On 19/08/2011 09:27, "Lin-Bao Zhang" <2004.zhang@xxxxxxxxx> wrote:
> 2011/8/18 Lin-Bao Zhang <2004.zhang@xxxxxxxxx>
>>
>> 2011/8/18 Keir Fraser <keir@xxxxxxx>:
>>> It's almost certainly this. In latest Xen we run the trampoline code at
>>> 0x7c000. So you can try rebuilding your hypervisor with the following
>>> changes:
>>> 1. Change BOOT_TRAMPOLINE to 0x7c000
>>> 2. In arch/x86/boot/head.S, change the line 'mov $0x98000,%esp' to 'mov
>>> $0x80000,%esp'.
>>>
>>> That will probably fix your problem.
>>>
>>> -- Keir
>
> Hi Keir ,
> maybe 0x80000 is not very good, I check xen-4.1.0 code:
> /* Copy bootstrap trampoline to low memory, below 1MB. */
> mov $sym_phys(trampoline_start),%esi
> mov $bootsym_phys(trampoline_start),%edi
> mov $trampoline_end - trampoline_start,%ecx
> rep movsb
>
> mov $bootsym_phys(early_stack),%esp //eary_stack.
> call cmdline_parse_early
>
> /* Jump into the relocated trampoline. */
> jmp $BOOT_CS32,$bootsym_phys(trampoline_boot_cpu_entry)
>
> it seems that direct number is not very compatible .
I was just suggesting a minimal patch for you to try.
> 1, another point : it seems that this code should copy code (from
> trampoline_start to trampoline_end) to BOOT_TRAMPOLINE(0x90000)) , I
> will consider if trampoline code size bigger than
> 0x95400-0x90000=21KBytes .
> but my binary xen.gz doesn't include symbols , so I can't see address
> of trampoline_start and trampoline_end. how to get them ? must
> recompile it ?I think maybe redhat has deleted symbols . yet I don't
> trampoline will be over 21KBytes,because it seems small code.
You could not move the trampoline code, and just move the stack, which is th
thing that is actually overlapping your EBDA currently. I.e., just change
the mov $0x98000,%esp instruction.
>
> 2, about definition of trampoline_start and trampoline_end ?
> I just can see:
> #define BOOT_TRAMPOLINE 0x90000
> #define bootsym_phys(sym) \
> (((unsigned long)&(sym)-(unsigned long)&trampoline_start)+BOOT_TRAMPOLINE)
> #define bootsym(sym) \
> (*RELOC_HIDE((typeof(&(sym)))__va(__pa(&(sym))), \
> BOOT_TRAMPOLINE-__pa(trampoline_start)))
> #ifndef __ASSEMBLY__
> extern char trampoline_start[], trampoline_end[];
> extern char trampoline_realmode_entry[];
> extern unsigned int trampoline_xen_phys_start;
> extern unsigned char trampoline_cpu_started;
> #endif
>
> 。。。。。。。。。。
> head.S :
> .globl trampoline_start, trampoline_end
> trampoline_start: // it seems that I must find this symbol to
> confirm its value , nm -a this file , displaying no symbols.
> #include "trampoline.S"
> trampoline_end:
>
>
> 3 , in sles11:
>
> I find sles11 again , I did the same steps .
> in sles11 x86_64 version ,there are also :
> xen-syms-3.3.1_18546_12-3.1 and xen.gz :
> nm -a xen(unzip xen.gz) ,can't trampoline_start/end
> while nm -a xen-syms-3.3.1_18546_12-3.1 , I can finally find
> "trampoline_start" and "trampoline_end" :
> --bob comment: I begin another email subject "about __XEN_VIRT_START
> and 0x100000" to ask the difference between these 2 files.
>
> from this output ,useful information:
> ffff828c80100000 A _start
> ...
> ffff828c8020d5b6 t trampoline_boot_cpu_entry
> ffff828c8020d549 T trampoline_cpu_started
> ffff828c8020fe50 T trampoline_end //here
> ffff828c8020d511 t trampoline_gdt
> ffff828c8020d54a t trampoline_protmode_entry
> ffff828c8020d4e0 T trampoline_realmode_entry
> ffff828c8020d4e0 T trampoline_start //here
> ffff828c8020d545 T trampoline_xen_phys_start
> ffff828c80172ab0 t transition_pstate
> ...
> from sles11 , trampoline code size 0xFFFF828C8_020FE50 -
> ffff828c8_020d4e0 = 0x2970 (about 10KBytes)
> if BOOT_TRAMPOLINE == 0x90000 , then trampoline end will be 0x92970
> ,while my EBDA is 0x954000 ~ 0xa0000 . so I think don't overlap.
> certainly , this number is from sles11 xen-kernel ,but I think their
> trampoline.S basily same ,length should be almost same.
Yes, you could probably leave the trampoline code where it is.
-- Keir
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home