BUG: Bad page state in process swapper pfn:00000

[Laura Abbott]

Hi,

Thanks for the bisect.

On 6/11/2014 4:40 AM, Geert Uytterhoeven wrote:
> With current mainline, I get an early crash on r8a7791/koelsch:
> 
> BUG: Bad page state in process swapper  pfn:00000
> page:ee20b000 count:0 mapcount:0 mapping:66756200 index:0x65726566
> page flags: 0x74656b63(locked|error|lru|active|owner_priv_1|arch_1|private|writeback|head|swapcache
> |reclaim|mlocked)
> page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
> bad because of flags:
> page flags: 0x212861(locked|lru|active|private|writeback|swapcache|mlocked)
> 
> I bisected it to
> 
> commit 1c2f87c22566cd057bc8cde10c37ae9da1a1bb76
> Author: Laura Abbott <lauraa at codeaurora.org>
> Date:   Sun Apr 13 22:54:58 2014 +0100
> 
>     ARM: 8025/1: Get rid of meminfo
> 
>     memblock is now fully integrated into the kernel and is the prefered
>     method for tracking memory. Rather than reinvent the wheel with
>     meminfo, migrate to using memblock directly instead of meminfo as
>     an intermediate.
> 
>     Acked-by: Jason Cooper <jason at lakedaemon.net>
>     Acked-by: Catalin Marinas <catalin.marinas at arm.com>
>     Acked-by: Santosh Shilimkar <santosh.shilimkar at ti.com>
>     Acked-by: Kukjin Kim <kgene.kim at samsung.com>
>     Tested-by: Marek Szyprowski <m.szyprowski at samsung.com>
>     Tested-by: Leif Lindholm <leif.lindholm at linaro.org>
>     Signed-off-by: Laura Abbott <lauraa at codeaurora.org>
>     Signed-off-by: Russell King <rmk+kernel at arm.linux.org.uk>
> 
> As this is a quite intrusive change, it cannot be reverted on top of mainline.
> 
> The commit before (1c8c3cf0b5239388e712508a85821f4718f4d889)
> does work. Dmesg difference between them:
> 
>  Uncompressing Linux... done, booting the kernel.
>  Booting Linux on physical CPU 0x0
> -Linux version 3.15.0-rc1-koelsch-reference-00027-g1c8c3cf0b523-dirty
> (geert at ramsan) (gcc version 4.6.3 (GCC) ) #174 SMP Wed Jun 11 13:19:00
> CEST 2014
> +Linux version 3.15.0-rc1-koelsch-reference-00028-g1c2f87c22566-dirty
> (geert at ramsan) (gcc version 4.6.3 (GCC) ) #175 SMP Wed Jun 11 13:20:28
> CEST 2014
>  CPU: ARMv7 Processor [413fc0f2] revision 2 (ARMv7), cr=10c5347d
>  CPU: PIPT / VIPT nonaliasing data cache, PIPT instruction cache
> -Ignoring memory at 0x200000000 outside 32-bit physical address space
>  Machine model: Koelsch
>  bootconsole [earlycon0] enabled
>  debug: ignoring loglevel setting.
> -Truncating RAM at 40000000-bfffffff to -6f7fffff (vmalloc region overlap).
> +Truncating RAM at 0x00000000-0xc0000000 to -0x6f800000

I'm guessing this is the issue right there.

        memory at 40000000 {
                device_type = "memory";
                reg = <0 0x40000000 0 0x40000000>;
        };

        memory at 200000000 {
                device_type = "memory";
                reg = <2 0x00000000 0 0x40000000>;
        };

Those are the memory nodes from r8a7791-koelsch.dts. It looks like the memory
outside 32-bit address range is not being dropped. It was suggested to drop
early_init_dt_add_memory_arch which called arm_add_memory and just use the
generic of code directly but the problem is arm_add_memory does additional
bounds checking. It looks like early_init_dt_add_memory_arch in
drivers/of/fdt.c checks for overflow on u64 types but not for overflow
on phys_addr_t (32 bits) which is what memblock_add actually uses. 

For a quick test, can you try bringing back early_init_dt_add_memory_arch
and see if that fixes the problem:

diff --git a/arch/arm/kernel/devtree.c b/arch/arm/kernel/devtree.c
index e94a157..ea9ce92 100644
--- a/arch/arm/kernel/devtree.c
+++ b/arch/arm/kernel/devtree.c
@@ -27,6 +27,10 @@
 #include <asm/mach/arch.h>
 #include <asm/mach-types.h>
 
+void __init early_init_dt_add_memory_arch(u64 base, u64 size)
+{
+       arm_add_memory(base, size);
+}
 
 #ifdef CONFIG_SMP
 extern struct of_cpu_method __cpu_method_of_table[];

Thanks,
Laura


-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation