[Qemu-devel] Question about softmmu

[Qemu-devel] Question about softmmu

[Ye Wen]

Hi, Guys:

I'm reading the QEMU code. I have a question haunting my mind for some time
about tb flushing due to MMU change. Generally, whenever the page table has
any change, we should flush the tb hash table. But I only see in the code that
when CR3 is changed, the tb_flush is called. What if the CR3 does not change,
but some level 2 page table in the memory change, should the tb also be flushed?

Thanks,

--
Ye Wen
wen@umail.ucsb.edu

Re: [Qemu-devel] Question about softmmu

[olivier cozette]

Hello,

>I'm reading the QEMU code. I have a question haunting my mind for some time
>about tb flushing due to MMU change. Generally, whenever the page table has
>any change, we should flush the tb hash table. But I only see in the code that
>when CR3 is changed, the tb_flush is called. What if the CR3 does not change,
>but some level 2 page table in the memory change, should the tb also be
>flushed?


In this case, in the real processor (see Intel Manual), the processor TLBs (it's
like the Qemu tb) are not flushed, so Qemu act as a real processor.

Olivier

Re: [Qemu-devel] Question about softmmu

[Blue Swirl]

Hi,

I'm also suspecting the flushes cause problems in the Sparc system emulator. 
I don't fully understand the difference between various flushes available 
and what is being issued. For example:

1. CPU issues flush instructions to synch I/D caches: NOP?
2. MMU table changes: flush what, TLBs?
3. MMU translation cache flushes: NOP?
4. HW-issued memory access changes memory contents including code, flush 
TBs?

There is also a problem with GDB breakpoints, the debug instructions added 
for the breakpoint aren't removed when deleting the breakpoint. TB 
invalidate is called, but that doesn't work?

About your question, perhaps Qemu should track MMU tables like it does for 
self-modifying code.

_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today it's FREE! 
http://messenger.msn.com/