fix generic build when CONFIG_SMP=n
PATCH 1/1 fix generic build when CONFIG_SMP=n
This patch fixes the issues I saw with CONFIG_SMP=n when building a generic kernel with CONFIG_NUMA=y and CONFIG_DISCONTIGMEM=y. It's still pretty rough, but I'd appreciate comments and testing on other platforms. The per-cpu fixes are still pretty ugly--I'm sure I'm missing something that would make abstracting out per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data much nicer looking.
Fixing the build of generic kernels w/o CONFIG_NUMA or CONFIG_DISCONTIGMEM is left as an exercise for the reader. :)
Thanks,
Jesse
arch/ia64/kernel/Makefile +1 -0
@@ -15,6 +15,7 @@ obj-$(CONFIG_IOSAPIC) += iosapic.o obj-$(CONFIG_MODULES) += module.o obj-$(CONFIG_SMP) += smp.o smpboot.o +obj-$(CONFIG_NUMA) += numa.o obj-$(CONFIG_PERFMON) += perfmon_default_smpl.o obj-$(CONFIG_IA64_CYCLONE) += cyclone.o
arch/ia64/kernel/acpi.c +3 -1
@@ -618,8 +618,10 @@if (smp_boot_data.cpu_phys_id[cpu] !=hard_smp_processor_id ()) node_cpuid[i++].phys_id = smp_boot_data.cpu_phys_id[cpu]; } -build_cpu_to_node_map ();# endif +#endif +#ifdef CONFIG_ACPI_NUMA +build_cpu_to_node_map ();#endif /* Make boot-up look pretty */ printk (KERN_INFO "%d CPUs available, %d CPUs total\n" , available_cpus, total_cpus);
arch/ia64/kernel/numa.c new file +57 -0
57 lines not on this page · raw message
arch/ia64/kernel/setup.c +0 -2
@@ -317,11 +317,9 @@machvec_init (acpi_get_sysname ());#endif -#ifdef CONFIG_SMP /* If we register an early console, allow CPU 0 to printk */ if (!early_console_setup ())cpu_set (smp_processor_id (), cpu_online_map); -#endif #ifdef CONFIG_ACPI_BOOT /* Initialize the ACPI boot-time table parser */
arch/ia64/kernel/smpboot.c +0 -41
47 lines not on this page · raw message
arch/ia64/mm/discontig.c +48 -23
122 lines not on this page · raw message
include/asm-ia64/smp.h +1 -0
@@ -126,6 +126,7 @@#else #define cpu_logical_id (cpuid) 0 +#define cpu_physical_id (i) ((ia64_getreg(_IA64_REG_CR_LID) >> 16) & 0xffff)#endif /* CONFIG_SMP */ #endif /* _ASM_IA64_SMP_H */
include/asm-ia64/sn/sn_cpuid.h +0 -4
@@ -83,10 +83,6 @@* */ -#ifndef CONFIG_SMP -#define cpu_physical_id (cpuid) ((ia64_getreg(_IA64_REG_CR_LID) >> 16) & 0xffff) -#endif -/* * macros for some of these exist in sn/addrs.h & sn/arch.h, etc. However, * trying #include these files here causes circular dependencies.
