Boot failure on ia64.
Oops, missed sending this to the ia64 mailing list.
Robin
----- Forwarded message from Robin Holt <[email protected]> -----
Date: Tue, 24 Jun 2008 07:30:14 -0500
From: Robin Holt <[email protected]>
To: [email protected], Alex Chiang <[email protected]>
Cc: [email protected]
Subject: [BISECT] Boot failure on ia64.
I bisected to this commit 3463a93def55c309f3c0d0a8aaf216be3be42d64
3463a93def55c309f3c0d0a8aaf216be3be42d64 is first bad commit commit 3463a93def55c309f3c0d0a8aaf216be3be42d64
Author: Alex Chiang <[email protected]> Date: Wed Jun 11 17:29:27 2008 -0600
[IA64] Update check_sal_cache_flush to use platform_send_ipi()
...
This fails to boot on any sn2 ia64 with the sn2_defconfig.
Here is the output from that boot.
fs0:\efi\SuSE> elilo net0:holt/v1 root=/dev/sda7 console=ttySG0
ELILO
Uncompressing Linux... done
Linux version 2.6.26-rc5-00223-g3463a93 (holt@attica) (gcc version 4.1.2 20070115 (prerelease) (SUSE Linux)) #14 SMP Tue Jun 24 07:27:34 CDT 2008 EFI v1.10 by INTEL: SALsystab=0x6002c25f10 ACPI 2.0=0x6002c26000
console [sn_sal0] enabled
ACPI: RSDP 6002C26000, 0024 (r2 SGI)
ACPI: XSDT 6002C297F0, 0044 (r1 SGI XSDTSN2 10001 7C)
ACPI: APIC 6002C26870, 032C (r1 SGI APICSN2 10001 1)
ACPI: SRAT 6002C26BB0, 06B0 (r1 SGI SRATSN2 10001 1)
ACPI: SLIT 6002C27270, 012C (r1 SGI SLITSN2 10001 1)
ACPI: FACP 6002C27400, 00F4 (r3 SGI FACPSN2 30001 1)
ACPI: DSDT 6002C2AAF0, 0024 (r2 SGI DSDTSN2 20001 AAC)
ACPI: FACS 6002C273B0, 0040
Number of logical nodes in system = 16
Number of memory chunks in system = 16
SAL 3.2: SGI SN2 version 1.50
SAL Platform features: ITC_Drift
SAL: AP wakeup using external interrupt vector 0x12
Unable to handle kernel NULL pointer dereference (address 00000000000044b8) swapper[0]: Oops 8813272891392 [1] Modules linked in:
Pid: 0, CPU 0, comm: swapper
psr : 00001010084a2010 ifs : 8000000000000491 ip : [<a000000100087020>] Not tainted (2.6.26-rc5-00223-g3463a93)
ip is at sn2_send_IPI+0x80/0x240
unat: 0000000000000000 pfs : 0000000000000491 rsc : 0000000000000003
rnat: 000000000000afc8 bsps: 000000000001003e pr : 65691ba55aa68599
ldrs: 0000000000000000 ccv : 0000000000ff03ff fpsr: 0009804c8a70433f csd : 0000000000000000 ssd : 0000000000000000
b0 : a000000100942870 b6 : 00000000ff5423b0 b7 : e000000001fffc00 f6 : 1003e0000000000000000 f7 : 1003e0000000000000001 f8 : 1003e0000000000000000 f9 : 1003e0000000000000000 f10 : 100068fffffffff700000 f11 : 1003e0000000000000090 r1 : a000000100e8bd10 r2 : 00000000000044b8 r3 : 0000000000000000 r8 : 0000000000000000 r9 : 0000000000000000 r10 : ffffffffffff6298 r11 : 0000000000000000 r12 : a000000100adfc30 r13 : a000000100ad0000 r14 : 0000000000000000 r15 : e000006003106298 r16 : e000006003110000 r17 : a000000100d0dce8 r18 : a000000100d0dce8 r19 : a000000100d0dce8 r20 : 0000000000000000 r21 : ffffffffffff0420 r22 : 0000000000000800 r23 : 0000000000000007 r24 : e0000060030b0000 r25 : 000000000004ffff r26 : a00000010097c460 r27 : e0000060030b0010 r28 : e0000060030b0000 r29 : e0000060030b0020 r30 : 0000000000000000 r31 : 00000000000007ff
Unable to handle kernel NULL pointer dereference (address 0000000000000000) swapper[0]: Oops 8813272891392 [2] Modules linked in:
Pid: 0, CPU 0, comm: swapper
psr : 0000101008022018 ifs : 800000000000038c ip : [<a000000100175b30>] Not tainted (2.6.26-rc5-00223-g3463a93)
