[PATCH] x86_64: use apic core id bits for core nums

[PATCH] x86_64: use apic core id bits for core nums

[Yinghai Lu]

[-- Attachment #1: Type: text/plain, Size: 391 bytes --]

please check my version, we should use core id bits instead.

it should replace
[PATCH for review] [37/48] x86_64: make k8topology multi-core aware

YH

[PATCH] x86_64: use apic core id bits for core nums

we need to use apic core id bits to get core num and use that get apic_to_node
correctly.
it will work with quad core downcores cpus.

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>

[-- Attachment #2: k8_core.diff --]
[-- Type: text/x-patch, Size: 1689 bytes --]

diff --git a/arch/x86_64/mm/k8topology.c b/arch/x86_64/mm/k8topology.c
index f983c75..3e8e47f 100644
--- a/arch/x86_64/mm/k8topology.c
+++ b/arch/x86_64/mm/k8topology.c
@@ -49,7 +49,10 @@ int __init k8_scan_nodes(unsigned long start, unsigned long end)
 	int found = 0;
 	u32 reg;
 	unsigned numnodes;
-	unsigned dualcore = 0;
+	unsigned cores;
+	u32 ecx;
+	unsigned bits;
+	int j;
 
 	if (!early_pci_allowed())
 		return -1;
@@ -67,14 +70,26 @@ int __init k8_scan_nodes(unsigned long start, unsigned long end)
 
 	printk(KERN_INFO "Number of nodes %d\n", numnodes);
 
+        ecx = cpuid_ecx(0x80000008);
+        cores = (ecx & 0xff) + 1;
+
+        /* CPU telling us the core id bits shift? */
+        bits = (ecx >> 12) & 0xF;
+
+        /* Otherwise recompute */
+        if (bits == 0) {
+                while ((1 << bits) < cores)
+                        bits++;
+        }
+
+	cores = (1<<bits);
+
 	memset(&nodes,0,sizeof(nodes)); 
 	prevbase = 0;
 	for (i = 0; i < 8; i++) { 
 		unsigned long base,limit; 
 		u32 nodeid;
 		
-		/* Undefined before E stepping, but hopefully 0 */
-		dualcore |= ((read_pci_config(0, nb, 3, 0xe8) >> 12) & 3) == 1;
 		base = read_pci_config(0, nb, 1, 0x40 + i*8);
 		limit = read_pci_config(0, nb, 1, 0x44 + i*8);
 
@@ -170,8 +185,8 @@ int __init k8_scan_nodes(unsigned long start, unsigned long end)
 	for (i = 0; i < 8; i++) {
 		if (nodes[i].start != nodes[i].end) { 
 			nodeid = nodeids[i];
-			apicid_to_node[nodeid << dualcore] = i;
-			apicid_to_node[(nodeid << dualcore) + dualcore] = i;
+			for (j = 0; j < cores; j++)
+				apicid_to_node[(nodeid << bits) + j] = i;
 			setup_node_bootmem(i, nodes[i].start, nodes[i].end); 
 		} 
 	}

Re: [PATCH] x86_64: use apic core id bits for core nums

[Andi Kleen]

On Saturday 21 July 2007 09:56:15 Yinghai Lu wrote:
> please check my version, we should use core id bits instead.
> 
> it should replace
> [PATCH for review] [37/48] x86_64: make k8topology multi-core aware

Yes, you're right the patch is not good. It should probably use 
information from the regular CPU setup code, not touch cpuid
directly

I'll drop it for now

-Andi

Re: [PATCH] x86_64: use apic core id bits for core nums

[Yinghai Lu]

On 7/21/07, Andi Kleen <ak@suse.de> wrote:
> On Saturday 21 July 2007 09:56:15 Yinghai Lu wrote:
> > please check my version, we should use core id bits instead.
> >
> > it should replace
> > [PATCH for review] [37/48] x86_64: make k8topology multi-core aware
>
> Yes, you're right the patch is not good. It should probably use
> information from the regular CPU setup code, not touch cpuid
> directly

need move some code into early_identify_cpu from identify_cpu in
arch/x86_64/kernel/setup.c
then could use that bits in bsp cpu info.

BTW:
it seems for BSP only early_identify_cpu is called, and identify_cpu
is only use in smp_callin for AP. Do i miss sth.

