) + 1;
> @@ -288,7 +288,7 @@ static void __cpuinit amd_get_topology(struct cpuinfo_x86 *c)
>  
>  		rdmsrl(MSR_FAM10H_NODE_ID, value);
>  		nodes = ((value >> 3) & 7) + 1;
> -		node_id = value & 7;
> +		node_id = (value & 7) + (c->phys_proc_id << 3);
>  	} else
>  		return;
>  
> @@ -323,9 +323,9 @@ static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
>  
>  	bits = c->x86_coreid_bits;
>  	/* Low order bits define the core id (index of core in socket) */
> -	c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
> -	/* Convert the initial APIC ID into the socket ID */
> -	c->phys_proc_id = c->initial_apicid >> bits;
> +	c->cpu_core_id = c->apicid & ((1 << bits)-1);
> +	/* Convert the APIC ID into the socket ID */
> +	c->phys_proc_id = c->apicid >> bits;

Extracting phys_proc_id from apicid will result in wrong socket
enumeration on several systems. E.g. they would start with package_id
1 instead of 0.


Thanks,

Andreas



^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2012-04-17 17:26 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-04-17 10:06 [PATCH] x86, amd: Probe CPUs by APICID instead of initial APICID Vlad Zolotarov
2012-04-17 17:26 ` Andreas Herrmann

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®