ip is at kmem_cache_alloc+0x70/0x180
unat: 0000000000000000 pfs : 0000000000000610 rsc : 0000000000000003
rnat: 0000000000000000 bsps: 0000000000000000 pr : 65691ba55aa69aa5
ldrs: 0000000000000000 ccv : 0000000000000000 fpsr: 0009804c8a70033f csd : 0000000000000000 ssd : 0000000000000000
b0 : a000000100040bc0 b6 : a000000100040e00 b7 : a00000010000b730 f6 : 1003e45b3373c16c02344 f7 : 1003e9e3779b97f4a7c16 f8 : 1003e0a00000010001426 f9 : 10006c7fffffffd73ea5c f10 : 100068fffffffff700000 f11 : 1003e0000000000000090 r1 : a000000100e8bd10 r2 : a000000100bae950 r3 : a000000100bac860 r8 : 0000000000000000 r9 : 0000000000000000 r10 : a000000100ad0c54 r11 : 0000000000000000 r12 : a000000100adf100 r13 : a000000100ad0000 r14 : 0000000000000014 r15 : a000000100adf190 r16 : a000000100adf198 r17 : a000000100ca1480 r18 : a000000100adf17c r19 : a000000100adf170 r20 : 0000000000000000 r21 : 0000000000000000 r22 : a000000100adf170 r23 : a000000100adf174 r24 : 000000000000000c r25 : a000000100adf180 r26 : a000000100adf174 r27 : 0000000000000000 r28 : 0000000000000000 r29 : a000000100adf178 r30 : 000000007fffffff r31 : 000000000000000c
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
----- End forwarded message -----
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Robin" = Robin Holt <[email protected]> writes:
Robin> Oops, missed sending this to the ia64 mailing list. Robin
Hi Robin,
Just hit the same problem and did a little digging. It's because platform_send_ipi() ends up doing a cpuid_to_nasid() on sn2, which relies on NUMA information etc. being setup.
In fact, check_sal_cache_flush() is called a fair bit before platform_setup() in arch/ia64/kernel/setup.c, which I would claim is completely broken.
Either check_sal_cache_flush() needs to be moved to after platform_setup() or Alex's patch should be reverted until a better solution is found. I am attaching a patch that does the former, but I don't know if this is safe on HP's systems.
This boots on SN2.
Cheers,
Jes
Call check_sal_cache_flush() after platform_setup() as check_sal_cache_flush() now relies on being able to call platform vector code.
--- arch/ia64/kernel/setup.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
Index: linux-2.6.git/arch/ia64/kernel/setup.c =================================--- linux-2.6.git.orig/arch/ia64/kernel/setup.c +++ linux-2.6.git/arch/ia64/kernel/setup.c
@@ -578,8 +578,6 @@ setup_arch (char **cmdline_p)
cpu_init(); /* initialize the bootstrap CPU */
mmu_context_init(); /* initialize context_id bitmap */
- check_sal_cache_flush();
-
#ifdef CONFIG_ACPI
acpi_boot_init();
#endif
@@ -607,6 +605,7 @@ setup_arch (char **cmdline_p)
ia64_mca_init();
platform_setup(cmdline_p);
+ check_sal_cache_flush();
paging_init();
}
Alex Chiang
Tested-by* Jes Sorensen <[email protected]>:
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Robin" = Robin Holt <[email protected]> writes:
Robin> Oops, missed sending this to the ia64 mailing list. Robin
Hi Robin,
Quoted 7 lines
Just hit the same problem and did a little digging. It's because
platform_send_ipi()ends up doing acpuid_to_nasid()on sn2, which relies on NUMA information etc. being setup.In fact,
check_sal_cache_flush()is called a fair bit beforeplatform_setup()inarch/ia64/kernel/setup.c, which I would claim is completely broken.
