Re: [Qemu-devel] qemu softmmu inlined lookup sequence

Re: [Qemu-devel] qemu softmmu inlined lookup sequence

[陳韋任 (Wei-Ren Chen)]

On Sat, Apr 21, 2012 at 05:28:36PM -0400, Xin Tong wrote:
> On Mon, Apr 16, 2012 at 10:51 PM, 陳韋任 <chenwj@iis.sinica.edu.tw> wrote:
> >> what does the inline sequence look like ? what kind of things (other
> >> than refill tlb) performed in callout but not the inlined sequence ?
> >
> >  What do you mean by the inline sequence, the host binary? If so,
> >
> > ---
> >  ---- 0xe86c8
> >  mov_i32 tmp2,edi
> >  qemu_ld8u tmp0,tmp2,$0x0
> >  ext8u_i32 tmp12,tmp0
> >  movi_i32 tmp13,$0xffffff00
> >  and_i32 edx,edx,tmp13
> >  or_i32 edx,edx,tmp12
> >
> > OUT: [size=172]
> > 0x40000ce0:  mov    0x1c(%r14),%ebp
> > 0x40000ce4:  mov    %ebp,%esi               <--- (a)
> > 0x40000ce6:  mov    %ebp,%edi
> > 0x40000ce8:  shr    $0x7,%esi
> > 0x40000ceb:  and    $0xfffff000,%edi
> > 0x40000cf1:  and    $0x1fe0,%esi
> > 0x40000cf7:  lea    0x348(%r14,%rsi,1),%rsi
> > 0x40000cff:  cmp    (%rsi),%edi
> > 0x40000d01:  mov    %ebp,%edi
> > 0x40000d03:  jne    0x40000d0e
> > 0x40000d05:  add    0x10(%rsi),%rdi
> > 0x40000d09:  movzbl (%rdi),%ebp
> > 0x40000d0c:  jmp    0x40000d18
> > 0x40000d0e:  xor    %esi,%esi
> > 0x40000d10:  callq  0x54cf8e                <--- (b)
> > 0x40000d15:  movzbl %al,%ebp
> > 0x40000d18:  movzbl %bpl,%ebp
> > 0x40000d1c:  mov    0x8(%r14),%ebx
> > ---
> >
> 
> 
> I had anotated the inlined TLB lookup sequence. can you please help me
> verify whether it is correct ? I am putting this in a paper... must
> make sure it is correct ...
> 
> 
> # %ebp is the guest virtual address.
> 0x40000ce4:  mov    %ebp,%esi
> # %ebp is the guest virtual address.
> 0x40000ce6:  mov    %ebp,%edi
> 
> # take the page address of the guest virtual address.
> # take the bit 12 - 19 of the guest virtual address.
> 0x40000ce8:  shr    $0x7,%esi
> 0x40000ceb:  and    $0xfffff000,%edi
> 0x40000cf1:  and    $0x1fe0,%esi
> 
> # index into the software tlb using the bit 12 - 19.
> 0x40000cf7:  lea    0x348(%r14,%rsi,1),%rsi
> 
> # check for a possible match
> 0x40000cff:  cmp    (%rsi),%edi
> 
> 0x40000d01:  mov    %ebp,%edi
> 0x40000d03:  jne    0x40000d0e
> 
> # there is a match, calculate the translated address.
> 0x40000d05:  add    0x10(%rsi),%rdi
> 0x40000d09:  movzbl (%rdi),%ebp
> 0x40000d0c:  jmp    0x40000d18
> 
> # can not find a match, perform the callout.
> 0x40000d0e:  xor    %esi,%esi
> 0x40000d10:  callq  callout_TLBLookup
> 
> # other emulation code
> 0x40000d18:  movzbl %bpl,%ebp
> 0x40000d1c:  mov    0x8(%r14),%ebx

  I believe this reply is too late, but it's LGTM. :) 
> 
> >  The code between (a) and (b) is (almost) what TLB lookup hit/miss looks like.
> > If TLB miss, it will call something like __ldb_mmu (b). __ldb_mmu will try to
> > walk guest page table, then fill TLB entry if page table hit, or raise a guest
> > page fault exception if page table miss.

Regards,
chenwj

-- 
Wei-Ren Chen (陳韋任)
Computer Systems Lab, Institute of Information Science,
Academia Sinica, Taiwan (R.O.C.)
Tel:886-2-2788-3799 #1667
Homepage: http://people.cs.nctu.edu.tw/~chenwj