> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/base_dyn_inst.hh, line 229
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5387#file5387line229>
> >
> >     Please change to "Is the physical address valid (translation was 
> > successful)?"

fixed


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/base_dyn_inst.hh, line 232
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5387#file5387line232>
> >
> >     Please change to "Is address translation finished, faulting or not?"

fixed


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/base_dyn_inst.hh, line 973
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5387#file5387line973>
> >
> >     data may not need to be on the next line. It might need to be if the 
> > lines above are juuuust on the very edge.

fixed


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/base_dyn_inst.hh, line 1025
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5387#file5387line1025>
> >
> >     MachineCheckFaults aren't for uncached loads, they're for accesses that 
> > fall off the end of the world and cause a machine check. To the best of my 
> > knowledge these won't be returned immediately since there would be a delay 
> > getting to them through the memory system, but something in the middle 
> > might use the fault to signal a strange error.

It's used to make sure that the instruction doesn't execute early, so yes it is 
in this case. See lsq_unit.hh:561 and blame Kevin if you don't like it.


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/o3/lsq_unit_impl.hh, line 458
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5394#file5394line458>
> >
> >     I don't think this is true. A fault could have happened by this point, 
> > especially when there isn't a table walker. initiateAcc itself could fault 
> > without even performing an access if, for instance, it was a privileged 
> > instruction executed in a non-privileged mode.

True... fixed.


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/o3/lsq_unit_impl.hh, line 476
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5394#file5394line476>
> >
> >     Why is this code being removed?

It's being re-located.... See the memdepordering change. Since the translation 
might not be done yet we can't check for a violation here. 


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/o3/lsq_unit_impl.hh, line 565
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5394#file5394line565>
> >
> >     I don't totally follow what's going on here. Why are you putting these 
> > in a new structure instead of checking them in place? This might be fine, 
> > but I don't understand what's going on so I'm not sure if there's a flaw in 
> > the logic. One thing I'd be especially careful of here are translations 
> > finishing out of order and breaking the ordering checks.

You asked where the dependency checking was going and here it is.


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/o3/lsq_unit_impl.hh, line 582
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5394#file5394line582>
> >
> >     The else and the if shouldn't be on two different lines.

fixed...


> On 2010-11-21 01:19:58, Gabe Black wrote:
> > src/cpu/o3/lsq_unit_impl.hh, line 627
> > <http://reviews.m5sim.org/r/321/diff/1/?file=5394#file5394line627>
> >
> >     Shouldn't be on a new line.

fixed.


- Ali


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://reviews.m5sim.org/r/321/#review500
-----------------------------------------------------------


On 2010-11-19 16:13:39, Ali Saidi wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviews.m5sim.org/r/321/
> -----------------------------------------------------------
> 
> (Updated 2010-11-19 16:13:39)
> 
> 
> Review request for Default.
> 
> 
> Summary
> -------
> 
> O3: Make the DTLB translateTiming() a split transaction.
> ISAs that have hardware page table walkers need this. For those ISAs,
> initiateTranslation can return with NoFault, but with the translation
> unfinished.
> 
> In the BaseDynInst::read() function, actual memory(cache) access calling
> cpu->read() used to occur right after initiateTranslation() function call. 
> This
> access has been moved to the finishTranslation() so that it occurs when the
> translation finishes, either by TLB hit or TLB Miss followed by hardware
> page-walk. Faults occured during translation is detected in 
> finishTranslation()
> and handled during the IEW stage tick()->handleTranslationFault().
> 
> This is a work in progress, but I want to make sure there aren't any issues...
> 
> 
> Diffs
> -----
> 
>   src/arch/arm/tlb.cc 6286bb50127e 
>   src/cpu/base_dyn_inst.hh 6286bb50127e 
>   src/cpu/base_dyn_inst_impl.hh 6286bb50127e 
>   src/cpu/o3/commit_impl.hh 6286bb50127e 
>   src/cpu/o3/iew_impl.hh 6286bb50127e 
>   src/cpu/o3/lsq.hh 6286bb50127e 
>   src/cpu/o3/lsq_impl.hh 6286bb50127e 
>   src/cpu/o3/lsq_unit.hh 6286bb50127e 
>   src/cpu/o3/lsq_unit_impl.hh 6286bb50127e 
> 
> Diff: http://reviews.m5sim.org/r/321/diff
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Ali
> 
>

_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev

Reply via email to