[Qemu-devel] [4454] i386 specific TODO

[Qemu-devel] [4454] i386 specific TODO

[Fabrice Bellard]

Revision: 4454
          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=4454
Author:   bellard
Date:     2008-05-13 18:27:16 +0000 (Tue, 13 May 2008)

Log Message:
-----------
i386 specific TODO

Added Paths:
-----------
    trunk/target-i386/TODO

Added: trunk/target-i386/TODO
===================================================================
--- trunk/target-i386/TODO	                        (rev 0)
+++ trunk/target-i386/TODO	2008-05-13 18:27:16 UTC (rev 4454)
@@ -0,0 +1,29 @@
+Correctness issues:
+
+- rework eflags optimization (will be a consequence of TCG port)
+- SVM: rework the implementation: simplify code, move most intercept
+  tests as dynamic, correct segment access, verify exception safety,
+  remove most of the added CPU state.
+- arpl eflags computation is invalid
+- x86_64: fxsave/fxrestore intel/amd differences
+- x86_64: lcall/ljmp intel/amd differences ?
+- x86_64: cmpxchgl intel/amd differences ?
+- x86_64: cmovl bug intel/amd differences ?
+- x86: monitor invalid 
+- better code fetch (different exception handling + CS.limit support)
+- user/kernel PUSHL/POPL in helper.c
+- add missing cpuid tests
+- return UD exception if LOCK prefix incorrectly used
+- test ldt limit < 7 ?
+- fix some 16 bit sp push/pop overflow (pusha/popa, lcall lret)
+- full support of segment limit/rights 
+- full x87 exception support
+- improve x87 bit exactness (use bochs code ?)
+
+Optimizations/Features:
+
+- finish TCG port
+- evaluate x87 stack pointer statically
+- find a way to avoid translating several time the same TB if CR0.TS
+  is set or not.
+- move kqemu support outside target-i386.

[Qemu-devel] Re: [4454] i386 specific TODO

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1967 bytes --]

Fabrice Bellard wrote:
> Revision: 4454
>           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=4454
> Author:   bellard
> Date:     2008-05-13 18:27:16 +0000 (Tue, 13 May 2008)
> 
> Log Message:
> -----------
> i386 specific TODO
> 
> Added Paths:
> -----------
>     trunk/target-i386/TODO
> 
> Added: trunk/target-i386/TODO
> ===================================================================
> --- trunk/target-i386/TODO	                        (rev 0)
> +++ trunk/target-i386/TODO	2008-05-13 18:27:16 UTC (rev 4454)
> @@ -0,0 +1,29 @@
> +Correctness issues:
> +
> +- rework eflags optimization (will be a consequence of TCG port)
> +- SVM: rework the implementation: simplify code, move most intercept
> +  tests as dynamic, correct segment access, verify exception safety,
> +  remove most of the added CPU state.
> +- arpl eflags computation is invalid
> +- x86_64: fxsave/fxrestore intel/amd differences
> +- x86_64: lcall/ljmp intel/amd differences ?
> +- x86_64: cmpxchgl intel/amd differences ?
> +- x86_64: cmovl bug intel/amd differences ?
> +- x86: monitor invalid 
> +- better code fetch (different exception handling + CS.limit support)
> +- user/kernel PUSHL/POPL in helper.c
> +- add missing cpuid tests
> +- return UD exception if LOCK prefix incorrectly used
> +- test ldt limit < 7 ?
> +- fix some 16 bit sp push/pop overflow (pusha/popa, lcall lret)
> +- full support of segment limit/rights 

This topic is widely completed here (well... @work), just MMX/SSE
operations are missing (the customer doesn't use it) - and, of course,
all those corner cases I surely oversaw. As those checks do not make
things immediately faster, I added a switch so that you can turn them on
when starting qemu.

Haven't tried to reapply the patch, but I'm afraid some rebase work is
now required. Would it make sense to start this soon, or do you plan
further TCG rework in the near future?

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 254 bytes --]

Re: [Qemu-devel] [4454] i386 specific TODO

[Alexander Graf]

On May 13, 2008, at 8:27 PM, Fabrice Bellard wrote:

> Revision: 4454
>          http://svn.sv.gnu.org/viewvc/? 
> view=rev&root=qemu&revision=4454
> Author:   bellard
> Date:     2008-05-13 18:27:16 +0000 (Tue, 13 May 2008)
>
> Log Message:
> -----------
> i386 specific TODO
>
> Added Paths:
> -----------
>    trunk/target-i386/TODO
>
> Added: trunk/target-i386/TODO
> ===================================================================
> --- trunk/target-i386/TODO	                        (rev 0)
> +++ trunk/target-i386/TODO	2008-05-13 18:27:16 UTC (rev 4454)
> @@ -0,0 +1,29 @@
> +Correctness issues:
> +
> +- rework eflags optimization (will be a consequence of TCG port)
> +- SVM: rework the implementation: simplify code, move most intercept

What exactly do you mean here? Simplify it in what respect? Move them  
where?

Additional todos:

SVM: add NPT support
SVM: verify 32-bit support

>
> +  tests as dynamic, correct segment access, verify exception safety,
> +  remove most of the added CPU state.
> +- arpl eflags computation is invalid
> +- x86_64: fxsave/fxrestore intel/amd differences
> +- x86_64: lcall/ljmp intel/amd differences ?
> +- x86_64: cmpxchgl intel/amd differences ?
> +- x86_64: cmovl bug intel/amd differences ?
> +- x86: monitor invalid
> +- better code fetch (different exception handling + CS.limit support)
> +- user/kernel PUSHL/POPL in helper.c
> +- add missing cpuid tests
> +- return UD exception if LOCK prefix incorrectly used
> +- test ldt limit < 7 ?
> +- fix some 16 bit sp push/pop overflow (pusha/popa, lcall lret)
> +- full support of segment limit/rights
> +- full x87 exception support
> +- improve x87 bit exactness (use bochs code ?)
> +
> +Optimizations/Features:
> +
> +- finish TCG port
> +- evaluate x87 stack pointer statically
> +- find a way to avoid translating several time the same TB if CR0.TS
> +  is set or not.
> +- move kqemu support outside target-i386.
>
>
>
>