Yeah, using platform_* before platform_setup() completes is probably a bad idea. :-/
Quoted 6 lines
Either
check_sal_cache_flush()needs to be moved to afterplatform_setup()or Alex's patch should be reverted until a better solution is found. I am attaching a patch that does the former, but I don't know if this is safe on HP's systems.This boots on SN2.
This patch works on
- rx5670 (the original buggy HP platform)
- rx6600 (low-end HP platform)
- rx7620 (mid-range HP platform)
Thanks for fixing this.
/ac
Quoted 38 lines
Cheers,
JesCall
check_sal_cache_flush()afterplatform_setup()ascheck_sal_cache_flush()now relies on being able to call platform vector code.Signed-off-by Jes Sorensen <[email protected]>--- arch/ia64/kernel/setup.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)Index:
linux-2.6.git/arch/ia64/kernel/setup.c=================================> ---linux-2.6.git.orig/arch/ia64/kernel/setup.c+++linux-2.6.git/arch/ia64/kernel/setup.c(quoted diff) +1
-3@@ -578,8 +578,6 @@ setup_arch (char **cmdline_p) cpu_init(); /* initialize the bootstrap CPU */ mmu_context_init(); /* initialize context_id bitmap */ - check_sal_cache_flush(); - #ifdef CONFIG_ACPI acpi_boot_init(); #endif@@ -607,6 +605,7 @@ setup_arch (char **cmdline_p) ia64_mca_init(); platform_setup(cmdline_p); + check_sal_cache_flush(); paging_init(); } --To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Yep, that gets it.
Thanks,
Robin
Quoted 57 lines · Jes Sorensen
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Robin" = Robin Holt <[email protected]> writes:
Robin> Oops, missed sending this to the ia64 mailing list. Robin
Hi Robin,
Just hit the same problem and did a little digging. It's because
platform_send_ipi()ends up doing acpuid_to_nasid()on sn2, which relies on NUMA information etc. being setup.In fact,
check_sal_cache_flush()is called a fair bit beforeplatform_setup()inarch/ia64/kernel/setup.c, which I would claim is completely broken.Either
check_sal_cache_flush()needs to be moved to afterplatform_setup()or Alex's patch should be reverted until a better solution is found. I am attaching a patch that does the former, but I don't know if this is safe on HP's systems.This boots on SN2.
Cheers,
JesCall
check_sal_cache_flush()afterplatform_setup()ascheck_sal_cache_flush()now relies on being able to call platform vector code.Signed-off-by Jes Sorensen <[email protected]>--- arch/ia64/kernel/setup.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)Index:
linux-2.6.git/arch/ia64/kernel/setup.c=================================> ---linux-2.6.git.orig/arch/ia64/kernel/setup.c+++linux-2.6.git/arch/ia64/kernel/setup.c(quoted diff) +1
-3@@ -578,8 +578,6 @@ setup_arch (char **cmdline_p) cpu_init(); /* initialize the bootstrap CPU */ mmu_context_init(); /* initialize context_id bitmap */ - check_sal_cache_flush(); - #ifdef CONFIG_ACPI acpi_boot_init(); #endif@@ -607,6 +605,7 @@ setup_arch (char **cmdline_p) ia64_mca_init(); platform_setup(cmdline_p); + check_sal_cache_flush(); paging_init(); } --To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html