Hello
I worked lately on the bug 6745357 (kernel crash during startup at
page_ctr_add_internal).
I have observed the following :
The bug was introduced from nv_88.
The changeset that breaks the boot is this one :
Issues Resolved:
BUG/RFE:6594519Need support for ACPI System Resource Affinity Table (SRAT)
BUG/RFE:6621201Need support to read ACPI System Locality Information Table
(SLIT)
BUG/RFE:6688471x86/x64 lgroup platform support code needs cleaning
Files Changed:
update:usr/src/uts/i86pc/os/acpi_fw.h
update:usr/src/uts/i86pc/os/cpuid.c
update:usr/src/uts/i86pc/os/fakebop.c
update:usr/src/uts/i86pc/os/lgrpplat.c
update:usr/src/uts/i86pc/os/mlsetup.c
update:usr/src/uts/intel/sys/x86_archext.h
I haven't found a fix yet to resolve this issue, but I tested a workaround :
I modified the lgrpplat.c to add a new boot option (-B disable-numa-srat=y),
that (when detected) bypasses the call to lgrp_plat_process_srat.
<code>
# diff -c lgrpplat.c_orig usr/src/uts/i86pc/os/lgrpplat.c
*** lgrpplat.c_orig Sun Feb 22 15:36:42 2009
--- usr/src/uts/i86pc/os/lgrpplat.c Mon Feb 23 14:13:04 2009
***************
*** 150,155 ****
--- 150,157 ----
#define MAX_NODES 8
#define NLGRP (MAX_NODES * (MAX_NODES - 1) + 1)
+ #define BP_DISABLE_NUMA_SRAT "disable-numa-srat"
+
/*
* Constants for configuring probing
*/
***************
*** 684,690 ****
--- 686,695 ----
lgrp_plat_node_cnt = max_mem_nodes = 1;
#else /* __xpv */
uint_t probe_op;
+ int boot_prop_len;
+ char *boot_prop_name = BP_DISABLE_NUMA_SRAT;
+
/*
* Initialize as a UMA machine
*/
***************
*** 700,710 ****
lgrp_plat_apic_ncpus =
lgrp_plat_process_cpu_apicids(lgrp_plat_cpu_node);
/*
* Determine which CPUs and memory are local to each other and number
* of NUMA nodes by reading ACPI System Resource Affinity Table (SRAT)
*/
! if (lgrp_plat_apic_ncpus > 0) {
int retval;
retval = lgrp_plat_process_srat(srat_ptr,
--- 705,720 ----
lgrp_plat_apic_ncpus =
lgrp_plat_process_cpu_apicids(lgrp_plat_cpu_node);
+ boot_prop_len = BOP_GETPROPLEN(bootops, boot_prop_name);
+ if (boot_prop_len > 0) {
+ lgrp_plat_srat_error=-1;
+ }
+
/*
* Determine which CPUs and memory are local to each other and number
* of NUMA nodes by reading ACPI System Resource Affinity Table (SRAT)
*/
! if ((boot_prop_len <= 0) && lgrp_plat_apic_ncpus > 0) {
int retval;
retval = lgrp_plat_process_srat(srat_ptr,
</code>
This change has low impact, since it does nothing when no additional boot
option is added.
It allows to boot nevada on impacted platform (boot broken since build 88 !).
What do think about it ?
Can this be included in the source ?
Regards
Guy
--
This message posted from opensolaris.org# diff -c lgrpplat.c_orig usr/src/uts/i86pc/os/lgrpplat.c
*** lgrpplat.c_orig Sun Feb 22 15:36:42 2009
--- usr/src/uts/i86pc/os/lgrpplat.c Mon Feb 23 14:13:04 2009
***************
*** 150,155 ****
--- 150,157 ----
#define MAX_NODES 8
#define NLGRP (MAX_NODES * (MAX_NODES - 1) + 1)
+ #define BP_DISABLE_NUMA_SRAT "disable-numa-srat"
+
/*
* Constants for configuring probing
*/
***************
*** 684,690 ****
--- 686,695 ----
lgrp_plat_node_cnt = max_mem_nodes = 1;
#else /* __xpv */
uint_t probe_op;
+ int boot_prop_len;
+ char *boot_prop_name = BP_DISABLE_NUMA_SRAT;
+
/*
* Initialize as a UMA machine
*/
***************
*** 700,710 ****
lgrp_plat_apic_ncpus =
lgrp_plat_process_cpu_apicids(lgrp_plat_cpu_node);
/*
* Determine which CPUs and memory are local to each other and number
* of NUMA nodes by reading ACPI System Resource Affinity Table (SRAT)
*/
! if (lgrp_plat_apic_ncpus > 0) {
int retval;
retval = lgrp_plat_process_srat(srat_ptr,
--- 705,720 ----
lgrp_plat_apic_ncpus =
lgrp_plat_process_cpu_apicids(lgrp_plat_cpu_node);
+ boot_prop_len = BOP_GETPROPLEN(bootops, boot_prop_name);
+ if (boot_prop_len > 0) {
+ lgrp_plat_srat_error=-1;
+ }
+
/*
* Determine which CPUs and memory are local to each other and number
* of NUMA nodes by reading ACPI System Resource Affinity Table (SRAT)
*/
! if ((boot_prop_len <= 0) && lgrp_plat_apic_ncpus > 0) {
int retval;
retval = lgrp_plat_process_srat(srat_ptr,
_______________________________________________
opensolaris-code mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
