Re: [Qemu-devel] [PATCH 0/6] Multi-level page tables and userland mapping fixes, v3

[Martin Mohring <martin.mohring@opensuse.org>]

I am sorry to bring us all down to earth, we all had wished that this stuff gets finally fixed, but it seems that those patches applied to QEMU master have killed QEMU user mode.

I am referring to the user mode fixes after commit 0aef4261ac0ec9089ade0e3a92f986cb4ba7317e

I had QEMU working on PowerPC and ARM in chroots from the Linux versions:

- Fedora 11 / 12
- Ubuntu 9.04, 9.10, 10.04
- Debian 4, 5 and Sid

for ARM and PowerPC. My host OS is openSUSE 11.2 using a 2.6.31 kernel, or openSUSE 11.1. All machines are 64 Bit machines.

Now I have the situation that all the listed target chroots fail, except:
- Fedora 11 / 12 @ ARM

As you might guess (or even know), I am one of the Open Build Service (http://en.opensuse.org/Build_Service/Testing) developers and can automatedly setup and test such a big number of Targets and automatically run testsuites or compile jobs on them. The last fixes in the QEMU user mode memory allocator before your rewrite came from Jan-Simon and me.

I think its long overdue to fix the user mode stuff, so I am happy that now someone that knows more about QEMU than me cares about. SO my offering is to help to track this down and fix it. But I would be happy if you give me a hint about what your changes are really doing to track down the problem.

Aurelien Jarno wrote:

> On Fri, Mar 12, 2010 at 06:51:41PM +0000, Paul Brook wrote:
>   
>>> Richard Henderson (6):
>>>   Move TARGET_PHYS_ADDR_SPACE_BITS to target-*/cpu.h.
>>>   Use TARGET_VIRT_ADDR_SPACE_BITS in h2g_valid.
>>>   linux-user: Use h2g_valid in qemu_vmalloc.
>>>   linux-user: Fix mmap_find_vma returning invalid addresses.
>>>   Implement multi-level page tables.
>>>   Fix last page errors in page_check_range and page_set_flags.
>>>       
>> Applied, plus a bunch of follow-on patches that isolate the userspace 
>> emulation from any concept of physical address space.
>>
>>     
>
> The patch "Implement multi-level page tables" breaks i386/x86_64 system
> emulation. Using the image on [1], qemu now segfaults during the initrd
> parts. This is not 100% reproducible, a few times it hanged instead of
> crashing.
>
> [1] http://people.debian.org/~aurel32/qemu/i386/debian_lenny_i386_small.qcow2
>
>