[PATCH] target/arm: Set ISSIs16Bit in make_issinfo

[PATCH] target/arm: Set ISSIs16Bit in make_issinfo

[Richard Henderson]

During the conversion to decodetree, the setting of
ISSIs16Bit got lost.  This causes the guest os to
incorrectly adjust trapping memory operations.

Fixes: 46beb58efbb8a2a32
Cc: qemu-stable@nongnu.org
Reported-by: Jeff Kubascik <jeff.kubascik@dornerworks.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/target/arm/translate.c b/target/arm/translate.c
index 2b6c1f91bf..9f0afbdb75 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -8556,6 +8556,9 @@ static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
     /* ISS not valid if writeback */
     if (p && !w) {
         ret = rd;
+        if (s->base.pc_next - s->pc_curr == 2) {
+            ret |= ISSIs16Bit;
+        }
     } else {
         ret = ISSInvalid;
     }
-- 
2.20.1

Re: [PATCH] target/arm: Set ISSIs16Bit in make_issinfo

[Alex Bennée]

Richard Henderson <richard.henderson@linaro.org> writes:

> During the conversion to decodetree, the setting of
> ISSIs16Bit got lost.  This causes the guest os to
> incorrectly adjust trapping memory operations.
>
> Fixes: 46beb58efbb8a2a32

It's not really obvious from this commit where we end up now calling...


> Cc: qemu-stable@nongnu.org
> Reported-by: Jeff Kubascik <jeff.kubascik@dornerworks.com>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/arm/translate.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/target/arm/translate.c b/target/arm/translate.c
> index 2b6c1f91bf..9f0afbdb75 100644
> --- a/target/arm/translate.c
> +++ b/target/arm/translate.c
> @@ -8556,6 +8556,9 @@ static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
>      /* ISS not valid if writeback */
>      if (p && !w) {
>          ret = rd;
> +        if (s->base.pc_next - s->pc_curr == 2) {
> +            ret |= ISSIs16Bit;
> +        }

this function.

Should I be seeing op_load_rr/ri in the included generated functions?

>      } else {
>          ret = ISSInvalid;
>      }


-- 
Alex Bennée

Re: [PATCH] target/arm: Set ISSIs16Bit in make_issinfo

[Peter Maydell]

On Fri, 20 Dec 2019 at 13:35, Alex Bennée <alex.bennee@linaro.org> wrote:
>
>
> Richard Henderson <richard.henderson@linaro.org> writes:
>
> > During the conversion to decodetree, the setting of
> > ISSIs16Bit got lost.  This causes the guest os to
> > incorrectly adjust trapping memory operations.
> >
> > Fixes: 46beb58efbb8a2a32
>
> It's not really obvious from this commit where we end up now calling...
>
>
> > Cc: qemu-stable@nongnu.org
> > Reported-by: Jeff Kubascik <jeff.kubascik@dornerworks.com>
> > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > ---
> >  target/arm/translate.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/target/arm/translate.c b/target/arm/translate.c
> > index 2b6c1f91bf..9f0afbdb75 100644
> > --- a/target/arm/translate.c
> > +++ b/target/arm/translate.c
> > @@ -8556,6 +8556,9 @@ static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
> >      /* ISS not valid if writeback */
> >      if (p && !w) {
> >          ret = rd;
> > +        if (s->base.pc_next - s->pc_curr == 2) {
> > +            ret |= ISSIs16Bit;
> > +        }
>
> this function.

Yeah, the combination of decodetree and the C preprocessor
is pretty confusing. There's a macro DO_LDST in translate.c
which creates trans_LDR_rr() functions which call op_load_rr()
which then calls make_issinfo(); and the decodetree generator
is what generates the code that calls trans_LDR_rr() and friends.

thanks
-- PMM

Re: [PATCH] target/arm: Set ISSIs16Bit in make_issinfo

[Peter Maydell]

On Thu, 19 Dec 2019 at 18:56, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> During the conversion to decodetree, the setting of
> ISSIs16Bit got lost.  This causes the guest os to
> incorrectly adjust trapping memory operations.
>
> Fixes: 46beb58efbb8a2a32
> Cc: qemu-stable@nongnu.org
> Reported-by: Jeff Kubascik <jeff.kubascik@dornerworks.com>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/arm/translate.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/target/arm/translate.c b/target/arm/translate.c
> index 2b6c1f91bf..9f0afbdb75 100644
> --- a/target/arm/translate.c
> +++ b/target/arm/translate.c
> @@ -8556,6 +8556,9 @@ static ISSInfo make_issinfo(DisasContext *s, int rd, bool p, bool w)
>      /* ISS not valid if writeback */
>      if (p && !w) {
>          ret = rd;
> +        if (s->base.pc_next - s->pc_curr == 2) {
> +            ret |= ISSIs16Bit;
> +        }
>      } else {
>          ret = ISSInvalid;
>      }
> --
> 2.20.1

This is correct, and fixes the regression vs 46beb58efbb8a,
but I agree with Jeff that it's not sufficient (and in fact
we've always been reporting wrong ISS info as a result)
because of the wrong sense of the bool argument to
syn_data_abort_with_iss() in merge_syn_data_abort().

thanks
-- PMM