patch [1/1] convert i386 summit subarch to use SRAT data for apicid_to_node
RFC 1/1 patch [1/1] convert i386 summit subarch to use SRAT data for apicid_to_node
Hello All,
This patch converts the i386 summit subarch apicid_to_node to use node
information provided by the SRAT. The current way of obtaining the nodeid
static inline int apicid_to_node(int logical_apicid) {
return logical_apicid >> 5;
}
is just not correct for all summit systems/bios. Assuming the apicid matches the Linux node number require a leap of faith that the bios lay- ed out the apicids a set way. Modern summit HW does not layout its bios in the manner for various reasons and is unable to boot i386 numa.
The best way to get the correct apicid to node information is from the
SRAT table. It lays out what apicid belongs to what node. I use this information to create a table for use at run time.
The attached patch was built against 2.6.17-rc2 but applies and runs
against 2.6.17 just fine. It only changes the summit subarch although since a global data structure is created it might make sense to change all the subarches. All comments welcome.
arch/i386/kernel/smpboot.c +4 -1
@@ -108,6 +108,8 @@ { [0 ... NR_CPUS-1] = 0xff };EXPORT_SYMBOL (x86_cpu_to_apicid); +u8 apicid_2_node[MAX_APICID ] = { [0 ...MAX_APICID -1 ] =0 }; +/* * Trampoline 80x86 program as an array. */
@@ -640,7 +642,7 @@int apicid =logical_smp_processor_id (); cpu_2_logical_apicid[cpu] = apicid; -map_cpu_to_node (cpu,apicid_to_node (apicid)); +map_cpu_to_node (cpu,apicid_to_node (hard_smp_processor_id ())); }static void unmap_cpu_to_logical_apicid (int cpu)
@@ -943,6 +945,7 @@irq_ctx_init (cpu); + x86_cpu_to_apicid[cpu] = apicid;/* * This grunge runs the startup process for * the targeted processor.
arch/i386/kernel/srat.c +8 -0
@@ -58,6 +58,8 @@static int num_memory_chunks;/* total number of memory chunks */ static int zholes_size_init;static unsigned long zholes_size[MAX_NUMNODES *MAX_NR_ZONES ]; +static u8 apicid_to_pxm[MAX_APICID ] = { [0 ...MAX_APICID -1 ] =0 }; +extern u8 apicid_2_node[];extern void *boot_ioremap (unsigned long ,unsigned long );
@@ -73,6 +75,8 @@/* mark this node as "seen" in node bitmap */ BMAP_SET (pxm_bitmap, cpu_affinity->proximity_domain); + apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain; +printk ("CPU 0x%02X in proximity domain 0x%02X\n" , cpu_affinity->apic_id, cpu_affinity->proximity_domain); }
@@ -298,6 +302,10 @@printk ("Number of logical nodes in system = %d\n" ,num_online_nodes ());printk ("Number of memory chunks in system = %d\n" , num_memory_chunks); +for (i =0 ; i <MAX_APICID ; i++) { + apicid_2_node[i] = pxm_to_nid_map[apicid_to_pxm[i]]; + } +for (j =0 ; j < num_memory_chunks; j++){struct node_memory_chunk_s * chunk = &node_memory_chunk[j]; printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
include/asm-i386/mach-summit/mach_apic.h +2 -1
@@ -43,6 +43,7 @@extern u8 bios_cpu_apicid[];extern u8 cpu_2_logical_apicid[]; +extern u8 apicid_2_node[];static inline void init_apic_ldr (void ) {
@@ -86,7 +87,7 @@static inline int apicid_to_node (int logical_apicid) { -return logical_apicid >>5 ;/* 2 clusterids per CEC */ +return apicid_2_node[logical_apicid]; }/* Mapping from cpu number to logical apicid */
Quoted 10 lines · keith mannthey
Hello All,
This patch converts the i386 summit subarchapicid_to_nodeto use node
information provided by the SRAT. The current way of obtaining the nodeidstatic inline int
apicid_to_node(int logical_apicid){ return logical_apicid >> 5;}
Quoted 7 lines
is just not correct for all summit systems/bios. Assuming the apicid matches the Linux node number require a leap of faith that the bios lay- ed out the apicids a set way. Modern summit HW does not layout its bios in the manner for various reasons and is unable to boot i386 numa.