YH

Re: [PATCH] x86_64: use apic core id bits for core nums

[Andi Kleen]

On Saturday 21 July 2007 13:21:34 Yinghai Lu wrote:
> On 7/21/07, Andi Kleen <ak@suse.de> wrote:
> > On Saturday 21 July 2007 09:56:15 Yinghai Lu wrote:
> > > please check my version, we should use core id bits instead.
> > >
> > > it should replace
> > > [PATCH for review] [37/48] x86_64: make k8topology multi-core aware
> >
> > Yes, you're right the patch is not good. It should probably use
> > information from the regular CPU setup code, not touch cpuid
> > directly
> 
> need move some code into early_identify_cpu from identify_cpu in
> arch/x86_64/kernel/setup.c
> then could use that bits in bsp cpu info.

Yes something like this
 
> BTW:
> it seems for BSP only early_identify_cpu is called, and identify_cpu
> is only use in smp_callin for AP. Do i miss sth.

identify_cpu calls early_identify_cpu as first thing.

-Andi

Re: [PATCH] x86_64: use apic core id bits for core nums

[Yinghai Lu]

On 7/21/07, Andi Kleen <ak@suse.de> wrote:
> On Saturday 21 July 2007 13:21:34 Yinghai Lu wrote:
> > On 7/21/07, Andi Kleen <ak@suse.de> wrote:
> > > On Saturday 21 July 2007 09:56:15 Yinghai Lu wrote:
> > > > please check my version, we should use core id bits instead.
> > > >
> > > > it should replace
> > > > [PATCH for review] [37/48] x86_64: make k8topology multi-core aware
> > >
> > > Yes, you're right the patch is not good. It should probably use
> > > information from the regular CPU setup code, not touch cpuid
> > > directly
> >
> > need move some code into early_identify_cpu from identify_cpu in
> > arch/x86_64/kernel/setup.c
> > then could use that bits in bsp cpu info.
>
> Yes something like this
>
> > BTW:
> > it seems for BSP only early_identify_cpu is called, and identify_cpu
> > is only use in smp_callin for AP. Do i miss sth.
>
> identify_cpu calls early_identify_cpu as first thing.

the calling to identify_cpu in BSP is in check_bugs(). It is easy to
ignore it...
maybe we could move that to setup_arch directly.

YH

Re: [PATCH] x86_64: use apic core id bits for core nums

[Yinghai Lu]

On 7/21/07, Andi Kleen <ak@suse.de> wrote:
> On Saturday 21 July 2007 09:56:15 Yinghai Lu wrote:
> > please check my version, we should use core id bits instead.
> >
> > it should replace
> > [PATCH for review] [37/48] x86_64: make k8topology multi-core aware
>
> Yes, you're right the patch is not good. It should probably use
> information from the regular CPU setup code, not touch cpuid
> directly
>
> I'll drop it for now

too bad, just figure out you dropped wrong one.

following one 36/48 is right. please apply it again.

37/48 got into the mainline. that is wrong. please use
http://lkml.org/lkml/2007/7/22/7
[PATCH 2/3] x86_64: use core id bits for apicid_to_node initialization v2
that is rebased to 37/48.

YH


From: "Joachim Deguara" <joachim.deguara@amd.com>
This just adds the PCI IDs of AMD's family 10h and 11h CPU's northbridges to
k8topology discovery.

Signed-off-by: Joachim Deguara <joachim.deguara@amd.com>
Signed-off-by: Andi Kleen <ak@suse.de>

--
---
 arch/x86_64/mm/k8topology.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

Index: linux/arch/x86_64/mm/k8topology.c
===================================================================
--- linux.orig/arch/x86_64/mm/k8topology.c
+++ linux/arch/x86_64/mm/k8topology.c
@@ -28,11 +28,15 @@ static __init int find_northbridge(void)
               u32 header;

               header = read_pci_config(0, num, 0, 0x00);
-               if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)))
+               if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)) &&
+                       header != (PCI_VENDOR_ID_AMD | (0x1200<<16)) &&
+                       header != (PCI_VENDOR_ID_AMD | (0x1300<<16)) )
                       continue;

               header = read_pci_config(0, num, 1, 0x00);
-               if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)))
+               if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)) &&
+                       header != (PCI_VENDOR_ID_AMD | (0x1201<<16)) &&
+                       header != (PCI_VENDOR_ID_AMD | (0x1301<<16)) )
                       continue;
               return num;
       }
-