Re: [Qemu-devel] [PATCH 0/2] make guest-base support mandatory for TCG backends

From: Paolo Bonzini <pbonzini@redhat.com>
To: Laurent Desnogues <laurent.desnogues@gmail.com>
Cc: Peter Maydell <peter.maydell@linaro.org>,
	patches@linaro.org, Riku Voipio <riku.voipio@iki.fi>,
	qemu-devel@nongnu.org, Aurelien Jarno <aurelien@aurel32.net>,
	Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH 0/2] make guest-base support mandatory for TCG backends
Date: Tue, 09 Oct 2012 16:58:50 +0200	[thread overview]
Message-ID: <50743BAA.8050306@redhat.com> (raw)
In-Reply-To: <CABoDooPOSX2N=Ya_D9TYRBZAFghFkKPa7f6RMV7DsN9DQpqz3g@mail.gmail.com>

Il 09/10/2012 16:50, Laurent Desnogues ha scritto:
> On Tue, Oct 9, 2012 at 3:28 PM, Richard Henderson <rth@twiddle.net> wrote:
>> On 10/09/2012 06:16 AM, Peter Maydell wrote:
>>> That is, we could drop CONFIG_USER_GUEST_BASE. Does anybody have
>>> a practical use case for the --disable-guest-base configuration?
>>
>> Nope, because the backends are good at eliminating the overhead
>> during tcg_target_qemu_prologue if guest_base turns out to be zero.
> 
> I had some difficulties getting the same code generated with
> --enable-guest-base as the one generated with --disable-guest-base,
> for ARM on x86_64.  One has to play with -B 0 and -R, and that was
> not obvious, while --disable-guest-base was obvious for me :-).
> 
> Guest ARM, host x86_64.
> 
> $ ./arm-linux-user/qemu-arm -singlestep -d in_asm,out_asm
> /work/qemu/Tests/arm/gcc-a9-O3 x
> 
> IN:
> 0x00008138:  e59fc010      ldr	ip, [pc, #16]	; 0x8150
> 
> OUT: [size=53]
> 0x602296c0:  mov    $0x8150,%ebp
> 0x602296c5:  mov    $0x2ac1f9b61000,%rdi
> 0x602296cf:  add    %rbp,%rdi
> 0x602296d2:  mov    (%rdi),%ebp
> 0x602296d4:  mov    %ebp,0x30(%r14)
> 0x602296d8:  jmpq   0x602296dd
> 0x602296dd:  mov    $0x813c,%ebp
> 0x602296e2:  mov    %ebp,0x3c(%r14)
> 0x602296e6:  mov    $0x2ac1f6760281,%rax
> 0x602296f0:  jmpq   0x6226d6b6
> 
> Running with -B 0 and playing with -R values until the program
> succeeds in loading finally produces the same code as when
> compiling QEMU with --disable-guest-base
> 
> OUT: [size=41]
> 0x602281a0:  mov    $0x8150,%ebp
> 0x602281a5:  mov    0x0(%rbp),%ebp
> 0x602281a8:  mov    %ebp,0x30(%r14)
> 0x602281ac:  jmpq   0x602281b1
> 0x602281b1:  mov    $0x813c,%ebp
> 0x602281b6:  mov    %ebp,0x3c(%r14)
> 0x602281ba:  mov    $0x2b9b26096281,%rax
> 0x602281c4:  jmpq   0x6226c1b6
> 
> I know the reason why reserved_va was set to the current value
> (cf 288e65b9eea0c9b3cbe21be46f3e24e4e8b2a090), but I
> thought it was interesting to point out the above.

Perhaps you want to revive this patch:

http://lists.gnu.org/archive/html/qemu-devel/2010-06/msg00509.html

Paolo