The best way to get the correct apicid to node information is from the
SRAT table.
Do all summit's have SRAT tables ?
I was under the impression the early ones were around before the invention of SRAT.
Dave
Signature · 1 line
http://www.codemonkey.org.uk
Quoted 16 lines · Dave Jones
Quoted 15 lines · keith mannthey
Hello All,
This patch converts the i386 summit subarchapicid_to_nodeto use node
information provided by the SRAT. The current way of obtaining the nodeidstatic inline int
apicid_to_node(int logical_apicid){ return logical_apicid >> 5;}
is just not correct for all summit systems/bios. Assuming the apicid matches the Linux node number require a leap of faith that the bios lay- ed out the apicids a set way. Modern summit HW does not layout its bios in the manner for various reasons and is unable to boot i386 numa.
Quoted 6 lines
Quoted 2 lines
The best way to get the correct apicid to node information is from the
SRAT table.Do all summit's have SRAT tables ?
I was under the impression the early ones were around before the invention of SRAT.
That is a good point. Let me check into the x440 (1st gen). x445 x460 (2nd,3rd gen) uses SRAT for sure (these patches have been tested on these systems).
The x440 lists an srat but maybe it is using some special bios area. I will build a test kernel give it a whirl.
Quoted 2 lines
Dave
Signature · 2 lines
keith mannthey <[email protected]> Linux Technology Center IBM
Quoted 26 lines · keith mannthey
Quoted 16 lines · Dave Jones
Quoted 15 lines · keith mannthey
Hello All,
This patch converts the i386 summit subarchapicid_to_nodeto use node
information provided by the SRAT. The current way of obtaining the nodeidstatic inline int
apicid_to_node(int logical_apicid){ return logical_apicid >> 5;}
is just not correct for all summit systems/bios. Assuming the apicid matches the Linux node number require a leap of faith that the bios lay- ed out the apicids a set way. Modern summit HW does not layout its bios in the manner for various reasons and is unable to boot i386 numa.
Quoted 6 lines
Quoted 2 lines
The best way to get the correct apicid to node information is from the
SRAT table.Do all summit's have SRAT tables ?
I was under the impression the early ones were around before the invention of SRAT.
Quoted 6 lines
That is a good point. Let me check into the x440 (1st gen). x445 x460 (2nd,3rd gen) uses SRAT for sure (these patches have been tested on these systems).
The x440 lists an srat but maybe it is using some special bios area. I will build a test kernel give it a whirl.
I'm pretty sure they all had SRAT tables - the test machine we use regularly for test.kernel.org (elm3b67) does. The NUMA-Q (x430) doesn't, but that's a separate subarch.
M.
Quoted 31 lines · Martin J. Bligh
Quoted 30 lines · keith mannthey
Quoted 16 lines · Dave Jones
Quoted 15 lines · keith mannthey
Hello All,
This patch converts the i386 summit subarchapicid_to_nodeto use node
information provided by the SRAT. The current way of obtaining the nodeidstatic inline int
apicid_to_node(int logical_apicid){ return logical_apicid >> 5;}
is just not correct for all summit systems/bios. Assuming the apicid matches the Linux node number require a leap of faith that the bios lay- ed out the apicids a set way. Modern summit HW does not layout its bios in the manner for various reasons and is unable to boot i386 numa.
Quoted 6 lines
Quoted 2 lines
The best way to get the correct apicid to node information is from the
SRAT table.Do all summit's have SRAT tables ?
I was under the impression the early ones were around before the invention of SRAT.That is a good point. Let me check into the x440 (1st gen). x445 x460 (2nd,3rd gen) uses SRAT for sure (these patches have been tested on these systems).
Quoted 6 lines
Quoted 2 lines
The x440 lists an srat but maybe it is using some special bios area. I will build a test kernel give it a whirl.
I'm pretty sure they all had SRAT tables - the test machine we use regularly for test.kernel.org (elm3b67) does. The NUMA-Q (x430) doesn't, but that's a separate subarch.
Tested ontop of x440 just fine it does use the SRAT. It doesn't do anything NUMA-Q like :)
Thanks,
keith mannthey <[email protected]>
Linux Technology Center IBM