Quoted 9 lines · Jesse Barnes
This patch fixes the issues I saw with
CONFIG_SMP=nwhen building a generic kernel withCONFIG_NUMA=yandCONFIG_DISCONTIGMEM=y. It's still pretty rough, but I'd appreciate comments and testing on other platforms. The per-cpu fixes are still pretty ugly--I'm sure I'm missing something that would make abstracting outper_cpu(cpu_info, cpu).node_data = mem_data[node].node_data much nicer looking.Fixing the build of generic kernels w/o
CONFIG_NUMAorCONFIG_DISCONTIGMEMis left as an exercise for the reader. :)
Here's the results of some 2.6.8 boot testing:
-----------------------------------------------------------------------------
generic-no-smp-2 generic_defconfig
patch? delta machine results
-----------------------------------------------------------------------------
No ITANIUM=y HP i2000 OK
-----------------------------------------------------------------------------
Yes ITANIUM=y HP i2000 OK
-----------------------------------------------------------------------------
Yes ITANIUM=y N/A build fails [1]
SMP=n
-----------------------------------------------------------------------------
Yes ITANIUM=y N/A build fails [2]
SMP=n
CYCLONE=n
-----------------------------------------------------------------------------
Yes ITANIUM=y N/A OK
SMP=n
CYCLONE=n HP i2000
SERIAL_SGI_L1_CONSOLE=n
-----------------------------------------------------------------------------
Yes HP rx2600 OK
-----------------------------------------------------------------------------
Yes SMP=n HP rx2600 OK
CYCLONE=n
SERIAL_SGI_L1_CONSOLE=n
[1]
CC arch/ia64/kernel/cyclone.oarch/ia64/kernel/cyclone.c:13: error: parse error before "cyclone_setup"arch/ia64/kernel/cyclone.c:14: warning: return type defaults to `int'arch/ia64/kernel/cyclone.c: In function `get_offset_cyclone':arch/ia64/kernel/cyclone.c:29: warning: implicit declaration of function `readl'arch/ia64/kernel/cyclone.c: At top level:arch/ia64/kernel/cyclone.c:78: error: parse error before "init_cyclone_clock"arch/ia64/kernel/cyclone.c:79: warning: return type defaults to `int'arch/ia64/kernel/cyclone.c: In function `init_cyclone_clock':arch/ia64/kernel/cyclone.c:92: warning: implicit declaration of function `ioremap_nocache'arch/ia64/kernel/cyclone.c:92: warning: cast to pointer from integer of different sizearch/ia64/kernel/cyclone.c:98: warning: implicit declaration of function readq'arch/ia64/kernel/cyclone.c:104: warning: implicit declaration of function iounmap' arch/ia64/kernel/cyclone.c:108: warning: cast to pointer from integer of different sizearch/ia64/kernel/cyclone.c:114: warning: implicit declaration of function `writel'arch/ia64/kernel/cyclone.c:119: warning: cast to pointer from integer of different sizearch/ia64/kernel/cyclone.c:130: warning: cast to pointer from integer of different sizearch/ia64/kernel/cyclone.c: At top level:arch/ia64/kernel/cyclone.c:157: warning: type defaults to int' in declaration of __initcall'arch/ia64/kernel/cyclone.c:157: warning: parameter names (without types) in function declarationarch/ia64/kernel/cyclone.c:157: warning: data definition has no type or storage class
make[1]: *** [arch/ia64/kernel/cyclone.o] Error 1
make: *** [arch/ia64/kernel] Error 2
[2]
CC drivers/serial/sn_console.odrivers/serial/sn_console.c: In function `snt_sim_puts':drivers/serial/sn_console.c:216: warning: implicit declaration of function writeb' drivers/serial/sn_console.c: In function snt_sim_getc':drivers/serial/sn_console.c:230: warning: implicit declaration of function `readb' drivers/serial/sn_console.c: At top level:drivers/serial/sn_console.c:1090: error: parse error before "else" make[2]: *** [drivers/serial/sn_console.o] Error 1
make[1]: *** [drivers/serial] Error 2
make: *** [drivers] Error 2
On Sun, 15 Aug 2004 21:29:37 -0600 dann frazier wrote:
| On Fri, Aug 13, 2004 at 03:59:59PM -0700, Jesse Barnes wrote: | > This patch fixes the issues I saw with CONFIG_SMP=n when building a generic | > kernel with CONFIG_NUMA=y and CONFIG_DISCONTIGMEM=y. It's still pretty | > rough, but I'd appreciate comments and testing on other platforms. The | > per-cpu fixes are still pretty ugly--I'm sure I'm missing something that | > would make abstracting out per_cpu(cpu_info, cpu).node_data = | > mem_data[node].node_data much nicer looking. | > | > Fixing the build of generic kernels w/o CONFIG_NUMA or CONFIG_DISCONTIGMEM is | > left as an exercise for the reader. :) | | Here's the results of some 2.6.8 boot testing: | | ----------------------------------------------------------------------------- | generic-no-smp-2 generic_defconfig | patch? delta machine results | ----------------------------------------------------------------------------- | No ITANIUM=y HP i2000 OK | ----------------------------------------------------------------------------- | Yes ITANIUM=y HP i2000 OK | ----------------------------------------------------------------------------- | Yes ITANIUM=y N/A build fails [1] | SMP=n | ----------------------------------------------------------------------------- | Yes ITANIUM=y N/A build fails [2] | SMP=n | CYCLONE=n | ----------------------------------------------------------------------------- | Yes ITANIUM=y N/A OK | SMP=n | CYCLONE=n HP i2000 | SERIAL_SGI_L1_CONSOLE=n | ----------------------------------------------------------------------------- | Yes HP rx2600 OK | ----------------------------------------------------------------------------- | Yes SMP=n HP rx2600 OK | CYCLONE=n | SERIAL_SGI_L1_CONSOLE=n | [1]
is fixed by the attached patch...
