For the Arm target, we have a 3-operand tcg_insn_start,
where the 3 arguments are the PC, condexec bits, and
a syndrome value. We set it up like this:

    tcg_gen_insn_start(dc->pc,
                       (dc->condexec_cond << 4) | (dc->condexec_mask >> 1),
                       0);
    dc->insn_start = tcg_last_op();

and then we patch in the 3rd operand later in disas_set_insn_syndrome():
    tcg_set_insn_param(s->insn_start, 2, syn);

Unfortunately, if we're running on a setup where
TARGET_LONG_BITS > TCG_TARGET_REG_BITS (ie 64 bit guest
on 32 bit host), tcg_gen_insn_start() has under the hood
split the 3 operands we gave it into 6, and so we end
up patching a syndrome value into the condexec bits.
This means we'll end up with corrupted guest condexec
state if we have to do a cpu_restore_state(), which happens
often when using icount and occasionally for load/store
instructions that fault.

Fix the bug by using the correct operand offset for the
64-on-32 case.

Cc: qemu-stable@nongnu.org
Reported-by: alarson@ddci.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
---
This doesn't apply as-is to the stable branch, but the
difference is minor (insn_start was insn_start_idx, but
the 2 vs 4 for argument 2 is still the same.)
---
 target/arm/translate.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/target/arm/translate.h b/target/arm/translate.h
index c47febf99d..f04ece9cfd 100644
--- a/target/arm/translate.h
+++ b/target/arm/translate.h
@@ -120,7 +120,15 @@ static inline void disas_set_insn_syndrome(DisasContext *s, uint32_t syn)
 
     /* We check and clear insn_start_idx to catch multiple updates.  */
     assert(s->insn_start != NULL);
+#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
     tcg_set_insn_param(s->insn_start, 2, syn);
+#else
+    /* tcg_gen_insn_start has split every target_ulong argument to
+     * op_insn_start into two 32-bit arguments, so we want the low
+     * half of the 3rd argument, which is at index 4.
+     */
+    tcg_set_insn_param(s->insn_start, 4, syn);
+#endif
     s->insn_start = NULL;
 }
 
-- 
2.16.2

On 04/09/2018 08:38 PM, Peter Maydell wrote:
> +#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
>      tcg_set_insn_param(s->insn_start, 2, syn);
> +#else
> +    /* tcg_gen_insn_start has split every target_ulong argument to
> +     * op_insn_start into two 32-bit arguments, so we want the low
> +     * half of the 3rd argument, which is at index 4.
> +     */
> +    tcg_set_insn_param(s->insn_start, 4, syn);
> +#endif
>   

Ouch, good catch.

I think we should fix this in tcg_set_insn_param instead,
as several other targets are also affected by this.


r~

On 9 April 2018 at 23:09, Richard Henderson <rth@twiddle.net> wrote:
> On 04/09/2018 08:38 PM, Peter Maydell wrote:
>> +#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
>>      tcg_set_insn_param(s->insn_start, 2, syn);
>> +#else
>> +    /* tcg_gen_insn_start has split every target_ulong argument to
>> +     * op_insn_start into two 32-bit arguments, so we want the low
>> +     * half of the 3rd argument, which is at index 4.
>> +     */
>> +    tcg_set_insn_param(s->insn_start, 4, syn);
>> +#endif
>>
>
> Ouch, good catch.
>
> I think we should fix this in tcg_set_insn_param instead,
> as several other targets are also affected by this.

Are they? My grep didn't find anybody else using
tcg_set_insn_param() except the gen-icount.h code,
which isn't using target_long arguments.

If we can fix it in the tcg generic code instead that
would be nicer than an ifdef here, though.

thanks
-- PMM

On 04/10/2018 08:13 AM, Peter Maydell wrote:
> On 9 April 2018 at 23:09, Richard Henderson <rth@twiddle.net> wrote:
>> On 04/09/2018 08:38 PM, Peter Maydell wrote:
>>> +#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
>>>      tcg_set_insn_param(s->insn_start, 2, syn);
>>> +#else
>>> +    /* tcg_gen_insn_start has split every target_ulong argument to
>>> +     * op_insn_start into two 32-bit arguments, so we want the low
>>> +     * half of the 3rd argument, which is at index 4.
>>> +     */
>>> +    tcg_set_insn_param(s->insn_start, 4, syn);
>>> +#endif
>>>
>>
>> Ouch, good catch.
>>
>> I think we should fix this in tcg_set_insn_param instead,
>> as several other targets are also affected by this.
> 
> Are they? My grep didn't find anybody else using
> tcg_set_insn_param() except the gen-icount.h code,
> which isn't using target_long arguments.

Ah, the s390 code I thought I remembered is still sitting on a branch.  I shall
have to revive that for the next devel cycle...

> If we can fix it in the tcg generic code instead that
> would be nicer than an ifdef here, though.

I am working on this now.


r~