Re: [Qemu-devel] [PATCH v2 3/3] target-arm: Implement the S2 MMU inputsize > pamax check

["Edgar E. Iglesias" <edgar.iglesias@gmail.com>]

On Fri, Jan 22, 2016 at 11:45:57AM +0000, Alex Bennée wrote:
> 
> Edgar E. Iglesias <edgar.iglesias@gmail.com> writes:
> 
> > On Fri, Jan 22, 2016 at 10:28:43AM +0000, Alex Bennée wrote:
> >>
> >> Edgar E. Iglesias <edgar.iglesias@gmail.com> writes:
> >>
> >> > From: "Edgar E. Iglesias" <edgar.iglesias@xilinx.com>
> >> >
> >> > Implement the inputsize > pamax check for Stage 2 translations.
> >> > We have multiple choices for how to respond to errors and
> >> > choose to fault.
> >> >
> >> > Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
> >> > ---
> >> >  target-arm/helper.c | 16 ++++++++++++----
> >> >  1 file changed, 12 insertions(+), 4 deletions(-)
> >> >
> >> > diff --git a/target-arm/helper.c b/target-arm/helper.c
> >> > index 4abeb4d..9a7ff5e 100644
> >> > --- a/target-arm/helper.c
> >> > +++ b/target-arm/helper.c
> >> > @@ -6808,7 +6808,7 @@ static bool get_phys_addr_lpae(CPUARMState *env, target_ulong address,
> >> >           */
> >> >          int startlevel = extract32(tcr->raw_tcr, 6, 2);
> >> >          unsigned int pamax = arm_pamax(cpu);
> >> > -        bool ok;
> >> > +        bool ok = true;
> >> >
> >> >          if (va_size == 32 || stride == 9) {
> >> >              /* AArch32 or 4KB pages */
> >> > @@ -6818,9 +6818,17 @@ static bool get_phys_addr_lpae(CPUARMState *env, target_ulong address,
> >> >              level = 3 - startlevel;
> >> >          }
> >> >
> >> > -        /* Check that the starting level is valid. */
> >> > -        ok = check_s2_startlevel(cpu, va_size == 64, level,
> >> > -                                 inputsize, stride, pamax);
> >> > +        if (va_size == 64 &&
> >> > +            inputsize > pamax &&
> >> > +            (arm_el_is_aa64(env, 1) || inputsize > 40)) {
> >>
> >> If va_size == 64 doesn't that imply arm_el_is_aa64(env, 1)? Looking
> >> further up the function it seems that is what sets va_size in the first
> >> place. I think that makes the inputsize > 40 check redundant.
> >
> > va_size == 64 is true if the EL corresponding to the translation _regime_
> > is in 64 bit mode (in this case EL2).
> >
> > EL1 may very well be in 32bit mode.
> 
> Ahh yes, I missed that on the first reading. I think it might be clearer
> when reading the code to have the:
> 
>     bool is_aarch64_regime = (va_size == 64);
> 
> And use that to make it clear. And then comment on later check that it's
> incompatible with EL1 being aarch32.
> 
> >
> >>
> >> > +            /* We have multiple choices but choose to fault.  */
> >> > +            ok = false;
> >> > +        }
> >> > +        if (ok) {
> >> > +            /* Check that the starting level is valid. */
> >> > +            ok = check_s2_startlevel(cpu, va_size == 64, level,
> >> > +                                     inputsize, stride, pamax);
> >> > +        }
> >> >          if (!ok) {
> >> >              /* AArch64 reports these as level 0 faults.
> >> >               * AArch32 reports these as level 1 faults.
> >>
> >> I'm not a fan of the ok = true / ok = false / ok =
> >> check_s2_start_level() / if (!ok) ping-pong here as it is hard to
> >> follow. I'm not sure how you could make it cleaner to follow though.
> >> Maybe something like:
> >>
> >>         /* For stage 2 translations the starting level is specified by the
> >>          * VTCR_EL2.SL0 field (whose interpretation depends on the page size)
> >>          */
> >>         int startlevel = extract32(tcr->raw_tcr, 6, 2);
> >>         unsigned int pamax = arm_pamax(cpu);
> >>         bool is_aarch64_regime = (va_size == 64);
> >>         bool ok;
> >>
> >>         if (va_size == 32 || stride == 9) {
> >>             /* AArch32 or 4KB pages */
> >>             level = 2 - startlevel;
> >>         } else {
> >>             /* 16KB or 64KB pages */
> >>             level = 3 - startlevel;
> >>         }
> >>
> >>         if (is_aarch64_regime &&
> >>             inputsize > pamax) {
> >>             /* We have multiple choices but choose to fault.  */
> >>             ok = false;
> >>         } else {
> >>             /* Check that the starting level is valid. */
> >>             ok = check_s2_startlevel(cpu, is_aarch64_regime, level,
> >>                                      inputsize, stride, pamax);
> >>         }
> >>         if (!ok) {
> >>             /* AArch64 reports these as level 0 faults.
> >>              * AArch32 reports these as level 1 faults.
> >>              */
> >>             level = is_aarch64_regime ? 0 : 1;
> >>             fault_type = translation_fault;
> >>             goto do_fault;
> >>         }
> >>
> >> But I'm wondering if it just makes more sense to push the:
> >>
> >>     is_aarch64_regime && inputsize > pamax
> >>
> >> Check into check_s2_startlevel? Then you could just have a simple call
> >> which succeeds or falls through to a fault?
> >
> > Yeah, I guess we could rename check_s2_startlevel to something more generic
> > and move all the checks there. I don't feel very strongly about either way...
> 
> I think it would be cleaner to follow. get_phys_addr_lpae is already a
> bit of a monster so the less conditions to keep track of while reading
> it the better IMHO.


OK, I'll have a look at that for v4.

Thanks!
Edgar


> 
> > Thanks,
> > Edgar
> >
> >
> >
> >>
> >>         /* Check that the starting level is valid. */
> >>         if (!check_s2_startlevel(cpu, is_aarch64_regime, level,
> >>                                      inputsize, stride, pamax) ){
> >>             /* AArch64 reports these as level 0 faults.
> >>              * AArch32 reports these as level 1 faults.
> >>              */
> >>             level = is_aarch64_regime ? 0 : 1;
> >>             fault_type = translation_fault;
> >>             goto do_fault;
> >>         }
> >>
> >> --
> >> Alex Bennée
> 
> 
> --
> Alex Bennée