| [2]
[2] is fixed by
http://www.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.8.1/2.6.8.1-mm1/broken-out/fix-sn_console-for-config_smp=n.patch
--
~Randy
fix ia64 cyclone build errors
diffstat:=
arch/ia64/kernel/cyclone.c | 3 +++ 1 files changed, 3 insertions(+)
arch/ia64/kernel/cyclone.c renamed from arch/ia64/kernel/cyclone.c~ia64_no_smp +3 -0
@@ -1,7 +1,10 @@ +#include <linux/init.h> #include <linux/smp.h> #include <linux/time.h> #include <linux/errno.h> +#include <asm/io.h> +/* IBM Summit (EXA) Cyclone counter code*/ #define CYCLONE_CBAR_ADDR 0xFEB00CD0#define CYCLONE_PMCC_OFFSET 0x51A0
Quoted 28 lines · dann frazier
Quoted 8 lines · Jesse Barnes
This patch fixes the issues I saw with
CONFIG_SMP=nwhen building a
generic kernel withCONFIG_NUMA=yandCONFIG_DISCONTIGMEM=y. It's still
pretty rough, but I'd appreciate comments and testing on other platforms.
The per-cpu fixes are still pretty ugly--I'm sure I'm missing something
that would make abstracting outper_cpu(cpu_info, cpu).node_data > > mem_data[node].node_data much nicer looking.Fixing the build of generic kernels w/o
CONFIG_NUMAorCONFIG_DISCONTIGMEMis left as an exercise for the reader. :)Here's the results of some 2.6.8 boot testing:
---------------------------------------------------------------------------
-- generic-no-smp-2generic_defconfig
patch? delta machine results
---------------------------------------------------------------------------
-- No ITANIUM=y HP i2000 OK
---------------------------------------------------------------------------
-- Yes ITANIUM=y HP i2000 OK
---------------------------------------------------------------------------
-- Yes ITANIUM=y N/A build fails [1]
SMP=n
---------------------------------------------------------------------------
-- Yes ITANIUM=y N/A build fails [2]
SMP=n
CYCLONE=n
---------------------------------------------------------------------------
-- Yes ITANIUM=y N/A OK SMP=n
Quoted 8 lines
CYCLONE=n HP i2000
SERIAL_SGI_L1_CONSOLE=n
---------------------------------------------------------------------------
-- Yes HP rx2600 OK
---------------------------------------------------------------------------
-- Yes SMP=n HP rx2600 OK
CYCLONE=n
SERIAL_SGI_L1_CONSOLE=n
I posted fixes for cyclone and the sgi console driver separately. They should get in soon.
Jesse
Quoted 8 lines · Jesse Barnes
This patch fixes the issues I saw with
CONFIG_SMP=nwhen building a generic
kernel withCONFIG_NUMA=yandCONFIG_DISCONTIGMEM=y. It's still pretty
rough, but I'd appreciate comments and testing on other platforms. The
per-cpu fixes are still pretty ugly--I'm sure I'm missing something that
would make abstracting outper_cpu(cpu_info, cpu).node_data > mem_data[node].node_data much nicer looking.Fixing the build of generic kernels w/o
CONFIG_NUMAorCONFIG_DISCONTIGMEMis left as an exercise for the reader. :)
No comments? Tony, do you think it's ready to apply?
Thanks,
Jesse