qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH] Revived GUEST_BASE support for usermode emulati


From: Martin Mohring
Subject: Re: [Qemu-devel] [PATCH] Revived GUEST_BASE support for usermode emulation targets [v4]
Date: Fri, 15 May 2009 17:25:25 +0200
User-agent: Thunderbird 2.0.0.19 (X11/20081227)

Riku Voipio wrote:
> On Fri, May 15, 2009 at 12:12:50PM +0200, Martin Mohring wrote:
>   
>> Riku Voipio wrote:
>>     
>>> On Fri, May 15, 2009 at 03:25:31AM +0100, Paul Brook wrote:
>>>       
>>>> I'm pretty sure it's still broken for large offsets.
>>>>         
>
>   
>>> It is indeed broken. Since tcg_out_addi maps on x86_64 to tgen_arithi64 
>>> which
>>> doesn't accept 64bit values. But as Martin said, the currrent use for this
>>> patch is just to add very minimal offsets.
>>>       
>
>   
>> Since i am using guest base, what does its brokenness mean?
>>     
>
> Well, you are not using guest_base with large offsets, so it's not broken
> for you.
>
>   
>> Riku, will you talk with the author to fix it according to Pauls 
>> suggestions? Or
>> will the patch be accepted under the assumptions/constraints it was
>> written. Or is the request that a generic solution fixes other
>> longstanding issues?
>>     
>
> Well.. the patch fixes a real-life issue: qemu linux-user not working
> at all with recent kernel defaults. With this patch, there is still
> the potential issue - someone can set GUEST_BASE too large by hand.
> That of course didn't work before this patch either.
>
> I think short term it's best to put a size cap for guest_base offset
> and look later into supporting greater offsets. So at least, qemu
> linux-user will be usable out of box for most users.
>
> But, with current "Quod licet Iovi, non licet bovi" philosophy it seems
> it is not enough for bovine to provide patches that make qemu better,
> but perfect patches are required.
>
>
>   
Yes. you are right. The most important fact, that currently for hosts
where /proc/sys/vm/mmap_min_addr is 65536 (which is on all current linux
distros for x86 - Ubuntu 8.10, Debian Lenny, Fedora 9, openSUSE 11.0 ...
or higher - just to name a few) in this sense *QEMU user mode is in fact
broken*.

But even worse is the fact that this issue is on the table since many
months. Todays 5th rebase since our submission reminded me hard toward
this fact. Please Paul release us from this torture.

Martin





reply via email to

[Prev in Thread] Current Thread [Next in Thread]