[Crash-utility] Re: Cannot open dump when using 2G/2G memory split with 2.6.27.21 SMP x86 kernel
----- "Tero Pirkkanen" <tp(a)pirkkanen.net> wrote:
Hello
I've been trying to open crashdump produced by 2.6.27.21 SMP kernel
with 2G memory split without luck.
I tried to generate the dump by copying from /proc/vmcore and with
makedumpfile 1.3.3. Both without success.
The crash works fine with uniprocessor kernel as well as SMP kernel
using standard memory split. Architecture is 32bit x86 and the crash
was generated inside vmware server 2.0.
Is this supported configuration? Have anybody tried crash dump with 2G
split in recent 2.6.x kernels?
Apparently not -- at least I've never tried it...
But looking at the output, it's failing to read simple unity-mapped
memory, which I presume to be due to a failure to calculate the
correct machdep->kvbase in x86.c:x86_init() here at line 1740:
machdep->kvbase = symbol_value("_stext") & ~KVBASE_MASK;
And based upon that value, whenever an x86 unity-mapped kernel
virtual address is requested, it simply does this to find the physical
address:
#define VTOP(X) ((unsigned long)(X)-(machdep->kvbase))
And given that the initial set of data accesses are all simply
unity-mapped addresses and the data that is being read is obviously
junk (presuming that you don't have 32 online cpus):
.... skip to end
<readmem: 8053f700, KVADDR, "kernel_config_data", 32768, (ROE), a5f17c0>
WARNING: could not find MAGIC_START!
<readmem: 807348c0, KVADDR, "cpu_possible_map", 4, (ROE), 8430680>
cpu_possible_map: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
21 22 23 24 25 26 27 28 29 30 31
<readmem: 806c478c, KVADDR, "cpu_present_map", 4, (ROE), 8430680>
cpu_present_map: 3 7 8 9 10 14 18 19 21 22 31
<readmem: 806c42a0, KVADDR, "cpu_online_map", 4, (ROE), 8430680>
cpu_online_map: 8 9 14 18 19 21 22 31
<readmem: 807506e0, KVADDR, "xtime", 8, (FOE), 8408d38>
<readmem: 80679e84, KVADDR, "init_uts_ns", 390, (ROE), 840931c>
crash: cannot determine base kernel version
<readmem: 8053a000, KVADDR, "accessible check", 4, (ROE|Q), ff96df04>
<readmem: 8053a000, KVADDR, "readstring characters", 1499, (ROE|Q),
ff96cf00>
linux_banner:
&y �&y �&y �(z �]z �jz ��z �a� �f� ��� ��
��� �9� �� �0� �6� �<� �Њ
�B� �H� �N� �
crash: s3/boot/vmlinux-2.6.27.21+smp-syms and
kerneldump-20090526-140502 do not match
So it looks to be a miscalculation of machdep->kvbase.
If you take the same dumpfile above, but use -d8 instead of -d7, you'll
also see the translated physical address after each "readmem" line
above.
In any case, first determine what's machdep->kvbase is being set
to -- I'm presuming that it should be 0x8000000, but maybe not?
Dave